High-precision eye-tracking system using the center of eye rotation

Abstract

This system automatically identifies error factors and provides an eye-tracking system that enables highly accurate and consistent gaze point determination even in dynamic usage environments and long-term operation of wearable devices. [Solution] The system comprises an eye-tracking sensor that acquires the user's eyeball information, means for forming an external reference intersection, and information analysis means for geometrically calculating the center of eye rotation (ERC) from the intersection and eyeball information. The information analysis means verifies whether the ERC calculated by the physiological validity determination unit satisfies anatomical constraints and eliminates falsification and miscalculation. It also evaluates the error between the assumed gaze point derived from the ERC and the external reference intersection (measured value), and dynamically updates or determines the ERC to reduce this error. Furthermore, it selectively compensates for corrections to the ERC based on inertial measurement unit (IMU) information mounted on the wearable device and error factor identification (mechanical deviation or physiological drift) by comparing actual virtual intersection coordinates.

Description

Technical Field

[0001] The present invention relates to a gaze tracking system that tracks the direction of a user's gaze, and in particular, to a technique for dynamically calculating the center of eyeball rotation to correct the gaze direction with high accuracy, and a technique for expressing gaze tracking results in an absolute coordinate system. The present invention is applicable to a wide range of fields including head-mounted displays for VR (virtual reality), AR (augmented reality), MR (mixed reality), automobile driving assistance systems, industrial robot control, gaze input interfaces, neuro-ophthalmological research, medical diagnosis assistance, drone control, and the like.

Background Art

[0002] Gaze tracking technology is a technology that measures the direction of a human's gaze in real time, and has recently gained increasing importance in fields such as human-machine interfaces, driving assistance, and medical diagnosis. Conventional gaze tracking technologies are broadly classified into two categories: appearance-based estimation and geometric methods.

[0003] Appearance-based estimation is a method that directly estimates the gaze direction from an eyeball image using a machine learning model. This method has the advantage of relatively simple calibration, but its gaze tracking accuracy is as low as about 1-5 degrees to 5 degrees, making it unsuitable for applications requiring high precision. Additionally, it is susceptible to the influence of lighting conditions and individual differences, and has the problem of high environmental dependency.

[0004] On the other hand, the geometric method is a method that calculates the gaze direction based on a geometric model of the eyeball. In conventional geometric methods, the center of eyeball rotation is either treated as a fixed value or estimated through statistical correction. However, the actual center of eyeball rotation varies greatly between individuals (a variation of about 3 mm in the anterior-posterior direction and about 2 mm in the up-down, left-right directions), and apparently changes according to changes in the gaze direction and head posture. For this reason, conventional geometric methods can only achieve a gaze tracking accuracy of about 1 degree to 3 degrees, which has the problems of causing VR motion sickness and being unsuitable for precision work.

[0005] Furthermore, conventional eye-tracking technology required 60 to 120 seconds for calibration, which was a significant burden on the user. Additionally, if the glasses shifted position or the headset's placement changed during use, recalibration was necessary, posing practical problems.

[0006] Furthermore, conventional eye-tracking systems represent the direction of gaze only in a device-specific relative coordinate system, making it difficult to share gaze information between multiple devices or to determine the position of objects in the real world's absolute coordinate system. In particular, when integrating with GNSS (Global Navigation Satellite System) in outdoor environments, the geoid height is not adequately considered, resulting in positioning accuracy of only about ±3m to 5m.

[0007] In particular, in the field of VR goggles (head-mounted displays: HMDs), a type of wearable image display device, high-performance graphics processing must be performed in real time. However, rendering the entire human field of view at high resolution (e.g., 4K or 8K) places an extremely high load on current computing resources, posing significant challenges such as increased power consumption, heat generation, and rendering delays (latency). Foveated rendering technology is attracting attention as a method to alleviate this rendering load. Foveated rendering is a technique that renders the central field of view (fovea), which the user perceives most clearly, at high resolution, while intentionally rendering the rest of the peripheral field of view at a lower resolution. In order to effectively apply this technology, it is necessary to measure the user's gaze direction (point of gaze) with high precision and low latency (typically 100Hz or more) and to accurately and instantaneously track the center of rendering. As mentioned earlier, conventional eye-tracking systems have insufficient accuracy (an error of about 1 to 3 degrees), resulting in a visible discrepancy between the rendering center and the actual point of focus. This creates a new problem: it causes discomfort and unease for the user, preventing them from fully enjoying the benefits of foveated rendering.

[0008] Table 1 below shows a comparison between conventional eye-tracking technology and the present invention. Table 1: Comparison of Eye-Tracking Technologies |Item |Appearance-based estimation|Conventional geometric methods |This invention| |Center of eye rotation| Not used |Fixed value or statistical correction |Dynamic calculation and update based on geometric constraints| |Reference point |Not required |Fixed marker |External reference intersection (dynamically formed)| |Eyesight Accuracy| 1-5 degrees to 5 degrees | 1 degree to 3 degrees | 0-3 degrees to 0.7 degrees (approximately 5 times improvement)| |Calibration Time| 30 to 60 seconds | 60 to 120 seconds | 5 to 25 seconds (approximately 80 percent reduction) |Environmental dependence |High |Medium |Low (automatic correction function) | |Real-time performance| High | Medium | High (Processing time 7.7ms-120Hz)| |Absolute coordinate system support| Not possible | Not possible | Possible

[0009] Furthermore, Table 2 below shows a comparison of absolute coordinate system integration techniques. Table 2 Comparison of absolute coordinate system integration |Item |GNSS alone |Eye-tracking alone |This invention | |Outdoor positioning accuracy| ±3m to 5m |Not applicable | ±0-5m (geoid correction applied)| |Indoor positioning accuracy| Not applicable |Relative |High-precision positioning through coordinate-only SLAM integration| |Device-to-device communication| Not possible | Not possible | Possible (coordinate sharing function) | |Real-time performance|Low (update cycle 1Hz)|High (120Hz)|High (120Hz)| |Environmental adaptability|Outdoor only |Suitable for both indoor and outdoor use|Seamless indoor and outdoor | [Prior art documents] [Patent Documents]

[0010] [Patent Document 1] Japanese Patent Publication No. 2025-027566 [Patent Document 2] Japanese Patent Publication No. 2025-169318 [Patent Document 3] Japanese Patent Publication No. 2005-13752 [Patent Document 4] Japanese Patent Publication No. 2014-52758 [Patent Document 5] Japanese Patent Publication No. 2020-74952 [Overview of the project] [Problems that the invention aims to solve]

[0011] The first challenge is to improve the accuracy of eye-tracking based on the geometric determination of the center of rotation (ERC). Conventional geometric methods either treat the ERC as a fixed value or estimate it through statistical correction, making it unavoidable that accuracy would decrease due to individual differences, misalignment of glasses or headsets, and changes in gaze direction. This invention utilizes a physically known three-dimensional reference point called an external reference intersection to geometrically inversely calculate the ERC based on the geometric constraints of the eyeball, and dynamically updates it in real time in response to changes in gaze direction. The aim is to achieve highly accurate eye-tracking with an accuracy of 0 to 3 degrees at the center of the field of view and 0.7 degrees at the edge of the field of view. This represents an accuracy improvement of approximately five times compared to conventional technology and will serve as a basis for reducing VR sickness and optimizing foveated rendering, especially in VR / AR applications.

[0012] The second challenge is to simplify calibration and improve system robustness. Conventional eye-tracking systems require users to sequentially gaze at numerous fixed markers, resulting in calibration times ranging from 60 to 120 seconds. The present invention aims to reduce the initial calibration time from 5 to 25 seconds by dynamically forming external reference intersections and automatically calculating the optimal intersection placement. Furthermore, the present invention has a real-virtual intersection coordinate matching function that constantly monitors the spatial difference between the external reference intersections and virtual intersection coordinates, and aims to maintain accuracy against displacement of eyeglasses or headsets, as well as drift in system parameters due to internal thermal expansion and aging, through automatic recalibration during use.

[0013] A third object of the present invention is to represent gaze tracking information in an absolute coordinate system and expand the technology to a wide range of application fields. Conventional gaze tracking systems only represent gaze directions in a device-specific relative coordinate system, which makes it difficult to share gaze information between multiple devices and identify the positions of objects in the real world. The present invention applies geoid height correction to GNSS positioning data and integrates the corrected data with SLAM (simultaneous localization and mapping) technology, so as to represent the gaze direction and gaze point in an absolute coordinate system regardless of whether the location is indoor or outdoor. This enables sharing of gaze information among multiple users, remote control of devices such as drones, and accurate placement of virtual objects in augmented reality. Furthermore, another object of the present invention is to expand the application as a platform that comprehensively digitizes the geometric structure of the eyeball (such as the center of corneal curvature and the center of the pupil plane) based on this high-precision ERC. [Means for Solving the Problems]

[0014] To solve the above problems, the gaze tracking system of the present invention has the following configuration.

[0015] A first means of the present invention is a gaze tracking system comprising: a gaze tracking sensor that acquires a user's eyeball information; a means for forming an external reference intersection point; and an information analysis means that geometrically calculates an eyeball rotation center by associating the external reference intersection point with the eyeball information, wherein the information analysis means determines the physiological validity of the calculated eyeball rotation center, evaluates the error between an assumed gaze point derived from the eyeball rotation center and the external reference intersection point, and determines or updates the eyeball rotation center to reduce the error while satisfying the geometric constraint conditions in an eyeball model.

[0016] A second means of the present invention is the gaze tracking system according to the first means above, wherein the information analysis means but , in addition to the calibration using the external reference intersection point, hand, based on a reference point of a calibration jig having known three-dimensional coordinates, calculates the eyeball rotation center , which is characterized by the above.

[0017] A third aspect of the present invention provides the gaze tracking system according to the first aspect, wherein said means for forming the external reference intersection is supported by a holding member so as to maintain a known geometric arrangement relationship, By continuous wave or pulse wave a plurality of oscillation means capable of emitting a beam having rectilinearity or directivity; and control means for causing non-parallel beams emitted from said plurality of oscillation means to intersect in a space, comprising: said beams include electromagnetic wave beams and acoustic wave beams at least one of the that are a combination of those having the same or different physical properties .

[0018] A fourth aspect of the present invention provides the gaze tracking system according to the third aspect, wherein said control means is capable of controlling the irradiation timing of said beam constantly, intermittently, or based on a user's instruction.

[0019] A fifth aspect of the present invention provides the gaze tracking system according to the first aspect, wherein said information analysis means includes calibration means for comparing and collating spatial differences among coordinates of a calibration target intentionally gazed by a user, coordinates of said external reference intersection, and coordinates of an assumed gaze point calculated by said gaze tracking sensor, and correcting or re-determining said center of eye rotation based on a result of said comparison and collation.

[0020] A sixth aspect of the present invention provides the gaze tracking system according to the first aspect, wherein said information analysis means dynamically corrects at least any one of three-dimensional coordinates of said center of eye rotation and three-dimensional coordinates of feature points of an eye included in said eye information based on a correction amount obtained from an error with respect to said external reference intersection.

[0021] The seventh means of the present invention is characterized in that, in the eye-tracking system described in the first means above, the information analysis means analyzes the spatial difference between the coordinates of the external reference intersection and the coordinates of the assumed gaze point calculated by the eye-tracking sensor as a three-dimensional coordinate vector.

[0022] The eighth means of the present invention is characterized in that, in the eye-tracking system described in the first means above, the information analysis means stores the relative positional relationship between the center of eye rotation and the feature points of the eye as a personal template for biometric authentication and uses it for authentication.

[0023] The ninth means of the present invention is, in the eye-tracking system described in the first means above, Means for forming the external reference intersection but , the above Forming an external reference intersection The system includes means for adding unique identification information to the beam, The aforementioned Identification information includes temporal modulation patterns, frequency characteristics, or Includes at least one of the encoded pieces of information, The information analysis means, based on the identification information, By verification The authenticity of the aforementioned external reference intersection is determined. , eliminate irregular signals It is characterized by its ability to detect counterfeit products or external interference.

[0024] A tenth means of the present invention is a gaze tracking system described in the first means above, characterized in that the information analysis means converts the coordinates of the eyeball rotation center and the point of fixation to a fixed coordinate system established by an Earth-fixed coordinate system, a global coordinate system, or a local environment mapping based on information from an external coordinate assignment system, and cooperates with an external system.

[0025] An eleventh means of the present invention is the eye-tracking system described in the first means above, further comprising a depth measuring means, wherein the information analysis means synchronously controls the gaze direction calculated by the eye-tracking sensor and the depth information acquired by the depth measuring means to determine the gaze coordinates in three-dimensional space.

[0026] The twelfth means of the present invention is, in the eye-tracking system described in the first means above, The information analysis means determines the actual intersection of the external reference intersection coordinate and, The aforementioned Position of the oscillator that forms the external reference intersection and Irradiation direction to Based Geometrically extend the central axis of the beam hand Guidance Distributed virtual intersection coordinate Compare and Based on that difference, the error factors are as follows Forming an external reference intersection The displacement of the oscillation means or the drift of the eyeball model either That is To estimate or determine whether Selectively correct Self-diagnostic function It is characterized by the following:

[0027] A thirteenth means of the present invention is the eye-tracking system described in the first means above, further comprising an inertial measurement unit (IMU), wherein the information analysis means compensates for the correction of the eyeball rotation center based on positional displacement or deformation information of the mounting device detected by the inertial measurement unit.

[0028] The fourteenth means of the present invention is, in the eye-tracking system described in the first means above, The information analysis means includes a temperature sensor, a humidity sensor, or Based on long-term fluctuation information of the internal clock, The aforementioned Eye-tracking system Thermal expansion, contraction, or It detects errors caused by age-related drift in electronic components. The aforementioned error is due to the misalignment of the oscillation means that forms the external reference intersection. If this is the cause, the arrangement coordinates or irradiation direction of the oscillation means Correct, If the aforementioned error is due to the drift of the eyeball model The aforementioned center of rotation of the eyeball coordinate To compensate Its distinguishing feature is the eye-tracking system.

[0029] The fifteenth means of the present invention is the eye-tracking system according to claim 1, wherein the information analysis means comprises a model that dynamically corrects the baseline length between the centers of eye rotation based on the gaze direction of both eyes and the rotational movement (torsion) of the eyeballs.

[0030] The sixteenth means of the present invention is a self-correction means in which the information analysis means recalculates the center of eye rotation using a calibration target on which known coordinates are set.

[0031] The seventeenth means of the present invention is characterized in that, in the eye-tracking system described in the first means above, the information analysis means analyzes the time-series fluctuations of the eyeball rotation center or the long-term fluctuation patterns of the external reference intersection using statistical filtering or a machine learning algorithm, and corrects the eyeball rotation center.

[0032] The eighteenth means of the present invention is an eye-tracking system according to claim 1, wherein the system is mounted on a wearable image display device, and the information analysis means further includes a foveated rendering control unit that divides the display screen into a central field of view, a peripheral field of view, and an outer field of view based on the gaze direction derived from the eyeball rotation center, and renders the central field of view at high resolution, the peripheral field of view at medium resolution, and the outer field of view at low resolution, and a VR sickness reduction control unit that predicts the vestibulo-ocular reflex from the time-series fluctuation pattern of the eyeball rotation center and corrects the motion vector of the displayed image.

[0033] The 19th means of the present invention is an eye-tracking method, A process of forming an external reference intersection by intersecting beams having straight-line or directional properties, emitted from multiple oscillating means, A process of acquiring the user's eyeball information using an eye-tracking sensor, A step of geometrically calculating the center of rotation of the eyeball by associating the aforementioned external reference intersection with the aforementioned eyeball information, A step of determining the physiological validity of the calculated center of eye rotation, A step of evaluating the error between the assumed point of fixation derived from the center of eye rotation and the external reference intersection, A step of determining or updating the center of rotation of the eyeball in such a way as to reduce the error while satisfying the geometric constraints in the eyeball model, Based on the line of sight direction derived from the eyeball rotation center, the display screen is divided into central field of view, peripheral field of view, and outer field of view, and the rendering resolution of each field of view region is dynamically adjusted. It is characterized by including. The 20th means of the present invention is, in the eye-tracking system described in the first means above, The information analysis means is identified by the eye-tracking sensor Information on the point of focus and instructions based on the user's active will. Combine these to analyze commands sent to external devices. It further includes a command input analysis unit, The command input analysis unit handles operation of the control panel, voice commands, gesture input, Or at least two of the following: specific eye movement patterns or blinking rhythm patterns. The input modalities occur simultaneously or sequentially. As authentication requirements, The actual intersection coordinates of the external reference intersection and the corresponding point of gaze It was determined that the geometric consistency constraint with the virtual intersection coordinates was satisfied. A signal that controls the operation of the external device corresponding to the point of focus. It is characterized by transmitting. [Effects of the Invention]

[0034] The effect of claim 1 is that by establishing a dual verification criterion—geometrically calculating the user's eye rotation center (ERC) using a physically known criterion called an external reference intersection, and further determining the physiological validity of the calculation result—it achieves an extremely high eye-tracking accuracy of 0.3 to 0.7 degrees compared to conventional methods that rely on statistical estimation. Furthermore, by making the determination of physiological validity a mandatory requirement, even if competitors attempt structural circumvention using purely AI / statistical methods, the resulting verification process will infringe the essential requirements of this patent, thus maximizing the defense in enforcing the patent.

[0035] The effect of claim 2 is that, in addition to dynamic calibration using external reference intersections, or alternatively, by calculating the ERC based on reference points of a calibration jig having known three-dimensional coordinates, static and highly accurate initial calibration that strictly conforms to physical standards becomes possible during manufacturing and maintenance, improving the absolute accuracy and reliability of the entire system.

[0036] The effect of claim 3 is that by using a combination of different physical properties, including an electromagnetic wave beam (light) and an acoustic wave beam, in the beam that forms the external reference intersection, the stability and reliability of the intersection can be highly ensured even in situations where intersection formation is difficult with a single beam, such as environments with a lot of optical noise or acoustic interference, and the robustness of the system in real-world environments is greatly improved.

[0037] The effect of claim 4 is that, because the control means can control the beam irradiation timing continuously, intermittently, or based on user instructions, unnecessary power consumption can be suppressed and heat generation can be minimized, thereby improving the long-term reliability of the device and the safety of the user.

[0038] The effect of claim 5 is that by providing a calibration means that compares and verifies the coordinates of a calibration target that the user has intentionally focused on with the coordinates of an external reference intersection and an assumed point of focus, the ERC can be easily and quickly modified or re-determined based on the user's active operation, and initial setup and accuracy verification become easier.

[0039] The effect of claim 6 is that, by dynamically correcting the three-dimensional coordinates of the ERC or eyeball feature points based on the correction amount obtained from the error with the external reference intersection, it becomes possible to perform highly accurate self-correction in three-dimensional space including the depth direction (Z-axis), which was difficult with conventional two-dimensional correction, and the stability of eye tracking is dramatically improved.

[0040] The effect of claim 7 is that by analyzing the spatial difference between the external reference intersection and the assumed gaze point as a three-dimensional coordinate vector, the direction and magnitude of the error can be precisely grasped, enabling the identification of the cause of the error and the calculation of a precise correction amount to cancel it out, thereby contributing to ensuring depth accuracy in high-precision AR / VR applications.

[0041] The effect of claim 8 is to store the relative positional relationship between the center of eye rotation (ERC) and the characteristic points of the eye as a personal template for biometric authentication and use it for authentication, thereby realizing highly secure passive biometric authentication (authentication that does not require the user to consciously perform an authentication operation) using information that is extremely unique to the individual and difficult to duplicate, namely the parameters of the eye model.

[0042] The effect of claim 9 is that by adding unique identification information (serial code or modulated signal) to the beam of the external reference intersection and having the function of eliminating unauthorized signals, it is possible to detect and prevent misuse of system functions by counterfeit products and external interference from third parties, thereby strongly protecting the intellectual property rights of the external reference intersection technology, which is the core technology of the present invention, and the security of the entire system.

[0043] The effect of claim 10 is that by converting the coordinates of the ERC and the gaze point to a fixed coordinate system established by the Earth's fixed coordinate system or local environment mapping based on external information such as GNSS and SLAM, the eye-tracking results are freed from the device's specific relative coordinate system, enabling seamless representation in an absolute coordinate system regardless of whether it is indoors or outdoors, and enabling high-precision sharing of eye-tracking information between multiple devices.

[0044] The effect of claim 11 is that by further providing depth measuring means and synchronously controlling the line of sight direction and depth information, it becomes possible to determine the gaze coordinate in three-dimensional space based on measured depth information, without relying solely on monocular or binocular line of sight information, and the accuracy of depth estimation for nearby objects is greatly improved.

[0045] The effect of claim 12 is that by comparing the actual intersection coordinates of the external reference intersection with the virtual intersection coordinates based on the eyeball model and analyzing the difference based on spatial coordinates, it is possible to self-diagnose whether the error cause is due to hardware (deviation of the oscillation means) or software (drift of the eyeball model). This enables selective correction according to the cause, and the robustness and long-term stability of the system are greatly improved.

[0046] The effect of claim 13 is to compensate for ERC corrections based on positional displacement and deformation information of the worn device detected by the inertial measurement unit (IMU), thereby absorbing temporary errors caused by user head movements and changes in the device's wearing state in real time and maintaining highly accurate eye-tracking results at all times.

[0047] The effect of claim 14 is to detect and compensate for errors caused by thermal expansion of components or aging drift of electronic components based on long-term fluctuation information of a temperature sensor, humidity sensor, or internal clock. This effectively suppresses ERC drift caused not only by external influences but also by unstructured long-term fluctuation elements within the system, thereby ensuring accuracy throughout the entire product lifecycle.

[0048] The effect of claim 15 is to provide a model that dynamically corrects the intercenter baseline length (IPD) based not only on the direction of gaze of both eyes but also on the rotational movement (torsion) of the eyeballs. This accurately corrects physiological errors that tend to occur when fixating on the peripheral areas of the visual field, and uniformly improves the accuracy of eye tracking across the entire visual field.

[0049] The effect of claim 16 is that by providing a self-correction means that recalculates the ERC using a calibration target with known coordinates, it becomes possible to automatically readjust to a high-precision ERC with minimal user intervention when the system is started up or when the accuracy falls below a certain threshold, thereby improving practicality.

[0050] The effect of claim 17 is that by analyzing and correcting the time-series fluctuations of the ERC and the long-term fluctuation patterns of the external reference intersection using statistical filtering or machine learning algorithms, measurement noise can be removed and advanced predictive corrections based on complex environmental and aging patterns can be made, further improving the stability and reliability of eye-tracking results.

[0051] The effect of claim 18 is that by using an absolute coordinate standard called an external reference intersection, it becomes possible to determine the center of eye rotation with extremely high precision and without being affected by positional displacement or individual differences of the worn device. This significantly improves the accuracy of eye tracking, eliminates the unnatural feeling of foveated rendering in VR goggles and the like, and suppresses VR sickness. Furthermore, by integrating IMU compensation, machine learning, dynamic baseline length correction, and a secure signal authentication mechanism, long-term robustness and security in real-world usage environments, as well as secure cooperation with external devices, are established, dramatically expanding the practical application range of high-precision eye tracking systems.

[0052] The effect of claim 19 is that, by protecting not only the system invention but also the method invention which includes the essential steps of forming an external reference intersection or calibration using a calibration jig and determining physiological validity, strong rights are established for the core procedure (method) of the present invention, making it difficult for competitors to circumvent the method steps even if they change the system configuration, thus providing a double layer of legal protection. [Brief explanation of the drawing]

[0053] [Figure 1] This is a block diagram showing the overall configuration of an eye-tracking system according to an embodiment of the present invention. [Figure 2] This is a block diagram showing the overall configuration of the claims of the present invention. [Figure 3] This is an explanatory diagram showing the geometric relationship between the eyeball model and the center of rotation of the eyeball. [Figure 4] This is a diagram illustrating the configuration of the external reference intersection forming means. [Figure 5] This is a diagram illustrating the geometric relationship between the laser light source and the intersection locking mechanism. [Figure 6] This is a functional block diagram of the information analysis means. [Figure 7] This is a block diagram of the elements of the system control means. [Figure 8] This is a conceptual diagram illustrating coordinate transformations performed using an absolute coordinate system integration process (GNSS, geoid height correction, SLAM). [Figure 9] This is a schematic diagram illustrating the identification of a three-dimensional point of fixation by integrating depth measurement means and line-of-sight information (Claim 11). [Figure 10] This is a flowchart for identifying and selectively correcting error factors by comparing actual virtual intersection coordinates (Claim 12). [Figure 11] (a) is a schematic diagram showing the geometric relationship between the oscillation means and the beam intersection in the eye-tracking system of the present invention, and (b) is a perspective view showing a specific embodiment of the eye-tracking system (1) as a wearable device. [Figure 12] (a) is a schematic diagram showing the geometric relationship between the center of eye rotation and the fixation point (15) in binocular eye-tracking, and (b) is a schematic diagram showing the arrangement of the external reference intersection (6) and the fixation point (15) in three-dimensional space in the eye-tracking system. [Figure 13] This is a schematic diagram illustrating the principle of calculating the center of eye rotation (13), least squares estimation based on geometric constraints, and the principle of physiological validity determination (Claim 1). [Figure 14] (a) is a conceptual diagram of static calibration using a reference point of a calibration jig (Claim 2), and (b) is a diagram showing the geometric calibration principle using the beam intersection (6) as a reference point. [Figure 15] This is a schematic diagram showing the sensor arrangement (IMU, temperature sensor, humidity sensor) and application examples such as drone control. [Figure 16] This is a flowchart illustrating a method for determining the center of eye rotation based on geometric constraints. [Figure 17] These are the main mathematical formulas used in this specification. [Modes for carrying out the invention]

[0054] The embodiments of the eye-tracking system of the present invention will be described in detail below with reference to the drawings. The terms used in this invention are defined as follows, to the extent necessary to realize the technical concept of the present invention. These definitions do not limit the specific embodiments or components of the present invention, but rather are overarching conceptual definitions intended to maximize the technical scope to which the present invention applies and to protect future technological innovations and alternative means. Furthermore, the specific formulas, algorithms, and control laws for geometric calculations, coordinate transformations, and predictive control described herein are presented as representative methods to illustrate specific embodiments and do not limit the scope of the invention to their strict forms. These formulas are broadly interpreted as being replaceable by known alternative formulas, approximations, and optimization methods commonly used by those skilled in the art, as long as they are based on fundamental scientific principles such as triangulation, geometry, linear algebra, or statistical methods. This prevents imitators from formally circumventing the claims by making minor changes to the formulas or substituting them with known alternative algorithms, thereby maximizing the exclusive scope of the present invention. Furthermore, even without the use of terms like "approximately" or "about," the numerical values, physical quantities, dimensions, angles, etc., described herein are not strictly limited to exact values ​​within the scope necessary to solve the problem and achieve the intended effect of the invention. They should be understood to include errors and design variations that are ordinarily acceptable to those skilled in the art. Quantitative parameters such as numerical values, dimensions, and angles are described to illustrate specific embodiments and do not limit the technical scope of the present invention to these specific numerical values. These numerical values ​​should be understood to vary depending on design changes, manufacturing tolerances, measurement errors, or the operating environment, within the scope necessary to achieve the intended effect of the invention. Furthermore, unless otherwise specified, terms such as "includes," "equip," and "possess" in this specification are used in an open-ended form and do not exclude additional components or processes other than those described. Accordingly, the technical scope of the present invention is to be interpreted as not being limited to the explicitly described elements, but also including auxiliary components that a person skilled in the art would ordinarily add, and functionally equivalent alternative means that may be developed in the future through technological advancements. The eye-tracking sensor (7), oscillation means (2), physical sensors (IMU (35), etc.), and associated instruments and equipment incorporated in this eye-tracking system are not limited to the embodiments described herein, but include technologies and devices newly developed or realized after the filing of this patent application. In particular, the application of future sensing technologies that enable more accurate acquisition of eye information, such as quantum sensing technology and bioimpedance measurement, is also included in the technical scope of the present invention. Important mathematical formulas are shown in text, and each formula is assigned a unique, unified number (e.g., Formula 1, Formula 2). The formal mathematical notation of these formulas is summarized in Figure 17, and the formula numbers in the text correspond to the formula numbers in the figure.

[0055] Figure 1 is a block diagram showing the overall configuration of an eye-tracking system according to an embodiment of the "High-Precision Eye-Tracking System Using the Center of Eye Rotation" of the present invention. The "eye-tracking system (1)" refers to the entire device that detects the user's (9) gaze direction and calculates the point of gaze (15) in three-dimensional space (18). The system may take any form, such as a wearable device, a stationary device, a mobile device, or an environment-installed device, a chair, or a capsule, and its physical configuration and implementation are not limited. The eye-tracking system (1) may be an integrated type in which all components are housed within a single enclosure (a wearable device or a mobile object such as a car), or it may be a distributed type (drone, personal computer, etc.) in which some or all of the information analysis means are run on a local server, an edge computing environment, or a cloud server and connected via communication means (19). It may also be configured as part of a broader user interface system or biometric system that includes additional information processing functions for the purpose of estimating the user's behavior, cognition, or health status. The wearable device (34) worn by the user can be a smart glass, an AR (augmented reality) headset, a VR (virtual reality) goggle, or an HMD (head-mounted display), helmet, or any other device equipped with an optical or transmissive display (see Figures 11(b) and 15(b)). These devices include a holding member that integrally holds an eye-tracking sensor (7) for acquiring eye information and an oscillation means (2) for forming an external reference intersection that serves as a physical reference, near the user's eyes. This makes it possible to use the coordinates of the gaze point, calculated with high accuracy, for the placement of virtual objects displayed on the display and for controlling the user interface. The eye-tracking system (1) of this embodiment solves the eye-tracking error caused by the inaccuracy of the eye rotation center (13), which could not be solved in the prior art, by matching and correcting using an external reference intersection, thereby realizing high-precision and dynamically updateable eye-tracking. (See Figure 4)

[0056] This eye-tracking system (1) solves the problem of increased errors due to the fixed eyeball rotation center (13) in conventional technology by dynamically updating the position of the eyeball rotation center in real time while maintaining the geometric characteristics of the eyeball (10) (Claim 1). This update process essentially includes constrained optimization based on geometric calculations, physiological validity assessment, and error evaluation. Furthermore, it performs matching and correction using the geometric relationship with external reference points (23) having known three-dimensional coordinates and external reference intersection points (6) formed by multiple beams. As a result, it minimizes errors caused by minute positional shifts of the mounting device (34) or holding members (3a) that hold and constitute the system, as well as expansion, contraction, and individual differences, enabling extremely robust and ultra-high-precision three-dimensional eye-tracking. This system has clear advantages over conventional eye-tracking systems in the following three points. 1. Dynamic updating of the eyeball rotation center: While conventional systems fix the eyeball rotation center (13) once it is determined, this system dynamically updates the eyeball rotation center in real time even during use. This continuously absorbs minute misalignments of the mounting device (34) and drift errors due to prolonged use, dramatically improving long-term stability. 2. High-precision self-diagnosis and correction using an external reference intersection: An external reference intersection (6) formed in space by beams emitted from multiple oscillation means (2) is used as the absolute geometric reference of the system. By comparing and verifying this reference with the observation results of the eye-tracking sensor (see paragraph 0146), the system's health is self-diagnosed, and if an anomaly is detected, automatic correction is transparently performed by software or hardware. This provides robustness not found in conventional systems using a fixed target (19). 3. Calibration-Free Operation: The combination of the dynamic updates and self-diagnostic functions described above allows users to perform a quick initial calibration (a few seconds to tens of seconds) and then utilize high-precision eye tracking for extended periods without conscious recalibration. This will significantly improve convenience and availability in practical environments such as industrial, medical, and VR / AR applications.

[0057] "Oscillation means (2)" is a broad concept encompassing any device that generates a beam that becomes a directional wave. Oscillation means (2) is a fundamental element for forming the external reference intersection (6), which is the core technology of the present invention, and the oscillated beam provides reference information for determining the coordinates of the external reference intersection. By selecting its physical characteristics, stable intersection formation under diverse environmental conditions is possible. Oscillation means (2) includes laser oscillators (2a), sound wave oscillators (2b), ultrasonic oscillators, electromagnetic wave oscillators, optical oscillators, acoustic oscillators, and other wave sources. As for the laser oscillator (2a), semiconductor lasers, solid-state lasers, gas lasers, fiber lasers, or quantum cascade lasers can be used. The oscillation principle can be any method, such as semiconductor, solid, gas, fiber, quantum cascade, piezoelectric, structured optical rF, microwave, terahertz, etc. Electromagnetic wave beams include visible light lasers (wavelengths 450nm to 700nm), infrared lasers (wavelengths 700nm to 3000nm), LED light (visible or infrared), ultraviolet beams, millimeter-wave beams, etc. The oscillation wavelength may be in the visible light region (380nm to 780nm), near-infrared region (780nm to 2500nm), mid-infrared region, or ultraviolet region. The near-infrared region (850nm, 940nm) in particular is suitable for eye-tracking applications because it is highly safe for the human eye and corresponds to the sensitivity peak of silicon photodiodes. As the sound wave oscillator (2b), piezoelectric elements, electromagnetic speakers, electrostatic speakers, or ultrasonic transducers can be used. The oscillation frequency may be in the audible range (20Hz to 20kHz) or the ultrasonic range (20kHz or higher, especially 40kHz). There are no restrictions whatsoever on the range of oscillation wavelength and frequency. The versatility of the physical properties of the oscillation means (2) provides not only design freedom but also environmental adaptability and difficulty in patent circumvention. Even if competitors attempt to circumvent patents with embodiments limited to specific wavelength bands (e.g., visible light only), these are included within the scope of the patent rights under the broad definition of this application. Furthermore, the combined configuration of electromagnetic waves and sound waves provides complementary environmental resistance (optical noise / acoustic noise) and achieves high reliability that is difficult to achieve with prior art. The oscillation means (2) may oscillate a continuous wave or a pulsed wave, as described in claim 3.Pulse characteristics such as pulse width, pulse repetition frequency, and duty cycle can be arbitrarily set according to the application. The oscillation means (2) may be a monochromatic light source that oscillates a single wavelength, or a multicolor light source that oscillates multiple wavelengths simultaneously. The beams used in this system include electromagnetic wave beams and sound wave beams, and combinations of beams with the same or different physical characteristics are also possible. Beams with directionality and straight propagation may be formed from diffuse beams (LEDs, etc.) using lenses, waveguides, array structures, metalens, acoustic horns, etc. Here, directionality means the characteristic that the beam propagates concentrated within a specific angular range (e.g., diffusion angle of less than 10 milliradians), and does not include simply wide-angle diffuse light sources. The output intensity of the oscillation means (2) can be arbitrarily adjusted within a range that satisfies safety standards (IEC60825-1, etc.), and may be dynamically controlled according to the distance to the irradiation target, ambient light conditions, or detection accuracy requirements. In particular, the beam intensity, pulse characteristics, or irradiation direction of the oscillation means (2) are dynamically adjusted based on feedback to the line-of-sight tracking error or ambient light conditions calculated by the information analysis means (8). This dynamic control is performed by the beam intensity control unit (2f) based on information from the ambient light condition acquisition unit (56). It is also possible to increase the output intensity within a safe range depending on the application.

[0058] "Beam (5)" refers to a directional wave radiated from the oscillation means (2), and is a general term for wave energy having a spatially limited propagation path. Beam (5) refers to all directional waves radiated from the oscillation means (2) and broadly includes waves that have linear propagation properties and form a line, axis, or a region near thereto. Beam (5) includes laser beams, acoustic beams, ultrasonic beams, electromagnetic wave beams, or combinations thereof. Beam (5) has the property of propagating substantially linearly and defines a geometric line or axis in three-dimensional space (18). The linear propagation property of beam (5) is a fundamental prerequisite for calculating the coordinates of the external reference intersection (6) based on rigorous geometric principles such as triangulation. The beam divergence angle of beam (5) is determined by diffraction limit, collimation accuracy, or intentional beam shaping, and is preferably in the range of 0.01 milliradians to 10 milliradians, but is not limited thereto. The beam (5) may be irradiated as a single beam from a single oscillation means (2), or it may be split into multiple beams by a beam splitter or array configuration. The propagation characteristics of the beam (5) may be affected by the refractive index, temperature distribution, pressure distribution, or humidity distribution of the medium, and corrections based on these environmental parameters may be applied as needed. In particular, in the case of a sound wave beam, the speed of sound is temperature-dependent (increasing by approximately 0.6 m / s for every 1°C increase in temperature), so a correction based on temperature data measured by a temperature sensor (122) is performed by an environmental factor error detection unit (127). This correction aims to maintain the coordinate accuracy of the external reference intersection (6) to the high precision (sub-millimeter order) required for updating the center of rotation of the eyeball (ERC). Note that since there are technical difficulties in the reach of linearly propagating sound wave / acoustic beams, it is preferable to use a laser beam, ultrasonic beam, or electromagnetic wave beam. The term "external reference intersection (6)" or "beam intersection (6)" refers to a point or region where multiple beams (5) irradiated from spatially different positions intersect. The external reference intersection (6) is the most fundamental technical element of the present invention and simultaneously performs the following three roles: (1) a geometric reference point for calculating the eyeball rotation center (13), (2) an absolute coordinate anchor for verifying eye-tracking accuracy, and (3) a true reference point for system self-diagnosis. The external reference intersection (6) is a core feature of the present invention and always functions as a "reference point with known three-dimensional coordinates." The "geometric determination of the center of eye rotation (ERC) using an external reference intersection (physical reference)" is distinctly different from the internal reflection analysis and statistical corrections found in patents from Apple and Huawei. The external reference intersection (6), depending on its definition, encompasses two concepts: the actual intersection coordinate (24) and the virtual intersection coordinate (25). This dual structure enables the self-diagnostic function unique to the present invention. The actual intersection coordinate (24) is the region where the energies of multiple beams (5) physically overlap. In the actual intersection coordinate (24), the energies of multiple beams (5) overlap, and interference phenomena (in the case of the same wavelength) and nonlinear optical effects may occur. The position of the actual intersection coordinate (24) is measured by a physical detector such as a photodiode array-CCD sensor or an ultrasonic microphone. In particular, when using an acoustic beam, the point of overlap of the acoustic wave energy (actual intersection coordinate) is detected by an ultrasonic transducer array or an acoustic receiver array. The centroid coordinate of the detected energy density distribution is determined as the coordinate of the actual intersection coordinate (24).

[0059] The virtual intersection coordinate (25) is the point where the lines drawn by geometrically extending the central axes of each beam theoretically intersect. Although the virtual intersection coordinate (25) may be a point in space that the beams do not actually reach, it is calculated with high accuracy by the triangulation calculation unit (8a) from the known position of the oscillation means (2) (oscillation center point 2g) and the irradiation angle. The calculation of the virtual intersection coordinate (25) does not require physical detection and functions as a theoretical reference coordinate. Self-diagnosis by comparing actual and virtual intersection coordinates: The spatial difference between the measured coordinates of the actual intersection coordinates (24) and the theoretical coordinates of the virtual intersection coordinates (25) is evaluated by the following equation: (Figure 16, Equation 27) Δ=||Vreal-Vvirtual|| If this difference Δ exceeds a predetermined threshold (typical values ​​from 1.0 mm to 3.0 mm), it indicates that mechanical misalignment of the oscillation means (2), optical axis drift, or slight deformation of the holding part (3) has occurred. The self-diagnosis function by actual virtual intersection coordinate matching is a means of verifying the integrity of the system components themselves, which is performed prior to the error evaluation process of the eyeball rotation center (ERC) in claim 1. The actual virtual intersection coordinate matching unit (8s) continuously monitors this difference and performs automatic correction processing or warning notification when an abnormality is detected. Technological Advantage: The dual structure of the actual virtual intersection coordinates provides a unique concept of a "self-verifying reference point" that does not exist in conventional eye-tracking systems. In conventional technology, there is no means to verify the positional accuracy of the reference point (marker, etc.) itself, and the deviation of the reference point directly results in eye-tracking errors. In this invention, the soundness of the reference point itself can be verified in real time, which is a factor that significantly enhances patentability. The coordinates of the external reference intersection (6) are calculated by the intersection coordinate calculation unit (6a) based on the position coordinates of the oscillation means (2), the irradiation angle, and the propagation characteristics of the beam (5) (using equation 24, etc.). The position of the external reference intersection (6) may be fixed, or it can be moved to any position in space by dynamically changing the irradiation angle of the oscillation means (2) using the oscillation means angle adjustment mechanism (2d). When multiple external reference intersections (6) are formed, they are spatially dispersed to improve measurement accuracy (see Figures 11, 12, and 13). The intersection stabilization processing unit (6b) performs control to compensate for temporary fluctuations in the irradiation direction or intensity of the beam (5) due to external factors such as ambient light fluctuations, temperature changes, and device vibrations, thereby maintaining the stability of the spatial position of the external reference intersection (6). This process includes a temperature compensation algorithm or dynamic correction based on vibration data from an inertial sensor (10). The external reference intersection (6) is a core feature of the present invention and functions as a "reference point with known three-dimensional coordinates," providing an absolute reference point for improving the accuracy of eye-tracking, compensating for errors, and compensating for temperature fluctuations. As shown in Figure 12, the control means (44) intersects multiple mutually non-parallel, linear or directional light rays or sound wave beams (5) emitted from the oscillating means (2) in space to form the external reference intersection (6). The oscillation means (2) are preferably arranged as a pair, left and right, with the first beam (5a) irradiated from the left oscillation means (2) and the second beam (5b) irradiated from the right oscillation means (2). The oscillation sources for both beams are spaced apart by the baseline length of the holding part (16, typically 40 mm to 150 mm), and this geometric arrangement information is stored in the control means (44) as a fixed parameter. Furthermore, the oscillation means (2) is equipped with an oscillation means angle adjustment mechanism (2d) to realize "dynamic synchronous control," and the beam irradiation direction can be dynamically controlled independently in the vertical and horizontal directions based on commands from the control means (44). This makes it possible to move the external reference intersection (6) to coincide with or approach the user's (9) assumed point of focus, thereby expanding the calibration verification range to the entire three-dimensional space (18). This dynamic tracking control technically enables the implementation of "automatic and intermittent calibration (dynamic calibration) that does not require the user to perform calibration work," as claimed in claim 5.

[0060] The oscillator holder (3) is a structure for physically holding the oscillator (2) in a single housing (wearable device or fixed / stationary device, or a mobile object such as a car) into which the eye-tracking system (1) is incorporated, or in a distributed housing (drone, personal computer, etc.) in which multiple devices are connected via communication means (19). The holder defines the baseline length (16) of the holder between the held oscillator (2) and the eye-tracking sensor (7), and provides a basis for establishing the device's own reference coordinate system. The oscillator holder (3) is fixed to a wearable device (34) such as smart glasses and is designed to maintain a predetermined positional relationship with the user's head (see Figure 11-12-13). In housings such as personal computers and cars, the eye-tracking sensor that detects the state of the user's binoculars is provided in a position opposite the user, and the oscillator is held in the oscillator holder (3) so that it can project a beam toward the front of the user. When operating drones or robots, as shown in Figure 15b, it is possible to control them using a controller while monitoring the surrounding environment through cameras mounted on the drone or robot using goggles, and to form a beam intersection at the point of focus that the user has confirmed and recognized through the goggles, using a beam oscillator mounted on the drone or robot. (See Figure 15b) The holding member (3a) refers to a structural element that supports the oscillator holding part (3) and maintains the oscillation means (2) and other optical components (e.g., lenses, mirrors) in a predetermined geometric arrangement. This holding member (3a) is preferably made of a highly rigid material with a low coefficient of thermal expansion (e.g., carbon fiber reinforced plastic or low thermal expansion alloy) to suppress minute deformations due to temperature changes and external vibrations. The holding member (3a) plays a role in holding the oscillation means (2) with high precision so that the light emitted from the oscillation means (2) is accurately directed in the intended irradiation direction. The retaining member fixing mechanism (3b) refers to a mechanical mechanism for firmly fixing the oscillator retaining part (3) and the retaining member (3a) to the mounting device (34) or to each other. This mechanism is essential for preventing fluctuations in the position of the oscillator means (2) and the base length (16) of the retaining part due to the user's head movements or external vibrations, and for maintaining eye-tracking accuracy. The position adjustment rail section (3c) refers to a rail-shaped mechanism that allows the oscillator holding section (3) or holding member (3a) to slide precisely relative to the mounting device (34) and adjust its position. This adjustment function is used to optimize the placement of the oscillator means (2), including the base length (16) of the holding section, according to the individual user's interpupillary distance (IPD) and the distance between the center of eye rotation (13) and the mounting device (34), and is controlled manually or automatically by an actuator. This automatic adjustment function is feedback-controlled based on calibration information from the information analysis means (8) to compensate for changes in the user's head posture during wear and subtle aging changes in the holding member due to long-term use. The adjustment accuracy of the position adjustment rail section (3c) is preferably within ±0.1 mm, which enables adaptation to individual differences and shortens the initial calibration time. This system has a device local coordinate system (device-specific coordinate system). The "holding unit coordinate origin (4)" is strictly defined as the reference point of the device local coordinate system (device-specific coordinate system). Specifically, this point (4), which is provided on the holding member that supports the multiple oscillation means (2), is set as the origin O=(0,0,0) of the local coordinate system. (See Figure 5) This origin serves as the reference point for coordinate measurement and geometric calibration of the entire eye-tracking system, and the positions of all components, such as the oscillation means, eye-tracking sensor (7), external reference intersection (6), and eye rotation center (13), are expressed as relative coordinates from this origin (see Figure 12(a)). This holding unit coordinate system is a fixed and common reference coordinate system for dynamic updating of the eye rotation center (13), calculation of assumed gaze points, and coordinate linkage with external systems (world coordinate system, display coordinate system, etc.). The coordinate axes are preferably defined as a right-handed coordinate system with the origin (4) as the reference point, with the X-axis typically representing the baseline direction of the holding part (horizontal direction), the Y-axis representing the direction forward of the device (line of sight measurement axis), and the Z-axis representing the vertical upward direction. The most recommended position for the origin (4) is the center point of the baseline length (16) of the holding part, which ensures symmetry in coordinate calculations and improves the stability of numerical calculations. However, due to system limitations, it is possible to set the origin (4) to any position other than the baseline length center. What is important is that the set origin position maintains the same definition consistently across manufacturing lots and over time, and that this definition is stored in the calibration data storage unit (130a). This strict definition ensures the consistency and reproducibility of the system coordinate system, guaranteeing high-precision calibration and compatibility in coordinate integration with external systems (claim 10).

[0061] The oscillator control unit (2c) is a logic processing block that comprehensively controls the operation of the oscillation means (2). Primarily, it is responsible for setting a target angle for the oscillation means angle adjustment mechanism (2d) based on the irradiation direction command sent from the system overall control means (44), and outputting a control signal that instructs movement to that angle. The oscillator control unit (2c) performs high-speed dynamic tracking control based on feedback control laws such as proportional-integral-derivative (PID) control and state-space models. It also has a management function that dynamically adjusts the output intensity and pulse modulation frequency of the laser or beam (5) emitted from the oscillation means (2). The oscillator control unit (2c) comprehensively performs the following control functions: (1) Irradiation angle control: The target irradiation angle (pan angle θ_pan, tilt angle θ_tilt) is commanded to the oscillation means angle adjustment mechanism (2d). (2) Output intensity control: Based on information from the ambient light condition acquisition unit (56), the output intensity of the beam (5) is dynamically adjusted via the beam intensity control unit (2f). (3) Pulse modulation control: Pulse modulation is performed in order to superimpose the identification information (serial code, modulation signal) corresponding to claim 9 onto the beam (5). (4) Timing synchronization control: When using multiple oscillation means (2), their irradiation timings are synchronized or time-division controlled. The oscillation means angle adjustment mechanism (2d) is a mechanical and electrical mechanism for physically changing the optical axis or oscillation angle of the oscillation means (2). It is often built into the holding unit (3) and has the function of adjusting the oscillation angle of the beam oscillator (Figure 11a). Specifically, it consists of an actuator system that receives a signal from the oscillator control unit (2c) and drives an angle adjustment motor (45, e.g., galvanometer mirror, MEMS mirror, stepping motor, or ultrasonic motor). In particular, this mechanism also includes cases where a piezo actuator using a piezoelectric element or a microactuator using MEMS technology is employed for ultra-high-precision fine adjustment. This mechanism makes it possible to adjust the irradiation direction of the oscillation means (2) in the pan (horizontal) and tilt (vertical) directions with high precision. The oscillation means angle adjustment mechanism (2d) maintains a constant distance between the optical axes of the two oscillators, or their extensions, or between the lateral rotation pivot points, and ensures that the relative illumination and oscillation angles are precise. It is desirable to provide a controllable position and to set the line at which this position becomes constant as the baseline length (16) of the holding part. The base length (16) of the holding part is fixedly provided on the holding part (3), so that its physical distance remains constant even when the angle of the oscillation means angle adjustment mechanism (2d) is changed. Due to the permanent fixation of this base length, the calculation of the three-dimensional coordinates of the external reference intersection (6) by the oscillation beam (5) is always performed based on a stable absolute reference. An example of the performance specifications for the oscillation means angle adjustment mechanism (2d) is as follows: • Angle adjustment accuracy: ±0.1 degrees or more (when using a stepping motor), ±0.01 degrees or more (when using a galvanometer mirror) • Response time - less than 100ms (from command to reaching target angle) • Resolution: 0.01 degrees (when using a 16-bit encoder) • Drive range: ±15 degrees in the pan direction, ±10 degrees in the tilt direction (typical values, can be changed depending on the application) The adjustable angle range is a crucial factor in determining the extent of the three-dimensional space that forms the external reference intersection (6). A wider drive range allows for the dynamic positioning of the external reference intersection (6) to cover the entire user's field of view, improving the uniformity of eye-tracking accuracy across the entire field of view. The irradiation direction correction mechanism (2e) is a closed-loop feedback control mechanism that matches the actual irradiation direction of the oscillation means (2) to the target irradiation direction in real time, based on spatial errors and prediction errors calculated by the information analysis means (8) or the dynamic synchronization control unit (71). This mechanism includes a logic process that finely corrects the irradiation angle of the laser or beam by taking into account the amount of error when the three-dimensional coordinates of the external reference intersection (6) deviate from the assumed point of fixation, and controlling the oscillation means angle adjustment mechanism (2d). This makes it possible to accurately and dynamically synchronize the external reference intersection (6) with the point of fixation (15). The correction algorithm for the irradiation direction correction mechanism (2e) is expressed by the following formula: (Equation 29) Δθ correction = J^(-1)·δ Here, Δθ_correction is the correction amount for the irradiation angle (pan angle correction Δθ_pan, tilt angle correction Δθ_tilt), J is defined by the Jacobian matrix (Equation 34, which describes the relationship between the irradiation angle θ and the position V of the external reference intersection (6)), and δ is the spatial error vector between the external reference intersection (6) and the assumed gaze point (Equation 28). This Δθ_correction is calculated to converge the geometric error between the vector from the corrected eyeball rotation center (13) to the external reference intersection (6) and the actual beam irradiation direction to zero.

[0062] The logic for determining the location the user (9) is currently looking at is established by the following components. First, based on the coordinates of the eyeball rotation center (13) corrected by the eyeball rotation center dynamic update unit (8m) and the latest pupil center (11c) information detected by the gaze tracking sensor (7), the gaze direction calculation unit (8n) calculates the user's current gaze direction vector (14): (Formula 2)v_gaze(t)=[P_pupil(t)-C_eye(t)]||P_pupil(t)-C_eye(t)||+δ_calib Here, C_eye(t) is the coordinate of the eyeball rotation center (13), which has been updated in real time through geometric matching and physiological validity determination using an external reference intersection (6), and is fundamentally different from the conventional fixed value. Next, based on this line of sight direction vector (14) and associated depth information, the geometric calculation unit (8c) calculates the three-dimensional coordinate of the assumed point of fixation as the point where the current line of sight vectors intersect in space. In the case of binocular vision, the intersection or closest point of the left eye line of sight vector (14a) and the right eye line of sight vector (14b) is calculated as the assumed point of fixation (Equations 7 and 22). If the left and right line of sight vectors do not intersect spatially, the virtual intersection coordinate, optimized by the least squares method, is determined as the assumed point of fixation as the closest point of fixation or a point on the perpendicular bisector between the rays formed by the lines of sight of both eyes. This assumed point of fixation becomes the virtual coordinate of the position the user is currently looking at. The calculation accuracy of the assumed point of fixation directly depends on the accuracy of the eyeball rotation center (13). In this invention, the eyeball rotation center (13) is continuously corrected using an absolute coordinate reference called the external reference intersection (6), so that the spatial coordinate error of the assumed point of fixation is suppressed to within ±3 mm.

[0063] The main source of the irradiation direction command consists of the control function of the system control means (44), which integrates the assumed gaze point coordinates established in the previous paragraph with the following components. Gaze Point Prediction Algorithm (22): Based on the time-series data of the latest gaze direction vector (14) calculated by the gaze direction calculation unit (8n), the algorithm predicts the three-dimensional coordinates of the next gaze point that the user is likely to focus on. Time-series prediction methods such as Kalman filters, extended Kalman filters, particle filters, or LSTM (Long Short-Term Memory) neural networks can be used as prediction methods. In particular, recurrent neural networks (RNNs) and transformer models are used for prediction, and prediction accuracy can be improved by learning the transition patterns of fast eye movements such as saccades. The target irradiation direction (angle command θ_target) for forming an external reference intersection (6) in advance at these predicted coordinates is calculated as the main command to be sent to the oscillation means angle adjustment mechanism (2d). Furthermore, the target irradiation direction for accurately synchronizing the external reference intersection (6) and following the current assumed gaze point coordinates without delay is also determined by this algorithm. The gaze point prediction algorithm (22) enables proactive positioning, compensating for the tracking delay of the external reference intersection (6) associated with gaze movement, and maintaining synchronization accuracy between the external reference intersection (6) and the gaze point (15) even during high-speed eye movements such as saccade motion (angular velocity of 300 degrees / second to 700 degrees / second). The target illumination direction θtarget is calculated based on coordinates predicted in advance by time Δt to offset the total system delay time (latency) required for sensor acquisition, image processing, information analysis, and actuator driving. Inverse calculation processing unit (8e) and synchronization control unit (8f): When the eyeball rotation center (13) is corrected by geometric inverse calculation, or when a spatial error is detected between the assumed gaze point and the external reference intersection (6), these units calculate a fine angle correction amount to eliminate the error in real time. This correction amount functions as a feedback command that is added to the main command. Based on the spatial error (Equation 27) Δ = ||V_real - V_virtual|| calculated by the difference analysis unit (8d), the irradiation direction correction amount Δθ_feedback is calculated. This equation is a feedback control law for rapidly and stably converging the error between the external reference intersection (6) and the assumed gaze point to zero.

[0064] As a function of the system control means (44), the oscillator control unit (2c) strictly synchronizes the target angle command θ_target received from the gaze point prediction algorithm (22) and the correction command Δθ_feedback received from the information analysis means (8) (via the error correction integration unit (8g), etc.) on the time axis, and integrates them in the irradiation direction correction mechanism (2e). This integration process is executed on a dedicated real-time operating system (RTOS) or FPGA (Field-programmable Gate Array) at an extremely high frequency (100Hz to 1000Hz or higher) in synchronization with the system clock. This integrated final irradiation direction command θ_final is expressed by the following equation: θ_final=θ_target+Δθ_feedback+Δθ_environment correction Here, Δθ_environmental correction is a correction term based on environmental factors detected by a temperature sensor (122), a humidity sensor (123), etc., and is calculated by an environmental factor error detection unit (127). Δθ_feedback is a feedback term that comprehensively minimizes a wide range of error factors, including changes in geometric constraints associated with the update of the eyeball rotation center (ERC). This integrated final irradiation direction command θ_final is sent to the oscillation means angle adjustment mechanism (2d) via the oscillator control unit (2c) to achieve accurate dynamic synchronization of the external reference intersection (6). To ensure real-time performance, the integration processing of the irradiation direction command is preferably performed within a processing cycle of 10ms or less (100Hz or higher).

[0065] The "holding section baseline length (16)" refers to the physical distance between the optical axis center of the oscillator means (2) and the optical axis center of the gaze tracking sensor (7), as defined by the oscillator holding section (3). (See Figure 5) More precisely, when multiple oscillation means (2) are arranged, the distance between their oscillation center points (2g) is defined as the holding part baseline length (16). This holding part baseline length (16) is defined as a fundamental geometric parameter for establishing a unique reference coordinate system for the eye-tracking system (1) incorporated into the mounting device (34). This parameter is recalibrated intermittently or continuously based on the system's self-diagnostic function (such as actual virtual intersection coordinate matching) to compensate for deformation of the holding member (3a) due to environmental changes and aging, in addition to high-precision calibration during manufacturing. Typical values ​​for the base length of the retaining part (16) are in the range of 40 mm to 150 mm for wearable devices, with 60 mm to 70 mm being most preferable as it is close to the interpupillary distance (IPD) of humans. The longer the base length of the retaining part (16), the less the influence of angle measurement errors in triangulation is reduced, and the accuracy of calculating the coordinates of the external reference intersection (6) improves. On the other hand, if the base length of the retaining part (16) is too long, it becomes difficult to miniaturize the wearable device (34). This length is used as the basis for calculating the three-dimensional coordinates of the external reference intersection (6) in the triangulation calculation unit (8a) (e.g., Equation 24), and for estimating the eyeball rotation center (13) in the geometric calculation unit (8c). The measurement accuracy of the holding unit baseline length (16) is required to be within ±0.01 mm and is calibrated during manufacturing using an optical measuring instrument or CMM (coordinate measuring machine).

[0066] The holder coordinate origin (4) is the origin of the three-dimensional coordinate system fixed to the oscillator holder (3), and defines the reference coordinate system for the entire system. The holder coordinate system is defined as a right-handed system with the X-axis as the baseline connecting the left and right oscillation means (2), the Y-axis in front of the user (field of view), and the Z-axis vertically upward. The position of each oscillation means (2), the coordinates of the external reference intersection (6), and the position of the eyeball rotation center (13) are all expressed as relative coordinates with respect to this holder coordinate origin (4). This unified coordinate system ensures consistency in geometric calculations in the triangulation calculation unit (8a) and the sine rule calculation unit (8b), and enables integration with other sensor data via the coordinate system transformation module (8h). The position of the holder coordinate origin (4) is precisely calibrated during the calibration of the eye-tracking sensor (7) and stored in the system storage means (130). This holding unit coordinate system is used as the sole invariant reference coordinate when comparing the measured coordinates (actual intersection coordinates) and theoretical coordinates (virtual intersection coordinates) of the external reference intersection (6) (see paragraph 0059). Furthermore, the coordinates for displaying the user's body coordinates, such as the coordinates of the center of eye rotation, are also given based on this origin. Note that the position of the holding unit coordinate origin (4) differs depending on the shape of the oscillator holding unit (3) and the holding unit baseline length (16) of each system housing.

[0067] The triangulation calculation unit (8a) calculates the three-dimensional coordinates and distance of the external reference intersection (6) using the direction vectors of the beams (5) irradiated from the left and right oscillation means (2) and the baseline length (16) of the holding unit. Specifically, when the left and right beams (5) irradiate at known angles θ_L and θ_R, the distance d from the coordinate origin (4) of the holding unit is obtained by (Equation 24). This triangulation is a classical surveying method that utilizes the relationship between distance and angle from two known points (positions of the left and right oscillation means (2)) to an unknown point (external reference intersection (6)). However, in this invention, high-speed calculations are performed hundreds of times per second to track the dynamically changing line of sight direction. When multiple beams (N≧3) are used and the system becomes an over-determined system that theoretically does not intersect, the triangulation calculation unit (8a) determines the most likely three-dimensional coordinates of the external reference intersection (6) using the least squares method (LSM) or an iterative optimization method that minimizes the sum of the squares of the distances between all beams. Furthermore, by calculating inversely the position of the center of rotation of both eyes, the baseline length of the center of rotation of the eyeball can also be calculated. If the left and right line-of-sight vectors of both eyes are set to angles θ_eyeL and θ_eyeR with respect to the baseline length of the center of rotation of the eyeball, the distance to the point of fixation (15) can be obtained with high precision at the intersection of the left and right beams synchronized with the binocular vectors. Based on this distance, it becomes possible to correct and determine the angles of θ_eyeL and θ_eyeR more accurately. The coordinate information (known absolute coordinates) of the external reference intersection (6) from the oscillation means (2) and the observation information (estimated coordinates) of the gaze point (15) from the gaze tracking sensor (7) become mutually available via this distance information, which further improves the accuracy of the gaze tracking system.

[0068] The sine rule calculation unit (8b) works in cooperation with the triangulation calculation unit (8a) to calculate the distance to the external reference intersection (6) from the base length (16) and irradiation angle of the holding unit, particularly in the triangle formed by the left and right beams (5). According to the sine rule, if the base length L and the left and right irradiation angles θ_L and θ_R are known, the distance from each oscillation means (2) to the external reference intersection (6) is uniquely determined. This calculation is combined with coordinate transformation in the geometric calculation unit (8c) and output as the three-dimensional position (X, YZ) of the external reference intersection (6) in the holding unit coordinate system. To improve robustness against angle measurement errors, the sine rule calculation unit (8b) performs an averaging process (equation 11) across multiple frames. This distance calculation result is cross-validated with the coordinate calculation result from the triangulation calculation unit (8a), and if it does not satisfy the geometric constraints of the sine rule, it is detected as an outlier.

[0069] The geometric calculation unit (8c) verifies the geometric consistency of the coordinate data obtained from the triangulation calculation unit (8a) and the sine rule calculation unit (8b). Specifically, it evaluates the difference (equation 14) Δ=GB between the point where the direction vectors of the left and right beams (5) should theoretically intersect and the actually calculated external reference intersection (6), and if this difference exceeds a predetermined threshold, it notifies the environmental factor error detection unit (127). In addition, the geometric constraint determination unit (8k) verifies whether the distance from the eyeball rotation center (13) to the external reference intersection (6) is within a physiologically reasonable range (e.g., 10 mm to 100 mm) and eliminates outliers. In parallel with this verification process, the geometric calculation unit (8c) applies statistical filtering such as a Kalman filter or particle filter and fuses the time-series data of the pupil center position, corneal reflection point, and external reference intersection to suppress instantaneous measurement noise and ensure the stability and accuracy of the dynamically updated eyeball rotation center (13). This suppresses false detections caused by measurement noise and temporary occlusion, enabling stable eye-tracking. (See Figure 6)

[0070] The display unit (50) is a display device that presents visual information to the user and takes the form of either AR glasses, a VR headset, or a projection type. The display unit (50) receives eye-tracking results, the coordinates of the point of fixation (15), and eye-tracking trajectory data from the system control means (44), and this information is overloaded. The system displays a ray or highlights the target area. Furthermore, during the calibration process using a calibration jig (120), the system guides the position of the jig reference point (121) to assist in guiding the user's gaze. Additionally, based on the gaze tracking results, the system reduces the overall computational load and achieves low latency by performing foveated rendering, which renders the gaze area at high resolution and intentionally reduces the resolution of the surrounding area. The display unit (50) integrates with the three-dimensional shape data of the environment acquired from the depth sensor (7b) and visualizes the output of the SLAM processing unit (8p) that places virtual objects in real space. The display unit can also display three-dimensional coordinates including the true altitude of the gaze point, distance, information from the internet, acquired images, text, etc.

[0071] The body-worn device (48) is a wearable device attached to the user's head, wrist, fingers, or torso, and is connected to the eye-tracking system (1) by wireless communication. The body-worn device (48) is equipped with an accelerometer, a gyroscope, and a geomagnetic sensor to detect the user's head posture (Roll-Yaw-pitch) in real time. This posture information is used in the coordinate system transformation module (8h) for the transformation from the holding unit coordinate system to the global coordinate system (Equation 10) P_external=R·P_device+T, enabling accurate tracking of the absolute coordinates of the gaze point (15) even when the user moves their head. This posture information determines the spatial position and posture (6DoF) of the holding member (3a) that holds the eye-tracking sensor (7) and the oscillation means (2), and compensates for the change in the relative positional relationship between the eye-tracking sensor and the external reference intersection in the global coordinate system due to head movement. Furthermore, the smartwatch-type wearable device (48) measures heart rate and skin electrical activity, which are used as biometric information to estimate the user's cognitive load state. In addition, it integrates data from physiological sensors such as electromyography (EMG) sensors and electroencephalography (EEG) sensors to provide a multifaceted sensing platform for estimating the user's (9) intentions, level of concentration, or level of fatigue.

[0072] The smart ring (48a) is a ring-shaped wearable device worn on the user's finger and incorporates a touch sensor, a gesture sensor, and a haptic feedback mechanism. The smart ring (48a) functions as an input device for performing selection, confirmation, and scrolling operations on a gaze point (15) identified by eye tracking. For example, by tapping the smart ring (48a) while gazing at a specific object, the selection of that object is confirmed. This combination of gaze and gesture operation is more intuitive than conventional mice or touch panels and improves work efficiency in hands-free environments. In particular, it implements hierarchical input control in which input from the smart ring (48a) (tap, swipe, or air gesture) is enabled only when data from the eye tracking sensor (7) detects a dwell time (gaze duration) or a cessation of gaze movement (fixation) at the gaze point (15). The input signal from the smart ring (48a) is analyzed by the command input analysis unit (44g) of the system control means (44) and converted into appropriate application control. (See Figure 7)

[0073] The calibration jig (120) is a set of physical or virtual reference points used for the initial calibration and periodic recalibration of the eye-tracking system (1). The calibration jig (120) has multiple jig reference points (121) with known three-dimensional coordinates, and the user is instructed to sequentially gaze at these points. The coordinates of the jig reference points (121) are pre-measured and verified with sub-millimeter accuracy using a laser tracker or coordinate measuring machine (CMM), and registered with high precision as absolute coordinates relative to the holding unit coordinate origin (4). From the correspondence between the line-of-sight vector (14) when the user gazes at each jig reference point (121) and the known coordinates of the jig reference point (121), the calibration parameter δ_calib (Equation 2) is calculated and applied as a correction coefficient to the line-of-sight direction calculation unit (8n). (See Figure 14)

[0074] The fixture reference points (121) are implemented as physical LED markers, printed chart patterns, or virtual markers on an AR display. The physical fixture reference points (121) have their three-dimensional coordinates measured by a depth sensor (7b), allowing for high-precision calibration. The measured coordinates of these physical fixture reference points (121) also function as a second absolute reference point to verify the accuracy of the coordinate calculation of the external reference intersection (6), doubling the metric accuracy of the entire system. On the other hand, virtual fixture reference points (121) can be placed at any position on the display unit (50), enabling a flexible calibration process according to the operating environment. In the calibration process, the user gazes at at least five or more fixture reference points (121), and an optimization problem (Equation 1) is solved to minimize the error between the measured value and the theoretical value of the line-of-sight vector (14) at each point. In the calibration process, the error to be minimized is strictly defined as the angular difference (Angular Error) between the line of sight direction derived from the calculated eyeball rotation center (13) and the true direction vector toward the fixture reference point (121). This calibration corrects for individual differences in eyeball shape, misalignment of the device's position, and the effects of ambient light.

[0075] The eyeball rotation center (13) is the geometric center point when the eyeball (10) rotates, and is defined as the starting point of the line of sight vector (14). In this invention, the eyeball rotation center (13) is not a fixed parameter, but a dynamic parameter that is continuously updated by the eyeball rotation center dynamic update unit (8m). This update is performed by a geometric matching process that uses known coordinates of the external reference intersection (6) as an absolute reference, in addition to physiological measurement data from the eye-tracking sensor (7). Specifically, the three-dimensional coordinates of the eyeball rotation center (13) are estimated by geometric inverse calculation processing from time-series data of pupil position detected by the eye-tracking sensor (7) (Equations 1 and 5). This dynamic update maintains high-precision line of sight tracking even with respect to changes in the user's head posture, displacement of eyeglasses, and minute movements of the device due to prolonged use. The update of the eyeball rotation center (13) is verified for physiological validity by the geometric constraint determination unit (8k), and abnormal estimates are rejected. Physiological validity is constrained by the fact that the distance from the estimated center of ocular rotation (13) to the surface of the eyeball (ocular radius) is within a known physiological range (typically 11 mm to 13 mm), and the distance between the centers of ocular rotation of both eyes (inter-ERC baseline length) is within the physiological IPD range.

[0076] The line of sight vector (14) is a unit direction vector that starts at the eyeball rotation center (13), passes through the pupil center, and extends in the direction of the line of sight. The line of sight direction calculation unit (8n) calculates the line of sight vector (14) using (Equation 2) from the pupil position P_pupil(t) detected by the eye-tracking sensor (7) and the eyeball rotation center C_eye(t) which has been updated with high precision using the external reference intersection (6) by the eyeball rotation center dynamic update unit (8m). A correction coefficient δ_calib obtained in the calibration process using the calibration jig (120) is added to this calculation to correct for the line of sight direction offset due to individual differences. The line of sight vector (14) is calculated independently for the left and right eyes, and the fixation point (15) is determined by integrating the lines of sight of both eyes. To improve the temporal stability of the eye-line vector (14), a Kalman filter (Equation 17) or a low-pass filter (Equation 3) C_eye(t)=(1-α)C_eye(t-1)+αC_meas(t) is applied, and a statistical prediction model that takes into account the momentum of eye movement is also applied to compensate for the calculation delay.

[0077] The point of fixation (15) is defined as the point in three-dimensional space where the left and right line-of-sight vectors (14) are closest or virtually intersect. The triangulation unit (8a) calculates the parameters s and t that minimize the distance between the left eye's line-of-sight vector L_left(s, Equation 12) L_left(s) = C_eye^(L,s) + s·v_gaze^(L,s) and the right eye's line-of-sight vector L_right(t, Equation 13) L_right(t) = C_eye^(R,t) + t·v_gaze^(R,t), starting from the dynamically updated eyeball rotation center (13), using (Equation 7) min||L_left(s)-L_right(t)||. In actual living eyes, it is rare for the left and right line-of-sight vectors (14) to intersect perfectly at a single point, so the midpoint of the nearest point of tangency is adopted as the point of fixation (15) (Equation 22). The depth of the point of focus (15) is estimated from the convergence angle and interpupillary distance using (Equation 8) Z ≈ L_eye / [2tan(α / 2)] and integrated with the measurement value from the depth sensor (7b). This integration of the measurement value from the depth sensor (7b) and the convergence angle estimate is performed in the FusedGazeDepth calculation unit (8j) to correct the Triangulation error and improve the accuracy of the point of focus coordinates in the Z-axis direction. The coordinates of the point of focus (15) are output as relative coordinates with respect to the holding unit coordinate origin (4), or as absolute coordinates in the global coordinate system, and are provided to the application via the system output means (131) (see Figure 6).

[0078] The calibration jig (120) provides a set of reference points for optimizing eye-tracking accuracy during system startup and periodic calibration during use. The jig reference points (121) are markers with known coordinates placed within the user's field of view and are implemented as physical LED light sources, printed patterns, or virtual points on AR displays. During the calibration process, the user sequentially gazes at each jig reference point (121), and the estimated pupil position, corneal reflection position, and line-of-sight vector (14) are recorded. From this data, the eye rotation center calculation unit (8i) solves an optimization problem (Equation 1) to simultaneously estimate the three-dimensional coordinates of the eye rotation center (13) and the correction coefficient δ_calib for the line-of-sight direction. The initial estimate of the eye rotation center (13) obtained in this initial calibration serves as an initial parameter for a dynamic update algorithm based on external reference intersections (6) during subsequent use. Calibration accuracy depends on the number and spatial distribution of jig reference points (121), and a placement of 9 or more points covering the entire field of view is recommended. The calibration results are stored in the system storage means (130) and used as the initial values ​​for the next startup.

[0079] The eye-tracking sensor (7) is a composite sensor unit that measures the user's eye movements non-contact. It consists of an optical system using near-infrared (NIR) illumination, a high-resolution (HD or 4K) and high-frame-rate (240fps or higher) AI camera (7a), a depth sensor (7b), a pupil detection unit (7c), a corneal reflection detection unit (7d), an iris recognition unit (7e), etc. Alternatively, it is a broader concept that includes non-optical methods such as electrophysiological means (electrooculography), electromyography, electroencephalography, and magnetic sensors in addition to these similar means. These subsystems are time-synchronized by a synchronization control unit (8f) and perform high-speed image acquisition at 60 to 240 frames per second. The acquired images are geometrically corrected by a lens distortion correction module (7f) to ensure the accuracy of eye information extraction. The eye-tracking sensor (7) is fixed to the oscillator holder (3), and its relative position from the holder coordinate origin (4) is known, so the conversion between the sensor coordinate system and the holder coordinate system is uniquely determined by the coordinate system conversion module (8h). The eye-tracking sensor (7) uses infrared illumination, making it robust to fluctuations in the visible light environment and enabling stable pupil detection even in dark places or under strong ambient light. Furthermore, the eye-tracking sensor (7) also includes a scleral tracking function that tracks the pattern and blood vessel pattern of the white part of the eye (sclera) in order to estimate the roll angle (rotation) in the direction of gaze.

[0080] The AI ​​camera (7a) is a high-speed image processing camera with a built-in deep learning model that automatically detects eye features such as the user's pupil, iris, corneal reflection, and eyelids. This AI camera (7a) has the capability to capture images at high resolution (HD or 4K) and high frame rates (240fps or higher). The convolutional neural network (CNN) installed in the AI ​​camera (7a) is pre-trained on tens of thousands of eye images, enabling highly accurate feature point detection even in the presence of partial occlusion or reflection noise. For eye feature point detection, a segmentation model such as MaskR-CNN or an object detection model such as YOLO is applied to accurately extract the regions of the pupil, corneal reflection, and iris. The coordinates of the detected feature points are interpolated from pixel level to sub-pixel accuracy, stabilized by time-series filtering, and then transferred to the pupil detection unit (7c) and the corneal reflection detection unit (7d). The AI ​​camera (7a) classifies the user's blinks and types of eye movements (saccades, smooth pursuit, fixation tremors) in real time and notifies the system control unit (44). This information is used to evaluate the reliability of gaze detection and to recognize intentional gaze input commands.

[0081] The depth sensor (7b) measures the three-dimensional shape within the field of view in real time using either the Time-of-Flight method or the stereo vision method. The depth sensor (7b) also incorporates the StructuredLight method. The distance image acquired by the depth sensor (7b) is used in the SLAM processing unit (8p) to generate an environment map and determine whether the point of fixation (15) is on the surface of a real object or a virtual point in space. The depth sensor (7b) also provides an initial estimate of the depth coordinate of the eyeball rotation center (13) by measuring the distance to the surface of the eyeball. This initial estimate accelerates the convergence of the optimization calculation in the eyeball rotation center calculation unit (8i) and shortens the calibration time at system startup. The measurement data from the depth sensor (7b) is used to verify the difference in the Z-axis direction between the actual coordinates of the external reference intersection (6) and the theoretical coordinates of the virtual intersection coordinates (25), providing essential information for self-verification of high-precision geometric criteria. The data from the depth sensor (7b) is corrected for systematic errors in the depth calibration unit (8l) and integrated with the calculation results from the triangulation calculation unit (8a). (See Figures 6 and 9)

[0082] The pupil detection unit (7c) extracts the outline of the pupil from image data acquired from the AI ​​camera (7a) and calculates the pupil center coordinates and pupil diameter by elliptic fitting. The pupil is extracted by thresholding using the brightness difference with the iris or by edge detection, and sub-pixel accurate center coordinates are obtained. Elliptic fitting using the RANSAC (RandomSampleConsensus) algorithm, which has high noise tolerance, is applied to calculate the pupil center coordinates. The corneal reflection detection unit (7d) detects the specular reflection (Purkinje image) of the corneal surface due to infrared illumination and provides auxiliary information to estimate the orientation of the eyeball (10) from its position. In particular, the corneal reflection detection unit (7d) detects at least one of the first Purkinje image (P1) or the fourth Purkinje image (P4) and provides foundational information for a head movement-robust gaze tracking algorithm (Pupil-CornealReflection-pCR). The position of the corneal reflection is recorded as its relative position to the pupil center, improving the accuracy of the line-of-sight vector (14) calculation. The data from the pupil detection unit (7c) and the corneal reflection detection unit (7d) are time-synchronized by the synchronization control unit (8f) and processed independently for the left and right eyes.

[0083] The iris recognition unit (7e) extracts feature quantities from the iris pattern and performs personal authentication of the user. Each individual's iris has a unique texture pattern, making it suitable for high-precision biometric authentication. The iris recognition unit (7e) applies feature extraction using a Gabor filter bank or deep learning to the iris image acquired from the AI ​​camera (7a) and compares it with a registered template. The iris pattern is a bio-specific feature, which ensures that calibration parameters, eye rotation center (ERC) parameters, and IPD (interpupillary distance) are reliably linked to each user, preventing the application of incorrect calibration values. The authentication results are transmitted to the security management unit (44l) of the system control unit (44) and used for automatic loading of calibration parameters for each user, application of personal settings, and management of access rights. The iris recognition unit (7e) continuously performs authentication in the background during eye-tracking operation, and in addition to authenticating the user (9), it detects user changes and unauthorized access, and also detects minute shifts in the wearable device (34) (such as IPD shift) in real time, triggering automatic fine-tuning of eye-tracking parameters.

[0084] As shown in Figure 6, the system control unit (44) is a central control unit that oversees the operation of the entire eye-tracking system (1). It mediates the data flow between the information analysis unit (8), the oscillation unit (2), the eye-tracking sensor (7), the display unit (50), the system storage unit (130), and the system output unit (131), and controls the operating timing of each subsystem. The system control unit (44) synchronizes the entire cycle of eye-tracking data acquisition (sensor 7), data analysis (information analysis unit 8), and beam irradiation direction adjustment (oscillation unit 2) at a sub-millisecond level, performing strict real-time control to keep the total system latency (Total System Latency) below 10ms. Part of the control logic is implemented on an FPGA (Field-programmable Gate Array) or a dedicated ASIC for low-latency processing, and hierarchically distributed processing of computation tasks. The system control unit (44) operates on a real-time operating system and is responsible for the periodic execution of eye-tracking processing, management of the calibration process, error detection and recovery, and communication with external applications. Detailed control functions will be explained in paragraph 0104 and beyond, but this paragraph will clarify that the system control means (44) is responsible for issuing processing start commands to each calculation module of the information analysis means (8) and for collecting and integrating the calculation results. Some of the initial gaze tracking algorithms executed in the information analysis means (8) and the system control means (44) may be constructed based on existing known or open-source programs, libraries, or algorithms, taking development efficiency into consideration. Specifically, these programs are used for image extraction of the pupil center (11) and corneal reflection point (12), or for initial relative gaze direction calculation (Section 1). However, the core of the present invention lies in achieving high accuracy (visual angle error within 0.7 degrees), which was difficult to achieve with conventional techniques, by applying geometric inverse calculation (Claim 1) and dynamic optimization (Claim 16, 17) using a unique external reference intersection (6) to the output of these initial programs. Even if publicly known or open-source programs are used, the unique technical components of the present invention (utilization of external reference intersection (6), geometric inverse calculation of the eyeball rotation center (13), application of physiological constraints, foveated rendering control unit (133), and VR sickness reduction control unit) are considered independent and original works not subject to the open-source license. This ensures that the implementation of the present invention does not create an obligation to disclose the source code of the unique technical components. Therefore, it is considered that the patenting and implementation of the present invention will not be affected by the disclosure requirements or other usage restrictions imposed on open-source programs.

[0085] As shown in Figure 6, the information analysis means (8) is a hierarchical calculation system that processes raw data acquired from the eye-tracking sensor (7) and calculates the three-dimensional coordinates of the point of fixation (15), the line of sight vector (14), and the center of eye rotation (13). The information analysis means (8) is a unique hybrid calculation architecture of this invention that combines geometric calculation based on absolute coordinates using an external reference intersection (6) (triangulation) with estimation of a physiological model based on eye information (center of eye rotation), and continuously verifies the consistency between the two using a real virtual intersection coordinate matching unit (8s) and a geometric constraint determination unit (8k). The information analysis means (8) consists of a triangulation calculation unit (8a), a sine rule calculation unit (8b), a geometric calculation unit (8c), a difference analysis unit (8d), an inverse calculation processing unit (8e), a synchronization control unit (8f), an error correction integration unit (8g), a coordinate system transformation module (8h), an eyeball rotation center calculation unit (8i), a corneal curvature center calculation unit (8j), a geometric constraint condition determination unit (8k), a depth calibration unit (8l), an eyeball rotation center dynamic update unit (8m), a line of sight direction calculation unit (8n), a SLAM processing unit (8p), a real-virtual intersection coordinate matching unit (8s), and an environmental factor error detection unit (127). This system provides a self-verification function that diagnoses not only the basic parameters of eye-tracking (ERC, line of sight vector) but also the soundness of the system components themselves (deformation of holding members, drift of beam irradiation direction) in real time. These computing modules operate in parallel via pipeline processing, enabling high-speed eye-tracking at 60 to 240 frames per second. (See Figure 6)

[0086] The triangulation calculation unit (8a) calculates the three-dimensional coordinates of the external reference intersection (6) using the direction vectors of the beams (5) irradiated from the left and right oscillation means (2) and the baseline length (16) of the holding unit. The calculated coordinates of the external reference intersection (6) are the absolute geometric reference point in the holding unit coordinate system, and their accuracy forms the basis for all geometric calculations of the system. The distance to the external reference intersection (6) is determined from the irradiation angles θ_L and θ_R obtained from the oscillator control unit (2c) and the known baseline vector at the origin (4) of the holding unit coordinate system using (Equation 24). The triangulation calculation unit (8a) implements the nearest point calculation algorithms (Equation 21) d_min = ||O_L - O_R||·|d_L × d_R| / ||d_L × d_R|| and (Equation 26) d_min = ||(O_L - O_R)·(d_L × d_R)|| / ||d_L × d_R|| to handle cases where the left and right beams (5) do not intersect perfectly (i.e., when they are in a skew position). For the nearest point calculation in the case of skew positions (skewlines), a nonlinear optimization algorithm such as the Levenberg-Marquardt method is applied to determine the optimal coordinates in space where the sum of the squares of the distances to all beams is minimized. The calculated coordinates of the external reference intersection (6) are transferred to the geometric calculation unit (8c), where their consistency with the line-of-sight vector (14) is verified. The triangulation calculation unit (8a) reduces measurement noise by averaging multiple frames (Equation 11) F_avg = (1 / M)ΣF_k, thereby achieving stable coordinate output. (See Figure 4)

[0087] The sine rule calculation unit (8b) works in cooperation with the triangulation calculation unit (8a) to calculate the length of each side of a triangle formed by the holding unit baseline length (16) and the left and right illumination angles by applying the sine rule. Specifically, in triangle ABC where the positions of the left and right oscillation means (2) are A and B, and the external reference intersection (6) is C, if the baseline length AB=L, angles ∠CAB=θ_L, and ∠CBA=θ_R are known, the lengths of sides AC and BC can be determined by the sine rule. The sine rule calculation unit (8b) provides an independent mathematical verification path for the calculation of the external reference intersection coordinates by the triangulation calculation unit (8a), eliminating errors caused by measurement noise and rounding errors in numerical calculations through mutual verification. This mutual verification improves the reliability of the calculation of the external reference intersection (6) coordinates, and as a result, the stability of the dynamic update process of the eyeball rotation center (13) is guaranteed. This calculation provides redundancy for distance calculation in the triangulation calculation unit (8a) and is used for cross-verification of calculation results. The output of the sine rule calculation unit (8b) is compared with the result of the triangulation calculation unit (8a) in the difference analysis unit (8d), and if the difference between the two exceeds a threshold, it is detected as a measurement error.

[0088] The geometric calculation unit (8c) verifies the consistency of multiple geometric data obtained from the triangulation calculation unit (8a), the sine rule calculation unit (8b), and the gaze direction calculation unit (8n), and calculates an integrated three-dimensional coordinate. The main role of the geometric calculation unit (8c) is to verify whether the gaze vector (14) obtained from the gaze tracking sensor (7) accurately passes through (or approaches) a known geometric criterion called the external reference intersection (6), and the result of this verification forms the basis for evaluating the error of the dynamic update of the eyeball rotation center (13) (Claim 1). Specifically, it verifies whether the coordinates of the external reference intersection (6), the coordinates of the eyeball rotation center (13), and the direction of the gaze vector (14) are geometrically consistent. The external reference intersection (6) should lie on the extension of the gaze vector (14) starting from the eyeball rotation center (13), and the amount of deviation from this condition (Equation 14) is calculated. The geometric calculation unit (8c) performs coordinate transformations between the holding unit coordinate system, the sensor coordinate system, and the global coordinate system via the coordinate system transformation module (8h). It also has the function of applying a transformation matrix (homography matrix or affine transformation matrix) to compensate for minute manufacturing errors and mounting misalignments of the holding member (3a), and processes the data obtained from each subsystem in a unified coordinate system.

[0089] The difference analysis unit (8d) compares results obtained from multiple independent calculation paths to detect measurement errors and calculation anomalies. The difference analysis unit (8d) functions as a central diagnostic module that evaluates all geometric and temporal differences generated within the system, isolating and identifying the effects of estimation errors in the eyeball model, drift in the beam irradiation angle, mechanical deformation of the holding member, or environmental noise. For example, it evaluates the difference (Equations 27, 28) between the coordinates of the external reference intersection (6) calculated by the triangulation calculation unit (8a) and a point on the extension of the line of sight vector (14) obtained from the line of sight direction calculation unit (8n). If this difference exceeds a predetermined threshold, the environmental factor error detection unit (127) is notified, and a cause analysis is performed. The difference analysis unit (8d) continuously monitors the discrepancy between the actual intersection coordinates (24) and the virtual intersection coordinates (25). If the discrepancy is large, it estimates the fluctuation in the holding unit baseline length (16) caused by temperature, humidity, or aging through the environmental factor error detection unit (127), provides a correction value to the information analysis means (8) to compensate for this fluctuation, and commands the oscillator control unit (2c) to readjust the irradiation angle. It also evaluates the coordinate difference of the gaze points (15) calculated independently from the left and right eyes to verify the consistency of binocular vision.

[0090] The reverse calculation processing unit (8e) performs a process to calculate the three-dimensional coordinates of the eyeball rotation center (13) from the known coordinates of the external reference intersection (6) and the pupil position detected by the eye-tracking sensor (7). This reverse calculation process is formulated as an optimization problem to find a point that satisfies the geometric constraint (the distance from the eyeball rotation center to the pupil is a constant value r) (Equations 1 and 5). The external reference intersection (6) is a known point in space that the line of sight vector must pass through, and the constraint that the line of sight vector must pass through this point significantly improves the spatial estimation accuracy of the eyeball rotation center (13) compared to conventional methods based only on pupil position and corneal reflection (PCR). When estimating the eyeball rotation center (13) from multiple line of sight direction data, the reverse calculation processing unit (8e) uses the least squares method or an iterative optimization algorithm (Equation 6). This reverse calculation process automatically estimates the individual user's eye shape parameters during the calibration process using the calibration jig (120), optimizing eye-tracking accuracy. The detailed reverse calculation algorithm is described in paragraphs 0120 to 0122.

[0091] The synchronization control unit (8f) synchronizes the time of each subsystem of the eye-tracking sensor (7), the oscillation means (2), and the calculation modules of the information analysis means (8). In the eye-tracking system (1), image acquisition by the left and right AI cameras (7a), distance measurement by the depth sensor (7b), and beam (5) irradiation by the oscillation means (2) must be performed at the same time. The synchronization control unit (8f) synchronizes the data acquisition timing of each subsystem with an accuracy of less than 1 millisecond using a hardware trigger signal or software timestamp. In particular, by strictly synchronizing the start time of beam irradiation by the oscillation means (2) and the exposure start time of image acquisition of pupil and corneal reflection by the eye-tracking sensor (7) at the microsecond level, errors caused by the time difference between beam irradiation and eye information acquisition are eliminated. Furthermore, when integrating data from multiple sensors operating at different frame rates (e.g., camera 120fps, depth sensor 60fps), time interpolation is used to convert them into data at the same time. The operation of the synchronization control unit (8f) enables accurate eye-tracking even with high-speed eye movements (saccades).

[0092] The error correction integration unit (8g) comprehensively corrects various errors detected throughout the entire system. Error factors include sensor noise, calibration errors, ambient light fluctuations, thermal expansion of the device, and displacement of the mounting position due to prolonged use. The error correction integration unit (8g) integrates the depth measurement error provided by the depth calibration unit (8l), the environmental error detection unit (127), and the geometric consistency evaluation provided by the geometric constraint determination unit (8k) to calculate a comprehensive correction vector. This correction vector has a two-layer structure applied to the irradiation angle correction (θfeedback) of the oscillation means (2) and the coordinate correction (ΔCeye) of the eyeball rotation center (13), minimizing errors on both the input and output sides of eye-tracking. This correction vector is applied to the calculation of the eye-line vector (14) in the eye-line direction calculation unit (8n) and the coordinate calculation of the external reference intersection (6) in the triangulation calculation unit (8a), improving the accuracy of the entire system. The error correction integration unit (8g) calculates the optimal estimate from the time series data using (Δθfeedback (Equation 17)X^k+1=AX^k-1+Bu^k).

[0093] The coordinate system transformation module (8h) manages and performs inter-coordinate transformations between multiple coordinate systems used within the eye-tracking system (1). The main coordinate systems are the holder coordinate system (based on the holder coordinate origin (4)), the sensor coordinate system (based on the mounting position of each sensor), the eyeball coordinate system (based on the eyeball rotation center (13)), and the global coordinate system (environment-fixed absolute coordinates). The coordinate system transformation module (8h) applies transformation matrices between each coordinate system using (Equation 10) to represent data acquired by one sensor in another coordinate system. In particular, the global coordinate system is linked to a real-world environment map generated by the SLAM processing unit (8p), ensuring that the coordinates of the gaze point (15) are accurately mapped as a physical position in the real world. Specifically, head pose information (Roll-Yaw-pitch) acquired from the body-worn device (48) is used to perform a rotational transformation from the holder coordinate system to the global coordinate system, accurately tracking the absolute coordinates of the gaze point (15) even when the user moves their head.

[0094] The eyeball rotation center calculation unit (8i) estimates the three-dimensional coordinates of the eyeball rotation center (13) from pupil position data of multiple frames acquired by the gaze tracking sensor (7). The eyeball rotation center (13) is a geometric center that behaves as a fixed point when the eyeball (10) rotates, and utilizes the constraint condition (Equation 5) ||P_i-C||=r, which states that the distance from this point to the pupil center remains approximately constant. The eyeball rotation center calculation unit (8i) formulates the eyeball rotation center C that best satisfies the constraint condition from the pupil positions P_i (i=1-2,...,N) in multiple gaze directions as an optimization problem (Equation 1), and solves it using the iterative least squares method or gradient descent method. The solution to this optimization problem is given as an initial solution using a dataset (pupil position, corneal reflection position) obtained when fixating on a reference point (121) of a calibration jig (120), establishing a baseline for the accuracy and convergence speed of subsequent dynamic updates. The calculated eye rotation center (13) is verified for physiological validity by a geometric constraint determination unit (8k), and if deemed valid, is stored in the system storage means (130).

[0095] The corneal curvature center calculation unit (8j) estimates the corneal curvature center from the position of the corneal reflection image (Purkinje image) detected by the corneal reflection detection unit (7d). The cornea is generally spherical in shape, and its radius of curvature is approximately 7 to 8 mm, although there are individual differences. The corneal curvature center calculation unit (8j) geometrically calculates the coordinates of the corneal curvature center from the positional relationship of corneal reflection images from multiple illumination positions. A higher-order optical model that takes into account the asphericity of the cornea is applied to this calculation to estimate the curvature center more accurately. This information on the corneal curvature center is used as auxiliary information to improve the estimation accuracy of the eyeball rotation center (13) in the eyeball rotation center calculation unit (8i). Furthermore, since the distance between the corneal curvature center and the eyeball rotation center (13) is anatomically almost constant, using this relationship as a constraint improves the convergence speed of the estimation of the eyeball rotation center (13). The magnitude of the error ε is converted to the visual angle error θ_error and evaluated: θ_error≒arctan(||ε|| / D)≒||ε|| / D(Radin) (Formula 50) Here, D is the distance to the external reference intersection (6). By small-angle approximation, if the distance D = 500 mm and the spatial error ||ε|| = 0.5 mm, the visual angle error is approximately 0.057 degrees.

[0096] The geometric constraint determination unit (8k) verifies whether the geometric parameters output from each calculation module of the information analysis means (8) are within a physiologically and physically reasonable range. The constraints to be verified include that the distance from the center of eye rotation (13) to the pupil is within a physiological range (e.g., 10 to 15 mm), that the distance between the left and right centers of eye rotation (13) is consistent with the interpupillary distance (e.g., 55 to 75 mm), that the line of sight vector (14) and the external reference intersection (6) are geometrically consistent, and that the corneal curvature radius is within a physiological range (e.g., 7 to 9 mm). In particular, the determination that "the line of sight vector (14) and the external reference intersection (6) are geometrically consistent" means that the Euclidean distance (spatial error) between the extension of the line of sight vector (14) and the coordinates of the external reference intersection (6) is less than or equal to a predetermined threshold (e.g., 1 mm), or that the angle between the two is less than or equal to a threshold (e.g., 0, 1 degree). Data that violates these constraints is detected as a measurement error or computational anomaly, and the data in the corresponding frame is discarded. The geometric constraint determination unit (8k) also uses statistical outlier detection methods to exclude outliers that change abruptly in the time series data.

[0097] The physiological validity determination in the geometric constraint determination unit (8k) verifies constraints based on the anatomical structure of the eyeball (10). Specifically, it verifies that the diameter of the eyeball (10) is approximately 20 to 25 mm, that the center of rotation of the eyeball (13) is located posteriorly to the eyeball, that the distance from the corneal apex to the center of rotation of the eyeball (13) is approximately 12 to 16 mm, and that the pupil diameter varies within a range of approximately 2 to 8 mm depending on the light environment. By verifying the validity of these physiological parameters, abnormal values ​​due to measurement noise and temporary occlusion are eliminated, and the robustness of the system is enhanced. In particular, when the center of rotation of the eyeball (13) is updated by the eyeball rotation center dynamic update unit (8m), if the updated value does not satisfy physiological validity, the update is rejected and the previous value is maintained. This physiological validity determination functions as a final safety valve to prevent candidate values ​​of the center of rotation of the eyeball (13) obtained by geometric matching using the external reference intersection (6) from deviating from the actual living eyeball model. This conservative update strategy ensures long-term stable operation.

[0098] The depth calibration unit (8l) corrects systematic errors in distance data measured by the depth sensor (7b). Since the depth sensor (7b) has measurement errors that depend on the measurement distance, surface reflectivity, and incident angle, an error model is constructed through pre-calibration. The depth calibration unit (8l) applies this error model to correct the raw distance measurements and outputs highly accurate distance data. It also corrects the shift in measurement position caused by the parallax of the optical systems of the depth sensor (7b) and the eye-tracking sensor (7) via the coordinate system transformation module (8h). The depth calibration unit (8l) compensates in real time for the shift in the measurement coordinate system of the depth sensor (7b), especially when the wearable device (34) moves slightly relative to the head, based on posture information from the body-worn device (48). The distance data corrected by the depth calibration unit (8l) is integrated with the calculation of the coordinates of the external reference intersection (6) in the triangulation calculation unit (8a), contributing to improved eye-tracking accuracy. The depth calibration unit (8l) also monitors the characteristic changes of the depth sensor (7b) due to temperature changes and dynamically adjusts the correction parameters as needed.

[0099] The eye rotation center dynamic update unit (8m) continuously updates the coordinates of the eye rotation center (13) during eye-tracking operation to accommodate shifts in mounting position and changes in head posture during prolonged use. The eye rotation center dynamic update unit (8m) compares the latest candidate eye rotation center value calculated by the eye rotation center calculation unit (8i) with the value of the eye rotation center (13) currently held in the system memory means (130), and updates it stepwise using (Equations 31 and 32). The main trigger for this dynamic update is the increase in spatial error between the external reference intersection (6) detected by the actual virtual intersection coordinate matching unit (8s) and the assumed gaze point (25), and the coordinates of the eye rotation center (13) are corrected by a small amount ΔCeye to minimize this error. This stepwise update enables adaptive operation that follows true changes while suppressing abrupt fluctuations due to measurement noise. The eyeball rotation center dynamic update unit (8m) works in cooperation with the geometric constraint determination unit (8k) to perform updates only if the candidate update values ​​satisfy physiological validity. Equations 15 and 16 are used to determine the convergence of the updates, and after convergence, the update frequency is reduced to alleviate the computational load.

[0100] The gaze direction calculation unit (8n) calculates the gaze vector (14) using (Equation 2) from the eye rotation center (13), which has been updated by the eye rotation center dynamic update unit (8m) to maximize geometric consistency with the external reference intersection (6), and the pupil position detected by the pupil detection unit (7c). The calculation of the gaze vector (14) applies an individual correction coefficient δ_calib obtained in a calibration process using a calibration jig (120), correcting for individual differences in the gaze direction. This correction coefficient (δcalib) models the angular offset between the vector from the eye rotation center (13) to the pupil center and the true gaze direction. The gaze direction calculation unit (8n) independently calculates the gaze vector (14) for the left and right eyes and transfers the gaze vectors for both eyes to the triangulation calculation unit (8a). To improve the temporal stability of the line-of-sight vector (14), a low-pass filter (Equation 3) or a Kalman filter (Equation 17) is applied to remove high-frequency noise and integrate it with the predicted data from the prediction control unit (44j) (Predictive Filtering). The output of the line-of-sight direction calculation unit (8n) is used for calculating the gaze point (15), inputting the gaze point prediction algorithm (22), and generating irradiation direction commands to the oscillator control unit (2c).

[0101] The SLAM processing unit (8p) generates a three-dimensional map of the environment using image data acquired from the depth sensor (7b) and the AI ​​camera (7a), and simultaneously estimates the spatial position of the eye-tracking system (1). Through SLAM (Simultaneous Localization and Mapping) processing, the relative positional relationship between fixed environmental feature points and the eye-tracking system (1) is continuously updated as the user moves through the environment. The SLAM processing employs the Visual-Inertial Odometry (VIO) algorithm, which fuses visual information with inertial information from the body-worn device (48). The environment map generated by the SLAM processing unit (8p) is used to identify which object surface in the environment the gaze point (15) is located on, and also functions as a reference for the placement of virtual objects in AR display. In particular, by comparing the depth information of the gaze point (15) with the surface depth information of the SLAM map, semantic contextual information (gaze target ID, category, etc.) indicating which object the user is gazing at is generated and provided to the application linking unit (44f). The SLAM processing unit (8p) performs visual feature point tracking, loop closure detection, and map optimization to suppress cumulative errors during long-term operation. (See Figure 6)

[0102] The actual virtual intersection coordinate matching unit (8s) detects the discrepancy between the actual intersection coordinates (24) and the virtual intersection coordinates (25) at the external reference intersection (6) and performs a process to isolate the cause. The actual intersection coordinates (24) are the point where the left and right beams (5) physically intersect, and the virtual intersection coordinates (25) are, The central axis of the beam theoretically intersects, calculated from the known position and irradiation angle of the oscillation means (2) that forms the external reference intersection.This is the point. The amount of deviation between the two (Equation 27) is due to one of the following: eye-tracking error, beam irradiation error, or estimation error of the eyeball rotation center (13). The actual virtual intersection coordinate matching unit (8s) performs a geometric error analysis that decomposes the deviation vector based on the sensitivity matrix (Jacobian matrix) of the eye-tracking vector in order to separate these error factors, and distinguishes between systematic errors and accidental errors. If a systematic error is detected, it commands the eyeball rotation center dynamic update unit (8m) to re-estimate the eyeball rotation center (13), and if it is an accidental error, it discards the data for the corresponding frame. This error separation function separates the mechanical error of the beam irradiation mechanism from the physiological model error of eye tracking, making it possible to apply correction means optimized for each error factor (feedback control unit (44k) or eyeball rotation center dynamic update unit (8m)), and the present invention Conventional of Eye Tracking By differentiating itself from other technologies, it establishes a unique technological advantage. (See Figure 6)

[0103] The environmental factor error detection unit (127) detects errors caused by external factors such as ambient light fluctuations, temperature changes, device vibrations, and external electromagnetic interference, and takes appropriate countermeasures. The environmental factor error detection unit (127) utilizes data from temperature sensors, humidity sensors, and acceleration sensors built into the wearable device (34). The environmental factor error detection unit (127) receives anomaly detection signals from the difference analysis unit (8d), geometric constraint determination unit (8k), and depth calibration unit (8l), and determines whether the cause of the error is an environmental factor or an internal system factor. If the ambient light changes rapidly, it issues a command to adjust the exposure time of the AI ​​camera (7a) and the intensity of infrared illumination. If a temperature change is detected, it estimates the amount of minute geometric deformation of the holding unit coordinate system based on the coefficient of linear expansion of the members constituting the holding unit baseline length (16), and commands the depth calibration unit (8l) to update the temperature correction parameters. The environmental factor error detection unit (127) records the system's operation log in the system storage means (130) and supports preventive maintenance through long-term trend analysis. (See Figure 6)

[0104] The system control means (44) is a central control unit that oversees the operation of the entire eye-tracking system (1), and consists of a master scheduler unit (44a), a data flow management unit (44b), a calibration process control unit (44c), a power management unit (44d), a communication interface unit (44e), an application linkage unit (44f), a command input analysis unit (44g), an eye-tracking trajectory recording unit (44h), a gaze determination unit (44i), a prediction control unit (44j), a feedback control unit (44k), a security management unit (44l), a diagnostic unit (44m), and a log management unit (44n). These control modules are organized hierarchically, with the master scheduler unit (44a) performing the highest-level arbitration function and determining the operation timing and priority of each subsystem. The system control means (44) is implemented on a heterogeneous computing platform incorporating an FPGA or a dedicated ASIC to ensure real-time performance, and separates and executes a low-latency (<5ms) critical path from high-throughput background processing. The system control means (44) operates on a real-time operating system and guarantees deterministic response time.

[0105] The master scheduler unit (44a) controls the scheduling and execution timing of processing tasks for all subsystems of the eye-tracking system (1). The master scheduler unit (44a) uses the data acquisition period (e.g., 120Hz) of the eye-tracking sensor (7) as a reference clock and performs time-division multiplexing of the control of each arithmetic module of the information analysis means (8), the oscillation means (2), data writing to the system storage means (130), and data transmission from the system output means (131). Each task is assigned a priority, with high priority given to processes requiring real-time performance, such as calculating the gaze vector (14) and determining the gaze point (15), and low priority given to background processes such as logging and statistical processing. This scheduling is optimized to minimize the total delay between the output of the prediction control unit (44j) and the irradiation angle correction of the oscillation means (2) (output of the feedback control unit (44k)). When the processing load is high, the master scheduler unit (44a) temporarily postpones low-priority tasks to maintain real-time performance.

[0106] The data flow management unit (44b) manages the flow of data generated and transmitted within the eye-tracking system (1) and mediates data transfers between subsystems. The data flow management unit (44b) controls the paths of raw data acquired from the eye-tracking sensor (7), calculation results from the information analysis means (8), historical data stored in the system storage means (130), and data transmitted externally from the system output means (131). The data flow management unit (44b) performs data queuing, buffering, and priority control to prevent data loss and delays. In particular, to efficiently process the rapidly generated eye-tracking data (120 to 240 frames per second), a ring buffer structure is adopted to constantly hold the latest N frames of data. For raw image data, the data flow management unit (44b) separates the area required for real-time processing from the area required for post-processing and diagnosis, and applies lossless compression to post-processed data to save storage capacity. The data flow management unit (44b) monitors the capacity of the system storage means (130) and compresses or archives old data as needed.

[0107] The calibration process control unit (44c) manages the initial calibration, periodic recalibration, and dynamic calibration of the eye-tracking system (1) during use. When the system starts up, the calibration process control unit (44c) reads calibration parameters corresponding to the user identified by the iris recognition unit (7e) from the system storage means (130) and sets them in the information analysis means (8). For first-time users, it starts a calibration process using a calibration jig (120) and instructs the user via the display unit (50) to sequentially gaze at the jig reference points (121). The calibration process control unit (44c) stores individual parameters such as the eye rotation center (13) calculated by the eye rotation center calculation unit (8i), the gaze direction correction coefficient δ_calib, and the interpupillary distance in the system storage means (130). If the amount of deviation detected by the actual virtual intersection coordinate matching unit (8s) during use continuously exceeds the set threshold, or if removal and reattachment of the wearable device (34) is detected, the calibration process control unit (44c) automatically performs a simplified recalibration and updates the parameters.

[0108] The power management unit (44d) monitors the overall power consumption of the eye-tracking system (1) and performs power control to maximize battery life. The power management unit (44d) dynamically switches the operating mode of each subsystem according to the user's eye activity. For example, if the user is fixating for a long time, it reduces power consumption by decreasing the irradiation frequency of the oscillating means (2) and lowering the frame rate of the eye-tracking sensor (7). Conversely, if high-speed eye movements (saccades) are detected, it automatically switches to high-frame-rate operation. Based on biometric information such as heart rate and skin electrical activity provided by the body-worn device (48), the power management unit (44d) estimates that the user's (9) cognitive load is low or idle and switches the system to a low-power mode. The power management unit (44d) monitors the battery level and, if the level drops, temporarily stops non-essential functions (SLAM processing, high-precision calibration, etc.) to maintain basic eye-tracking functionality. The control of the power management unit (44d) improves continuous operating time by more than 30% compared to conventional technology.

[0109] The communication interface unit (44e) manages communication between the eye-tracking system (1) and external devices. The communication interface unit (44e) supports multiple communication protocols such as Bluetooth, Wi-Fi, USB, and wired Ethernet, and performs data exchange with body-worn devices (48), smart rings (48a), external computers, and cloud servers. In particular, dedicated wireless channels such as ultra-low latency, wideband UWB (UltraWideband) or WiGig (60GHz band) are preferentially allocated for real-time streaming of eye-tracking data (point of gaze coordinates, gaze vector). The communication interface unit (44e) supports real-time streaming of eye-tracking data, remote updates of system settings, and over-the-air (OTA) software updates. When transmitting data, it works in cooperation with the security management unit (44l) to encrypt the data and protect privacy. The communication interface unit (44e) monitors the quality of the network connection, and if the connection is unstable, it temporarily stores the data in the system storage means (130) and retransmits it after the connection is restored.

[0110] The application integration unit (44f) implements an application programming interface (API) to provide the functions of the eye-tracking system (1) to external applications. The application integration unit (44f) outputs the coordinates of the gaze point (15), the direction of the gaze vector (14), the gaze judgment result, and the type of eye movement (saccade, fixation, tracking, etc.) in a standardized data format. The output data format uses the industry standard OpenXR or a custom JSON-protobuf format, and a confidence score is added to the gaze data. External applications can acquire gaze tracking data in real time via this API and implement application functions such as gaze control interfaces, gaze analysis, attention state estimation, and advertising effectiveness measurement. The application integration unit (44f) enables bidirectional cooperation by receiving feedback from external applications (e.g., boundary coordinates of the gaze target object, task context information) and using this to improve the gaze judgment accuracy of the information analysis means (8). The Application Integration Unit (44f) manages simultaneous access from multiple applications, ensuring data consistency and exclusive control. It also sets access permissions for each application and works in conjunction with the Security Management Unit (44l) to prevent unauthorized access.

[0111] The command input analysis unit (44g) integrates eye tracking with other input modalities (gestures, voice, touch, button operation, etc.) to analyze the user's intent. The command input analysis unit (44g) receives tap operations from the smart ring (48a), gesture input from the body-worn device (48), and voice commands from the voice recognition system, and interprets them as an integrated command in combination with the current gaze point (15). For example, if the user double-tap the smart ring (48a) while gazing at a specific object, this is interpreted as a command to select and execute that object. This analysis applies a multimodal fusion model of "gaze-and-gesture" which is conditional on the time (dwell time) that the gaze point (15) occupies over a specific GUI element (e.g., button, icon). To prevent unintended errors, the command input analysis unit (44g) performs gaze time threshold determination and gesture reliability evaluation. The analyzed command is sent to an external application via the application integration unit (44f).

[0112] The gaze trajectory recording unit (44h) records time-series data such as gaze vector (14), point of fixation (15), type of eye movement, and fixation time in the system storage means (130). The gaze trajectory recording unit (44h) assigns a timestamp to each data and saves it in a format that allows for later analysis and playback. The recorded data includes the three-dimensional coordinates of the point of fixation (15), the ID of the object being fixed on, the start time of fixation, the duration of fixation, the saccade speed, and the number of blinks. Furthermore, for later offline calibration and diagnosis, raw data of the time-series pupil center coordinates and corneal reflection position, as well as image frames from the corresponding AI camera (7a), are recorded in a compressed format. The eye-tracking recording unit (44h) provides a function that allows users to control the start and stop of recording to protect privacy, separates personally identifiable information (PII), and, in cooperation with the security management unit (44l), assigns only an anonymized ID (PseudonymizedID) to eye-tracking data. The recorded eye-tracking data will be used for research purposes such as cognitive load analysis, attention pattern analysis, and interface improvement.

[0113] The gaze detection unit (44i) determines whether the user is gazing at a specific object based on time-series data of the gaze vector (14) and the gaze point (15). The gaze detection unit (44i) evaluates the spatial stability (dispersion is below a threshold) and temporal continuity (duration is above a threshold) of the gaze point (15) to detect fixation. For gaze detection, an algorithm based on a velocity threshold (I-VT: Velocity-based Thresholding) or an algorithm based on a dispersion threshold (I-DT: Dispersion-based Thresholding) is applied. Once fixation is detected, the object being gazed at is identified by matching it with an environmental map generated by the SLAM processing unit (8p). To distinguish between intentional gaze and accidental gaze fixation, the gaze detection unit (44i) comprehensively evaluates auxiliary indicators such as pupil diameter changes, blinking frequency, and the presence of microsaccades. Based on this evaluation, a gaze confidence score (GazeConfidenceScore) is calculated and output to the application integration unit (44f). The gaze judgment result is transmitted to an external application via the application integration unit (44f) and used for object selection, highlighting of the area of ​​interest, etc., in the gaze control interface.

[0114] The prediction control unit (44j) executes the gaze point prediction algorithm (22) to predict the user's next gaze direction. The prediction control unit (44j) predicts the short-term gaze direction (50 to 200 milliseconds ahead) from past gaze trajectory data, current eye movement velocity, object placement in the field of view, and the user's task context. The prediction model uses recurrent neural network (RNN) models such as LSTM (Long Short-Term Memory) or GRU (Gated Recurrent Unit) that have learned the time-series patterns of past gaze trajectories. The predicted gaze direction is transmitted in advance to the oscillator control unit (2c), and the direction of beam (5) illumination is pre-adjusted to the predicted position. This prediction control reduces the delay in gaze tracking during high-speed eye movements (saccades), bringing the total system delay of gaze tracking close to zero or negative (time cancellation by prediction). The prediction control unit (44j) performs predictions using a Kalman filter (Equation 17) or a machine learning model and continuously evaluates the prediction accuracy. If the prediction error is large, the parameters of the prediction model are adaptively adjusted.

[0115] The feedback control unit (44k) receives discrepancy information between the real intersection coordinates (24) and virtual intersection coordinates (25) provided by the real-virtual intersection coordinate matching unit (8s) as a feedback signal and dynamically corrects the irradiation direction of the oscillating means (2). The feedback control unit (44k) converts the discrepancy vector (Equation 28) δ = V_real - V_virtual into an irradiation angle correction amount (Equation 29) Δθ = J^(-1)·δ using the Jacobian matrix (Equation 34) and transmits it to the oscillator control unit (2c). This control loop is a closed-loop geometric stabilization control aimed at always making the position of the external reference intersection (6) coincide with the point through which the line of sight vector (14) passes. This closed-loop control ensures that the external reference intersection (6) is accurately positioned on the extension of the line of sight vector (14), improving the accuracy of line-of-sight tracking. The feedback control unit (44k) implements PID control or model predictive control to optimize control stability and response speed. The gain parameters of the feedback control are adaptively adjusted according to the operating environment and the user's eye movement characteristics.

[0116] The Security Management Department (44l) is responsible for protecting the privacy of eye-tracking data and preventing unauthorized access to the system. Based on the biometric authentication results from the iris recognition unit (7e), the Security Management Department (44l) verifies the user's identity and grants access to the system only to authenticated users. Because eye-tracking data is sensitive information that reflects an individual's inner state, the Security Management Department (44l) encrypts the data when it is stored in the system storage means (130) and when it is transmitted from the system output means (131). AES-256 or RSA encryption is used for encryption, and a different encryption key is generated for each user. In particular, when performing machine learning analysis using eye-tracking data (e.g., cognitive load estimation), homomorphic encryption or secure computation techniques are applied to enable computation while the data itself remains encrypted, thus protecting privacy during analysis. The Security Management Department (44l) records access logs in cooperation with the Log Management Department (44n) and notifies the system administrator if an attempt at unauthorized access is detected.

[0117] The diagnostic unit (44m) monitors the operating status of each subsystem of the eye-tracking system (1) and performs early detection of abnormalities and failures. The diagnostic unit (44m) periodically inspects the image quality of the eye-tracking sensor (7), the operating accuracy of the oscillation means (2), the consistency of the calculation results of the information analysis means (8), the available capacity of the system memory means (130), the battery level, and the communication connection status. The diagnostic unit (44m) implements a predictive maintenance algorithm that analyzes the noise characteristics obtained from the eye-tracking sensor (7), the drive current of the oscillation means (2), or the minute vibration patterns of the holding member in a time series to predict the lifespan of the main components. If an abnormality is detected, the diagnostic unit (44m) classifies the type and severity of the abnormality, attempts automatic repair in the case of a minor abnormality, and displays a warning to the user on the display unit (50) in the case of a serious abnormality. The diagnostic unit (44m) works in cooperation with the environmental factor error detection unit (127) to distinguish between temporary abnormalities caused by the environment and system failures. The diagnostic results are recorded by the log management unit (44n), and preventive maintenance is achieved through long-term trend analysis.

[0118] The log management unit (44n) records the operation history, error logs, and usage statistics of the eye-tracking system (1) in the system storage means (130). The information recorded by the log management unit (44n) includes system startup and shutdown times, calibration execution history, statistical values ​​of eye-tracking accuracy, error frequency, subsystem operating time, and power consumption trends. This log data is used for long-term performance evaluation of the system, analysis of failure causes, and software improvement. To guarantee the integrity and non-tampering of the recorded log data, the log management unit (44n) has the function of generating an immutable log using blockchain technology or hash chains, ensuring data reliability, especially in application fields where medical and legal evidence is required. To protect privacy, the log management unit (44n) manages detailed eye-trajectory data and operation logs separately, and the operation logs do not include personally identifiable information. The log data is periodically uploaded to a cloud server via the communication interface unit (44e), and aggregated analysis contributes to improving the overall quality of the system.

[0119] Each control module (44a to 44n) of the system central control means (44) works in cooperation with each other to realize the integrated operation of the eye-tracking system (1). A typical operation sequence is as follows: When the system starts up, the master scheduler unit (44a) starts the initialization process, and the security management unit (44l) performs user authentication by the iris recognition unit (7e). After authentication, the calibration process control unit (44c) reads user-specific calibration parameters from the system storage means (130) and sets them in the information analysis means (8). During eye-tracking operation, the data flow management unit (44b) coordinates the acquisition and calculation processing of sensor data, and the prediction control unit (44j) and feedback control unit (44k) optimize the irradiation direction of the oscillation means (2). In this process, the geometric error calculated by the actual virtual intersection coordinate matching unit (8s) is fed back to the oscillation means (2) via the feedback control unit (44k), while also being fed back to the eye rotation center dynamic update unit (8m). This performs dual closed-loop control, which adaptively corrects the physiological model parameter called the eye rotation center (13) based on a geometric absolute standard (external reference intersection (6)). When the gaze determination unit (44i) detects a gaze state, the command input analysis unit (44g) integrates it with the input from the smart ring (48a) and sends a command to an external application via the application linkage unit (44f). This series of coordinated operations realizes highly accurate and responsive gaze tracking.

[0120] Geometric inverse calculation is an inverse problem that estimates the three-dimensional coordinates of the eyeball rotation center (13) from time-series data of pupil position observed by the eye-tracking sensor (7). In normal eye-tracking, the eyeball rotation center (13) is assumed to be known, and the forward problem of calculating the gaze direction from the pupil position is solved. However, in the geometric inverse calculation of the present invention, this relationship is reversed, and the eyeball rotation center (13) is estimated from multiple pupil positions. The geometric inverse calculation of the present invention achieves higher accuracy and robustness than pure pupil-only tracking geometric inverse calculation by incorporating the physical constraint that the gaze vector (14) passes through a geometric absolute reference point in space called an external reference intersection (6). This inverse calculation process makes it possible to automatically adapt to individual differences in eyeball shape for each user, displacement of the wearing position, and positional changes over time. Geometric inverse calculations are performed not only during initial calibration using the calibration jig (120), but also continuously during use by the eyeball rotation center dynamic update unit (8m), enabling long-term high-precision tracking. The mathematical formulation and implementation algorithm of geometric inverse calculations are described in detail in the following paragraphs.

[0121] The estimation of the center of eye rotation (13) in geometric inverse calculation is formulated as a constrained optimization problem. Based on the geometric constraint (Equation 5) that the distance from the center of eye rotation C to the pupil position P_i remains approximately constant at value r (corresponding to the eyeball radius), an objective function (Equation 1) J(C) = Σ[||P_pupil-C||-r]^2 + λ||ModelConstraint(C)||^2 is constructed to minimize the sum of squared distance errors for multiple observation points i=1,2,...,N. The objective function (Equation 1) is defined to maximize geometric consistency by adding the errors of the external reference intersection (6) and assumed gaze point (25) (Equation 27) as penalty terms in addition to the sum of squared distance errors. A regularization term λ||ModelConstraint(C)|| is added to this objective function, constraining that the center of eye rotation (13) falls within an anatomically reasonable position range. In a more advanced formulation, the consistency between the observed gaze direction vector vi and the theoretical gaze direction calculated from the estimated eyeball rotation center C is also evaluated, and this is extended as a weighted least squares problem (Equation 6) J(C) = Σ{w_i·[||P_i-C||-r]^2+(1-w_i)·||v_i-v_i^obs||^2. This formulation makes it possible to handle measurement errors of pupil position and estimation errors of gaze direction in an integrated manner.

[0122] The eyeball rotation center calculation unit (8i) solves the optimization problem formulated by (Equation 1) or (Equation 6) using iterative least squares, the Levenberg-Marquardt method, or gradient descent. Initial values ​​are estimated from the distance to the eyeball surface measured by the depth sensor (7b) and standard eyeball dimensions (approximately 24 mm in diameter). These initial estimates are essential for significantly accelerating the initial convergence of the optimization calculation and reducing calibration time. The iterative calculation continues until the change in the objective function (Equation 15)|J(C_k+1)-J(C_k)|<ε1, or the change in the solution (Equation 16|C_k+1-C_k|<ε2) falls below the convergence thresholds ε1 and ε2. The calculated candidate values ​​for the eye rotation center are verified for physiological validity by the geometric constraint determination unit (8k), and are adopted if deemed valid. The eye rotation center dynamic update unit (8m) updates the current eye rotation center (13) stepwise as new gaze data is accumulated during use using (Equation 31)Ceye^new=Ceye^old+α·δC and (Equation 32)δ_C=β·ΔP-γ·(ΔP·v_gaze)·v_gaze. This update formula gradually updates the value by multiplying it by the difference δCα from the new estimate, thereby suppressing rapid fluctuations due to measurement noise while achieving adaptive operation that follows true changes.

[0123] In the calibration process using the calibration jig (120), the user sequentially gazes at jig reference points (121) with known coordinates, enabling highly accurate estimation of the eyeball rotation center (13). From the pupil position Pi when gazing at each jig reference point (121) and the known coordinate Ri of the jig reference point (121), the direction of the line-of-sight vector v_i = R_i - C)||R_i - C|| (Equation 4) is determined. By solving this relation simultaneously for multiple jig reference points (121), the eyeball rotation center C and eyeball radius r are estimated simultaneously. The calibration process control unit (44c) guides the jig reference points (121) to be distributed across the entire field of view to ensure the numerical stability of the estimation. The calibration jig (120) also has the function of simultaneously forming an external reference intersection (6) during the calibration process, and by using both the known coordinates of the jig reference points (121) and the geometric criteria of the external reference intersection (6), absolute assurance of calibration accuracy is provided. By using at least five or more jig reference points (121), an over-deterministic system is created, enabling robust estimation by the least squares method. The eyeball rotation center (13) and gaze direction correction coefficient δcalib estimated in the calibration process are stored in the system storage means (130) in association with the user ID.

[0124] To improve the accuracy of estimating the center of eye rotation (13), the inverse processing unit (8e) comprehensively processes pupil position data at multiple time points. As the user moves their gaze in various directions, diverse pupil positions Pi (i=1-2,...,N) are observed. When estimating the center of eye rotation C from this multi-viewpoint data, the unit seeks an optimal solution that simultaneously satisfies the constraints (Equation 5) for all data points. As the number of data points N increases, the statistical reliability of the estimation improves, but the computational load also increases. Therefore, the inverse processing unit (8e) uses a sliding window method to use only the most recent M data points (e.g., M=50 to 200). Older data is sequentially discarded, and new data is added, continuously updating the estimated value of the center of eye rotation (13). This sliding window method adaptively tracks the gradual change (drift) of the center of eye rotation (13) due to minute displacement of the wearable device (34) caused by prolonged use. This method automatically adapts to changes in the mounting position over time.

[0125] In estimating the center of eye rotation (13), the robustness of the estimation is improved by simultaneously applying multiple geometric and physiological constraints, not just a single constraint (Equation 5). Additional constraints include: (1) the center of eye rotation (13) being located posterior to the eyeball; (2) the distance between the left and right centers of eye rotation (13) being close to the interpupillary distance (e.g., 60 ± 10 mm); and (3) the distance between the corneal curvature center estimated by the corneal curvature center calculation unit (8j) and the center of eye rotation (13) being within an anatomically reasonable range (e.g., 5 ± 2 mm). Also included is the core geometric constraint of this invention: (4) the line-of-sight vector (14) calculated from the center of eye rotation (13) passing through the measured external reference intersection (6). The geometric constraint determination unit (8k) formulates these constraints as an evaluation function and searches for a solution space that satisfies all constraints. By simultaneously applying multiple constraints, a physiologically valid center of eye rotation (13) can be stably estimated even in the presence of measurement noise or temporary occlusion.

[0126] Geometric inverse calculation is a nonlinear optimization problem and is computationally intensive, so the inverse calculation processing unit (8e) implements methods to improve computational efficiency. These efficiency methods include: (1) reducing the number of iterations by appropriately setting initial values ​​from depth sensor (7b) data and standard eyeball dimensions; (2) analytically deriving the Jacobian matrix (Equation 34) J=(∂V / ∂θ) and avoiding numerical differentiation (Equation 35) ∂P_c / ∂θ_j=[P_2(θ_j+ε)-P_2(θ_j-ε)] / 2ε; (3) high-speed linear algebra calculations using a sparse matrix arithmetic library; (4) reducing unnecessary iterations by appropriately setting convergence criteria (Equations 15, 16); and (5) acceleration through GPU parallel computing. Furthermore, if the estimated eye rotation center (13) is within the convergence range, the estimation process for the next frame applies a reuse strategy (Warm-StartStrategy) that uses the current estimate as the initial value, optimizing the starting point of the iterative calculation. These efficiencies allow the update process in the eye rotation center dynamic update unit (8m) during real-time operation to be performed without hindering the eye-tracking frame rate (60 to 240 Hz). Adaptive computational strategies are employed depending on the situation, such as performing more accurate iterative calculations during the initial estimation in the calibration process and performing simpler updates with reduced computational load during dynamic updates in use.

[0127] The implementation of geometric inverse calculation in this invention is a clear differentiating factor from existing eye-tracking technologies. In conventional technology, the eye rotation center (13) is either treated as a fixed value once estimated during calibration or updated only by simple interpolation. In contrast, in this invention, the eye rotation center (13) is continuously optimized during use through the coordinated operation of the inverse calculation processing unit (8e) and the eye rotation center dynamic update unit (8m). This continuous optimization is driven by geometric error feedback (deviation between the external reference intersection and the gaze vector) provided by the actual virtual intersection coordinate matching unit (8s). Furthermore, the process (paragraph 0102) of separating the deviation information between the actual intersection coordinates (24) and virtual intersection coordinates (25) detected by the actual virtual intersection coordinate matching unit (8s) into the estimation error of the eye rotation center (13) and the illumination angle error is a clear differentiating point from Apple's patented technology. The technical originality of this invention is clearly asserted through the mathematical formulation of geometric inverse calculation (Equations 1, 5, and 6) and the detailed description of its implementation algorithm.

[0128] The system storage means (130) is a storage device that hierarchically stores all data generated and used by the eye-tracking system (1), and consists of a volatile memory (130a), a non-volatile memory (130b), and an external storage device (130c). The volatile memory (130a) is a high-speed memory such as DRAM, and holds raw data acquired from the eye-tracking sensor (7), intermediate calculation results of the information analysis means (8), and temporary data necessary for real-time processing. The non-volatile memory (130b) is flash memory or an SSD, and permanently stores calibration parameters, user profiles, system settings, and software programs. The external storage device (130c) is a microSD card or cloud storage, and stores long-term eye-trajectory logs, diagnostic logs, and backup data. The system storage means (130) performs strict hierarchical data management by the data flow management unit (44b) between the high-speed access layer (130a) that responds to the requirements of real-time performance and the low-speed access layers (130b-130c) for data persistence and large-capacity recording. The data flow management unit (44b) automatically places each data item into the appropriate storage hierarchy according to its importance and frequency of reference.

[0129] The volatile memory (130a) rapidly reads and writes data generated for each frame during the periodic processing of eye tracking (60 to 240 Hz). The data held in the volatile memory (130a) includes image data from the AI ​​camera (7a), distance images from the depth sensor (7b), pupil coordinates detected by the pupil detection unit (7c), eye-line vectors (14) calculated by the eye-line direction calculation unit (8n), and coordinates of the external reference intersection (6) calculated by the triangulation calculation unit (8a). This data is managed by a ring buffer structure, and the latest N frames (e.g., N=60, 1 second) of data are always held. Older data is overwritten sequentially, but if a gaze event is detected by the gaze determination unit (44i), the data for that frame is copied to the non-volatile memory (130b) for long-term storage. The volatile memory (130a) is positioned as close as possible to the processing unit (information analysis means (8)) and is designed to minimize data transfer latency. The access speed of the volatile memory (130a) is a crucial factor in determining the real-time performance of eye tracking.

[0130] The non-volatile memory (130b) stores data that needs to be retained even when the system is powered off. The main data stored in the non-volatile memory (130b) includes (1) user-specific calibration parameters (coordinates of the center of eye rotation (13), gaze correction coefficient δcalib, interpupillary distance), (2) authentication template for the iris recognition unit (7e), (3) system settings (frame rate, power saving mode settings, communication settings), (4) software programs and firmware, and (5) records of gaze events (object of gaze, gaze time, gaze duration). The calibration process control unit (44c) stores the estimated center of eye rotation (13) and correction coefficient in the non-volatile memory (130b) when calibration is complete, and reads them at the next startup, thereby eliminating the need for recalibration. Furthermore, baseline information for detecting minute displacements of the mounting position of the retaining member (34) (e.g., initial relative distances between each sensor when worn) is stored and used as a reference for correcting mounting position shifts after long-term use. Data in non-volatile memory (130b) is encrypted by the security management unit (44l) and protected from unauthorized access by a hardware security module (HSM) that meets FIPS 140-2 standards.

[0131] The external storage device (130c) archives long-term eye-tracking data, system operation logs, and diagnostic data. The eye-tracking recording unit (44h) saves time-series data of eye vectors (14), fixation points (15), and eye movement types to the external storage device (130c) when the user enables recording. The saved data also includes raw data of pupil center and corneal reflection position (RawSensorData), structured with frame numbers and timestamps, for later reapplication of calibration parameters or analysis with new algorithms. This data is saved in a timestamped structured format (e.g., CSV, JSON) for use in later analysis and research. The log management unit (44n) periodically writes system operation history, error logs, and usage statistics to the external storage device (130c). If the capacity of the external storage device (130c) reaches its limit, the data flow management unit (44b) compresses or uploads old log data to cloud storage to free up local storage space. The external storage device (130c) is implemented as a removable microSD card, making it easy to back up data and transfer it to other systems.

[0132] The system storage means (130) implements a database management system to enable efficient data retrieval and access. User profiles, calibration parameters, and gaze event records are structured as relational or NoSQL databases and can be quickly searched using keys such as user ID, timestamp, and event type. Eye-tracking data is managed as a time-series database (TSDB), enabling efficient queries by specifying time ranges. In particular, the design accommodates the time-series characteristics of eye-tracking data (high data density, write priority, sequential access), enabling data extraction in milliseconds. The data flow management unit (44b) automatically generates and updates database indexes to optimize search performance. Even when a large amount of data is accumulated, the index ensures that access time to the desired data remains constant. Database backups are regularly created on external storage devices (130c) or cloud storage to reduce the risk of data loss.

[0133] The eye-tracking system (1) continuously generates data at high frame rates (60 to 240 Hz), making efficient use of memory capacity crucial. The system memory means (130) optimizes memory capacity by applying data compression algorithms. Image data is compressed using lossless compression (PNG, lossless JPEG) or lossy compression (JPEG, H-264). Numerical data (eye-tracking vector (14), coordinates of the fixation point (15)) is compressed using differential encoding or floating-point compression, which optimizes and stores floating-point values ​​in binary to maintain high accuracy. Time-series data is reduced in size by adaptive sampling, which decimates redundant continuous values. For example, if the fixation point (15) is fixed for a long period, data is not saved for each frame, only the fixation start and end times are recorded. This adaptive sampling dynamically changes the sampling rate depending on the type of eye movement (fixation, saccade, tracking). The data flow management unit (44b) performs the compression process as a low-priority task so as not to affect real-time processing.

[0134] Because eye-tracking data includes personal information such as the user's attention state, interests, and cognitive load, the system storage means (130) implements privacy protection functions. In cooperation with the security management unit (44l), encryption is applied to the stored data to prevent information leakage due to unauthorized access. When data is provided externally for research purposes, data anonymization is performed, and information that can identify the user (iris pattern, facial image, user ID) is deleted or pseudonymized. This anonymization process employs a separation management method that separates personally identifiable information from eye-tracking data and allows data combination only through a hashed pseudo-user ID. The coordinates of the object being gazed at are retained, but the contents of the object (e.g., URL of the web page being viewed, document contents) are deleted. Users can control which data is recorded and stored through system settings, guaranteeing their right to self-determination regarding privacy. The log management unit (44n) records the history of data access and provides a function that allows users to audit data usage.

[0135] The system output means (131) is an interface for providing eye-tracking data calculated by the eye-tracking system (1) to an external system, and consists of a real-time streaming output unit (131a), a file output unit (131b), an API output unit (131c), and a visualization output unit (131d). The real-time streaming output unit (131a) transmits the eye-tracking vector (14), the coordinates of the gaze point (15), and the gaze determination result to an external application with low latency (1 to 10 milliseconds). The file output unit (131b) outputs the recorded eye-tracking trajectory data in standard file formats such as CSV, JSON, and HDF5. The API output unit (131c) works in cooperation with the application linkage unit (44f) to provide data in response to requests from external applications. These output means provide geometric error information indicating the accuracy of the gaze point (15) (e.g., the magnitude of the deviation calculated by the actual virtual intersection coordinate matching unit (8s)) as additional information. The visualization output unit (131d) visualizes the gaze trajectory as a heatmap, screen, and gaze time graph, and outputs them to the display unit (50) or an external display.

[0136] The real-time streaming output unit (131a) provides eye-tracking data to real-time applications (VR games, eye-tracking interfaces, attention state monitoring) with low latency. The output data includes the eye-tracking vector (14) at the current time, the three-dimensional coordinates of the gaze point (15), the ID of the gazed object, the type of eye movement (fixation, saccade, smooth pursuit), pupil diameter, and gaze confidence score. The real-time streaming output unit (131a) minimizes the delay from data acquisition to transmission by using low-latency transmission via the UDP protocol or inter-process communication via shared memory. In particular, the coordinates of the gaze point (15) are output in a geometrically fixed coordinate system (global coordinate system) by the dynamic placement of the external reference intersection (6) by the oscillation means (2), providing stable eye-tracking data that is independent of the movement of the worn device. The output frame rate is adjustable according to the application requirements, set to 240Hz when high-speed response is required, and to 30Hz when power saving is prioritized. The communication interface unit (44e) manages simultaneous connections from multiple applications and provides each application with an independent data stream.

[0137] The file output unit (131b) outputs the data recorded by the eye-tracking recording unit (44h) in a standard file format for research and analysis. In the CV format, each line represents data for one frame, and comma-separated information such as the timestamp, the X, Y, and Z components of the eye-tracking vector (14), the coordinates of the point of fixation (15), and the pupil diameter are recorded. In the JSON format, a hierarchical data structure (session, event, frame) is represented, and metadata (user ID, calibration parameters, environmental conditions) is also included. The HDF5 format can efficiently store and load large amounts of time-series data and is suitable for analysis using Python or MATLAB. The file output unit (131b) adds the coordinates of the eye-rotation center (13) and the correction coefficient (δcalib) obtained in the calibration process as a file header during output, allowing analysts to recalibrate the data. The file output unit (131b) also supports conversion to standard eye-tracking data formats (e.g., Tobii, SMI-pupilLabs) to ensure compatibility with existing analysis tools. The output file includes quality information such as data collection conditions, calibration accuracy, and missing frame rate.

[0138] The API output unit (131c) provides an application programming interface for external application developers to programmatically access eye-tracking data. The API output unit (131c) offers r-ESTfulAPI, WebSocketAPI, and native SDKs (C++, Python, C#), enabling flexible integration according to the developer's technology stack. Data obtainable via the API includes real-time eye-tracking data, historical data queries, gaze event searches, and statistical information (average gaze time, saccadic frequency). The API output unit (131c) features bidirectional communication capabilities, enabling system configuration changes (e.g., switching to power-saving mode, initiating dynamic calibration) and the provision of geometric hints for gaze targets from external applications. The API output unit (131c) works in conjunction with the application integration unit (44f) to authenticate and authorize API calls, preventing unauthorized access. API rate limiting protects the system from excessive requests. API specifications and sample code are provided to developers to facilitate application development utilizing eye-tracking functionality.

[0139] The visualization output unit (131d) converts the gaze trajectory data into an intuitively understandable visual representation and outputs it to the display unit (50) or an external display. The main visualization formats include: (1) Heatmap: displaying areas with long gaze durations in warm colors; (2) Scampus: displaying the trajectory of gaze movement as lines in chronological order; (3) Gaze point plot: displaying each gaze point as a circle and representing the gaze duration by the size of the circle; (4) Timeline: displaying the type of eye movement (fixation, saccade) on a time axis; and (5) 3D gaze trajectory: displaying the trajectory of the gaze vector (14) in three-dimensional space in 3D. In particular, the visualization of the 3D gaze trajectory displays the actual coordinates of the external reference intersection (6) and the virtual intersection coordinates (25) simultaneously, enabling visual verification of geometric consistency. The visualization output unit (131d) superimposes the environmental map generated by the SLAM processing unit (8p) with the gaze trajectory to visually show which objects the user gazed at. Visualization parameters (color, transparency, time range) can be adjusted by the user or application. The visualized eye-tracking data can be used for applications such as usability evaluation, advertising effectiveness measurement, and learning support.

[0140] The system output means (131) evaluates the quality of the output data and assigns a confidence score. The elements of the confidence evaluation include (1) the success rate of pupil detection, (2) the consistency of the left and right gaze vectors (14), (3) the validity judgment result by the geometric constraint condition determination unit (8k), (4) the anomaly detection result by the environmental factor error detection unit (127), and (5) the elapsed time of calibration parameter updates. Each frame of data is assigned a confidence score ranging from 0 to 1, and frames with low scores are marked as low-quality data. The application can determine whether the data is usable based on this confidence score. If consecutive low-confidence frames are detected, the system output means (131) notifies the diagnostic unit (44m) to prompt a re-execution of calibration or a system restart. Quality metrics such as calibration accuracy (e.g., angular error of 0.5 degrees), sampling rate, and missing frame rate are attached to the output data.

[0141] When integrating eye-tracking data with an external system, the accuracy of time synchronization is crucial. The system output means (131) assigns a high-precision timestamp to each data point, enabling time synchronization with the external system. The timestamp is recorded with microsecond precision based on a reference clock managed by the synchronization control unit (8f). When synchronization with an external system (motion capture, electroencephalograph, eye-tracking application) is required, the clocks between systems are synchronized using NetworkTimeProtocol (NTP) or PrecisionTimeProtocol (PTP). The system output means (131) also supports input and output of hardware synchronization signals (TTL pulses), enabling millisecond-precision synchronization with external devices. The reference time for the timestamp (UNIX time, elapsed time since system startup) is clearly stated in the metadata of the output data, clarifying time reference during data analysis.

[0142] The system output means (131) provides a function to filter output data according to the application's requirements. Filtering conditions include (1) a confidence score threshold, (2) outputting only gaze events, (3) outputting only gaze within a specific object area, (4) outputting only saccades or fixations, and (5) specifying a time range. For example, a usability evaluation application may only require fixation events with a confidence score of 0.8 or higher, and this filtering excludes noisy data. The API output unit (131c) allows filtering conditions to be specified by query parameters, enabling efficient acquisition of only the necessary data. The filtering process is executed as a low-priority task by the data flow management unit (44b) and does not affect real-time eye-tracking performance. The filtered data is stored in a cache and provides a fast response when the same conditions are requested again.

[0143] The coordinate system of the eye-tracking data can be transformed according to the application's requirements. The system output means (131) works in conjunction with the coordinate system transformation module (8h) to allow selection of the coordinate system of the output data. The provided coordinate system options include: (1) Holding part coordinate system (based on the origin of the holding part coordinate system (4)) (2) Global coordinate system (environment-fixed absolute coordinates) (3) Screen coordinate system (2D coordinates of the display unit (50)) (4) The object coordinate system (local coordinates of the object being focused on) is included. AR applications require a global coordinate system, while UI applications require a screen coordinate system. The system output means (131) performs a coordinate transformation using (equation 10) and outputs the data in the requested coordinate system. The coordinate system transformation matrix is ​​dynamically updated based on head pose information from the body-worn device (48) and the environment map from the SLAM processing unit (8p). The metadata of the output data clearly defines the coordinate system being used, eliminating ambiguity in data interpretation.

[0144] Configuration and function of calibration jigs (including material specifications) The calibration jig (120) in Figure 14 is a device for holding multiple jig reference points (121) having known three-dimensional coordinates in precise positions in space, and consists of a calibration surface on which the jig reference points (121) are arranged and a holding mechanism that supports it. The calibration surface is implemented as a planar check sheet (49a), a spherical check sheet (49b), or a monitor-type check sheet (49c). Material specifications for calibration jig (120) The calibration jig (120) is made of a material that has high dimensional stability against changes in ambient temperature, aging, and mechanical stress. As a result, the three-dimensional coordinate accuracy (within ±0.1 mm) of the jig reference point (121), which was calibrated at the factory, is maintained even after long-term use. Recommended materials for high-precision calibration fixtures: (Material 1) Invar alloy (Fe-Ni alloy) An iron-nickel alloy with an extremely low coefficient of thermal expansion (approximately 1.2 × 10⁻⁶ / K). Dimensional changes are kept below 0.01 mm / m in the temperature range of 50°C to +150°C. It is a material widely used as a reference gauge for precision measuring instruments and is ideal as the main material for the holding mechanism of the calibration jig (120) of the present invention. (Material 2) Low thermal expansion ceramics Crystallized glass ceramics such as zero-dür or Ceravit. They have an extremely low coefficient of thermal expansion (approximately ±0.05 × 10⁻⁶ / K) and are used as reference bases in optical systems. They are suitable as substrate materials for spherical check sheets (49b). (Material 3) Quartz glass (fused silica) A material with low thermal expansion (approximately 0.55 × 10⁻⁶ / K) and optically transparent properties. Suitable for detecting jig reference points (121) using a transmissive optical system in a planar check sheet (49a). (Material 4) Carbon Fiber Reinforced Plastic (CFRP) It is lightweight yet possesses high rigidity and low thermal expansion properties. The temperature coefficient is approximately -0.5 × 10⁻⁶ / K in the fiber direction. It is effective as a material that achieves both weight reduction and dimensional stability in portable calibration jigs (120). Distinction from a simplified checklist On the other hand, separate from the calibration jig (120) of the present invention, it is also possible to use a check sheet made of a general-purpose plastic material such as acrylic resin, polycarbonate resin, or ABS resin, or a paper check sheet, as a simple calibration means. These simple check sheets have the following characteristics: (Features of the simplified version) Large coefficient of thermal expansion (approximately 50 to 100 × 10⁻⁶ / K) Dimensions may vary due to temperature and humidity changes (e.g., ±0.5mm / mat ±20℃). Low-cost manufacturing possible It is intended to be disposable or replaced regularly. Suitable for home, educational, and simple eye-tracking systems. Simplified coordinate correction When using a simplified checklist, it is necessary to correct for the thermal and humidity expansion of the material based on the environmental conditions measured by the temperature sensor (122) and humidity sensor (123). The correction formula is as follows: L_corrected=L_nominal×[1+α_T(T-T_ref)+α_H(H-H_ref)] Here, L_corrected: Corrected distance between fixture reference points L_nominal: Nominal distance at standard temperature and humidity α_T: Linear expansion coefficient [1 / K] α_H: Moisture expansion coefficient [1 / %RH] T,T_ref: Current temperature, reference temperature H,H_ref: Current humidity, reference humidity This correction allows even a simple check sheet to achieve a practical calibration accuracy (approximately ±0.5 mm). Material composition of calibration jig (120) by form Flat-type checklist (49a) Substrate: Low thermal expansion glass, quartz glass, or Invar alloy plate, etc. Jig reference point (121): Microholes created by high-contrast ceramic markers or etching. Retaining frame: Invar alloy, aluminum alloy (with temperature compensation mechanism) Spherical checklist (49b) Substrate: Low thermal expansion ceramics (zero dühl, etc.), or finely polished quartz glass, etc. Jig reference point (121): Micro-marking by laser etching, or vapor-deposited metal markers Holding mechanism: High-precision holding of spherical orientation through three-point support. Monitor-type checklist (49c) Display unit (50): Liquid crystal display, organic EL display Display unit position measurement: Real-time position measurement by inertial measurement unit (35) and SLAM processing unit (8p) Coordinate accuracy: Depends on the position measurement accuracy of the display unit (50) (typically ±1 mm to ±2 mm) How to use the calibration jig (120) and the simple check sheet 1. Differences in materials and performance Calibration jigs (120) and simple check sheets are clearly distinguished in terms of their material, performance, and application. The calibration jig (120) is made of a material with extremely low thermal expansion coefficients and excellent dimensional stability, such as Invar, ceramics, or quartz. This ensures that coordinate accuracy remains at a high level of ±0.1 mm over the long term, with a service life of over 10 years. While its manufacturing cost is high (tens to hundreds of thousands of yen), its precision is essential for applications requiring extremely precise measurements, such as industrial, medical, and research applications. Therefore, calibration frequency can be limited to periodic performance checks, typically once a year. On the other hand, simple check sheets are made of inexpensive and easily processed materials such as acrylic, plastic, or paper. They have low dimensional stability and their coordinates fluctuate with temperature and humidity, so even with temperature compensation, their coordinate accuracy remains at around ±0.5 mm. Manufacturing costs are low (a few hundred to a few thousand yen), and their service life is short, ranging from six months to two years. Their use is limited to household, educational, or simple applications. Due to their simplicity, frequent calibration, such as before each use or once a month, is recommended. 2. Applicable fields and the role of standards In the system of the present invention, the calibration jig (120) is positioned as the absolute calibration standard and is used for the initial setup of the system and for verifying the accuracy of the reference coordinate system after long-term use (annual calibration). This establishes a highly accurate measurement standard throughout the entire system. In contrast, the simplified check sheet is used for simple checks by everyday users (9), basic operational verification of the eye-tracking algorithm, and relative accuracy adjustment. By using the jig (120) when high-precision calibration is required, and the simplified check sheet for everyday operation, both operational efficiency and accuracy maintenance of the system can be achieved. This distinction allows for the selection of the optimal calibration method according to the required accuracy and cost. In particular, in order to achieve high-precision calculation of the eyeball rotation center (13) of the present invention (within ±0.5 mm), it is strongly recommended to use the calibration jig (120) during the initial calibration. On the other hand, for everyday simple checks or applications with low accuracy requirements (e.g., calibration of eye-tracking interfaces), sufficient performance can be obtained with the simplified check sheet. Supplement: Correspondence with Claim 2 The "reference point (121) of a calibration jig (120) having known three-dimensional coordinates" described in claim 2 primarily refers to the high-precision calibration jig described above, but the wording of the claim is broad enough to also include a simplified check sheet. As a result, the scope of the claim broadly covers systems ranging from high-precision to simplified systems. However, in order to achieve the greatest technical effect of the present invention, which is the target value of eye-tracking accuracy of ±0.5 mm (viewing angle within 0.5 degrees), the use of the high-precision calibration jig (120) described in paragraph 0144 is substantially essential.

[0145] The calibration jig (120) has three main forms depending on the intended use and environment. Form 1: The planar check sheet (49a) is a physical sheet having multiple jig reference points (18a, 18b, etc.) arranged in a regular grid or concentric pattern on a planar substrate. Each jig reference point (121) is printed or affixed as a high-contrast marker (e.g., black dot on a white background, or fluorescent marker) for visibility, and its two-dimensional coordinates are measured with high precision during manufacturing and recorded in the calibration data storage unit (130a). The planar check sheet (49a) is mainly suitable for applications where line-of-sight accuracy in the XY plane is important, such as operating a two-dimensional user interface, inputting text, and selecting icons on a screen. The calibration procedure involves the user (9) wearing the wearable device (48), placing the planar check sheet (49a) at a known distance (e.g., 50 cm) in front of them, and sequentially gazing at each jig reference point (121) according to the system's instructions. The eye-tracking sensor (7) records eyeball information for each moment of fixation, and the information analysis means (8) calculates the eyeball rotation center (13) from the known coordinates of the jig reference point (121) and the eyeball information. Form 2: The spherical check sheet (49b) is a jig having multiple jig reference points (121) arranged on a spherical or hemispherical substrate, capable of calibrating line-of-sight accuracy in all directions in three-dimensional space. The jig reference points (121) on the spherical check sheet (49b) are defined in a spherical coordinate system (radius r, zenith angle θ, azimuth angle φ), and their three-dimensional Cartesian coordinates (x, y, z) are recorded in the calibration data storage unit (130a). The spherical check sheet (49b) is suitable for applications requiring high-precision line-of-sight tracking in three-dimensional space, such as VR / AR applications, 3D CAD operations, and surgical support systems. Particularly important is that the spherical check sheet (49b) has superior calibration accuracy in the depth direction (Z axis) compared to the planar check sheet (49a), and can determine the Z coordinate of the eyeball rotation center (13) with high precision. This is because the jig reference point (121) on the sphere has a wide range of Z coordinate values, thus providing abundant parallax information in the Z-axis direction. The spherical check sheet (49b) plays a particularly important role in the parallelism calibration of the dual baselines. When the user (9) fixates on the center point on the sphere (corresponding to the neutral position 18c), the line-of-sight angles θ_eyeL and θ_eyeR of the left and right eyes should theoretically be equal, and the accuracy of the system's geometric arrangement can be verified from this symmetry. Specifically, the measured value of the baseline length between the centers of eye rotation (17) is calculated using equation 70 from the distance d_known from the center of the sphere and the measured line-of-sight angle, and the geometric consistency with the base length of the holding part (16) is further verified using equation 74. If the symmetry is compromised (Δ_parallel > threshold in equation 73), the parallelism of the dual baselines is restored by correcting the tilt of the mounting device (48) by referring to the data of the inertial measurement unit (35), or by fine-tuning the beam angle with the oscillation means angle adjustment mechanism (2d). Note that the neutral position of First Note By looking External references An intersection point is formed, but if that intersection point does not coincide with the neutral position, the eyeballs move. hand The intersection point coincides with the neutral position. Adjust to do so By doing so, period error amount To seek but It is possible. If an intersection point is not formed, do it one eye at a time. Neutral position Difference and Error sequentially Request ru Therefore 、 Each eyeball line of sight Vectors and intersections and Measure the difference Then, identify the intersection point. to do but This becomes possible. This involves assigning the point of focus and the intersection point to each coordinate point. Transfer By doing so, the error can be determined. Embodiment 3: The monitor-type check sheet (49c) is a form that provides dynamic jig reference points (121) electronically displayed on a liquid crystal display or OLED display. The position of the jig reference points (121) displayed on the display unit (50) is dynamically controlled by the system control means (44) and is sequentially presented to the optimal position according to the user's (9) calibration progress. The advantages of the monitor-type check sheet (49c) are that there is no need to carry physical jigs and the calibration pattern can be flexibly changed by software updates. In addition, the blinking and color changes of the jig reference points (121) easily attract the user's (9) attention, improving the efficiency of the calibration work. However, since the coordinate accuracy of the monitor-type check sheet (49c) depends on the measurement accuracy of the position and angle of the display unit (50), the position of the calibration jig (120) itself must be measured with high precision by an inertial measurement unit (35) or a SLAM processing unit (8p). Each type of calibration jig (120) is selected according to the operating environment and required accuracy. As a general recommendation, a spherical check sheet (49b) is used for initial calibration or when the highest accuracy is required, while a flat check sheet (49a) or monitor check sheet (49c) is used for routine, simple calibration.

[0146] The system output means (131) works in conjunction with the security management unit (44l) to implement security and access control for the output of eye-tracking data. Since eye-tracking data is sensitive information related to individual privacy, it is necessary to prevent unauthorized access and leakage. Access control functions include (1) setting access permissions for each application, (2) setting output permission for each data type, (3) enabling / disabling output by the user, and (4) recording encrypted communication (TLS / SSL-5) access logs. Users can control which data is provided to which application through system settings. The API output unit (131c) verifies the legitimacy of the application using API keys or OAuth authentication. Attempts to access unauthorized data are detected by the security management unit (44l) and recorded by the log management unit (44n). These security functions ensure the privacy protection of eye-tracking data.

[0147] The system memory means (130) and the system output means (131) operate in coordination under the mediation of the data flow management unit (44b). A typical data flow is as follows: Raw data acquired from the eye-tracking sensor (7) is temporarily stored in volatile memory (130a) and processed by the information analysis means (8). The processing results (eye-view vector (14), gaze point (15)) are immediately transmitted to an external application via the real-time streaming output unit (131a) and simultaneously held in the ring buffer of the volatile memory (130a). When a gaze event is detected by the gaze determination unit (44i), the data for the corresponding frame is stored in non-volatile memory (130b). If the user has enabled eye-tracking trajectory recording, data for all frames is streamed and stored in the external storage device (130c). The recorded data can later be retrieved via the file output unit (131b) or the API output unit (131c). This hierarchical data management and diverse output formats enable efficient real-time application and post-analysis. Claim 1 is an independent term that defines the basic configuration of the eye-tracking system (1) and the technical features of geometrically calculating the eyeball rotation center (13) and dynamically optimizing it through physiological validity determination. Claim 1 identifies an information analysis means (8) as a core component that combines eyeball information acquired by an eye-tracking sensor (7) with means for forming an external reference intersection (6) (oscillating means (2)), and geometrically calculates the eyeball rotation center (13) by associating the two. Furthermore, it clearly differentiates itself from conventional technology by determining the physiological validity of the calculated eyeball rotation center (13), evaluating the error between the assumed gaze point and the external reference intersection (6), and determining or updating the eyeball rotation center (13) to reduce the error while satisfying geometric constraints. This dynamic optimization process provides a function that automatically continues to adapt to deviations in the mounting position of the wearable device, individual differences in eyeball shape for each user, and positional changes (drift) over time, without requiring the user to perform calibration work. The components of Claim 1 will be described in detail below.

[0148] Claims 1 through 18 will be explained in order below. Figure 2 is a block diagram showing the overall configuration of the eye-tracking system of the present invention, illustrating a structure in which multiple technical elements are hierarchically integrated, starting from the "absolute coordinate reference based on an external reference intersection," which is the core technical feature of the present invention. The patentability of this system is based on the following three technical hierarchical structures: The first layer is the "absolute coordinate formation unit," which generates an external reference intersection (6) by the intersection formation of beams irradiated from multiple oscillation means, establishing an observable geometric reference point in the space outside the eyeball. Unlike relative references such as image coordinate systems and camera coordinate systems on which conventional technology relied, this external reference intersection (6) functions as a physically measurable absolute coordinate, which constitutes the fundamental inventiveness of the present invention. The second layer is the "eyeball rotation center (ERC) high-precision calculation unit," which determines the eyeball rotation center (13) by geometric inverse calculation, using the external reference intersection (6) as a known true value. Unlike conventional statistical estimation methods, this inverse calculation process involves verification of anatomical constraints (such as the physiological range of the baseline length between the centers of eye rotation, with an eyeball radius of 11 mm to 14.5 mm) by the physiological validity determination unit (claims 1 to 18), achieving high accuracy of approximately one-tenth of the conventional error (within 0.3 to 0.7 degrees of the visual angle). This fusion of physiological constraints and geometric inverse calculation prevents structural evasion by pure AI methods and ensures the robustness of the patent scope. The third layer is a "multilayered defense structure block," in which three independent technical elements are arranged in parallel: (1) Anti-counterfeiting / signal authentication unit (claim 9): Adds unique identification information to the beam and provides a cryptographic security layer that detects counterfeit products and external interference. (2) Geometric hybrid unit (claims 2 to 16): In addition to dynamic calibration by an external reference intersection (6), static calibration using a calibration jig is possible, enabling flexible selection of calibration methods according to the usage environment and accuracy requirements. This combination of dynamic and static calibration achieves both high initial accuracy setting and accuracy maintenance during continuous use. (3) Multi-physical characteristic inclusion section (Claim 3): By combining an electromagnetic wave beam (laser light) and a sound wave beam (ultrasound), adaptability according to the operating environment (underwater, vacuum, electromagnetic shielding environment, etc.) is ensured, and robustness to environmental changes is provided.Below these multi-layered defense structures is a "self-diagnosis / error factor discrimination unit (claim 12)" which separates and discriminates the deviation of the oscillation means and the drift of the eyeball model by comparing the actual intersection point and the virtual intersection point, and performs selective correction. Furthermore, an "environment / aging deterioration compensation unit (claim 14)" compensates for thermal expansion and aging drift based on information from a temperature sensor and an internal clock. In the lowest layer, the "ERC calculation mathematical processing unit" calculates the optimal solution for the eyeball rotation center (13) using spherical fitting and nonlinear least squares method (WLS / Levenberg-Marquardt method). This mathematical processing is integrated into a "hierarchical data processing structure (digital twin)" which manages a three-dimensional eyeball model including the corneal curvature center (C_v) and pupil plane center (P_p), enabling development into VR / AR optimization, VR sickness reduction, foveated rendering, etc. in the "application possibilities block". The essence of the invention's patentability lies in the fact that, through this hierarchical integrated structure, it is fundamentally impossible to partially circumvent the invention through the processing of individual elements of the prior art (statistical estimation, machine learning, etc.), and it establishes a comprehensive scope of rights centered on the core technology of "absolute coordinate standards by external reference intersections (6)."

[0149] The "eye-tracking sensor for acquiring user eyeball information" in claim 1 corresponds to the eye-tracking sensor (7) and is a device that non-contactively measures physical and geometric information about the user's (9) eyeballs (10) (left eyeball (10a), right eyeball (10b)). As shown in Figures 11(b) and 12(a), the eye-tracking sensor (7) is mounted on a wearable device (48) (smart glasses, VR goggles) or on a computer or in front of the driver's seat of a car, etc., and observes the user's (9) eyeballs (10) from the front. (See paragraphs 0079 to 0083).

[0150] The “means for forming an external reference intersection” in claim 1 corresponds to the oscillation means (2) (see paragraphs 0057 to 0061). The external reference intersection (6) has a dual structure of actual intersection coordinates (24) and virtual intersection coordinates (25) as shown in Figure 13(a). The means for detecting the actual intersection coordinates (24) monitors the beam deviation due to the physical distortion of the system in real time, and the virtual intersection coordinates (25) provide the theoretical and absolute geometric true value of the external reference intersection (6).

[0151] The "information analysis means for geometrically calculating the center of eye rotation by associating the external reference intersection with eyeball information" in claim 1 corresponds to the information analysis means (8) and is a calculation system that geometrically associates the eyeball information (pupil position) acquired from the gaze tracking sensor (7) with the coordinates of the external reference intersection (6) formed by the oscillation means (2) to calculate the three-dimensional coordinates of the center of eye rotation (13) (left eye rotation center (13a), right eye rotation center (13b)). As shown in Figure 12(a), there are eyeball rotation centers (13a-13b) for the left and right eyeballs (10a-10b), and the distance between them is defined as the baseline length between the centers of eye rotation (17). The basic principle of the geometric calculation utilizes the constraint condition (Equation 5) that when the eyeball (10) rotates, the center of rotation of the eyeball (13) behaves as a fixed point, and the distance from the center of rotation of the eyeball (13) to the pupil center (11c) remains approximately constant, r (radius of the eyeball). The eyeball rotation center calculation unit (8i) of the information analysis means (8) formulates the eyeball rotation center C that best satisfies this constraint condition as an optimization problem from the pupil positions Pi (P1, P2, P3, P4, 5 etc. in Figure 13(b)) in multiple line-of-sight directions, and solves it using the least squares calculation unit (23b). This basic formulation is expressed as follows (original form of Equation 1). min_{C,r}Σ_{i=1}^N[||Pi-C||-r]^2 (Formula 1 prototype) In this system, this formulation is further extended as a constrained least-squares problem incorporating physiological constraints as penalty terms. Its basic form is represented by (Equation 1) below. J(C)=Σ[||Ppupil-C||-r]^2+λ||ModelConstraint(C)||^2 (Formula 1) Here, the first term is the geometric error (the degree of deviation of the eyeball from its spherical orientation), and the second term (λ||ModelConstraint(C)||^2) is a penalty term for deviations from physiological constraints (such as the acceptable range of eyeball radius and the stability of the baseline length between the left and right eyeball rotation centers). This equation A functions as a regularization to prevent convergence to unrealistic solutions caused by measurement errors. Furthermore, this formulation is ultimately extended as a constrained least-squares problem, where the coordinates of the external reference intersection (6) are known true values, and a geometric consistency constraint—that the line-of-sight vector from the pupil position passes through the external reference intersection (6)—is incorporated into the objective function as a weighted penalty term. The external reference intersection (6) should be positioned on the extension of the line-of-sight vectors (14) (left eye line-of-sight vector (14a), right eye line-of-sight vector (14b)) and serves as a reference point for verifying the accuracy of the estimation of the center of eye rotation (13). (See Figure 3) In particular, this system enhances residual evaluation during the weighted least squares (WLS) optimization process. While conventional techniques tend to focus only on the simple spatial positional difference (shift on the XY plane) between the line-of-sight vector and the external reference point, this system incorporates distance information (depth Z coordinate), obtained with high accuracy through the aforementioned line-of-sight vector synchronization, into the comparison and matching process. As a result, the residual evaluation function evolves from a conventional positional difference such as |Vgaze-Vref| to a multidimensional error function that increases the weight of the error in the distance direction. Specifically, a large weight WZ is assigned to the difference |Zgaze-Zref| between the distance (Zref) of the external reference intersection (6) and the distance of the estimated gaze intersection (point of fixation) (Zgaze). Since the positional error of the eye rotation center (13), especially the error in the depth direction (Z-axis), proportionally affects the estimated distance of the gaze intersection, maximizing this distance error penalty significantly improves the accuracy of the Z-axis estimation of the eye rotation center. This WLS augmentation is an important technical feature that determines the overall system robustness in environments where Z-axis mounting errors are likely to occur, such as head-mounted displays (HMDs).

[0152] The "determination of the physiological validity of the calculated center of eye rotation" in claim 1 is an important verification process performed by the geometric constraint determination unit (8k). As shown in Figure 13(b), it is determined whether the geometrically calculated candidate values ​​of the center of eye rotation (13) (C1, C2, C3, C4, C5, etc.) are valid in light of the anatomical structure of the human eyeball (10). For the determination of physiological validity, the physiological range determination unit (116a) and the consistency verification unit (116b) are used and compared with the physiological constraint range (119) stored in the physiological database (118). (See Figure 6) The criteria for judgment include: (1) the distance from the center of ocular rotation (13) to the pupillary center (11c) being within the physiological range (10 to 15 mm, shown by the radius circle in Figure 13(b)); (2) the center of ocular rotation (13) being located posterior to the eyeball (10); (3) the distance between the left and right centers of ocular rotation (13a, 13b) (intercenter of ocular rotation baseline length (17)) being consistent with the interpupillary distance (55 to 75 mm); and (4) the distance between the corneal curvature center (114) (left corneal curvature center (114a), right corneal curvature center (114b)) and the center of ocular rotation (13) being within the anatomically reasonable range (5 ± 2 mm). Furthermore, (5) the relative positional relationship (coordination of both eyes) in three-dimensional space of the estimated left and right eyeball rotation centers (13a, 13b) is also verified to be within a physiologically acceptable range, thereby improving the stability of the estimation. These criteria eliminate abnormal estimates due to measurement noise or temporary occlusion, and only physiologically feasible eyeball rotation centers (13) are adopted.

[0153] The "evaluation of the error between the assumed gaze point derived from the eyeball rotation center and the external reference intersection" in claim 1 is an error evaluation process performed by the actual virtual intersection coordinate matching unit (8s) and the difference analysis unit (8d). As shown in Figure 13(a), a point that should theoretically be gazed upon (assumed gaze point (15)) lies on the line extended from the line of sight vectors (14a, 14b) calculated from the eyeball rotation center (13a, 13b) and the pupil center (11a, 11b). This assumed gaze point (15) is This is a theoretical fixation point derived from the eyeball side based on the eyeball rotation center and line of sight vector, and is a different concept from the virtual intersection coordinates (25) derived from the illumination direction of the oscillating means. In the ideal state where the user is precisely fixating on the external reference intersection (6), the assumed fixation point (15) and the virtual intersection coordinates (25) spatially coincide.On the other hand, the external reference intersection (6) (actual intersection coordinates (24), indicated as PC: 6 in the figure) physically formed by the oscillation means (2a, 2b) has independently measured coordinates. The actual virtual intersection coordinate matching unit (8s) (also indicated as reference numeral 27 in the figure) calculates the spatial deviation Δi (the difference shown at the top of Figure 13(a)) between the virtual intersection coordinates (25) and the actual intersection coordinates (24), and the three-dimensional spatial error detection unit (67) evaluates this deviation as an error. As shown in the calibration correction interval area at the top of Figure 13(a), this error Δi is quantified as the difference between the coordinates (X, Y, Z) of the point of interest (15) and the external reference intersection (6). The differential analysis unit (8d) simultaneously performs a process (see paragraph 0102) to separate and identify the cause of this discrepancy Δi into estimation error of the eyeball rotation center (13) and irradiation angle error of the beam (5), and uses this as a feedback signal for dynamic updates. The cause of the discrepancy is one of the following: (1) estimation error of the eyeball rotation center (13), (2) irradiation angle error of the beam (5), or (3) calculation error of the line of sight vector (14).

[0154] The phrase "determining or updating the center of rotation of the eyeball to reduce the error while satisfying the geometric constraints in the eyeball model" in claim 1 is an optimization process performed by the eyeball rotation center dynamic update unit (8m). As shown in Figure 13(b), the geometric constraints in the eyeball model refer to the condition that pupil positions P1 to P5 lie on a sphere of radius r centered on the eyeball rotation center (13) (Equation 5 ||P_i-C||=r), and the anatomical constraints verified by the physiological validity assessment. The optimization formula at the bottom of Figure 13(b) is "Copt=argminΣwi·||projected(Pg,i,C,di)-Pg,i||^2". As shown in the text box on the right, "minΣ[||Ci-13||-r]^2", the eyeball rotation center dynamic update unit (8m) calculates the correction amount δ_C for the eyeball rotation center (13) using (Equation 32) based on the error information (Equation 28) provided by the actual virtual intersection coordinate matching unit (8s). This correction amount is set by the constraint application unit (115c) in a direction that minimizes the error within the range that satisfies the geometric constraints, and the optimization calculation unit (115b) adds it step by step to the current eyeball rotation center (13) using (Equation 31). The candidate point calculation unit (115a) generates C5 from a plurality of candidate points C1 shown in Figure 13(b), and the convergence determination unit (68) determines convergence to the optimal solution. The sequential correction control unit (69) performs gradual updates controlled by the learning rate α, suppressing rapid fluctuations due to measurement noise while following true changes. This optimization process improves computational efficiency through GPU parallel computing and analytical Jacobian matrix derivation (paragraph 0126), and is executed in real time without interfering with the eye-tracking frame rate (240Hz).

[0155] The following technical effects are achieved by the configuration of the eye-tracking system (1) as defined in claim 1. Firstly, by utilizing an observable external reference point called the external reference intersection (6), the center of eye rotation (13) inside the eyeball can be estimated with high accuracy. Secondly, by combining geometric calculation and physiological validity determination, highly reliable estimation becomes possible. Thirdly, by continuously evaluating the error between the assumed gaze point (15) and the external reference intersection (6) and dynamically updating the center of eye rotation (13), accuracy degradation is suppressed even during long-term use. Claims 1 and 18 integrate the core requirements of the present invention's inventive step: determining the physiological validity of the geometrically calculated center of eye rotation (ERC) and determining or updating it to reduce errors. This results in a self-correcting system based on geometric constraints, rather than mere statistical estimation.

[0156] Each component of claim 1 corresponds to a specific embodiment described in detail in the modes for carrying out the invention. "Eye-tracking sensor" corresponds to the eye-tracking sensor (7) in paragraphs 0079 to 0083, shown in Figures 11(b) and 12(a). "Means for forming an external reference intersection" corresponds to the oscillation means (2) in paragraphs 0057 to 0061, shown in Figure 11(a). "Information analysis means" corresponds to the information analysis means (8) in paragraphs 0085 to 0103. "Geometrically calculating the eye rotation center" corresponds to the eye rotation center calculation unit (8i) in paragraphs 0094 and 0120 to 0122. "Determining physiological validity" corresponds to the geometric constraint determination unit (8k) in paragraphs 0096 to 0097. "Evaluating the error between the assumed gaze point and the external reference intersection" corresponds to the actual virtual intersection coordinate matching unit (8s) in paragraphs 0102 and 0153. "Determining or updating the center of eye rotation" corresponds to the dynamic updating section (8m) of the center of eye rotation in paragraphs 0099 and 0122.

[0157] Claim 2 is a claim dependent on Claim 1 and defines the diversity of calibration means in the eye-tracking system (1). Claim 2 includes a static calibration means that calculates the eye rotation center (13) based on a reference point (fixture reference point (121)) of a calibration fixture (120) having known three-dimensional coordinates, in addition to dynamic calibration using an external reference intersection (6), or in place of the external reference intersection (6). Static calibration is used for initial high-precision calibration when the system is started or when a user changes, while dynamic calibration is used for compensating for accuracy drift over time during continuous use. The components of Claim 2 will be described below with reference to Figures 14 and 16.

[0158] The phrase "in the eye-tracking system described in Claim 1" in Claim 2 means that Claim 2 inherits all the components of Claim 1. Claim 2 is a dependent claim that extends Claim 1 by adding an option for calibration methods in addition to this basic configuration. This hierarchical structure makes it possible to patent the technical concept of applying a high-precision geometric calibration method using a calibration jig (120) even to systems that do not have an external reference intersection (6).

[0159] The phrase "in addition to calibration using an external reference intersection, or instead of an external reference intersection" in claim 2 is an important constituent element that specifies the selective implementation of the calibration method. The word "in addition" includes embodiments that use both dynamic calibration using an external reference intersection (6) and static calibration using a calibration jig (120). In this combined embodiment, initial calibration is performed using the calibration jig (120) when the system starts up, and the calibration parameters are continuously updated using the external reference intersection (6) during subsequent use. The word "instead" also includes embodiments that do not use an external reference intersection (6) and use only the calibration jig (120). This alternative embodiment demonstrates that even in a simplified eye-tracking system that does not incorporate an oscillator (2), high-precision estimation of the eyeball rotation center (13) by geometric inverse calculation is possible. Thus, claim 2, through the selective wording of "or", encompasses a wide range of embodiments.

[0160] The "reference points of a calibration jig having known three-dimensional coordinates" in claim 2 correspond to the calibration jig (120) and the jig reference points (121), and are implemented in multiple forms as shown in Figure 14(a) (see paragraphs 0144 to 0145). The target presentation unit (23d) displays multiple jig reference points (121), including the neutral position (18c), in three-dimensional space (18) using a virtual intersection coordinate system. The physical spherical check sheet (49b), the planar check sheet (49a), and the monitor-type check sheet (49c) also provide jig reference points (121) with known coordinates. The coordinates of these jig reference points (121) are pre-registered in the calibration data storage unit (130a) of the system storage means (130) as absolute coordinates with respect to the holding unit coordinate origin (4).

[0161] The calibration process for "calculating the center of eye rotation" in claim 2 is managed by a calibration process control unit (44c) and an automatic calibration means (91) (see paragraph 0116). In the calibration process shown in Figure 14(b), the user (9) sequentially gazes at a plurality of jig reference points (121) displayed on the target presentation unit (23d). First, the user (9) gazes at the central neutral position (18c), and the eye-tracking sensor (7) records the pupil position P_0 at this time. Next, the user (9) sequentially gazes at the surrounding jig reference points (18a-18b), and the pupil positions P_1, P_2,..., P_n at each point are recorded. The center of eye rotation calculation unit (8i) solves the optimization problem shown in Figure 13(b) from this pupil position data and estimates the three-dimensional coordinates of the center of eye rotation (13).

[0162] The neutral position (18c) shown in Figure 14(a) is a key fixture reference point (121) that defines the reference line of sight direction in the calibration process. The neutral position (18c) is located on the line of sight direction when the user (9) is facing forward while wearing the wearable device (48), and is indicated as the center of the sphere (NP18c). Setting this reference point eliminates the uncertainty of the coordinate system in estimating the center of eye rotation (13).

[0163] The arc fitting unit (23a) shown in the processing flow of Figure 14(b) is a processing unit that fits the trajectory of the pupil position as an arc when the user (9) sequentially gazes at multiple jig reference points (121) and estimates the eyeball rotation center (13). When the eyeball (10) rotates, the pupil center (11c) moves on a sphere centered on the eyeball rotation center (13), so the projected trajectory of the pupil position draws an arc. The arc fitting unit (23a) fits the multiple pupil positions observed by the gaze tracking sensor (7) into an arc using the least squares method and estimates the center of this arc as the projection point of the eyeball rotation center (13).

[0164] While the arc fitting unit (23a) performs processing on a two-dimensional plane, the eyeball rotation center calculation unit (8i) performs spherical fitting in three-dimensional space. As shown in the spherical display at the top of Figure 14(a), multiple jig reference points (18a, 18b) are distributed in three-dimensional space. When the user (9) fixates on these points, the pupil position is distributed on a sphere of radius r centered on the eyeball rotation center (13). The eyeball rotation center calculation unit (8i) estimates the sphere that best approximates these point clusters using the least squares method and determines its center coordinates as the eyeball rotation center (13). This spherical fitting is formulated as an optimization problem that minimizes the objective function shown in (Equation 1).

[0165] The difference analysis unit (8d) incorporated into the information analysis means (8) of the present invention, when evaluating the spatial deviation amount Δi between the assumed gaze point (15) and the external reference intersection (6) (see paragraph 0153), improves the separation and discrimination of error factors by integrating a machine learning algorithm (claim 17) in addition to simple geometric inverse calculation. Specifically, multidimensional sensor data such as the quality of the pupil image obtained from the gaze tracking sensor (7), the shape distortion of corneal reflected light, ambient light noise, and the temperature history of the wearable device (48) are used as the input layer, and the estimation error of the eyeball rotation center (13), the irradiation angle error of the oscillation means (2), the gaze error due to the user's fatigue level, and the residual of the geometric calibration are learned as intermediate layers. As a result, the eyeball rotation center dynamic update unit (8m) instantly separates and discriminates from Δi the error component that should be compensated for by the calibration parameter and the error component that should be ignored as temporary noise. This AI-integrated error factor separation significantly improves the robustness of dynamic calibration.

[0166] The static calibration process using the calibration jig (120) defined in claim 2 is enhanced in terms of estimation reliability, particularly in the depth (Z-axis) direction, through synchronous control with the depth measuring means (110) (claim 11). Since the jig reference point (121) of the calibration jig (120) has a known absolute Z coordinate, the distance from the user (9) to the jig is measured by the depth measuring means (110) at the same time as the calibration jig (120) is presented. The depth noise and scale error of the depth measuring means (110) itself are simultaneously calibrated by comparing and verifying whether the line-of-sight vector calculated using the eye rotation center (13) estimated by the calibration jig (120) matches the measured distance (Z coordinate). Through this mutual verification of depth sensors via the calibration jig (120), the system can ensure highly reliable Z-axis information (depth of the point of fixation) even if the beam accuracy of the oscillating means (2) that forms the external reference intersection (6) decreases. To ensure the accuracy of three-dimensional estimation, the calibration process control unit (44c) strictly tracks and records the user's head posture during planar calibration using an inertial sensor, and issues a warning if the head tilt exceeds a certain range during calibration.

[0167] The automatic calibration control unit (23e) works in cooperation with the calibration process control unit (44c) to automate the entire calibration process using the calibration jig (120) (see paragraph 0116). As shown in the processing flow of Figure 14(b), the automatic calibration control unit (23e) sequentially performs the following processes: (1) sequentially displays the jig reference points (121) on the target presentation unit (23d), (2) the gaze stability is detected at each jig reference point (121) by the gaze determination unit (44i), (3) the pupil position at the time of gaze stability is recorded, (4) after data collection is completed for all jig reference points, the eye rotation center (13) is calculated, and (5) the calculated eye rotation center (13) is verified and stored in the calibration data storage unit (130a). Furthermore, automatic calibration in this system is enhanced by the mutual use of geometric information based on the external reference intersection (6). Because high-precision distance information (depth Z coordinate of the point of fixation) is continuously obtained through gaze vector synchronization (see paragraph 0067), the automatic calibration control unit (23e) can automatically estimate and track the eyeball rotation center baseline length (17), the relative positions of the left and right eyeball rotation centers (13a, 13b), and the eyeball radius r, not only during calibration but also during normal use.

[0168] In addition to automatic calibration by the automatic calibration control unit (23e), the calibration correction processing unit (23) provides a manual calibration correction function. Manual calibration correction is a function that fine-tunes the calibration parameters when the results of automatic calibration do not match the user's (9) subjective gaze perception. The estimated position of the gaze point (15) is displayed on the calibration presentation display unit (50a) of the display unit (50), and the user (9) inputs the deviation from the position they are actually looking at via the operation panel (51). The gaze direction vector correction unit (70) adjusts the gaze direction correction coefficient δ_calib based on this input, and the eye rotation center correction unit (117b) fine-tunes the coordinates of the eye rotation center (13). Manual calibration correction is effective when the accuracy of automatic calibration is insufficient in a specific gaze direction (e.g., peripheral field of view). The correction values ​​obtained by manual correction are stored in the calibration data storage unit (130a) and applied the next time the system is started. This combination of manual and automatic calibration enables high-precision calibration that accommodates the individual differences of a wide range of users (9).

[0169] When using the calibration jig (120), the formation of an external reference intersection (6) by the oscillating means (2) is not essential, but the oscillating means angle adjustment mechanism (2d) is used auxiliaryly in the calibration process. The oscillating means (2) is adjusted to the horizontal irradiation reference position by the angle adjustment motor (45). The horizontal irradiation reference is the orientation in which the left and right oscillating means (2a, 2b) irradiate the beam (5a, 5b) in a direction parallel to the holding part baseline length (16), and the irradiation angle θ_L=θ_R=0 degrees in this orientation is defined as the reference angle.

[0170] As specified in claim 2, "in addition to calibration using an external reference intersection," the highest calibration accuracy is achieved by using the calibration jig (120) and the external reference intersection (6) in combination. In the combined calibration process, first, an initial estimate of the eyeball rotation center (13) and a correction coefficient δ_calib for the line of sight direction are calculated by static calibration using the calibration jig (120). Next, the oscillation means (2) is activated to form an external reference intersection (6) and perform dynamic calibration. The actual virtual intersection coordinate matching unit (8s) evaluates the discrepancy between the virtual intersection coordinates (25) obtained by static calibration and the actual intersection coordinates (24). If this discrepancy is below a threshold, calibration is completed; if the discrepancy is large, the eyeball rotation center dynamic update unit (8m) readjusts the eyeball rotation center (13).

[0171] The following technical effects are achieved by selectively implementing the calibration method specified in claim 2. First, by including static calibration using a calibration jig (120), highly accurate estimation of the eyeball rotation center (13) becomes possible even in a simplified system that does not have an oscillator (2). Second, by selecting from a variety of calibration jigs (120) (spherical, planar, physical, and virtual), flexible calibration according to the operating environment and accuracy requirements is realized. Third, automation by the automatic calibration control unit (23e) reduces the burden on the user (9) and improves the reproducibility of calibration.

[0172] Each component of claim 2 corresponds to a specific embodiment described in detail in the modes for carrying out the invention. "Reference point of the calibration jig" corresponds to the calibration jig (120) and jig reference point (121) described in paragraphs 0073 to 0074 and 0078, and is shown in Figure 14(a). The process of "calculating the center of eye rotation" corresponds to the process of the center of eye rotation calculation unit (8i) described in paragraphs 0094 and 0120 to 0123, and is implemented by the arc fitting unit (23a) and the least squares calculation unit (23b) shown in the processing flow of Figure 14(b). "In addition to calibration using an external reference intersection" corresponds to the combined calibration described in paragraph 0170, and "instead of an external reference intersection" corresponds to the alternative calibration described in paragraph 0159. This clearly demonstrates that, in addition to the inventive step of dynamic calibration, static calibration is not merely a conventional calibration method, but a high-precision estimation technique based on a geometric model.

[0173] Claim 3 is a claim dependent on Claim 1 and enhances patent defense by specifying in detail the specific configuration of the means for forming the external reference intersection (6). Claim 3 further limits four technical features: (1) a support structure by a holding member (3) that maintains a known geometric arrangement relationship; (2) irradiation of a directional beam (5) by a plurality of oscillating means (2); (3) spatial crossing control of mutually non-parallel beams; and (4) selective use of electromagnetic and acoustic beams. This detail clarifies the differentiation from a mere laser pointer or diffuse light source. The components of Claim 3 are described below. (See Figures 4 and 5)

[0174] The constituent requirement in claim 3, "supported by a holding member so as to maintain a known geometric arrangement relationship," corresponds to the oscillator holding unit (3) and holding member (3a) detailed in paragraph 0060. As shown in Figures 11(a) and 11(b), the multiple oscillating means (2) are fixed on the holding member (3a), and the holding unit baseline length (16), which is the distance between each oscillation center point (2g), is maintained as a known design value. This known value allows the triangulation calculation unit (8a) and the sine rule calculation unit (8b) to use the positions of the oscillating means (2) as known parameters and calculate the coordinates of the external reference intersection (6) with high accuracy.

[0175] The “multiple oscillating means capable of emitting a beam having straightness or directionality” in claim 3 corresponds to the oscillating means (2) detailed in paragraph 0057. The oscillating means (2) includes a laser oscillator (2a) and an acoustic oscillator (2b), and each oscillating means (2) emits a beam (5) from an oscillation center point (2g). The arrangement patterns of the “multiple oscillating means” include a typical arrangement of two left and right (Figure 11(a)), as well as advanced embodiments in which three or more are arranged. The arrangement patterns for the "multiple oscillation means" include a typical arrangement of two on the left and right (Figure 11(a)), as well as more advanced embodiments that include three, a third beam, for example, four or more arranged on the circumference or at the vertices of a rectangular circle in an equilateral triangle arrangement. In any arrangement, the relative position of each oscillation center point (2g) from the coordinate origin (4) of the holding unit is managed as known by the method described in paragraph 0060. The advantages of multiple arrangements are ensuring redundancy when some of the oscillation means (2) are shielded and improving the accuracy of three-dimensional coordinate measurement. In particular, by arranging three or more oscillation means (2), redundancy is ensured in determining the external reference intersection (6) in three-dimensional space, and even if any two beams do not intersect, the nearest point of contact (intersection center) can be calculated by the least squares method.

[0176] The "control means for intersecting mutually non-parallel beams in space" in claim 3 corresponds to the oscillator control unit (2c) and the oscillator angle adjustment mechanism (2d) described in paragraph 0061. As shown in the geometric diagram at the bottom of Figure 11(a), the first beam (5a) and the second beam (5b) irradiated from the left and right oscillator means (2) are irradiated in non-parallel directions on the same plane and intersect at a point in space (external reference intersection (6)). The oscillator control unit (2c) independently adjusts the irradiation angles θ_L and θ_R and dynamically intersects the beams (5) to follow the assumed point of fixation (15).

[0177] The principle of generating a straight beam (5) from multiple oscillation means (2) and forming a beam intersection (6) in space is based on triangulation and the law of sines, which are detailed in paragraphs 0067 to 0069. The formula "d=L / [tan(θ_L)+tan(θ_R)]" shown in the diagram at the bottom of Figure 11(a) is referred to as (formula 24) in paragraph 0067 and is the basic formula for calculating the perpendicular distance d from the base length (16):L of the holding part and the irradiation angles θ_L and θ_R. The triangulation calculation unit (8a) (paragraph 0086) uses this geometric relationship to calculate the three-dimensional coordinates of the beam intersection (6). When using three or more oscillation means (2), the distance is calculated using the distance between two specific oscillation means (2) as the reference baseline length, and the beams (5) from the remaining oscillation means (2) are used for coordinate verification or to improve redundancy. This calculation based on geometric principles has the absolute advantage of not being affected by image processing errors or lens distortion aberrations of the eye-tracking sensor (7) because it relies solely on pure beam angle measurements.

[0178] The statement in claim 3 that "the beam includes an electromagnetic wave beam and an acoustic wave beam" defines the diversity of physical properties of the beam (5) as defined in paragraphs 0057 and 0058. The "electromagnetic wave beam" includes infrared laser light (wavelengths 850 nm, 940 nm), visible light, and ultraviolet light emitted from the laser oscillator (2a), and the "acoustic wave beam" includes ultrasonic waves (frequency 20 kHz or higher) emitted from the acoustic oscillator (2b). It is also possible to use pulsed or continuous irradiation with intervals. This inclusion of diversity makes it possible to select the optimal beam type according to the usage environment (underwater, vacuum, electromagnetic shielding environment, etc.). Furthermore, by utilizing the difference in propagation speed between the electromagnetic wave beam and the acoustic wave beam and dually measuring the coordinates of the external reference intersection (6) in the same space, changes in propagation speed caused by the temperature and humidity (condition of the medium) of the measurement space can be detected in real time and used as a correction coefficient for triangulation calculations.

[0179] The phrase "a combination of beams (5) whose physical properties are the same or different" in claim 3 defines the flexibility of the combination of beams (5) irradiated from multiple oscillating means (2). A combination of "same physical properties" includes the basic embodiment (Figure 11(a)) in which laser light of the same wavelength is irradiated from both the left and right laser oscillators (2a). The advantages of this identical physical property are cost reduction through the commonality of parts and high-precision intersection detection using interference patterns. A combination of "different physical properties" includes (1) laser light on the left and ultrasound on the right, (2) infrared laser on the left and visible laser on the right, and (3) combinations of different wavelengths or frequencies. The advantages of a combination of different physical properties are improved environmental adaptability and measurement redundancy. For example, in a foggy environment, laser light is scattered but ultrasound is transmitted, so using them together allows for the formation of a stable external reference intersection (6). Details of the physical properties of the beams (5) are described in paragraphs 0057 and 0058. In particular, by irradiating an external reference intersection (6) formed in the same space with beams having different physical properties, errors between the electromagnetic wave sensor and the acoustic sensor in detecting the actual intersection coordinates (24) can be separated and identified, dramatically improving the accuracy and reliability of the system self-diagnosis (paragraph 0153).

[0180] According to the detailed configuration of the external reference intersection (6) forming means defined in claim 3, the following remarkable technical effects are achieved. First, the holding member (3) described in paragraph 0060 maintains the known geometric arrangement relationship, so that the position of the oscillating means (2) can be used as a known parameter that does not require calibration, and the accuracy of the triangulation calculation and the sine theorem calculation described in paragraphs 0086 and 0087 is improved. In particular, ensuring the thermal and mechanical stability of the holding member guarantees the absolute reliability of the coordinates of the external reference intersection (6). Second, the use of the beam (5) having rectilinearity and directivity defined in paragraphs 0057 and 0058 clearly defines the external reference intersection (6). Third, the control of mutually non-parallel beams (5) described in paragraph 0061 enables unique determination of three-dimensional coordinates. Fourth, the inclusion of electromagnetic wave beams and sound wave beams realizes a flexible system configuration according to the use environment. Fifth, the combination of beams (5) with the same or different physical properties achieves both simplification of the system and environmental adaptability. Sixth, the combined use of beams with different physical properties provides a self-compensation function that detects optical axis shifts and propagation velocity changes caused by medium changes (temperature, humidity, fog, etc.) in real time, and separates and compensates for error factors in intersection coordinate calculation. Based on these effects, claim 3 asserts the technical advantages in the implementation of the external reference intersection (6) forming means, and clarifies the differentiation from a simple light source.

[0181] Each component of claim 3 corresponds clearly to a specific embodiment described in detail in the modes for carrying out the invention. "Supported by a retaining member to maintain known geometric arrangement relationships" corresponds to the oscillator retaining unit (3) and retaining member (3a) described in paragraph 0060, shown in Figures 11(a) and (b). "Multiple oscillating means capable of emitting beams having straightness or directionality" corresponds to the oscillating means (2) described in paragraph 0057, which emits the beam (5) as defined in paragraph 0058. "Control means for intersecting mutually non-parallel beams in space" corresponds to the oscillator control unit (2c) and oscillating means angle adjustment mechanism (2d) described in paragraph 0061. "Including electromagnetic wave beams and sound wave beams" corresponds to the variety of beams (5) described in paragraph 0057. "A combination of those having the same or different physical properties" corresponds to the embodiments described in paragraph 0057 and this paragraph 0179. The principle of distance calculation is described in detail in paragraphs 0086 and 0087. This configuration ensures that claim 3 provides absolute accuracy and robustness to environmental changes in the three-dimensional coordinates of the external reference intersection (6) from both hardware and control perspectives. Thus, all components of claim 3 are specifically disclosed in the embodiments and satisfy the enablement requirement.

[0182] Claim 4 is a claim dependent on Claim 3 and defines a function for controlling the irradiation timing of the beam (5) irradiated from the oscillating means (2), thereby achieving power saving, countermeasures against imitation, and flexible operation mode switching based on user instructions. Claim 4 further limits the irradiation timing control modes to three: "continuous irradiation," "intermittent irradiation," and "user-instructed irradiation," in addition to the basic configuration of the external reference intersection (6) forming means defined in Claim 3. The continuous irradiation mode is suitable for applications requiring continuous eye tracking (VR games, driving assistance), the intermittent irradiation mode is suitable for applications requiring power-saving operation (long-term wear, battery operation), and the user-instructed mode is suitable for applications requiring the transmission of intentional eye-tracking commands or privacy protection. This diverse irradiation timing control enables optimal system operation according to usage conditions, making patent circumvention difficult. In particular, this control function modulates the pulse width or frequency of the beam (5), forming the basis of a communication protocol for optically transmitting digital commands independent of eye-tracking information. The components of Claim 4 will be described in detail below. (See Figure 4)

[0183] The structural requirement "capable of controlling beam irradiation timing" in claim 4 is implemented by the oscillator control section (2c) (paragraph 0061) and the beam intensity control section (2f). Irradiation timing control means dynamically controlling the irradiation time, duration and irradiation interval of the beam (5) emitted from the oscillation means (2). This control is integrated by the time-division control of the master scheduler section (44a) described in paragraph 0105 and the power saving control of the power management section (44d) described in paragraph 0108. The technical significance of irradiation timing control lies in: (1) reducing power consumption by stopping irradiation of the beam (5) when it is not needed; (2) anti-counterfeiting measures where the beam (5) is irradiated only during initial calibration and subsequent operation is performed only by the gaze tracking sensor (7); (3) transmitting gaze input commands based on the intentional instructions of the user (9); (4) realizing selective irradiation of the beam (5) for privacy protection. These functions improve the practicability, safety and anti-counterfeiting difficulty of the system. Furthermore, the irradiation timing functions as spatiotemporal duty cycle control to uniformize the thermal load of the system, and minimizes fluctuations in the geometric arrangement relationship caused by thermal expansion of the holding member (3) (see paragraph 0174).

[0184] The "continuous" illumination mode is an operating mode in which the oscillating means (2) continuously illuminates the beam (5) and constantly forms an external reference intersection (6). In this mode, the triangulation calculation unit (8a) described in paragraph 0086 and the sine rule calculation unit (8b) described in paragraph 0087 continuously calculate the coordinates of the beam intersection (6) in synchronization with the frame rate (60 to 240 Hz) of the eye-tracking sensor (7). The continuous illumination mode is suitable for applications that require tracking of high-speed eye movements (saccades (103)) or applications that require continuous error evaluation by the real virtual intersection coordinate matching unit (8s) described in paragraph 0102. The oscillator control unit (2c) performs illumination angle control described in paragraph 0061 every frame and moves the external reference intersection (6) in accordance with changes in the line of sight direction. In continuous illumination mode, the eye rotation center dynamic update unit (8m), described in paragraph 0099, operates most effectively, achieving continuous optimization of the eye rotation center (13). However, continuous illumination consumes the most power, so battery level monitoring by the power management unit (44d), described in paragraph 0108, is important. The true technological advantage of continuous illumination mode lies in the fact that the coordinates of the external reference intersection (6) are continuously referenced as absolute coordinates in response to image noise and ambient light fluctuations of the eye-tracking sensor (7), ensuring that tracking accuracy does not deteriorate even momentarily.

[0185] The "intermittent" irradiation mode is an operating mode in which the oscillating means (2) intermittently irradiates the beam (5), alternating between irradiation and non-irradiation periods. This mode is controlled by the power management unit (44d) described in paragraph 0108 to maximize battery life. Typical intermittent irradiation patterns include (1) intermittent irradiation at a fixed period (e.g., 100 milliseconds irradiation, 900 milliseconds pause, repeated), (2) irradiation only when the gaze tracking sensor (7) detects a gaze state, and (3) adaptive patterns in which the gaze determination unit (44i) described in paragraph 0113 reduces the irradiation frequency when it detects fixation and increases the irradiation frequency when it detects a saccade (103). In the intermittent irradiation mode, the gaze vector (14) is calculated only by the gaze tracking sensor (7) during non-irradiation periods, and the eye rotation center (13) is updated by matching with the external reference intersection (6) during irradiation periods. This method reduces power consumption by 30 to 70% compared to continuous irradiation while avoiding a significant degradation in eye-tracking accuracy. The time interval control unit (21a) of the time division control means (21) (see paragraph code list) manages the irradiation timing. In particular, in the adaptive irradiation pattern, the velocity or acceleration of the saccade (103) is predicted, and the irradiation frequency is dynamically increased immediately before the start of the predicted eye movement, thereby pinpointing and compensating for fluctuations in the center of eye rotation (13) during high-speed movement.

[0186] The configuration requirement that the irradiation timing can be controlled "based on user instructions" defines a function that controls the operation of the oscillating means (2) by the active will of the user (9). This function is realized by the command input analysis unit (44g) described in paragraph 0111 and supports multiple input modalities. The main input means include (1) tapping the smart ring (48a), (2) voice commands from the user (9) acquired from the voice input unit (51b), (3) blinking rhythm patterns detected by the eye-tracking sensor (7), (4) specific eye movement patterns (e.g., rapid left-right reciprocating motion), and (5) button operations on the operation panel (51). For example, when the user (9) double-tap the smart ring (48a), the oscillating means (2) is activated to irradiate the beam (5) and an external reference intersection (6) is formed. When a specific blinking pattern (e.g., two consecutive blinks) is performed, a command signal is transmitted to an external device (20) along with gaze direction information. These user-instructed controls ensure privacy protection (prevention of unintended gaze data transmission) and the transmission of intentional gaze input commands. In this control, the command input analysis unit (44g) includes a redundant command mechanism that requires the simultaneous or consecutive occurrence of two or more input modalities (e.g., tapping the smart ring and fixation on the object being gazed upon) as authentication conditions to ensure the reliability of user instructions.

[0187] The irradiation timing control function also plays an important role as a countermeasure against patent imitation. In certain embodiments, the beam (5) is irradiated only during the initial calibration at system startup (calibration by the calibration process control unit (44c) described in paragraph 0107), and the eyeball rotation center (13) is estimated with high accuracy using the calibration jig (120) described in paragraph 0123. After calibration is complete, the oscillation means (2) is stopped, and eye tracking continues using only the eye-tracking sensor (7). This operating mode leverages the advantages of dynamic calibration using an external reference intersection (6) (improved initial calibration accuracy) while eliminating the need for the oscillation means (2) during normal use, thereby differentiating it from cases where an imitator only imitates the "continuous beam irradiation" embodiment. Furthermore, the embodiment in which the beam (5) is irradiated only when explicitly instructed by the user (9) is superior from the standpoint of protecting privacy and serves as a clear differentiating factor from the continuous irradiation method. The security management unit (44l) (paragraph 0116) detects unauthorized irradiation requests and works in conjunction with the serial code generation unit (41) and the counterfeit detection unit (43) to prevent the use of counterfeit products. The highly accurate eyeball rotation center (13) obtained in this initial calibration is encrypted and stored as a geometric security key in the calibration data storage unit (130a). If the tracking accuracy subsequently falls below a threshold (drifts), the unit forces re-irradiation of the beam (5) and requests re-verification of the key.

[0188] The irradiation timing control based on user (9) instructions, when combined with line-of-sight direction information, enables the transmission of diverse information signals. For example, when a user (9) taps the smart ring (48a) while fixating on a specific object, (1) the coordinates of the current fixation point (15), (2) the ID of the object being fixed on, (3) the timing information of the tap, and (4) the beam (5) irradiation pattern (short-time irradiation, long-time irradiation, Morse code-like pattern) are integrated into a command signal and transmitted to an external device (20) via the communication means (19). The beam (5) irradiation pattern itself can also be used as a modulation signal, and the modulation signal assignment unit (42) encodes the information into an on / off pattern of irradiation. This method allows for diverse command representation in a three-dimensional or more information space by adding time-dimension information such as irradiation timing to two-dimensional information such as line-of-sight direction. The application linkage unit (44f) described in paragraph 0110 interprets these multi-dimensional commands and provides them to the external application in an appropriate format. In particular, the modulation signaling unit (42) encodes a command code using pulse width modulation (PWM) or pulse code modulation (PCM) of the beam (5), enabling low-latency optical communication of several kilobits per second (kbps).

[0189] The irradiation timing control function is integrated with the bidirectional communication control unit (19c) and feedback signal processing unit (19d) of the communication means (19) to realize bidirectional communication. The sensing means (62) (sensor) receives optical or acoustic signals transmitted from an external device (20) or another eye-tracking system (1). In addition, sensors such as light, temperature, pressure, position, physical, sound, and barometric pressure sensors can be installed according to various applications. The dedicated light receiving sensor (62) detects the modulated optical signal, and the receiving unit (19b) of the communication means (19) decodes it. The received information is displayed as text or a graphic on the calibration display unit (50a) of the display unit (50), or output as sound by the auditory output unit (131b) (see paragraph code list). This bidirectional communication enables sharing of eye-tracking information among multiple users (9), synchronization of gaze points in collaborative work, and communication with a remote controller (20f). For example, if user A briefly shines the beam (5) while fixating on a specific location, this information is transmitted to user B's system, and user B's display unit (50) displays "User A has indicated coordinates (X, Y, Z)." This two-way communication function can be applied to collaborative work in industrial settings, communication between surgeons in medical surgery, and sharing attention between teachers and students in educational settings. In this system, half-duplex or full-duplex optical communication using a single beam path structure is realized by controlling the beam irradiation (downlink) by the oscillation means (2) and the reception of external signals (uplink) by a dedicated light receiving sensor (62) using time-division multiplexing (TDM).

[0190] In addition to irradiation timing control, a function to dynamically switch the physical characteristics of the beam (5) at the instruction of the user (9) can also be implemented. The oscillator control unit (2c) and beam intensity control unit (2f) switch the directivity of the beam (5) between a "narrowband (high directivity)" mode and a "broadband (low directivity)" mode. In narrowband mode, the directivity and straightness described in paragraphs 0057 and 0058 are high, and the external reference intersection (6) is formed as a clear point. This mode is used when high-precision line-of-sight tracking is required. In broadband mode, the diffusion angle of the beam (5) is increased to illuminate a wider area. This mode is used when wide-area ambient illumination or simultaneous illumination of multiple gaze candidate points is required, rather than precise positioning of the external reference intersection (6). Mode switching is performed by operation of the control panel (51), voice commands, or requests from the application. Furthermore, in a system equipped with both a laser oscillator (2a) and a sound wave oscillator (2b), it is possible to switch between beams (5) with different physical characteristics as described in paragraph 0179. The environmental information acquisition unit (126) can also monitor ambient light conditions and humidity and automatically select the optimal beam type. Furthermore, by dynamically switching the wavelength or modulation frequency of the beam (5), interference between different systems in a multi-user environment can be prevented, and eye-tracking of multiple users can be performed in parallel in the same space.

[0191] The "beam irradiation only when necessary" operating mode is a safety enhancement function realized in conjunction with the eye protection system (38). The automatic shut-off unit (38a), safety distance determination unit (38b), and light intensity monitoring unit (38c), as described in the paragraph code list, continuously monitor the risk of the beam (5) directly irradiating the user's (9) eyeball (10). The face detection unit (37a) and proximity determination unit (37b) of the face recognition system (37) detect when another person's face approaches the front of the wearable device (48) or is present in the user's field of view, and notify the eye protection system (38). If it is determined that another person's (including animals, etc.) eyeball (10) or a designated object is located in the beam (5) irradiation path or is present within the user's field of view, the automatic shut-off unit (38a) immediately stops the irradiation of the beam (5). By combining such safety functions with irradiation timing control by user instruction, compliance with laser safety standards (Class 1 or Class 1M) is ensured. Furthermore, if the user (9) sets the "beam irradiation prohibited" mode for privacy protection, the oscillation means (2) is completely stopped, and eye tracking is performed solely by the eye-tracking sensor (7). The warning information output unit (79) constantly notifies the user (9) of the irradiation status of the beam (5) to prevent unintended irradiation. The eye protection system (38) is equipped with a dual system: a soft-stop function that instantly reduces the intensity of the beam (5) to a non-harmful level defined by the safety distance determination unit (38b), in addition to stopping the irradiation of the beam (5); and a hardware-based shutoff function using a mechanical shutter or electro-optic modulator.

[0192] In intermittent illumination mode and user-instructed illumination mode, eye tracking must continue even during periods when the beam (5) is not irradiating, so the operation of the eye rotation center dynamic update unit (8m) described in paragraph 0099 is important. During periods of non-irradiation, the eye-line vector (14) is calculated based on the current eye rotation center (13) from the pupil position acquired by the eye-tracking sensor (7). During illumination, the actual virtual intersection coordinate matching unit (8s) described in paragraph 0102 evaluates the discrepancy between the virtual intersection coordinates (25) on the extension of the eye-line vector (14) and the actual intersection coordinates (24) of the beam intersection (6). Based on this discrepancy information, the eye rotation center dynamic update unit (8m) updates the eye rotation center (13) using (Equation 31)(Equation 32). The updated eye rotation center (13) is used to calculate the eye-line vector (14) during the next non-irradiation period. In this way, by alternating between irradiation and non-irradiation periods, continuous optimization of the eyeball rotation center (13) and power-saving operation are achieved simultaneously. The dynamic synchronization control unit (71) synchronizes the irradiation timing with the frame rate of the gaze tracking sensor (7) and prioritizes processing data in the irradiated frames. The eyeball rotation center dynamic update unit (8m) applies statistical filtering techniques such as a Kalman filter or particle filter to predict the eyeball rotation center (13) during the non-irradiation period, and uses highly accurate external reference intersection data acquired during the irradiation period as a filter update step, thereby achieving both accuracy and power saving at the algorithmic level.

[0193] The irradiation timing control function defined in claim 4 achieves the following remarkable technical effects: Firstly, the intermittent irradiation control by the power management unit (44d) described in paragraph 0108 improves battery life by 30 to 70%. Secondly, the anti-imitation measure of irradiating the beam (5) only during the initial calibration clearly differentiates it from the continuous irradiation method, making patent circumvention difficult. In particular, a self-check function is realized that uses the eyeball rotation center as a geometric security key. Thirdly, the user-instructed irradiation control by the command input analysis unit (44g) described in paragraph 0111 enables multidimensional command input combining gaze direction and irradiation timing. Fourthly, the bidirectional communication function enables sharing of gaze information and collaborative work among multiple users. Fifthly, dynamic switching of beam characteristics enables optimal system operation according to usage conditions. Sixthly, by coordinating with the eye protection system (38), laser safety is improved by irradiating only when necessary. Seventh, by predicting the center of eye rotation during intermittent irradiation using a Kalman filter, etc., it is possible to minimize the decrease in accuracy during non-irradiation periods. *These technical effects provide claim 4 with superior protection in terms of practicality, safety, difficulty of imitation, and wide range of applications.

[0194] Each component of claim 4 clearly corresponds to a specific embodiment described in detail in the modes for carrying out the invention. "Controllable beam irradiation timing" is achieved by the oscillator control unit (2c) described in paragraph 0061 and the power management unit (44d) described in paragraph 0108. The "continuous" irradiation mode is described in paragraph 0184, and the "intermittent" irradiation mode is described in paragraph 0185. Control "based on user instructions" is achieved by the command input analysis unit (44g) described in paragraph 0111, which supports multiple input means such as a smart ring (48a), voice input unit (51b), and blink detection. Anti-counterfeiting functions are described in paragraph 0187, gaze input commands in paragraph 0188, and bidirectional communication in paragraph 0189. Switching of beam characteristics is described in paragraph 0190, coordination with safety functions in paragraph 0191, and coordination with dynamic updating of the eye rotation center in paragraph 0192. The synchronization of irradiation timing and gaze tracking is achieved by the synchronization control unit (8f) described in paragraph 0091. In particular, the coordinated operation of intermittent irradiation and dynamic updating is disclosed at a specific algorithmic level in the embodiment by applying a statistical prediction model such as a Kalman filter in the eye rotation center dynamic updating unit (8m). Thus, all components of claim 4 are specifically disclosed in the embodiment and satisfy the enablement requirement.

[0195] Claim 5 is a claim dependent on Claim 1 and provides a calibration means complementary to automatic calibration by defining a manual calibration function based on an active calibration operation by the user (9). In addition to the geometric calculation of the eyeball rotation center (13) defined in Claim 1, Claim 5 further defines a calibration means that spatially compares and matches three independent coordinate data: (1) the coordinates of a calibration target (Ptarget) that the user (9) has intentionally gazed upon, (2) the coordinates of an external reference intersection (Pbeam), and (3) the coordinates of an assumed gaze point (Pgaze) calculated by the gaze tracking sensor (7), and modifies or redetermines the eyeball rotation center (13) based on the difference. This manual calibration function improves adaptability to individual differences and special usage environments that are difficult to address with automatic calibration, and achieves agreement between the user's (9) subjective gaze perception and the system's estimated value. In particular, this tripartite comparison function acts as an advanced diagnostic tool that can geometrically isolate and quantitatively separate three major error factors: the estimation error of the eyeball rotation center (13), the irradiation angle error of the oscillating means (2), and the optical axis / lens distortion error of the gaze tracking sensor (7). This dramatically improves the reliability and difficulty of imitation of the calibration. The components of claim 5 will be described in detail below with reference to Figure 13.

[0196] The "coordinates of the calibration target that the user intentionally fixates on" in claim 5 corresponds to the fixture reference point (121) of the calibration fixture (120) described in paragraphs 0073 to 0074, or the calibration target displayed on the target presentation unit (23d) described in paragraphs 0160 to 0162. As shown in Figure 14(a), the calibration target is arranged in three-dimensional space (18) as a plurality of fixture reference points (18a, 18b), including a neutral position (18c). The requirement of "intentionally fixated" means the act of the user (9) voluntarily fixating on a particular target, which is confirmed by the fixation state being detected by the fixation determination unit (44i) described in paragraph 0113. The coordinates of the calibration targets are known as three-dimensional coordinates (X_target, Y_target, Z_target) relative to the coordinate origin (4) of the holding unit, and are pre-registered in the calibration data storage unit (130a) described in paragraph 0130. In the manual calibration process, the automatic calibration control unit (23e) described in paragraph 0167 sequentially presents the targets to the user (9), and records data when fixation on each target stabilizes. These known coordinates are used as reference values ​​in the comparison and matching process described later. In particular, the absolute coordinates (Ptarget) of the targets are determined as compensated absolute coordinates that compensate in real time for thermal expansion and minute deformation of the calibration jig based on the measurement results of the temperature sensor and humidity sensor built into the calibration jig (120), and the reliability of the reference point is guaranteed in a manner that does not depend on external high-precision measuring instruments.

[0197] The "coordinates of the external reference intersection" in claim 5 are the three-dimensional coordinates of the external reference intersection (6) formed by the oscillation means (2) described in paragraphs 0057 to 0061. As shown in Figure 13(a), the point where the beams (5a, 5b) irradiated from the left and right oscillation means (2a, 2b) intersect in space is formed as the external reference intersection (6) (beam intersection PC). The coordinates (X_beam, Y_beam, Z_beam) of this intersection are calculated from the holding unit baseline length (16) and irradiation angles θ_L, θ_R using (Equation 24) by the triangulation calculation unit (8a) described in paragraph 0086 and the sine rule calculation unit (8b) described in paragraph 0087. In the manual calibration process, when the user (9) gazes at the calibration target, the oscillator control unit (2c) illuminates the beam (5) in the direction of the line of sight using the illumination direction correction mechanism (2e) described in paragraph 0061. Ideally, the coordinates of the external reference intersection (6) should coincide with the coordinates of the calibration target, but spatial differences occur due to line-of-sight estimation errors and illumination angle errors. These differences become important information in the comparison and matching process described later. Furthermore, the coordinates of the external reference intersection (6) maintain high reliability as geometric absolute coordinates by incorporating the simultaneous measurement of the difference in propagation speed between the electromagnetic wave beam and the sound wave beam (see paragraph 0178), thereby compensating in real time for errors in beam refraction and propagation speed caused by the medial conditions (temperature, humidity) of the measurement space.

[0198] The "coordinates of the assumed gaze point calculated by the gaze tracking sensor" in claim 5 correspond to the calculation of the gaze vector (14) described in paragraph 0076 and the gaze point (15) described in paragraph 0077. As shown in Figure 13(a), the gaze direction calculation unit (8n) described in paragraph 0100 calculates the gaze vectors (14a, 14b) using (Equation 2) from the pupil position (11a, 11b) detected by the gaze tracking sensor (7) and the current eyeball rotation center (13a, 13b). The point where the extensions of the left and right gaze vectors (14) are closest or virtually intersect is calculated as the assumed gaze point (15), and its coordinates (X_gaze, Y_gaze, Z_gaze) are obtained by (Equation 22) X_gaze=(P_L+P_R) / 2. As indicated by the symbol “Δi error corrected interval” shown at the top of Figure 13(a), this assumed gaze point (15) may spatially deviate from the coordinates of the calibration target that the user (9) is actually fixating on. This deviation is due to estimation errors in the eyeball rotation center (13), inaccuracies in the gaze direction correction coefficient δ_calib, or measurement errors in the eye-tracking sensor (7). The calculation of the assumed gaze point (15) is an estimated coordinate that includes physiological and optical uncertainties such as the inaccuracy of the initial estimate of the eyeball rotation center (13), lens distortion aberration of the eye-tracking sensor (7), pupil ellipse approximation errors, and user-specific corneal shape heterogeneity. Therefore, a difference comparison with the absolute coordinates Ptarget of the calibration target is essential.

[0199] The process of "comparing and matching spatial differences" in claim 5 is performed by the difference analysis unit (8d) described in paragraph 0089 and the actual virtual intersection coordinate matching unit (8s) described in paragraph 0102. The spatial differences between the three coordinates shown in Figure 13(a), namely (1) known coordinates of the calibration target (Xtarget, Ytarget, Ztarget), (2) measured coordinates of the external reference intersection (6) (Xbeam, Ybeam, Zbeam), and (3) calculated coordinates of the assumed gaze point (15) (Xgaze, Ygaze, Zgaze), are quantitatively evaluated. The main difference vector includes the deviation Δi = (Xgaze-Xbeam-Ygaze-Ybeam-Zgaze-Zbeam) between the virtual intersection coordinates (25) and the actual intersection coordinates (24) described in paragraph 0153, and the magnitude of the deviation is calculated by (Equation 27). Furthermore, the difference between the calibration target and the assumed gaze point, ΔT=(Xtarget-Xgaze-Ytarget-Ygaze-Ztarget-Zgaze), and the difference between the calibration target and the external reference intersection, ΔB=(Xtarget-Xbeam-Ytarget-Ybeam-Ztarget-Zbeam), are also calculated. ΔT is the overall error of the eyeball model, mainly due to inaccuracies in the center of eyeball rotation (13). ΔB is the overall error of the external reference system, mainly due to inaccuracies in the irradiation angle control of the oscillation means (2). Δi is the internal consistency error and serves as the basis for determining whether ΔT or ΔB is dominant. By comparing and matching these vectors (ΔTvsΔBvsΔi) rather than using them individually, it becomes possible to geometrically separate two major error factors—the estimation error of the eyeball rotation center and the error of the beam irradiation angle—which is a sophisticated self-diagnostic function not seen in conventional techniques. This eliminates the risk of correcting incorrect parameters and maximizes the efficiency and accuracy of the correction.

[0200] Furthermore, the present configuration also improves dynamic tracking performance. The description of "adaptive irradiation pattern" indicates that the present technology is not limited to static calibration. Specifically, the prediction of saccades (103) (pre-estimation of velocity and acceleration) identifies the moment at which the center of rotation of the eyeball (13) varies the most and at the highest speed. The beam irradiation frequency (sampling rate) is dynamically increased immediately before such variation occurs. This is a technology for dramatically improving the accuracy of "dynamic calibration" or "dynamic update", and enables the variation of the center of rotation of the eyeball (13) to be pinpointed and reflected in the correction parameter (ΔT). This "prediction-based measurement optimization" ensures robust dynamic update of the center of rotation of the eyeball even during high-speed eye movements where tracking accuracy is most likely to decrease. This greatly supports the technical effect of the claimed invention in terms of "ensuring long-term stability of high-precision eye tracking".

[0201] The process of "correcting or redetermining the center of rotation of the eyeball based on the result of the comparison and collation" in claim 5 is executed by the back calculation processing unit (8e) described in paragraph 0090 and the dynamic center of eyeball rotation updating unit (8m) described in paragraph 0099. The correction direction and correction amount for the center of rotation of the eyeball (13) are determined based on difference information obtained through the comparison and collation. The "correction" is a process of applying stepwise correction to the current center of rotation of the eyeball (13), and is executed by the update formulas shown in (Formula 31) and (Formula 32). Specifically, the correction vector δ_C of the center of rotation of the eyeball (13) is calculated from the difference vector ΔT between the calibration target and the assumed gaze point, and updated according to C_eye^(new)=C_eye^(old)+α·δ_C. "Redetermination" is a process of discarding the current center of rotation of the eyeball (13), and re-estimating the center of rotation of the eyeball (13) from the initial state by solving the optimization problem (Formula 1) (Formula 6) described in paragraph 0121 again from pupil position data of a plurality of calibration targets. The reverse calculation processing unit (8e) performs correction or re-determination of the eyeball rotation center (13) only if the spatial coordinate error separation unit (129) determines that the estimation error of the eyeball rotation center (13) caused by the difference vector ΔT is the main error source (e.g., the total error contribution is 70% or more). If other errors (beam or sensor optical axis caused by the difference vector ΔB) are the main cause, the reverse calculation processing unit (8e) prioritizes oscillator control or sensor correction. The choice between correction and re-determination is adaptively determined based on whether the magnitude of the overall spatial difference (||ΔT||) exceeds the correction convergence threshold and the contribution of the error source.

[0202] The calibration means gradually improves the estimation accuracy of the eyeball rotation center (13) by iteratively performing a comparison and matching process on multiple calibration targets. The automatic calibration control unit (23e), described in paragraph 0167, presents the user (9) with a first calibration target and records three coordinates after gaze stabilization. After the eyeball rotation center (13) is corrected by comparison and matching, a second calibration target is presented and the same process is repeated. As shown in Figure 14(a), by sequentially gazing on multiple jig reference points (18a, 18b) distributed across the entire field of view, the error characteristics dependent on the gaze direction become apparent. The arc fitting unit (23a), described in paragraph 0163, and the spherical fitting, described in paragraph 0164, process data from multiple targets in an integrated manner to optimize the eyeball rotation center (13). The convergence determination unit (68) monitors the convergence conditions shown in (Equation 15) and (Equation 16), and determines that calibration is complete when the difference between the calibration target and the assumed gaze point falls below the threshold ε for all targets. This convergence determination, in addition to the spatial difference falling below the threshold, incorporates as an additional convergence condition that the user's (9) eye movement pattern during the calibration process (gaze stability, saccade speed, pupillary sway) does not show any abnormal tendencies, thereby detecting user fatigue and intentional misoperation and ensuring the reliability of the calibration results. After calibration is complete, the updated eye rotation center (13) and gaze direction correction coefficient δ_calib are stored in the calibration data storage unit (130a) described in paragraph 0130.

[0203] The calibration means defined in claim 5 can be integrated with the manual calibration correction described in paragraph 0168. After the eye rotation center (13) is determined by the automatic calibration process, if the user (9) subjectively feels that the gaze direction is off, the calibration target is presented again via the control panel (51) to confirm the display position of the assumed gaze point (15). The calibration presentation display unit (50a) of the display unit (50) visually displays the coordinates of the current assumed gaze point (15), allowing the user (9) to confirm the discrepancy with the position they are actually looking at. When the user (9) inputs "shifted upward," the gaze direction vector correction unit (70) adjusts the gaze direction correction coefficient δ_calib, and the eye rotation center correction unit (117b) fine-tunes the eye rotation center (13). This manual correction input by the user is reflected only if it is logically consistent with the contribution of the error factor calculated by error separation (paragraph 0199). If there is a contradiction (e.g., the system diagnoses beam error (ΔB), but the user points out ERC error (ΔT)), the diagnostic unit (44m) is notified, and a highly reliable feedback mechanism is formed that is used as training data for system self-diagnosis. The correction value obtained by this manual correction is also integrated into the comparison and matching results and used as the initial value for the next calibration. The combined use of manual correction and automatic calibration achieves high adaptability to individual differences and subjective differences in gaze perception.

[0204] The calibration means defined in claim 5 operates in coordination with the eye rotation center dynamic update unit (8m) described in paragraph 0099. Initial calibration by the calibration means determines a highly accurate initial value of the eye rotation center (13). During subsequent normal use, the dynamic update unit (8m) continuously optimizes the eye rotation center (13) based on error information from the actual virtual intersection coordinate matching unit (8s) described in paragraph 0102. If the mounting position shifts due to prolonged use, or if the eye characteristics change due to user (9) fatigue, large errors that are difficult to address by dynamic updates alone may accumulate. In such cases, the diagnostic unit (44m) described in paragraph 0117 detects a decrease in eye-tracking accuracy and commands the calibration process control unit (44c) to recalibrate. The highly accurate eye rotation center (13) determined by the calibration means functions as the initial state or a reliable constraint (reduction of noise dispersion) for statistical filtering algorithms such as Kalman filters and particle filters in the dynamic update process. Furthermore, in cooperation with the security management unit (44l), this highly accurate initial value is also used as a geometric reference point to detect unauthorized parameter changes during dynamic updates (e.g., external rewriting of calibration parameters), thereby enhancing both the security and reliability of the system. During recalibration, the calibration means is activated again, and the eye rotation center (13) is reset by comparison with the calibration target. This coordination between the calibration means and dynamic updates realizes a two-stage adaptive mechanism in which dynamic updates respond to short-term fluctuations, and the calibration means responds to long-term fluctuations or large deviations.

[0205] The performance of the calibration means is quantitatively evaluated by the magnitude of the spatial difference after calibration. At the time of calibration completion, the root mean square (RMS) of the difference between the calibration target and the assumed gaze point is calculated as an indicator of eye-tracking accuracy. Due to the output data quality assurance function described in paragraph 0140, the magnitude of the difference vector ||ΔT|| for each calibration target is recorded, and the average value for all targets is reported as the calibration accuracy. Typical calibration accuracy is 0.3 to 0.5 degrees in the center of the field of view and 0.5 to 1.0 degrees in the periphery of the field of view. If the calibration accuracy does not reach the target value, the calibration correction processing unit (23) takes measures such as increasing the number of calibration targets or changing the arrangement pattern of the targets. In particular, as evaluation metrics for calibration accuracy, in addition to the RMS of all targets, a spatial stability index that evaluates the directional consistency (variability) of the difference vector ΔT and a time-series stability evaluation that quantitatively tracks the rate of deterioration (drift rate) of eye-tracking accuracy over time after calibration completion are introduced. The history of calibration accuracy is stored in the measurement history storage unit (130d) described in paragraph 0130, and the time-series accuracy deterioration is monitored by the long-term variation detection unit (125) described in paragraph 0125. If a trend of decreasing calibration accuracy over time is detected, it suggests hardware deterioration or a change in the user's visual characteristics, and maintenance inspection or medical examination is recommended.

[0206] The manual calibration function defined in claim 5 achieves the following remarkable technical effects: Firstly, by comparing and matching three independent coordinate data sets—the known coordinates of the calibration target, the measured coordinates of the external reference intersection (6), and the calculated coordinates of the assumed gaze point (15)—the estimation error of the eyeball rotation center (13) and the beam irradiation error can be separated, improving the accuracy of spatial coordinate error separation as described in paragraph 0102. Secondly, by analyzing the three difference vectors ΔT, ΔB, and Δi, the system provides advanced self-diagnostic information that allows for the separation of the contribution of error factors as a percentage, thereby dramatically improving the system's self-correction (paragraph 0201) and diagnostic capabilities (diagnostic unit (44m)). Thirdly, manual calibration, in which the user (9) actively fixates on the calibration target, can accommodate individual differences and subjective differences in gaze perception that are difficult to address with automatic calibration. Fourth, the combination of arc fitting and spherical fitting described in paragraphs 0163 to 0164 enables the integrated processing of data from multiple visual targets and allows for the statistically robust estimation of the center of ocular rotation (13). Fifth, the coordination of the calibration means and dynamic updates ensures both initial calibration accuracy and long-term accuracy maintenance. Sixth, the quantitative evaluation of calibration accuracy and the evaluation of stability over time objectively guarantee the reliability of the system. These technical effects improve the practicality and reliability of high-precision eye tracking.

[0207] Each component of claim 5 corresponds clearly to a specific embodiment described in detail in the modes for carrying out the invention. "Coordinates of the calibration target that the user is intentionally fixated on" corresponds to the calibration jig (120) and jig reference point (121) described in paragraphs 0073 to 0074 and 0160 to 0162, and is shown in Figure 14(a). "Coordinates of the external reference intersection" corresponds to the external reference intersection (6) formed by the oscillating means (2) described in paragraphs 0057 to 0061, and is calculated by triangulation and sine rule calculations described in paragraphs 0086 and 0087. "Coordinates of the assumed gaze point calculated by the gaze tracking sensor" corresponds to the gaze vector (14) and gaze point (15) described in paragraphs 0076 to 0077, and is calculated by (Equation 2)(Equation 22). "Comparing and matching spatial differences" is realized by the difference analysis unit (8d) described in paragraph 0089 and the actual virtual intersection coordinate matching unit (8s) described in paragraph 0102, as shown in Figure 13(a). In particular, the multivariable error discrimination model based on three coordinates corresponds to the implementation of a discriminator using a linear algebraic method or machine learning in the spatial coordinate error separation unit (129) described in paragraph 0129. "Calibration means for correcting or redetermining the eyeball rotation center" corresponds to the inverse calculation processing unit (8e) described in paragraph 0090 and the eyeball rotation center dynamic update unit (8m) described in paragraph 0099, and is realized by (Equation 1), (Equation 31), and (Equation 32). Thus, all components of claim 5 are disclosed in concrete and technical chains in the embodiments, satisfying the enablement requirement.

[0208] Claim 6, dependent on Claim 1, specifies a configuration in which the information analysis means (8) dynamically corrects at least one of the three-dimensional coordinates of the eyeball rotation center (13) or the three-dimensional coordinates of the eyeball feature points included in the eyeball information, based on a correction amount obtained from the error with the external reference intersection (6). This claim enhances the system's self-correction function and protects technical features that enable long-term accuracy maintenance. This dynamic correction compensates for cumulative errors such as minute misalignment of the wearer and time-dependent drift of the eyeball model in real time, forming the core of closed-loop control that realizes the system's requirement for "no calibration required" or "reduced calibration frequency."

[0209] The main components of this claim are as follows: • Dynamic updating unit (8m) of the eyeball rotation center within the information analysis means (8) (see paragraph 0099) • Actual virtual intersection coordinate matching unit (8s) (see paragraph 0102) • Difference analysis unit (8d) (see paragraph 0090) • Error correction integration unit (8g) (see paragraph 0095) • Eye rotation center calculation unit (8i) (see paragraph 0097) These components form an ultra-low latency processing pipeline that performs error detection (8s) and parameter correction (8m) in each frame of eye tracking (typically less than 10ms), keeping eye tracking accuracy constantly optimal.

[0210] The basic principle of dynamic correction is explained below. The real-virtual intersection coordinate matching unit (8s) compares and matches the real intersection coordinates (24) physically formed by the oscillation means (2) with the virtual intersection coordinates (25) geometrically calculated from the current eyeball parameters (see paragraph 0154). The spatial difference vector δ obtained by this matching is expressed by (equation 28): δ = V_real - V_virtual, where V_real is the coordinate of the real intersection coordinates (24) and V_virtual is the coordinate of the virtual intersection coordinates (25). This difference vector $\delta$ functions as a feedback signal that quantitatively indicates in three-dimensional space the degree to which the parameters of the eyeball model do not satisfy the geometric constraint of the external reference intersection (absolute coordinates).

[0211] The difference analysis unit (8d) analyzes the spatial difference vector δ and determines whether its cause is due to a misalignment of the eyeball rotation center (13) or to detection errors of eyeball feature points (pupil center (11c), corneal curvature center (114), etc.). This determination involves matching the consistency of the gaze direction vector, the time-series pattern of errors across multiple frames, and physiological constraints (see paragraphs 0124 to 0125). This determination of the error cause is a sophisticated diagnosis that geometrically isolates the source of the error vector, enabling intelligent processing to logically determine which parameters of the eyeball model should be modified.

[0212] The dynamic correction of the eyeball rotation center (13) is described below. The eyeball rotation center dynamic update unit (8m) calculates the correction amount δ_C of the eyeball rotation center based on the difference vector δ. This correction amount is calculated according to (Equation 32): δ_C = β·ΔP - γ·(ΔP·v_gaze)·v_gaze, where ΔP is the measurement error vector, v_gaze is the line of sight direction vector, and β and γ are weighting coefficients. This equation decomposes the error vector into a line of sight direction component and a vertical component, and corrects each with different weightings to achieve a physiologically valid correction. In particular, by separating and correcting the line of sight direction component, the phenomenon of correction instability due to the strong correlation between the line of sight direction (gaze angle) and the eyeball rotation center (position) is prevented, and the stability and convergence of the correction calculation are guarantee...

Claims

1. An eye-tracking system comprising: an eye-tracking sensor for acquiring user eyeball information; means for forming an external reference intersection; and information analysis means for geometrically calculating the center of eyeball rotation by associating the external reference intersection with the eyeball information, wherein the information analysis means determines the physiological validity of the calculated center of eyeball rotation, evaluates the error between the assumed gaze point derived from the center of eyeball rotation and the external reference intersection, and determines or updates the center of eyeball rotation to reduce the error while satisfying geometric constraints in the eyeball model.

2. In the eye-tracking system according to claim 1, In addition to calibration using the external reference intersection, the information analysis means provides: Based on the reference point of a calibration jig with known three-dimensional coordinates, The center of rotation of the eyeball is calculated An eye-tracking system characterized by the following features.

3. In the eye-tracking system according to claim 1, The means for forming the external reference intersection is supported by a retaining member so as to maintain a known geometric arrangement relationship, Multiple oscillation means capable of irradiating a beam having straight-line propagation or directionality as a continuous wave or pulse wave, Control means for intersecting mutually non-parallel beams irradiated from the plurality of oscillation means in space, Includes, The beam includes at least one of an electromagnetic wave beam and a sound wave beam, and is a combination of those having the same or different physical properties. An eye-tracking system characterized by the following features.

4. An eye-tracking system according to claim 3, characterized in that the control means can control the irradiation timing of the beam continuously, intermittently, or based on user instructions.

5. An eye-tracking system according to claim 1, wherein the information analysis means comprises calibration means that compares and verifies the spatial difference between the coordinates of a calibration target that the user has intentionally gazed upon, the coordinates of the external reference intersection, and the coordinates of the assumed gaze point calculated by the eye-tracking sensor, and corrects or re-determines the eyeball rotation center based on the result of the comparison and verification.

6. An eye-tracking system according to claim 1, wherein the information analysis means dynamically corrects at least one of the three-dimensional coordinates of the eyeball rotation center and the three-dimensional coordinates of the eyeball feature points included in the eyeball information, based on a correction amount obtained from the error with the external reference intersection.

7. An eye-tracking system according to claim 1, wherein the information analysis means analyzes the spatial difference between the coordinates of the external reference intersection and the coordinates of the assumed gaze point calculated by the eye-tracking sensor as a three-dimensional coordinate vector.

8. An eye-tracking system according to claim 1, characterized in that the information analysis means stores the relative positional relationship between the center of eye rotation and the feature points of the eye as a personal template for biometric authentication and uses it for authentication.

9. In the eye-tracking system according to claim 1, The means for forming the external reference intersection forms the external reference intersection The system includes means for adding unique identification information to the beam, The aforementioned identification information includes a temporal modulation pattern, frequency characteristics, or Includes at least one of the encoded pieces of information, The information analysis means performs a matching based on the identification information The authenticity of the aforementioned external reference intersection is determined, and non-normal signals are excluded. Features include detecting counterfeit products or external interference. Eye-tracking system.

10. An eye-tracking system according to claim 1, wherein the information analysis means converts the coordinates of the eyeball rotation center and the point of fixation to a fixed coordinate system established by an Earth-fixed coordinate system, a global coordinate system, or a local environment mapping based on information from an external coordinate assignment system, and cooperates with an external system.

11. An eye-tracking system according to claim 1, further comprising depth measuring means, wherein the information analysis means synchronously controls the gaze direction calculated by the eye-tracking sensor and the depth information acquired by the depth measuring means to determine the gaze coordinates in three-dimensional space.

12. In the eye-tracking system according to claim 1, The information analysis means uses the actual intersection coordinates of the external reference intersection and The position and irradiation direction of the oscillation means that form the external reference intersection. Based on this, the central axis of the beam is geometrically extended and derived. Compare with the virtual intersection coordinates, Based on that difference, the error factors form the external reference intersection. The displacement of the oscillation means or the drift of the eyeball model To estimate or determine which one it is, An eye-tracking system characterized by having a self-diagnostic function that selectively corrects.

13. An eye-tracking system according to claim 1, further comprising an inertial measurement unit (IMU), wherein the information analysis means compensates for the correction of the eyeball rotation center based on positional displacement or deformation information of the mounting device detected by the inertial measurement unit.

14. In the eye-tracking system according to claim 1, The information analysis means includes a temperature sensor, a humidity sensor, or Based on long-term fluctuation information of the internal clock, Thermal expansion, contraction, or It detects errors caused by age-related drift in electronic components. The aforementioned error is due to the misalignment of the oscillation means that forms the external reference intersection. If this is the cause, the arrangement coordinates or irradiation direction of the oscillation means Correct, If the aforementioned error is due to the drift of the eyeball model To compensate for the coordinates of the center of rotation of the eyeball A key feature is its eye-tracking system.

15. The eye-tracking system according to claim 1, wherein the information analysis means comprises a model that dynamically corrects the baseline length between the centers of eye rotation based on the gaze direction of both eyes and the rotational movement (torsion) of the eyeballs.

16. An eye-tracking system according to claim 1, characterized in that the information analysis means recalculates the center of eye rotation using a calibration target on which known coordinates are set.

17. An eye-tracking system according to claim 1, wherein the information analysis means analyzes the time-series fluctuations of the eyeball rotation center or the long-term fluctuation patterns of the external reference intersection using statistical filtering or a machine learning algorithm, and corrects the eyeball rotation center.

18. An eye-tracking system according to claim 1, wherein the system is mounted on a wearable image display device, and the information analysis means further includes a foveated rendering control unit that divides the display screen into a central field of view, a peripheral field of view, and an outer field of view based on the gaze direction derived from the eyeball rotation center, and renders the central field of view at high resolution, the peripheral field of view at medium resolution, and the outer field of view at low resolution, and a VR sickness reduction control unit that predicts the vestibulo-ocular reflex from the time-series fluctuation pattern of the eyeball rotation center and corrects the motion vector of the displayed image.

19. It is an eye-tracking method, A process of forming an external reference intersection by intersecting beams having straight-line or directional properties, emitted from multiple oscillating means, A process of acquiring the user's eyeball information using an eye-tracking sensor, A step of geometrically calculating the center of rotation of the eyeball by associating the aforementioned external reference intersection with the aforementioned eyeball information, A step of determining the physiological validity of the calculated center of eye rotation, A step of evaluating the error between the assumed point of fixation derived from the center of eye rotation and the external reference intersection, A step of determining or updating the center of rotation of the eyeball in such a way as to reduce the error while satisfying the geometric constraints in the eyeball model, Based on the line of sight direction derived from the center of eye rotation, the display screen is divided into central field of view, peripheral field of view, and outer field of view, and the rendering resolution of each field of view region is dynamically adjusted. A method for tracking eye movements, characterized by including the following:

20. In the eye-tracking system according to Claim 1, The information analysis means is identified by the eye-tracking sensor Information on the point of focus and instructions based on the user's active will. Combine these to analyze commands sent to external devices. It further includes a command input analysis unit, The command input analysis unit handles operation of the control panel, voice commands, gesture input, Or at least two of the following: specific eye movement patterns or blinking rhythm patterns. The input modalities occur simultaneously or sequentially. As authentication requirements, The actual intersection coordinates of the external reference intersection and the corresponding point of gaze It was determined that the geometric consistency constraint with the virtual intersection coordinates was satisfied. A signal that controls the operation of the external device corresponding to the point of focus. An eye-tracking system characterized by its ability to transmit data.

Images