Multi-user eye tracking method, system, electronic device, and storage medium

Abstract

The application discloses a multi-user eye movement tracking method and system, electronic equipment and storage medium, relates to the technical field of eye movement tracking, and the method comprises the steps of: presetting a reference calibration point and a dynamic reference point; in each image acquisition period, an image of the eyeball of a current user is acquired through a camera module; the infrared camera module is provided with a reference calibration mark and a dynamic reference mark; iris texture features and corneal reflection spot data are extracted based on the eyeball image; a user identity code is calculated, and a three-dimensional line of sight coordinate is generated in combination with a corneal curvature parameter and a pupil center displacement amount; when a multi-user coexistence state is detected, a user association matrix is established, and a user differentiation weight value is calculated to distinguish different users; line of sight focus position data is generated based on the three-dimensional line of sight coordinate, and the infrared light supplement intensity is adjusted according to the dynamic reference mark. The application solves the problem that in a multi-user scene, the eye movement tracking system is limited in tracking accuracy due to the lack of an effective user differentiation mechanism.

Description

Technical Field

[0001] This application relates to the field of eye-tracking technology, and in particular to a multi-user eye-tracking method, system, electronic device, and storage medium. Background Technology

[0002] In the field of human-computer interaction technology, traditional facial recognition technology and mouse and keyboard input are gradually becoming insufficient to meet the needs of high precision and personalization, while eye-tracking technology has become a research hotspot due to its non-contact characteristics and accuracy.

[0003] Current eye-tracking systems are mostly limited to single-user use and cannot effectively identify and track the eye movements of multiple users. Furthermore, their accuracy drops significantly in low-light environments, resulting in a substantial decrease in tracking precision. This limits the widespread application of eye-tracking systems. Therefore, in multi-user scenarios, existing eye-tracking systems suffer from limited tracking precision due to the lack of an effective user differentiation mechanism, leaving room for improvement. Summary of the Invention

[0004] To address the issue of limited tracking accuracy in multi-user scenarios due to the lack of an effective user differentiation mechanism, this application provides a multi-user eye-tracking method, system, electronic device, and storage medium.

[0005] Firstly, the objective of this invention is achieved through the following technical solution: A multi-user eye-tracking method includes: Pre-set baseline calibration points and dynamic reference points in the eye-tracking system; During each frame image acquisition cycle, the current user's eye image is acquired through the infrared camera module; the infrared camera module includes a binocular parallax structure and is equipped with a reference calibration mark that matches the reference calibration point and a dynamic reference mark that adapts to the dynamic reference point; Iris texture features and corneal reflectance data are extracted based on the eyeball image; the benchmark calibration mark is adapted to the pupil center localization algorithm, and the dynamic reference mark is adapted to the gaze heatmap generation module; The user identification code is calculated based on the iris texture features, and three-dimensional line-of-sight coordinates are generated by combining corneal curvature parameters and pupil center displacement. When a multi-user coexistence state is detected, a user association matrix is ​​established by matching iris texture features, and user differentiation weight values ​​are calculated by combining the spatial distribution difference of gaze heatmap to distinguish different users. Based on the three-dimensional line-of-sight coordinates, generate line-of-sight focus position data, and adjust the infrared fill light intensity according to the exposure compensation parameters corresponding to the dynamic reference marker; The matching degree of the gaze focus position data is calculated with the preset interaction threshold area. When the gaze focus of multiple users overlaps, a hierarchical response mechanism is executed, giving priority to responding to the command of the user who first enters a stable gaze state.

[0006] By adopting the above technical solution, a multi-user eye-tracking technology suitable for multi-user interaction scenarios (such as eye-tracking analysis in online education platforms, player gaze positioning in augmented reality (AR) / virtual reality (VR) games, and participant attention monitoring in remote conferencing systems) is provided. The preset reference calibration point is a virtual coordinate origin or reference frame predefined by the system at the software or algorithm level, used as a reference for coordinate system alignment and user gaze calculation during subsequent image processing. The reference calibration mark is a fixed physical mark (such as a light spot of a specific shape, a coded pattern, etc.) on the infrared camera module, whose position and optical characteristics are known. By capturing an image of the reference calibration mark, the position of the physical mark is matched with the preset reference calibration point to complete the camera's intrinsic parameter calibration (such as focal length, distortion coefficient). The dynamic reference point is an auxiliary reference coordinate generated in real time during system operation, used to dynamically adjust tracking parameters (such as exposure time, supplementary light intensity). The dynamic reference mark is a variable light source or adjustable coded pattern (such as a ring-shaped coded light source, a light spot with adjustable brightness) on the infrared module, whose optical characteristics (such as brightness, frequency) can be adjusted in real time according to environmental changes.

[0007] This invention's technical solution possesses multi-user high-precision positioning and identity differentiation capabilities. It employs a dual structure of benchmark calibration points and dynamic reference points: combining a binocular parallax infrared module to eliminate depth ambiguity in monocular vision and improve pupil center positioning accuracy; iris texture features and dynamic weight calculation generate user identification codes based on the uniqueness of iris textures, and dynamically calculate user differentiation weight values ​​by combining the spatial distribution differences of gaze heatmaps (such as hotspot overlap rate and gaze direction angle), solving the identity confusion problem when multiple users coexist; the variable intensity ring-coded light source of the dynamic reference identifier is synchronized with the image acquisition frame rate (e.g., 12) through modulation frequency. The system operates at 0Hz, providing real-time compensation for ambient light interference (such as strong reflections and low light). Based on the brightness information fed back by the dynamic reference marker, it automatically adjusts the intensity of infrared supplementary light to improve the signal-to-noise ratio under low light conditions. Then, through the generation of three-dimensional gaze coordinates and a hierarchical response mechanism, when the gaze focus of multiple users overlaps, it prioritizes responding to the command of the user who first enters a stable gaze state (determined by timestamp and gaze duration threshold), avoiding interaction conflicts. This is suitable for eye-tracking scenarios with high concurrency and high precision requirements. Thus, this invention solves the problem of limited tracking accuracy in multi-user scenarios due to the lack of an effective user differentiation mechanism in eye-tracking systems.

[0008] In a preferred embodiment of this application: the reference calibration mark includes at least three non-collinear positioning markers, and the dynamic reference mark includes a ring-shaped coded light source with variable intensity, the modulation frequency of which is synchronized with the image acquisition frame rate; The execution-level response mechanism includes: Assign a unique timestamp to each user to record the start time of each user entering a stable gaze state; When multiple users' gazes overlap, compare the timestamp values ​​of each user and respond to the command of the user with the smallest timestamp value first. If the timestamp tolerance is less than the preset time tolerance, compare the iris texture matching degree of the user's identity code and select the one with the highest matching degree to execute the instruction.

[0009] By adopting the above technical solution, a hierarchical high-precision response mechanism is provided. The benchmark calibration mark includes at least three non-collinear positioning markers (triangular distribution). The non-collinear positioning markers eliminate calibration errors through geometric constraints, improving the timestamp synchronization accuracy. At the same time, to accurately resolve multi-user interaction conflicts, a timestamp priority mechanism is adopted to ensure that the user's command that first enters a stable gaze state is responded to immediately. Iris texture matching is double-checked within the timestamp tolerance (e.g., ΔT=500ms), reducing the false response rate. By synchronizing the modulation frequency of the ring-coded light source with the frame rate, image blurring or data loss caused by hardware asynchrony is eliminated, ensuring the stability of timestamp acquisition. At the same time, through the dual determination of timestamp and iris texture, user intentions are prioritized (e.g., teacher commands take precedence over student commands). This is suitable for scenarios requiring access control, such as education / medical care, and helps improve the consistency of user experience.

[0010] In a preferred embodiment of this application: when a multi-user coexistence state is detected, a user association matrix is ​​established through iris texture feature matching, and a user differentiation weight value is calculated by combining the spatial distribution difference of the gaze heatmap to distinguish different users, including: When a multi-user coexistence state is detected, a deep learning algorithm is used to match the iris texture features of each user to construct a user association matrix; Based on the gaze heatmap, the overlap rate of hotspot regions and the angle of gaze direction for multiple users are extracted; The spatial distribution difference is calculated by associating the overlap rate of the hotspot areas with the angle of view direction with a preset weight coefficient. The spatial distribution difference is compared with a preset weight threshold to generate a unique user differentiation weight value for each user and associate it to distinguish different users.

[0011] By adopting the above technical solution, the spatial distribution difference is calculated based on the extracted hotspot overlap rate (HOR) and gaze direction angle (GDA). This application uses a deep learning algorithm to match iris texture features, which can significantly improve the accuracy and speed of user recognition, especially in multi-user environments. By analyzing the hotspot overlap rate and gaze direction angle, the system can more finely distinguish the gaze behavior of different users, enhancing the accuracy of user differentiation. The spatial distribution difference, combined with a preset weight threshold, ensures that different users can be efficiently and accurately identified and differentiated even in complex interaction scenarios.

[0012] In a preferred embodiment of this application: the step of adjusting the infrared fill light intensity according to the exposure compensation parameters corresponding to the dynamic reference mark includes: Collect the brightness value of the ring-shaped coded light source of the dynamic reference mark, and calculate the brightness difference between the brightness value of the ring-shaped coded light source and the preset target brightness; The intensity of infrared supplementary light is dynamically adjusted based on the brightness difference and the PID control algorithm. When the brightness difference is less than the preset brightness difference threshold, the infrared supplementary light intensity is locked and adjustment is stopped.

[0013] By adopting the above technical solution, the PID control algorithm reduces the steady-state error of infrared illumination intensity through the coordinated adjustment of proportional (Kp=0.8), integral (Ki=0.2), and derivative (Kd=0.1) parameters. Dynamically adjusting the infrared illumination intensity ensures that the eye-tracking system can obtain high-quality eye images under different ambient lighting conditions. The application of the PID control algorithm makes the adjustment process of illumination intensity smoother and more precise, reducing image quality fluctuations caused by changes in light. The locking function stops adjusting after the ideal brightness is reached, avoiding unnecessary energy consumption.

[0014] In a preferred embodiment of this application: during the current frame image acquisition period, when a random change in the user's gaze focus position is detected, after extracting iris texture features and corneal reflective spot data based on the eye image, the method further includes: The eye image of the current frame is preprocessed and features are extracted to obtain eye image feature information, which includes color information, pixel distribution and brightness information; the current gaze feature position coordinates are obtained based on the eye image feature information. The gaze prediction matching information of the current frame is obtained, and the gaze prediction matching information of the current frame is associated and matched with the current gaze feature position coordinates to obtain the gaze feature judgment result; the gaze prediction matching information of the current frame is obtained by prediction matching based on the eye image of the previous frame and the current time node. The gaze feature judgment result is output to the control terminal of the eye-tracking system.

[0015] By adopting the above technical solutions, multi-dimensional feature extraction (such as color, pixel distribution, and brightness) eliminates noise from light fluctuations or eye movements, reduces the error in calculating gaze coordinates, and the prediction matching mechanism predicts gaze trajectory through Kalman filtering (such as a time window length of T=300ms), thereby improving prediction accuracy when the user moves their head quickly.

[0016] In a preferred embodiment of this application: the gaze feature judgment result includes gaze abnormality and gaze normality, and the method further includes: When the gaze feature judgment result indicates gaze abnormality, an information prompt instruction is generated and sent to the data monitoring terminal of the eye-tracking system; based on the information prompt instruction, the gaze feature judgment result and the time data when the information prompt instruction is issued are stored in a preset record list; When the gaze feature judgment result is that the gaze is normal, the stability of the eye image is evaluated based on a preset gaze stability analysis algorithm and a preset data monitoring frequency to obtain the gaze quality evaluation result.

[0017] By adopting the above technical solution, alarms and data storage are triggered when abnormalities occur, and the quality of vision is evaluated through stability analysis algorithms when normal conditions are met. The stability analysis algorithm (such as saccade amplitude and fixation duration variance) quantifies the user's focus and supports personalized interaction strategies (such as prioritizing highly focused users in educational scenarios). The use of the record list facilitates subsequent troubleshooting and system maintenance.

[0018] In a preferred embodiment of this application: when the modulation frequency of the dynamic reference identifier is dynamically adjusted, after adjusting the infrared supplementary light intensity, the method further includes: Obtain the infrared reflectance intensity distribution map of the current frame eye image, calculate the time synchronization error based on the synchronization relationship between the modulation frequency of the ring-coded light source with dynamic reference marker and the image acquisition frame rate; When the time synchronization error is greater than the image sampling time, the image preprocessing process is triggered; Perform high-speed filtering and histogram equalization on the current frame's eye image; Extract pupil contour feature points from eyeball images, calculate pupil center displacement, predict current pupil position using Kalman filtering based on eyeball movement trajectory in the previous frame within a preset time window, and calculate the matching degree between predicted pupil position and actual pupil position. If the matching degree is less than the preset matching degree threshold, a frequency compensation command is triggered based on the dynamic reference identifier to adjust the modulation frequency of the ring-coded light source.

[0019] By adopting the above technical solutions, the time synchronization error monitoring and correction mechanism ensures accurate synchronization between image acquisition and light source modulation, reducing image quality problems caused by asynchrony; the image preprocessing process effectively eliminates noise, enhances image contrast, and improves the accuracy of pupil positioning; the application of Kalman filtering makes pupil position prediction more accurate, while the frequency compensation mechanism ensures that optimal tracking performance can be maintained even in high dynamic environments.

[0020] Secondly, the objective of this invention is achieved through the following technical solution: A multi-user eye-tracking system, applied to any of the preceding multi-user eye-tracking methods, the system comprising: The eye-tracking system has preset baseline calibration points and dynamic reference points; An infrared camera module includes a binocular parallax structure and is equipped with a reference calibration mark that matches the reference calibration point and a dynamic reference mark that adapts to the dynamic reference point, for acquiring the current user's eye image in each frame image acquisition cycle; The iris texture feature extraction unit extracts iris texture features based on the eyeball image and adapts them to the reference calibration mark through the pupil center localization algorithm to calculate the user identification code; The corneal reflective spot data analysis unit combines corneal curvature parameters and pupil center displacement to generate three-dimensional line-of-sight coordinates. When the user differentiation processing unit detects a multi-user coexistence state, it establishes a user association matrix through iris texture feature matching and calculates user differentiation weight values ​​by combining the spatial distribution difference of the gaze heatmap in order to distinguish different users. The line-of-sight focus position data generation unit generates line-of-sight focus position data based on the three-dimensional line-of-sight coordinates, and adjusts the infrared fill light intensity according to the exposure compensation parameters corresponding to the dynamic reference marker. The interaction response control unit calculates the matching degree between the gaze focus position data and the preset interaction threshold area. When the gaze focus of multiple users overlaps, a hierarchical response mechanism is executed, prioritizing the response to the command of the user who first enters a stable gaze state.

[0021] By adopting the above technical solutions, combined with binocular parallax structure, iris texture feature extraction, and pupil center localization algorithm, efficient and accurate tracking of eye movements of multiple users is achieved.

[0022] Thirdly, the objective of this invention is achieved through the following technical solution: An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the multi-user eye-tracking method described above.

[0023] Fourthly, the objective of this invention is achieved through the following technical solution: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the aforementioned multi-user eye-tracking method.

[0024] In summary, this application includes at least one of the following beneficial technical effects: 1. This application features a dynamic weight-driven multi-user differentiation mechanism that breaks through the traditional fixed threshold method. It dynamically assigns weights based on heatmap differences and iris features to adapt to complex scenarios. Through hardware and algorithm-coordinated exposure compensation design, the ring-coded light source is synchronized with the frame rate, balancing real-time performance and imaging quality. The hierarchical response and three-dimensional coordinate fusion method combines biometrics, behavioral features, and spatial location to achieve dual priority determination of "identity-intention". 2. Using deep learning algorithms to match iris texture features can significantly improve the accuracy and speed of user recognition, especially in multi-user environments; by analyzing the overlap rate of hotspot areas and the angle of gaze direction, the system can more finely distinguish the gaze behavior of different users, enhancing the accuracy of user differentiation. Attached Figure Description

[0025] Figure 1 This is a flowchart of a multi-user eye-tracking method according to an embodiment of this application; Figure 2 This is a flowchart of step S6 in a multi-user eye-tracking method according to an embodiment of this application; Figure 3 This is a schematic diagram of a device according to one embodiment of this application. Detailed Implementation

[0026] The present application will be further described in detail below with reference to the accompanying drawings.

[0027] In one embodiment, such as Figure 1 As shown, this application discloses a multi-user eye-tracking method, which specifically includes the following steps: S1: Preset the baseline calibration point and dynamic reference point in the eye-tracking system.

[0028] In this embodiment, the preset reference calibration point is a virtual coordinate origin or reference frame predefined by the system at the software or algorithm level, used as a reference for coordinate system alignment and user line-of-sight calculation during subsequent image processing. The reference calibration mark is a fixed physical marker (such as a light spot of a specific shape, a coded pattern, etc.) on the infrared camera module, whose position and optical characteristics are known. By capturing an image of the reference calibration mark, the position of the physical mark is matched with the preset reference calibration point to complete the intrinsic parameter calibration of the camera (such as focal length, distortion coefficient). The dynamic reference point is an auxiliary reference coordinate generated in real time by the system during operation, used to dynamically adjust tracking parameters (such as exposure time, supplementary light intensity). The dynamic reference mark is a variable light source or adjustable coded pattern (such as a ring coded light source, a light spot with adjustable brightness) on the infrared module, whose optical characteristics (such as brightness, frequency) can be adjusted in real time according to environmental changes.

[0029] Specifically, a calibration board (containing a checkerboard pattern) is used to acquire multi-angle images, and the three-dimensional coordinates of the benchmark calibration points are calculated using a feature matching algorithm (such as OpenCV's findChessboardCorners) to establish a projection matrix from the world coordinate system to the image coordinate system.

[0030] S2: During each frame image acquisition cycle, the current user's eye image is acquired through the infrared camera module; the infrared camera module includes a binocular parallax structure, and is equipped with a reference calibration mark that matches the reference calibration point and a dynamic reference mark that adapts to the dynamic reference point.

[0031] In this embodiment, the reference calibration mark includes at least three non-collinear positioning markers, and the dynamic reference mark includes a ring-shaped coded light source with variable light intensity, the modulation frequency of which is synchronized with the image acquisition frame rate.

[0032] Specifically, a high-precision binocular infrared camera is used; three non-collinear reference calibration points (e.g., infrared reflection markers) are fixed on the camera bracket of the eye-tracking system, forming an equilateral triangle to cover the user's possible gaze range; four ring-shaped coded light sources (wavelength 850nm) are evenly distributed around the camera lens, with the modulation frequency of each light source synchronized with the camera frame rate (e.g., 120Hz), and timing consistency is ensured by hardware triggers. The ring-shaped light source of the dynamic reference marker flashes at a frequency of 120Hz, and a synchronization signal is generated by an FPGA to ensure that the light source's on / off state is aligned with the camera's exposure window.

[0033] S3: Extract iris texture features and corneal reflectance data based on eye images; adapt the benchmark calibration marker to the pupil center positioning algorithm, and adapt the dynamic reference marker to the gaze heatmap generation module.

[0034] In this embodiment, the pupil center localization algorithm refers to locating the pupil center using a deep learning model (such as U-Net); the gaze heatmap generation module visualizes the probability distribution of the user's gaze point as a heatmap and smooths it using Gaussian filtering.

[0035] Specifically, the adaptation of the benchmark calibration mark and the pupil center localization algorithm refers to transforming the iris center coordinates to the world coordinate system through the calibration matrix, and using the calibration parameters to perform distortion correction on the image to improve the localization accuracy.

[0036] S4: Calculate the user identification code based on the iris texture features, and generate three-dimensional line-of-sight coordinates by combining corneal curvature parameters and pupil center displacement.

[0037] In this embodiment, the corneal curvature parameter refers to the radius of curvature (default value 7.8mm) calculated by fitting the corneal boundary with an ellipse; the pupil center displacement refers to the motion vector of the pupil center between consecutive frames.

[0038] Specifically, iris texture feature points are projected into 3D space, and the 2D image coordinates are converted into world coordinates using a calibration matrix: Where Z is the depth; B is the focal length; B is the baseline distance. The horizontal pixel coordinates (i.e., horizontal axis position) of the object in the image captured by the left camera of the binocular infrared camera group. The horizontal pixel coordinates of the same object in the image captured by the right camera; Next, the refractive error is corrected by combining corneal curvature parameters to generate a three-dimensional line-of-sight vector; and the line-of-sight coordinates are adjusted according to the central displacement, using the following formula: = , The current visual coordinates, The coordinates are from the previous moment. The smoothing adjustment coefficient is set to 0.5.

[0039] S5: When a multi-user coexistence state is detected, a user association matrix is ​​established by matching iris texture features, and the user differentiation weight value is calculated by combining the spatial distribution difference of the gaze heatmap to distinguish different users.

[0040] In this embodiment, the user association matrix refers to an N×N matrix (N being the number of users) constructed using iris texture similarity, with element values ​​representing matching probabilities; the spatial distribution difference refers to a comprehensive index of the spatial overlap rate (HOR) of the gaze heatmap and the angle between the gaze direction (GDA).

[0041] Specifically, step S5 includes: S51: When a multi-user coexistence state is detected, a deep learning algorithm is used to match the iris texture features of each user to construct a user association matrix.

[0042] Specifically, multi-user coexistence refers to the system detecting that the iris features of two or more users are simultaneously present in the field of vision (determined by the iris detection algorithm); iris texture features refer to the unique textures (crypts, ridges, etc.) on the iris surface, which are high-dimensional feature vectors obtained through Gabor filters; the deep learning algorithm is the ResNet-50 model used for feature matching, which outputs an iris feature similarity matrix; the user association matrix is ​​an N×N matrix (N is the number of users), where the element values ​​represent the iris feature similarity between users (0~1). The cosine similarity between users is calculated to generate the iris feature vectors of each user, and then the user association matrix is ​​constructed.

[0043] S52: Extract the overlap rate of hotspot regions and the angle of gaze direction for multiple users based on gaze heatmap.

[0044] Specifically, the calculation steps for the spatial overlap rate (HOR) of the gaze heatmap are as follows: perform a pixel-by-pixel minimum value operation on the multi-user gaze heatmap, and then calculate the overlap area ratio by combining the maximum heat value of each pixel; the calculation steps for the gaze direction angle (GDA) are as follows: calculate the gaze direction vector based on the pupil center coordinates and head posture angle (obtained through IMU), and calculate the gaze angle between users A and B. Assuming that the probability of user A's gaze heatmap being in the region (30, 40) is 0.9, and the probability of user B in the same region is 0.85, then the value obtained by performing a pixel-by-pixel minimum value operation on the multi-user gaze heatmap is 0.85. At this time, the overlap area ratio = 0.85 / maximum heat value of each pixel.

[0045] S53: Based on the overlap rate of hotspot areas and the angle between the line of sight, a preset weight coefficient is associated, and the spatial distribution difference is calculated.

[0046] In this embodiment, a preset weight coefficient (λ) is used to dynamically adjust the weights of HOR and GDA, and the weights are adjusted according to the scene complexity (such as increasing λ when occlusion occurs); the spatial distribution difference refers to the normalized difference index that combines HOR and GDA.

[0047] Specifically, the default weight λ=0.7 (HOR accounts for 70%) was determined through experimental optimization.

[0048] Dynamic adjustment rules: When the user's head occlusion rate is >30%, λ=0.9; When the ambient light intensity is less than 50 lux, λ = 0.5.

[0049] The steps for calculating the difference are as follows: Normalized HOR and GDA: , The formula for calculating the weighted difference degree D is: ;in The normalized hotspot overlap rate reflects the degree of overlap between user heatmaps; the smaller the value, the more overlap there is. The maximum heat value of a single pixel in all user heatmaps (global maximum value); The minimum heat value of a single pixel in all user heatmaps (global minimum); The angle of the normalized line of sight reflects the difference in line of sight between users; the larger the value, the greater the difference.

[0050] S54: Compare the spatial distribution difference with the preset weight threshold to generate a unique user differentiation weight value for each user and associate it to distinguish different users.

[0051] Specifically, the preset weight threshold refers to the critical value that distinguishes user differences (determined by the ROC curve); the user differentiation weight value is a numerical value (0~1) that uniquely identifies user differences, and the larger the value, the more significant the difference.

[0052] For example, the user-differentiated weight value = .

[0053] S6: Generates the focal point position data based on the three-dimensional line-of-sight coordinates, and adjusts the infrared fill light intensity according to the exposure compensation parameters corresponding to the dynamic reference marker.

[0054] In this embodiment, the dynamic reference flag adapter links the PWM duty cycle of the ring light source with the camera exposure time to ensure that the light intensity signal is synchronized with the image acquisition.

[0055] Specifically, the intensity of the infrared fill light is adjusted according to the exposure compensation parameters corresponding to the dynamic reference marker, including: S61: Collect the brightness value of the ring-shaped coded light source of the dynamic reference mark, and calculate the brightness difference between the brightness value of the ring-shaped coded light source and the preset target brightness.

[0056] Specifically, the brightness difference refers to the actual brightness of the current ring light source ( ) and preset target brightness ( The difference between the two values ​​is used to measure whether the supplementary light intensity meets the standard.

[0057] In this embodiment, the dynamic reference marker uses four ring-shaped coded light sources (wavelength 940nm), evenly distributed around the camera lens. The PWM duty cycle of each light source is controlled by an FPGA. A Region of Interest (ROI) is defined within the camera's field of view, and only the brightness value of the dynamic reference marker area is collected (the collection process involves using OpenCV's cv2.mean function to calculate the average brightness value of the ROI area). (0~255 grayscale), and read the current value of the ring light source through the analog-to-digital converter, convert it into light intensity (unit: lux), and set the target brightness value to 120 lux.

[0058] S62: Dynamically adjust the intensity of infrared supplementary light based on the brightness difference and PID control algorithm.

[0059] Specifically, the formula for calculating the control quantity u(t) output by the PID control algorithm is: Based on the calculated control quantity u(t), it is mapped to the PWM duty cycle: in, =20%, =80%.

[0060] S63: When the brightness difference is less than the preset brightness difference threshold, lock the infrared supplementary light intensity and stop adjusting.

[0061] In this embodiment, the preset brightness difference threshold is the allowable brightness error range, such as 5 lux; the locking mechanism is to stop PID adjustment and keep the current PWM duty cycle unchanged. If the brightness difference changes abruptly (such as changing by more than 20% within 100ms), the lock is temporarily released and PID adjustment is reactivated.

[0062] Furthermore, such as Figure 2 As shown, when the modulation frequency of the dynamic reference marker is dynamically adjusted, after adjusting the intensity of the infrared supplementary light, a multi-user eye-tracking method further includes: S601: Obtain the infrared reflection intensity distribution map of the current frame eye image, calculate the time synchronization error based on the synchronization relationship between the modulation frequency of the ring-coded light source with dynamic reference marker and the image acquisition frame rate.

[0063] In this embodiment, the time synchronization error is calculated by analyzing the phase difference between the infrared light source modulation frequency and the camera exposure time to quantify the dynamic synchronization deviation. The time synchronization formula is as follows: ,in, Modulate the frequency of the light source. This represents the number of light source phase cycles (e.g., 3 samples per cycle). Calculate the frame rate of 3 consecutive frames. Standard deviation, as the time synchronization error.

[0064] S602: When the time synchronization error is greater than the image sampling time, the image preprocessing process is triggered.

[0065] In this embodiment, the image preprocessing stream is triggered when the time synchronization error is greater than the image sampling time; otherwise, it is skipped.

[0066] S603: Performs high-speed filtering and histogram equalization on the current frame's eye image.

[0067] In this embodiment, high-speed filtering denoising refers to using an improved bilateral filter (spatial domain σ=3, color domain σ=0.1) to preserve edge details while eliminating salt-and-pepper noise; histogram equalization refers to adaptive histogram equalization of the ROI (pupil region) (CLAHE algorithm, clip limit=0.03).

[0068] S604: Extract pupil contour feature points from the eye image, calculate the pupil center displacement, predict the current pupil position based on the eye movement trajectory of the previous frame within a preset time window length using Kalman filtering, and calculate the matching degree between the predicted pupil position and the actual pupil position.

[0069] In this embodiment, the Euclidean distance d between the predicted pupil position and the actual pupil position is calculated, and the matching degree is... .

[0070] S605: If the matching degree is less than the preset matching degree threshold, trigger a frequency compensation command based on the dynamic reference flag to adjust the modulation frequency of the ring-coded light source.

[0071] In this embodiment, if (Assuming) =0.7) Calculate the frequency compensation amount : It employs PI control, in which... To preset the matching threshold, =50Hz, =10Hz / s; The current modulation frequency will be superimposed, and a control signal will be output to the ring light source driver circuit via the DAC to re-acquire the next frame image. Steps S601-S605 will be repeated until M ≥ .

[0072] S7: Calculate the matching degree between the gaze focus position data and the preset interaction threshold area. When the gaze focus of multiple users overlaps, execute the hierarchical response mechanism and prioritize responding to the command of the user who first enters a stable gaze state.

[0073] In this embodiment, the timestamp (TS) refers to the timestamp (accurate to milliseconds) when the user enters a stable gaze state, which is used to identify the temporal priority of the user interaction. The criteria for determining a stable gaze are a gaze point displacement of <2° and a gaze duration of >500ms. The gaze point displacement is calculated using the gaze point displacement variance (e.g., calculating the standard deviation of displacement over 5 consecutive frames), and the gaze duration is the number of consecutive stable frames counted (≥500ms). The pupil center displacement used in stable gaze detection is the Euclidean distance of the pupil center displacement between consecutive frames. The gaze duration is counted using a sliding window to count the number of consecutive stable frames.

[0074] For example, user A first satisfies the stable gaze condition at t=1200ms, and is assigned TS=1200; user B satisfies it at t=1205ms, and is assigned TS=1205.

[0075] Specifically, the hierarchical response mechanism includes: S71: Assign a unique timestamp to each user to record the start time of each user entering a stable gaze state.

[0076] Specifically, a high-precision timer (such as the SysTick timer of ARM Cortex-M4) is used to record time, and a timestamp allocation is triggered when the user meets the stable gaze condition.

[0077] S72: When multiple users' gazes overlap, compare the timestamp values ​​of each user and prioritize responding to the command of the user with the smallest timestamp value.

[0078] Specifically, if the difference in TS values ​​among multiple users is less than ΔT (e.g., TS1=1200, TS2=1203, ΔT=500), then the iris matching process is initiated (see step S73).

[0079] S73: If the timestamp tolerance is less than the preset time tolerance, compare the iris texture matching degree of the user's identity code and select the one with the highest matching degree to execute the instruction.

[0080] In this embodiment, iris texture matching degree refers to the similarity (0~1) of iris feature vectors quantified by cosine similarity. Specifically, the iris feature extraction steps include: extracting 256-dimensional iris feature vectors using a ResNet-18 model, L2 normalizing them, and then calculating the cosine similarity between users. : ,in, Let i be the iris texture feature vector. and It is a high-dimensional dense vector (e.g., 256-dimensional); Let be the iris texture feature vector of user j; The dot product is the sum of element-wise products of two vectors, measuring the similarity of their projections in a direction. A larger value indicates greater similarity. The dot product is largest when the two vectors are in the same direction and smallest (negative) when they are in opposite directions. (Magnitude product) It is a normalization factor that scales the dot product result to the interval [0, 1]. If the eigenvectors have been L2 normalized (|f|=1), then the denominator is 1. Let be the Euclidean norm (modulus) of the feature vector of user i. Let be the magnitude of the feature vector of user j. The cosine similarity in this embodiment of the application. A distinction threshold of 0.7 is set, i.e. =0.85 indicates that the irises of users A and B are highly matched and are determined to be the same person.

[0081] In one embodiment, when a random change in the user's gaze focus position is detected during the current frame image acquisition period, after step S3, a multi-user eye-tracking method further includes: S31: Preprocess and extract features from the eye image of the current frame to obtain eye image feature information, which includes color information, pixel distribution and brightness information; obtain the current gaze feature position coordinates based on the eye image feature information.

[0082] In this embodiment, preprocessing refers to operations such as denoising the original eye image (using Gaussian filtering (kernel size 3×3) to remove image noise) and contrast enhancement (applying CLAHE (contrast-limited adaptive histogram equalization) to improve the clarity of iris texture) to improve the accuracy of feature extraction; color information refers to color distribution information (such as RGB / HSV channels) and pixel distribution (heatmap); gaze feature position coordinates refer to the gaze point coordinates (x, y) calculated based on the position of the iris center and the corneal reflective spot.

[0083] Specifically, the color information is obtained by: statistically analyzing the average RGB channel values ​​within the ROI (Region of Interest) to generate a color histogram; the pixel distribution is obtained by detecting the iris edge through Hough circle transform and generating a gaze heatmap (Gaussian kernel size 5×5); and the brightness information is obtained by calculating the average gray value of the ROI region.

[0084] S32: Obtain the gaze prediction matching information of the current frame, associate and match the gaze prediction matching information of the current frame with the current gaze feature position coordinates to obtain the gaze feature judgment result; the gaze prediction matching information of the current frame is obtained by prediction matching based on the eye image of the previous frame and the current time node.

[0085] In this embodiment, gaze prediction matching information refers to predicting the gaze position of the current frame based on the pupil movement trajectory of the previous frame (such as Kalman filter prediction); the time node refers to the time interval between the current frame and the previous frame; the time node is the reciprocal of the image acquisition frame rate.

[0086] Specifically, a line-of-sight prediction model is set up, which includes a state equation and a measurement equation, wherein the state equation is: ,in, and Let A represent the current state and the predicted next state (position, velocity), respectively; A is the state transition matrix. This is process noise.

[0087] The measurement equation is: ,in, H represents the actual observation values ​​of the current frame, and H is the observation matrix. To measure noise; then Kalman filtering is used to predict the pupil position in the current frame ( , Then calculate the predicted pupil position in the current frame and the current feature coordinates (( , The Euclidean distance is used to obtain the matching degree d: If d > 5 pixels, trigger recalibration (e.g., start active focus algorithm).

[0088] Then, the matching degree d is compared with the preset matching threshold (e.g., 3 pixels), and the matching result (match / no match) is output.

[0089] S33: Output the gaze feature judgment result to the control terminal of the eye-tracking system.

[0090] In this embodiment, the control end refers to the central processing unit of the eye-tracking system (such as an embedded computer or GPU server); the standardized data format is JSON or Protobuf, which includes information such as coordinates, confidence level, and timestamp.

[0091] In one embodiment, the gaze feature judgment result includes gaze abnormality and gaze normality; a multi-user eye tracking method further includes: S331: When the gaze feature judgment result is gaze abnormal, generate an information prompt command and send it to the data monitoring terminal of the eye-tracking system; based on the information prompt command, store the gaze feature judgment result and the time data when the information prompt command is issued into a preset record list.

[0092] In this embodiment, abnormal gaze refers to the user's eye movement exceeding a preset threshold (such as gaze point displacement > 5° or sudden change in gaze direction), or the detection of occlusion (such as eye occlusion rate > 30%); normal gaze refers to the user's eye movement being stable (displacement < 2°) and unobstructed, meeting the preset stability conditions; the data monitoring terminal is a server or embedded module responsible for receiving, storing, and analyzing abnormal events; generating prompt instructions refers to constructing alarm information in JSON format, including the abnormality type, timestamp, and user ID.

[0093] For example, the pupil center displacement and the gaze direction vector are extracted from the eye image. Eye occlusion rate (segmented by HSV threshold). If the user's head moves rapidly and causes the pupil center displacement to be 6 pixels, an anomaly marker is triggered.

[0094] S332: When the gaze feature judgment result is that the gaze is normal, the stability of the eye image is evaluated based on the preset gaze stability analysis algorithm and the preset data monitoring frequency to obtain the gaze quality evaluation result.

[0095] In this embodiment, the preset line-of-sight stability analysis algorithm includes stability assessment based on the residual (error) between Kalman filter prediction and actual displacement; the data monitoring frequency refers to dynamically adjusting the sampling rate (10Hz in normal state, increased to 30Hz in abnormal state).

[0096] Specifically, the prediction residuals in the Kalman filter model The calculation formula is: The smaller the residual, the higher the stability.

[0097] The dynamic weighting model used for the stability score is as follows: ,in, It is a dynamic weight used to balance the contribution ratio of residuals and time; It is the length of time a user can maintain a stable gaze (in milliseconds). This is a preset maximum fixation time threshold, setting the upper limit of the stability assessment threshold; for example, the parameter value is: =0.6, =1000ms.

[0098] The grading criteria are as follows: S≥0.8: High stability (prioritizes response to commands); 0.5≤S<0.8: Moderate stability (requires secondary confirmation); S<0.5: Low stability (triggers recalibration); if If the value approaches 0, a recalibration is triggered (e.g., prompting the user to "keep watching").

[0099] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0100] In one embodiment, a multi-user eye-tracking system is provided, which corresponds to a multi-user eye-tracking method in the above embodiments.

[0101] A multi-user eye-tracking system includes an infrared camera module, an iris texture feature extraction unit, a corneal reflective spot data analysis unit, a user differentiation processing unit, a gaze focus position data generation unit, and an interactive response control unit. Detailed descriptions of each functional module are as follows: The eye-tracking system has preset baseline calibration points and dynamic reference points; The infrared camera module includes a binocular parallax structure and is equipped with a reference calibration mark that matches the reference calibration point and a dynamic reference mark that adapts to the dynamic reference point, which is used to acquire the current user's eye image in each frame image acquisition cycle. The iris texture feature extraction unit extracts iris texture features based on eyeball images and adapts them to the reference calibration mark through the pupil center localization algorithm to calculate the user identification code; The corneal reflective spot data analysis unit combines corneal curvature parameters and pupil center displacement to generate three-dimensional line-of-sight coordinates. When the user differentiation processing unit detects a multi-user coexistence state, it establishes a user association matrix through iris texture feature matching and calculates user differentiation weight values ​​by combining the spatial distribution difference of the gaze heatmap in order to distinguish different users. The line-of-sight focus position data generation unit generates line-of-sight focus position data based on three-dimensional line-of-sight coordinates and adjusts the infrared fill light intensity according to the exposure compensation parameters corresponding to the dynamic reference marker. The interaction response control unit calculates the matching degree between the gaze focus position data and the preset interaction threshold area. When the gaze focus of multiple users overlaps, a hierarchical response mechanism is executed, prioritizing the response to the command of the user who first enters a stable gaze state.

[0102] For specific limitations regarding a multi-user eye-tracking system, please refer to the limitations of a multi-user eye-tracking method mentioned above, which will not be repeated here. Each module in the aforementioned multi-user eye-tracking system can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in the electronic device in hardware form, or it can be stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of each module.

[0103] In one embodiment, an electronic device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, this electronic device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores eye images, user identification codes, and user association matrices, etc. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a multi-user eye-tracking method.

[0104] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: S1: Preset baseline calibration points and dynamic reference points in the eye-tracking system; S2: During each frame image acquisition cycle, the current user's eye image is acquired through the infrared camera module; the infrared camera module includes a binocular parallax structure and is equipped with a reference calibration mark that matches the reference calibration point and a dynamic reference mark that adapts to the dynamic reference point. S3: Extract iris texture features and corneal reflective spot data based on eyeball images; adapt the benchmark calibration mark to the pupil center positioning algorithm, and adapt the dynamic reference mark to the gaze heatmap generation module; S4: Calculate the user identification code based on iris texture features, and generate three-dimensional line-of-sight coordinates by combining corneal curvature parameters and pupil center displacement. S5: When a multi-user coexistence state is detected, a user association matrix is ​​established by matching iris texture features, and the user differentiation weight value is calculated by combining the spatial distribution difference of the gaze heatmap to distinguish different users. S6: Generates the focal point position data based on the three-dimensional line-of-sight coordinates, and adjusts the infrared fill light intensity according to the exposure compensation parameters corresponding to the dynamic reference marker; S7: Calculate the matching degree between the gaze focus position data and the preset interaction threshold area. When the gaze focus of multiple users overlaps, execute the hierarchical response mechanism and prioritize responding to the command of the user who first enters a stable gaze state.

[0105] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: S1: Preset baseline calibration points and dynamic reference points in the eye-tracking system; S2: During each frame image acquisition cycle, the current user's eye image is acquired through the infrared camera module; the infrared camera module includes a binocular parallax structure and is equipped with a reference calibration mark that matches the reference calibration point and a dynamic reference mark that adapts to the dynamic reference point. S3: Extract iris texture features and corneal reflective spot data based on eyeball images; adapt the benchmark calibration mark to the pupil center positioning algorithm, and adapt the dynamic reference mark to the gaze heatmap generation module; S4: Calculate the user identification code based on iris texture features, and generate three-dimensional line-of-sight coordinates by combining corneal curvature parameters and pupil center displacement. S5: When a multi-user coexistence state is detected, a user association matrix is ​​established by matching iris texture features, and the user differentiation weight value is calculated by combining the spatial distribution difference of the gaze heatmap to distinguish different users. S6: Generates the focal point position data based on the three-dimensional line-of-sight coordinates, and adjusts the infrared fill light intensity according to the exposure compensation parameters corresponding to the dynamic reference marker; S7: Calculate the matching degree between the gaze focus position data and the preset interaction threshold area. When the gaze focus of multiple users overlaps, execute the hierarchical response mechanism and prioritize responding to the command of the user who first enters a stable gaze state.

[0106] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0107] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0108] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A multi-user eye-tracking method, characterized in that, include: Pre-set baseline calibration points and dynamic reference points in the eye-tracking system; During each frame image acquisition cycle, the current user's eye image is acquired through the infrared camera module; The infrared camera module includes a binocular parallax structure and is equipped with a reference calibration mark that matches the reference calibration point and a dynamic reference mark that adapts to the dynamic reference point. Iris texture features and corneal reflectance data are extracted from the eye images. The reference calibration identifier is adapted to the pupil center localization algorithm, and the dynamic reference identifier is adapted to the gaze heatmap generation module; The user identification code is calculated based on the iris texture features, and three-dimensional line-of-sight coordinates are generated by combining corneal curvature parameters and pupil center displacement. When a multi-user coexistence state is detected, a user association matrix is ​​established by matching iris texture features, and user differentiation weight values ​​are calculated by combining the spatial distribution difference of gaze heatmap to distinguish different users. Based on the three-dimensional line-of-sight coordinates, generate line-of-sight focus position data, and adjust the infrared fill light intensity according to the exposure compensation parameters corresponding to the dynamic reference marker; The matching degree of the gaze focus position data is calculated with the preset interaction threshold area. When the gaze focus of multiple users overlaps, a hierarchical response mechanism is executed, giving priority to responding to the command of the user who first enters a stable gaze state.

2. The multi-user eye-tracking method according to claim 1, characterized in that, The reference calibration mark includes at least three non-collinear positioning markers, and the dynamic reference mark includes a ring-shaped coded light source with variable intensity, the modulation frequency of which is synchronized with the image acquisition frame rate. The execution-level response mechanism includes: Assign a unique timestamp to each user to record the start time of each user entering a stable gaze state; When multiple users' gazes overlap, compare the timestamp values ​​of each user and respond to the command of the user with the smallest timestamp value first. If the timestamp tolerance is less than the preset time tolerance, compare the iris texture matching degree of the user's identity code and select the one with the highest matching degree to execute the instruction.

3. The multi-user eye-tracking method according to claim 1, characterized in that, When a multi-user coexistence state is detected, a user association matrix is ​​established through iris texture feature matching, and a user differentiation weight value is calculated by combining the spatial distribution difference of the gaze heatmap to distinguish different users, including: When a multi-user coexistence state is detected, a deep learning algorithm is used to match the iris texture features of each user to construct a user association matrix; Based on the gaze heatmap, the overlap rate of hotspot regions and the angle of gaze direction for multiple users are extracted; The spatial distribution difference is calculated by associating the overlap rate of the hotspot areas with the angle of view direction with a preset weight coefficient. The spatial distribution difference is compared with a preset weight threshold to generate a unique user differentiation weight value for each user and associate it to distinguish different users.

4. The multi-user eye-tracking method according to claim 1, characterized in that, The step of adjusting the infrared fill light intensity according to the exposure compensation parameters corresponding to the dynamic reference marker includes: Collect the brightness value of the ring-shaped coded light source of the dynamic reference mark, and calculate the brightness difference between the brightness value of the ring-shaped coded light source and the preset target brightness; The intensity of infrared supplementary light is dynamically adjusted based on the brightness difference and the PID control algorithm. When the brightness difference is less than the preset brightness difference threshold, the infrared supplementary light intensity is locked and adjustment is stopped.

5. The multi-user eye-tracking method according to claim 1, characterized in that, Within the current frame image acquisition period, when a random change in the user's gaze focus position is detected, after extracting iris texture features and corneal reflective spot data based on the eye image, the process further includes: The eye image of the current frame is preprocessed and features are extracted to obtain eye image feature information, which includes color information, pixel distribution and brightness information; the current gaze feature position coordinates are obtained based on the eye image feature information. The gaze prediction matching information of the current frame is obtained, and the gaze prediction matching information of the current frame is associated and matched with the current gaze feature position coordinates to obtain the gaze feature judgment result; the gaze prediction matching information of the current frame is obtained by prediction matching based on the eye image of the previous frame and the current time node. The gaze feature judgment result is output to the control terminal of the eye-tracking system.

6. The multi-user eye-tracking method according to claim 5, characterized in that, The gaze feature judgment result includes gaze abnormality and gaze normality, and the method further includes: When the gaze feature judgment result indicates gaze abnormality, an information prompt instruction is generated and sent to the data monitoring terminal of the eye-tracking system; based on the information prompt instruction, the gaze feature judgment result and the time data when the information prompt instruction is issued are stored in a preset record list; When the gaze feature judgment result is that the gaze is normal, the stability of the eye image is evaluated based on a preset gaze stability analysis algorithm and a preset data monitoring frequency to obtain the gaze quality evaluation result.

7. A multi-user eye-tracking method according to claim 4, characterized in that, When the modulation frequency of the dynamic reference identifier is dynamically adjusted, after adjusting the infrared supplementary light intensity, the method further includes: Obtain the infrared reflectance intensity distribution map of the current frame eye image, calculate the time synchronization error based on the synchronization relationship between the modulation frequency of the ring-coded light source with dynamic reference marker and the image acquisition frame rate; When the time synchronization error is greater than the image sampling time, the image preprocessing process is triggered; Perform high-speed filtering and histogram equalization on the current frame's eye image; Extract pupil contour feature points from eyeball images, calculate pupil center displacement, predict current pupil position using Kalman filtering based on eyeball movement trajectory in the previous frame within a preset time window, and calculate the matching degree between predicted pupil position and actual pupil position. If the matching degree is less than a preset matching degree threshold, a frequency compensation command is triggered based on the dynamic reference identifier to adjust the modulation frequency of the ring-coded light source.

8. A multi-user eye-tracking system, characterized in that, The system, applied to a multi-user eye-tracking method as described in any one of claims 1 to 7, comprises: The eye-tracking system has preset baseline calibration points and dynamic reference points; An infrared camera module includes a binocular parallax structure and is equipped with a reference calibration mark that matches the reference calibration point and a dynamic reference mark that adapts to the dynamic reference point, for acquiring the current user's eye image in each frame image acquisition cycle; The iris texture feature extraction unit extracts iris texture features based on the eyeball image and adapts them to the reference calibration mark through the pupil center localization algorithm to calculate the user identification code; The corneal reflective spot data analysis unit combines corneal curvature parameters and pupil center displacement to generate three-dimensional line-of-sight coordinates. When the user differentiation processing unit detects a multi-user coexistence state, it establishes a user association matrix through iris texture feature matching and calculates user differentiation weight values ​​by combining the spatial distribution difference of the gaze heatmap in order to distinguish different users. The line-of-sight focus position data generation unit generates line-of-sight focus position data based on the three-dimensional line-of-sight coordinates, and adjusts the infrared fill light intensity according to the exposure compensation parameters corresponding to the dynamic reference marker. The interaction response control unit calculates the matching degree between the gaze focus position data and the preset interaction threshold area. When the gaze focus of multiple users overlaps, a hierarchical response mechanism is executed, prioritizing the response to the command of the user who first enters a stable gaze state.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multi-user eye-tracking method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-user eye-tracking method as described in any one of claims 1 to 7.

Images