Training the eye tracking model

Abstract

A method (300) for training an eye tracking model (710), as well as a corresponding system (400) and a storage medium are disclosed. The eye tracking model is adapted to predict eye tracking data based on sensor data (709) from a first eye tracking sensor (411). The method comprises: receiving (301) sensor data obtained by the first eye tracking sensor at a certain moment; and receiving (302) reference eye tracking data generated for the moment by an eye tracking system (420) including a second eye tracking sensor (421). The reference eye tracking data is generated by the eye tracking system based on sensor data obtained by the second eye tracking sensor at the moment. The method comprises training (303) the eye tracking model based on the sensor data obtained by the first eye tracking sensor at the moment and the generated reference eye tracking data.

Description

[0001] This application is a divisional application of the Chinese invention patent application with application date of March 24, 2020, application number 202010212524.2, and invention name “Training Eye Tracking Model”. Technical Field

[0002] This disclosure generally relates to eye tracking. Background Art

[0003] A variety of different techniques have been developed to monitor which direction (or point on a display) a user is looking at. This is often referred to as gaze tracking. Such techniques typically involve detecting certain features in an image of the eye and then calculating the gaze direction or point of gaze based on the positions of these detected features. An example of such a gaze tracking technique is the pupil center corneal reflection (PCCR). PCCR-based gaze tracking uses the position of the pupil center and the position of the flash point (the reflection of a light-emitting device at the cornea) to calculate the eye's gaze direction or point of gaze on a display.

[0004] Another term often used in this context is "eye tracking." Although the term "eye tracking" may be used as an alias for gaze tracking in many cases, eye tracking does not necessarily involve tracking a user's gaze (e.g., in the form of gaze direction or gaze point). Eye tracking may, for example, involve tracking the position of an eye in space, rather than actually tracking the eye's gaze direction or gaze point.

[0005] As an alternative (or supplement) to conventional techniques such as PCCR-based eye tracking, machine learning can be used to train an algorithm for performing eye tracking. For example, machine learning can use training data in the form of eye images and associated known gaze points to train the algorithm so that the trained algorithm can perform eye tracking in real time based on the eye images. Such machine learning typically requires a large amount of training data to work properly. It may take considerable time and / or resources to collect the training data. In many cases, there may be certain requirements for the training data. The training data should, for example, preferably reflect all those types of situations / situations that the eye tracking algorithm should be able to handle. If only certain types of situations / situations are represented in the training data (e.g., only small gaze angles, or only well-illuminated images), the eye tracking algorithm may perform well for such situations / situations, but may not perform as well for other situations / situations that were not covered during the training phase.

[0006] It would be desirable to provide new approaches to address one or more of the above-mentioned problems. Summary of the Invention

[0007] A method, a system and a computer readable storage medium having the features defined in the independent claims are provided to solve one or more of the above problems. Preferred embodiments are defined in the dependent claims.

[0008] Therefore, a first aspect provides an embodiment of a method for training an eye-tracking model. The eye-tracking model is adapted to predict eye-tracking data based on sensor data from a first eye-tracking sensor. The method comprises: receiving sensor data obtained by the first eye-tracking sensor at a certain moment; and receiving reference eye-tracking data generated for the moment by an eye-tracking system including a second eye-tracking sensor. The reference eye-tracking data is generated by the eye-tracking system based on sensor data obtained by the second eye-tracking sensor at the moment. The method comprises training the eye-tracking model based on the sensor data obtained by the first eye-tracking sensor at the moment and the generated reference eye-tracking data.

[0009] As previously described in the Background section, it may take considerable time and / or resources to collect traditional training data. Reference eye tracking data from an established eye tracking system can be used as an alternative or in addition to such traditional training data to train an eye tracking model.

[0010] It will be appreciated that eye tracking data (such as predicted eye tracking data or reference eye tracking data) may, for example, indicate a gaze point of an eye on a display, and / or a gaze vector, and / or a position of the eye in space.

[0011] A second aspect provides an embodiment of a system for training an eye-tracking model. The eye-tracking model is adapted to predict eye-tracking data based on sensor data from a first eye-tracking sensor. The system includes a processing circuit system (or one or more processors) configured to: receive sensor data obtained by the first eye-tracking sensor at a certain moment; and receive reference eye-tracking data generated for the moment by an eye-tracking system including a second eye-tracking sensor. The reference eye-tracking data is generated by the eye-tracking system based on sensor data obtained by the second eye-tracking sensor at the moment. The processing circuit system is configured to train the eye-tracking model based on the sensor data obtained by the first eye-tracking sensor at the moment and the generated reference eye-tracking data.

[0012] The processing circuit system (or one or more processors) may, for example, be configured to perform the method as defined in any embodiment of the first aspect disclosed herein (in other words, in the claims, summary of the invention, detailed description or description of the figures). The system may, for example, include one or more non-transitory computer-readable storage media (or one or more memories) storing instructions that, when executed by the processing circuit system (or one or more processors), cause the system to perform the method as defined in any embodiment of the first aspect disclosed herein.

[0013] The effects and / or advantages presented in this disclosure for the embodiments of the method according to the first aspect may also be applicable to the corresponding embodiments of the system according to the second aspect.

[0014] A third aspect provides an embodiment of a non-transitory computer-readable storage medium storing instructions for training an eye-tracking model. The eye-tracking model is adapted to predict eye-tracking data based on sensor data from a first eye-tracking sensor. The instructions, when executed by a system, cause the system to:

[0015] Receiving sensor data obtained by the first eye tracking sensor at a certain moment;

[0016] receiving reference eye tracking data generated for the instant by an eye tracking system including a second eye tracking sensor, wherein the reference eye tracking data is generated by the eye tracking system based on sensor data obtained by the second eye tracking sensor at the instant; and

[0017] • Training the eye tracking model based on the sensor data obtained by the first eye tracking sensor at the moment and the generated reference eye tracking data.

[0018] The non-transitory computer-readable storage medium may, for example, store instructions that, when executed by the system (or by a processing circuit system contained in the system), cause the system to perform the method defined in any embodiment of the first aspect disclosed herein (in other words, the claims, the summary of the invention, the accompanying drawings, or the detailed description).

[0019] The non-transitory computer-readable storage medium may, for example, be provided in a computer program product. In other words, the computer program product may, for example, include a non-transitory computer-readable storage medium storing instructions that, when executed by a system, cause the system to perform the method as defined in any embodiment of the first aspect disclosed herein.

[0020] The effects and / or advantages presented in the present disclosure for the embodiments of the method according to the first aspect may also be applicable to the corresponding embodiments of the non-transitory computer-readable storage medium according to the third aspect.

[0021] It should be noted that embodiments of the present disclosure relate to all possible combinations of features recited in the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Hereinafter, example embodiments will be described in more detail with reference to the accompanying drawings, in which:

[0023] Figure 1 It is a front view of the eye;

[0024] Figure 2 Observe from the side of the eye Figure 1 Cross-sectional view of the eye;

[0025] Figure 3 is a flowchart of a method for training an eye tracking model according to an embodiment;

[0026] Figure 4 is a schematic overview of a system for training an eye tracking model according to an embodiment;

[0027] Figure 5 is a schematic overview of an example eye tracking system;

[0028] Figure 6 is a flowchart of a method for training an eye tracking model according to an embodiment, the method comprising using a method such as Figure 5 Eye tracking systems such as the eye tracking system in [1];

[0029] Figure 7 shows examples of eye tracking data that can be predicted by an eye tracking model;

[0030] Figure 8 shows how, according to one embodiment, Figure 3 and Figure 6 The scheme for training the eye tracking model in the method;

[0031] Figure 9 shows how, according to one embodiment, a condition may be met in response to Figure 3 and Figure 6 A scheme for performing training of an eye tracking model in a method;

[0032] Figure 10 A method of training an eye tracking model according to an embodiment is shown, the method comprising detecting a specific triggering action of the eye; and

[0033] Figure 11is a flow chart of a method for training an eye tracking model for a detected user according to an embodiment.

[0034] All figures are schematic, not necessarily drawn to scale, and generally show only the parts necessary to illustrate the various embodiments, while other parts may be omitted or only indicated. Unless otherwise indicated, any reference numerals that appear in multiple figures refer to the same object or feature in all figures. DETAILED DESCRIPTION

[0035] Throughout this disclosure, the term "eye tracking sensor" relates to a sensor adapted to obtain sensor data for eye tracking. Although an eye tracking sensor may, for example, be an imaging device (such as a camera), several other types of sensors may also be employed for eye tracking. For example, an eye tracking sensor may employ light, sound, a magnetic field, or an electric field to obtain sensor data, which sensor data (e.g., in combination with sensor data from other sensors) may be used to determine where the eye is located and / or the direction the eye is looking. An eye tracking sensor may, for example, be arranged to (or configured to) monitor the eye. An eye tracking sensor may, for example, be arranged to (or configured to) perform a measurement (or obtain sensor data) only when instructed to do so. In other words, an eye tracking sensor does not necessarily perform constant / continuous monitoring of the eye.

[0036] Throughout this disclosure, the term "imaging device" refers to a device adapted to capture images. The imaging device may be, for example, an image sensor or a camera, such as a charge coupled device (CCD) camera or a complementary metal oxide semiconductor (CMOS) camera. However, other types of imaging devices are also contemplated.

[0037] Embodiments of the method, system, and associated storage medium will be described below with reference to Figures 3 to 11 First, we will refer to Figures 1 to 2 Describe some features of the eye.

[0038] Figure 1 is a front view of the eye 100. Figure 2 is a cross-sectional view of the eye 100 as viewed from the side of the eye 100. Figure 2 Almost the entire eye 100 is shown, but Figure 1The front view shown in FIG. 1 shows only those portions of eye 100 that are typically visible from the front of a person's face. Eye 100 has a pupil 101 with a pupil center 102. Eye 100 also has an iris 103 and a cornea 104. Cornea 104 is located in front of pupil 101 and iris 103. Cornea 104 is curved and has a center of curvature 105, which is referred to as the center of corneal curvature 105 or simply corneal center 105. The radius of curvature 106 of cornea 104 is referred to as the radius 106 of cornea 104 or simply corneal radius 106. Eye 100 also has a sclera 107. Eye 100 has a center 108, which may also be referred to as the center of the eyeball 108 or simply the center of the eyeball 108. The visual axis 109 of eye 100 passes through center 108 of eye 100 to fovea 110 of eye 100. The optical axis 111 of the eye 100 passes through the pupil center 102 and the center 108 of the eye 100. The visual axis 109 forms an angle 112 with respect to the optical axis 111. The deviation or offset between the visual axis 109 and the optical axis 111 is generally referred to as the foveal offset 112. Figure 2 In the example shown, eye 100 is looking towards display 113 , and eye 100 is gazing at gaze point 114 on display 113 . Figure 1 Also shown is a reflection 115 of the light emitting device at the cornea 104. This reflection 115 is also referred to as a sparkle point 115.

[0039] Training machine learning (ML)-based eye tracking algorithms typically requires a very large number of eye images annotated with ground truth information, such as gaze origin (3D eye position), gaze direction, gaze point on the screen, etc. In traditional data collection, test subjects are required to look at points with known positions on the display to collect ground truth gaze data. This approach has several problems:

[0040] The test subject’s interaction with the computer often becomes unnatural, and therefore the training data fails to capture the real-life behavior that is important for the success of machine learning.

[0041] Tests must be supervised by a test manager who manages the data collection application and explains to test subjects how to behave during the test. This makes data collection quite expensive even without considering the compensation given to test subjects.

[0042] The wait time from “need data” to “completion of data collection” can be long (months in some cases) because traditional data collection requires setting up data collection tools, sequencing of data collection, and allocation of data collection resources.

[0043] Looking at tiny dots on a monitor is boring, and test subjects often lose focus, leading to poor data.

[0044] • During traditional data collection, only very short recordings can be made for each person (due to distractions and because the test subject cannot do anything else during the data collection period).

[0045] It may be difficult or even impossible to accurately measure certain information. For example, it may be difficult to measure the distance from the user to the eye tracker.

[0046] If a conventional, calibrated eye tracker is used instead to provide this ground truth data, data collection will be almost free of overhead and will generate a large amount of natural (real-life) training data. The user can be allowed to work normally at his / her computer while the ML-based eye tracker collects time-stamped sensor data (such as images) and the reference eye tracker collects ground truth information (such as gaze point, 3D position of the eye, gaze direction, etc.). Such a system can run in the background on the test subject's computer. At the end of the workday, a large amount of annotated data will be collected that can be used to train ML-based algorithms.

[0047] Therefore, a method for training an eye tracking model is proposed. Figure 3 3 is a flow chart of an embodiment of such a method 300. In this embodiment, an eye tracking model is adapted to predict eye tracking data based on sensor data from a first eye tracking sensor. In other words, the eye tracking model is configured to use sensor data from the first eye tracking sensor, or information derived from such sensor data, to predict or estimate eye tracking data. The eye tracking model can, for example, be viewed as a function (or mapping) that receives sensor data from the first eye tracking sensor as input (and optionally further input data) and provides predicted eye tracking data as output.

[0048] The eye tracking model trained in method 300 can be, for example, a machine learning (ML)-based eye tracking model. The eye tracking model can be, for example, based on an artificial neural network (such as a convolutional neural network). However, the eye tracking model can also be a more traditional model that can be trained, for example, by traditional optimization of a set of parameter values.

[0049] Method 300 includes receiving 301 sensor data obtained by a first eye-tracking sensor at a certain moment (or at a certain point in time). In other words, after the sensor data is obtained (or generated) by the first eye-tracking sensor at a certain moment or point in time, the sensor data is received 301. The first eye-tracking sensor can be, for example, an imaging device. However, as mentioned above, several other types of eye-tracking sensors are also contemplated.

[0050] Method 300 includes receiving 302 reference eye-tracking data generated for the instant by an eye-tracking system including a second eye-tracking sensor. The reference eye-tracking data is generated by the eye-tracking system based on sensor data obtained by the second eye-tracking sensor at the instant (in other words, at the point in time when the sensor data received 301 was obtained by the first eye-tracking sensor). The second eye-tracking sensor can, for example, be an imaging device. However, as described above, several other types of eye-tracking sensors are also contemplated. It will be understood that the second eye-tracking sensor is different from the first eye-tracking sensor. In other words, the first eye-tracking sensor and the second eye-tracking sensor are not identical, but they can, for example, be of similar types.

[0051] The method 300 includes training 303 an eye tracking model based on sensor data obtained by the first eye tracking sensor at the time instant and the generated reference eye tracking data. The training may, for example, include adjusting values ​​of one or more parameters of the eye tracking model.

[0052] It will be understood that the sensor data received at step 301 and the sensor data on which the reference eye-tracking data received at step 302 is based are not necessarily acquired by the first eye-tracking sensor and the second eye-tracking sensor at exactly the same time. In other words, the two sets of sensor data may be acquired by the corresponding eye-tracking sensors at approximately the same time, but there may of course be a slight deviation or time mismatch between the two sets of sensor data. It will be understood that as long as such deviation is so small that the eyes have not moved (or reoriented) much during this very short period of time, the step of training 303 the eye-tracking model will not be significantly affected by such mismatch.

[0053] Figure 4 is a schematic overview of a system 400 for training an eye tracking model according to an embodiment. The system 400 may, for example, perform the aforementioned Figure 3 Method 300 is described.

[0054] Consider the following scenario. You have a well-functioning eye-tracking system 420 that includes an eye-tracking sensor 421 and means for analyzing sensor data to generate eye-tracking data (such as an estimated position of the eye 100 in space, or an estimated gaze point of the eye 100). You have a new eye-tracking system 410 that includes an eye-tracking sensor 411, but the new eye-tracking system is not yet able to generate accurate gaze-tracking data based on the sensor data from the eye-tracking sensor 411. The software or algorithms used in the old eye-tracking system 420 are not as useful for the new eye-tracking system 410 for reasons such as:

[0055] The new eye tracking system 410 uses a different type of eye tracking sensor 411 than the old eye tracking system 420, or

[0056] The old eye tracking system 420 is equipped with a lighting device 422 that is not available in the new eye tracking system 410, or

[0057] The light emitting device 412 is available in the eye tracking system 410 , but is located at a different position relative to the eye tracking sensor 411 compared to the old eye tracking system 420 .

[0058] Therefore, rather than reusing software from the old eye tracking system 420 in the new eye tracking system 410, the old eye tracking system 420 is used to provide ground truth data for training the new eye tracking system 410. The new eye tracking system 410 is equipped with an eye tracking model adapted to predict eye tracking data based on sensor data obtained by the eye tracking sensor 411. Figure 3 Method 300 is described in which an eye-tracking model of a new eye-tracking system 410 is trained using reference eye-tracking data generated by an old eye-tracking system 420. In the terminology of method 300, eye-tracking sensor 411 is an example of a first eye-tracking sensor that obtained the sensor data received at step 301, and old eye-tracking system 420 is an example of an eye-tracking system that generated the reference eye-tracking data received at step 302. Further, eye-tracking sensor 421 is an example of a second eye-tracking system referenced in method 300.

[0059] System 400 includes processing circuitry 430 configured to perform method 300 to train an eye tracking model of a new eye tracking system 410 .

[0060] Processing circuitry 430 may, for example, include one or more processors 431. Processor(s) 431 may, for example, be an application-specific integrated circuit (ASIC) configured to perform a particular method (such as method 300). Alternatively, processor(s) 431 may be configured to execute instructions (e.g., in the form of a computer program) stored in one or more memories 432. Such one or more memories 432 may, for example, be included in processing circuitry 430 of system 400, or may be external to system 400 (e.g., remote from the system). The one or more memories 432 may, for example, store instructions for causing system 500 to perform method 300.

[0061] Processing circuitry 430 may be communicatively connected to old eye tracking system 420 and new eye tracking system 410 (or at least to eye tracking sensor 411 in new eye tracking system 410), for example, via a wired connection and / or a wireless connection.

[0062] The old eye tracking system 420 may be, for example, a PCCR-based eye tracking system. In other words, the eye tracking system 420 may have generated reference eye tracking data based on the image position of the corneal reflection of the light emitting device 422 at a known position relative to the eye tracking sensor 421 (which is an imaging device in this case) and the image position of the pupil center, and the reference eye tracking data was received at step 302 in method 300.

[0063] Old eye-tracking system 420 may, for example, include more eye-tracking sensors, or more advanced eye-tracking sensors, or more light-emitting devices than new eye-tracking system 410. By using more advanced eye-tracking system 420 to train new eye-tracking system 410, an eye-tracking system 410 having relatively cheaper components may be obtained, which is capable of working almost as well as more advanced eye-tracking system 420.

[0064] exist Figure 4In the example embodiment shown in FIG, eye tracking sensor 421 in old eye tracking system 420 is an imaging device (such as a camera), and one or more light emitting devices 422 to 424 are provided for illuminating eye 100. In this example embodiment, eye tracking sensor 411 in new eye tracking system 410 is also an imaging device. Light 428 from light emitting device 422 in old eye tracking system 420 reaches imaging device 421 in old eye tracking system 420 via reflection at the cornea of ​​eye 100. However, light 429 from light emitting device 422 in old eye tracking system 420 can also reach imaging device 411 in new eye tracking system 420, which may interfere with the image captured by imaging device 411 in new eye tracking system 410. Therefore, filter 415 can be used to prevent light 429 from light emitting device 422 from reaching imaging device 411.

[0065] Therefore, according to some embodiments, the old eye tracking system 420 includes a light emitting device 422 that outputs light 428 within a certain wavelength range to illuminate the eye 100, and the eye tracking sensor 421 of the old eye tracking system 420 provides sensor data based on the light within the wavelength range. The eye tracking sensor 411 of the new eye tracking system 410 can be provided with a filter 415 for suppressing light within the wavelength range.

[0066] The light emitted by light emitting device 422 may, for example, be light of relatively long wavelength, and optical filter 415 may be a filter that allows short wavelengths to pass through. Alternatively, the light emitted by light emitting device 422 may, for example, be light of relatively short wavelength, and optical filter 415 may be a filter that allows long wavelengths to pass through. If the light emitted by light emitting device 422 is not limited to a certain wavelength range, optical filter 426 may be provided in front of light emitting device 422 to suppress light outside of the aforementioned certain wavelength range.

[0067] Similarly, if the new eye-tracking system 410 includes one or more light-emitting devices 412-413 for illuminating the eye 100, the eye-tracking sensor 421 in the old eye-tracking system 420 can be provided with a filter 427 for suppressing light from the light-emitting device in the new eye-tracking system 410. The light emitted by the light-emitting device 412 can, for example, be light of relatively long wavelength, and the filter 427 can be a filter that allows short wavelengths to pass. Alternatively, the light emitted by the light-emitting device 412 can, for example, be light of relatively short wavelength, and the filter 427 can be a filter that allows long wavelengths to pass. If the light emitted by the light-emitting device 412 is not limited to a certain wavelength range, a filter 415 can be provided in front of the light-emitting device 412 to suppress light outside the aforementioned certain wavelength range.

[0068] In this way, the two eye tracking systems 410 and 420 are prevented from interfering with each other. In other words, the old eye tracking system 420 may employ light within a first wavelength range (e.g., approximately 940 nm), while the new eye tracking system employs light within a second wavelength range (e.g., approximately 850 nm) that does not overlap with the first wavelength range.

[0069] like Figure 4 As shown, the old eye tracking system 420 and / or the new eye tracking system 410 may include one or more light emitting devices. The light emitting devices may be, for example, infrared or near-infrared light emitting devices, such as in the form of light emitting diodes (LEDs). However, other types of light emitting devices are also contemplated.

[0070] like Figure 4 As shown, old eye tracking system 420 may include, for example, a member 425 (e.g., a circuit board (such as a printed circuit board PCB)) on which eye tracking sensor 421 and light emitting devices 422 to 424 are mounted. Similarly, new eye tracking system 410 may include, for example, a member 414 (e.g., a circuit board (such as a PCB)) on which eye tracking sensor 411 and light emitting devices 412 to 413 are mounted.

[0071] It will be understood that the system 400 does not necessarily include Figure 4 For example, system 400 may include only processing circuitry 431, while Figure 4 The remaining components shown may be considered to be external to system 400. In some embodiments, system 400 includes an eye tracking system 420 for generating reference eye tracking data received at step 302 in method 300. In some embodiments, system 400 includes an eye tracking sensor 411 for obtaining sensor data received at step 301 in method 300.

[0072] Eye tracking system 420 and eye tracking system 410 may, for example, be provided in the form of two separate units or devices, which may, for example, be mountable at a display device to perform eye tracking.

[0073] Figure 5 is a schematic overview of an example eye tracking system 500 . Figure 4 The old eye tracking system 420 and / or the new eye tracking system 410 in the embodiment may for example belong to the following references Figure 5 The type being described.

[0074] System 500 includes one or more lighting devices 501 for illuminating eye 100, and one or more imaging devices 502 for capturing images of eye 100 while eye 100 is looking at display 503. System 500 also includes processing circuitry 504 configured to estimate where eye 100 is located and / or where eye 100 is looking. Processing circuitry 504 may, for example, estimate eye tracking data, such as the gaze direction (or gaze vector) of eye 100 (corresponding to the direction of visual axis 109), or estimate a gaze point 508 of eye 100 on display 503. In other words, eye tracking system 500 may, for example, be a gaze tracking system.

[0075] Processing circuitry 504 is communicatively connected to lighting device 501 and imaging device 502, e.g., via a wired or wireless connection. Processing circuitry 504 may also be communicatively connected to display 503, e.g., for controlling (or triggering) display 503 to display test stimulus points for calibrating eye tracking system 500.

[0076] Figure 5 Example light emitting devices 501 are shown located on either side of the display 503, but the light emitting devices 501 may be located elsewhere. Figure 5 An example imaging device 502 is shown located above the display 503 , but the imaging device 502 may be located elsewhere, such as below the display 503 .

[0077] The display 503 may be, for example, a liquid crystal display (LCD) or an LED display. However, other types of displays are also contemplated. The display 503 may be, for example, a flat display or a curved display. The display 503 may be, for example, a television screen, a computer screen, or may be part of a head-mounted device (HMD) such as a virtual reality (VR) or augmented reality (AR) device. The display 503 may, for example, be placed in front of one of the user's eyes. In other words, separate displays 503 may be used for the left eye and the right eye. For example, separate eye tracking devices (such as a light emitting device 501 and an imaging device 502) may be used for the left eye and the right eye.

[0078] The processing circuitry 504 can be used for eye tracking of both eyes, or there can be separate processing circuitry 504 for the left and right eyes. The eye tracking system 500 can, for example, perform eye tracking for the left and right eyes separately, and can then determine a combined gaze point as the average of the gaze points of the left and right eyes.

[0079] Processing circuitry 504 may, for example, include one or more processors 506. Processor(s) 506 may, for example, be an application-specific integrated circuit (ASIC) configured to execute a particular eye tracking method. Alternatively, processor(s) 506 may be configured to execute instructions stored in one or more memories 507 (e.g., in the form of computer programs). Such memories 507 may, for example, be included in processing circuitry 504 of eye tracking system 500, or may be external to eye tracking system 500 (e.g., remote from the eye tracking system). Memory 507 may store instructions for causing eye tracking system 500 to execute the eye tracking method.

[0080] It will be understood that the foregoing reference Figure 5 The described eye-tracking system 500 is provided as an example, and many other eye-tracking systems are contemplated. For example, the light-emitting device 501 and / or the imaging device 502 need not be considered part of the eye-tracking system 500. The eye-tracking system 500 may, for example, consist solely of the processing circuitry 504. There are even eye-tracking systems that do not employ a light-emitting device at all. Furthermore, some eye-tracking systems employ other types of eye-tracking sensors in addition to imaging devices. In other words, the eye-tracking system 500 may employ other types of sensor data in addition to images to perform eye tracking. The display 503 may, for example, be included in the eye-tracking system 500, or may be considered separate from the eye-tracking system 500.

[0081] Previous reference Figure 3 The described method 300 may be, for example, passively received at steps 301 and 302 (at Figure 4 ) a first eye tracking sensor (exemplified by sensor 411 in Figure 4 4. The method 300 may further include receiving data from an eye-tracking system (exemplified by system 420 in step 301). However, the method 300 may further include using the first eye-tracking sensor to obtain sensor data at a certain moment (in other words, the sensor data received at step 301), and / or using the eye-tracking system to generate reference eye-tracking data for that moment (in other words, the eye-tracking data received at step 302). In other words, the method 300 may actively use the first eye-tracking sensor 411 and the eye-tracking system 420, for example, by controlling (or instructing) them to provide the necessary data.

[0082] Figure 6 is a flow chart of a method 600 for training an eye tracking model according to an embodiment, the method including such explicit use of the first eye tracking sensor 411 and the eye tracking system 420. Figure 4Method 600 is described using eye tracking sensor 411 and eye tracking system 420 as shown, but it will be understood that a different eye tracking sensor and / or a different eye tracking system may be employed in method 600 .

[0083] The method 600 comprises obtaining sensor data at a certain moment using 601 the first eye tracking sensor 411. The sensor data corresponds to the sensor data received at step 301 of the method 300.

[0084] Method 600 includes using 602 eye-tracking system 420 to generate reference eye-tracking data for the instant. Eye-tracking system 420 includes second eye-tracking sensor 421. The reference eye-tracking data is generated by eye-tracking system 420 based on sensor data obtained at the instant by second eye-tracking sensor 421. In other words, the generated reference eye-tracking data corresponds to the generated reference eye-tracking data received at step 302 in method 300.

[0085] The method 600 comprises training 303 an eye tracking model based on the sensor data obtained by the first eye tracking sensor 411 at the time instant and the generated reference eye tracking data. In other words, the method 600 comprises the same training step 303 as the method 300 .

[0086] According to some embodiments, the eye tracking data predicted by the eye tracking model in method 300 or method 600 indicates a predicted gaze point of the eye, and the generated reference eye tracking data (received at step 302 of method 300 or obtained at step 602 of method 600) indicates a reference gaze point of the eye. The predicted gaze point and the reference gaze point may, for example, be a gaze point on a display. This is in Figure 5 , where a predicted gaze point 508 and a reference gaze point 505 are shown on a display 503. Figure 5 Also shown in FIG. 5 is a distance 509 between the two gaze points.

[0087] According to some embodiments, the eye tracking data predicted by the eye tracking model in method 300 or method 600 indicates a predicted line of sight of the eye, and the generated reference eye tracking data (received at step 302 or obtained at step 602) indicates a reference line of sight of the eye. Figure 7, which shows two example sight lines. A first sight line 701 starts from a first estimated eye position 702 and is oriented along a first gaze vector 703. A second sight line 704 starts from a second estimated gaze point 705 and is oriented along a second gaze vector 706. The first sight line 701 may be, for example, a sight line predicted by the eye tracking model in method 300 or method 600, and the second sight line 704 may be, for example, a reference sight line indicated by the generated reference eye tracking data received at step 302 in method 300 or obtained at step 602 in method 600.

[0088] Figure 7 Also shown is the deviation between the line of sight 701 and 704, which may be measured, for example, via the angle 707 formed between the gaze vectors 703 and 706. A distance 708 may also be formed between the estimated eye positions 702 and 705. Figure 7 Also shown is that eye tracking model 710 can employ sensor data (such as eye image 709 ) to predict eye tracking data (such as gaze 701 ).

[0089] According to some embodiments, the eye tracking data predicted by the eye tracking model in method 300 or method 600 indicates a predicted position of the eye in space, and the generated reference eye tracking data (received at step 302 or obtained at step 602) indicates a reference position of the eye in space. Figure 7 exemplified in . Figure 7 The gaze starting point 702 of the first line of sight 701 in FIG. 7 may be the eye position predicted by the eye tracking model in method 300 or method 600 . Figure 7 The gaze starting point 705 of the second line of sight 704 in FIG. 7 may be a reference eye position indicated by the generated reference eye tracking data received at step 302 in method 300 or obtained at step 602 in method 600 .

[0090] Figure 8 It shows how the above reference Figure 3 and Figure 6 The scheme of training the eye tracking model in the described methods 300 and 600. In this embodiment, the step of training 303 the eye tracking model includes:

[0091] predicting 801 eye tracking data for the instant using the eye tracking model and sensor data obtained by the first eye tracking sensor at the instant;

[0092] applying 802 the objective function to at least the eye tracking data predicted by the eye tracking model for the instant and the generated reference eye tracking data; and

[0093] Updated 803 eye tracking model.

[0094] In other words, an objective function (such as a cost function or loss function) is employed to evaluate whether the predictions made by the eye tracking model appear compatible with the reference eye tracking data.The eye tracking model is updated 803 to improve its ability to make accurate predictions.

[0095] Applying the objective function step 802 may include inserting the eye tracking data predicted by the eye tracking model for the moment and the generated reference eye tracking data into the objective function. It will be understood that applying the objective function step 802 may also include, for example, inserting additional data into the objective function.

[0096] Step 803 of updating the eye tracking model may, for example, comprise modifying the value of at least one parameter of the eye tracking model. If the objective function is a cost function (or loss function) that should have a low value when the prediction 801 is accurate, the eye tracking model may, for example, be modified to reduce the value of the objective function (e.g., via grading). On the other hand, if the objective function is a function that should be maximized (e.g., if the objective function is the cost function multiplied by -1), the eye tracking model may, for example, be modified to increase the value of the objective function.

[0097] According to some embodiments, the step of applying 802 the objective function comprises forming a distance between the predicted gaze point indicated by the predicted eye tracking data for the moment in time and a reference gaze point indicated by the generated reference eye tracking data. Figure 5 This is exemplified in , where the distance 509 between the predicted gaze point 508 and the reference gaze point 505 is shown. If the prediction provided by the eye tracking model is accurate, this distance 509 should be small.

[0098] According to some embodiments, the step of applying 802 the objective function comprises forming a deviation between the predicted line of sight indicated by the predicted eye tracking data for the instant in time and the reference line of sight indicated by the generated reference eye tracking data. Figure 7 , where first line of sight 701 corresponds to the line of sight predicted by the eye tracking model in method 300 or method 600, and second line of sight 704 corresponds to the reference line of sight indicated by the generated reference eye tracking data received at step 302 in method 300 or obtained at step 602 in method 600. The deviation between predicted line of sight 701 and reference line of sight 704 can be represented, for example, in the form of an angle 707 formed between corresponding gaze vectors 703 and 706. If the prediction provided by the eye tracking model is accurate, this angle 707 should be small.

[0099] According to some embodiments, the step of applying 802 the objective function comprises forming a distance between the predicted eye position indicated by the predicted eye tracking data for the moment in time and the reference eye position indicated by the generated reference eye tracking data. Figure 7 , wherein a gaze origin 702 of a first line of sight 701 corresponds to an eye position predicted by the eye tracking model in method 300 or method 600, and wherein a gaze origin 705 of a second line of sight 704 corresponds to a reference eye position indicated by the generated reference eye tracking data received at step 302 in method 300 or obtained at step 602 in method 600. If the prediction provided by the eye tracking model is accurate, a distance 708 between the predicted eye position 702 and the reference eye position 705 should be small.

[0100] Figure 9 shows how the preceding reference may be made according to an embodiment Figure 6 and Figure 7 The scheme / method of training an eye tracking model in the described methods 300 and 600. In this embodiment, the step 303 of training the eye tracking model includes using the eye tracking model and the sensor data obtained by the first eye tracking sensor at the moment to predict 901 eye tracking data for the moment. A decision 902 is performed to determine whether the deviation between the eye tracking data predicted 901 by the eye tracking model for the moment and the generated reference eye tracking data (received at step 302 or generated at step 602) exceeds a threshold. If the deviation exceeds the threshold, the scheme / method continues to train 903 the eye tracking model based on the eye tracking data predicted by the eye tracking model for the moment and the generated reference eye tracking data. On the other hand, if the deviation does not exceed the threshold, the eye tracking data predicted by the eye tracking model for the moment and the generated reference eye tracking data may not be used to train the eye tracking model (such as by Figure 9 (as indicated by step 904 in FIG. 1 ).

[0101] In other words, as long as the predicted gaze tracking data agrees with (or matches) the reference gaze tracking data, the eye tracking model may not need to be trained. On the other hand, if it is detected that the predicted eye tracking data deviates from the reference eye tracking data, training may be required. The eye tracking model may, for example, perform well for certain types of input data (or certain situations or user activities) but perform poorly for other types of input data. Figure 9 The described scheme allows the eye tracking model to be trained on such input data that actually requires further training, rather than blindly training on all types of input data.

[0102] exist Figure 9The deviation employed at step 902 in can, for example, be a deviation between a predicted gaze point and a reference gaze point, a deviation between a predicted line of sight and a reference line of sight, or a deviation between a predicted eye position and a reference eye position.

[0103] It can be predefined, for example, in Figure 9 The threshold value used at step 902 in . However, embodiments are also contemplated in which this threshold value may be changed or modified.

[0104] Figure 10 A method 1000 for training an eye tracking model according to an embodiment is shown. The method 1000 is similar to the method described in the previous reference. Figure 3 The method 300 described above, but further comprising using 1001 (in Figure 4 The eye tracking system (exemplified by the eye tracking system 420) detects specific triggering actions of the eye 100. The specific triggering actions include:

[0105] Staring; and / or

[0106] saccades; and / or

[0107] Smooth pursuit.

[0108] In method 1000, step 301 of receiving sensor data obtained by the first eye tracking sensor at a given moment and / or step 303 of training the eye tracking model may be performed in response to detecting a specific triggering eye action. In other words, if a triggering action is detected, data collection step 302 and training step 303 may be performed. On the other hand, if no triggering action is detected, data collection step 302 and training step 303 may be skipped, or data collection step 302 may be performed and training step 303 may be skipped.

[0109] Previous reference Figure 10 The described method 1000 allows an eye tracking model to be trained to better handle certain situations (such as gaze, saccade, or smooth follow-through) rather than blindly training the eye tracking model for all types of input data. Certain situations may, for example, be particularly difficult for an eye tracking model to handle, or certain situations may require higher accuracy than other situations. Therefore, it may be useful to train the model specifically for such situations. The eye tracking model may be trained, for example, based on training data collected in association with a triggering action (such as during a triggering action, and / or shortly before a triggering action, and / or shortly after a triggering action).

[0110] In method 1000 , a triggering action may be detected, for example, by analyzing eye tracking data received from an eye tracking system, or an explicit indication of a triggering action may be received, for example, from an eye tracking system.

[0111] Previous reference Figure 10 The method 1000 described above is Figure 3 The method 300 is provided in the context of the method 300 described. It will be understood that in the previous reference Figure 6 In the context of the described method 600, the method 1000 can be easily modified rather than being used intactly. In other words, steps 602 and / or 303 in the method 600 can be conditioned on the detection of a specific triggering action of the eye 100, just like steps 302 and 303 in the method 300.

[0112] Figure 11 is a flow chart of a method 1100 for training an eye tracking model according to one embodiment. The method 1100 is similar to the method described above with reference to Figure 3 The method 300 is described above, but the method 1100 includes new steps 1101 to 1102, and step 303 is represented by the new step 1103. In this embodiment, the eye tracking model is one of several eye tracking models. These eye tracking models are associated with corresponding potential users or individuals.

[0113] Method 1100 includes detecting 1101 the presence of a user (or person) and selecting 1102 an eye tracking model associated with the user (or person). The user's presence can be detected 1101, for example, by an eye tracking system or via a first eye tracking sensor. The user's presence can be detected, for example, via biometric data (such as facial recognition, or a finger, or iris scan) or via some kind of credential (such as a smart card or wireless sensor tag). The eye tracking model can be selected 1102, for example, from a database of potential users and their corresponding eye tracking models.

[0114] Method 1100 includes training 1103 the selected eye tracking model based on sensor data obtained by the first eye tracking sensor at this moment (in other words, the sensor data received at step 301) and the generated reference eye tracking data (in other words, the reference eye tracking data received at step 302).

[0115] Previous reference Figure 11 The method 1100 described above is Figure 3 The method 300 is provided in the context of the method 300 described. It will be understood that in the previous reference Figure 6In the context of the described method 600, the method 1100 can be easily modified rather than adopted intactly. More specifically, steps 301 to 302 in the method 1100 can be replaced by steps 601 to 602 in the method 600, for example.

[0116] According to some embodiments, the previous reference Figure 3 The described method 300 can be extended to perform training on sensor data from a series of time instants. More specifically, the method 300 can include:

[0117] receiving sensor data acquired by the first eye-tracking sensor at a series of time instants;

[0118] receiving reference eye tracking data generated by the eye tracking system for the series of time instants, wherein the reference eye tracking data for the series of time instants is generated by the eye tracking system based on sensor data obtained by a second eye tracking sensor for the series of time instants; and

[0119] training an eye tracking model based on sensor data obtained by the first eye tracking sensor for the series of time instants and reference eye tracking data generated for the series of time instants, and / or storing sensor data obtained by the first eye tracking sensor for the series of time instants and reference eye tracking data generated for the series of time instants.

[0120] In other words, data for these moments in time can be used to train the eye tracking model, or can be stored for use in later training. The training data can be stored in a database, for example, or can be uploaded to the cloud. Training on the eye tracking data can be performed, for example, at a location remote from where the training data was collected. Training of the eye tracking model can be performed gradually, for example, as training data becomes available. Alternatively, a large amount of training data can be collected first, and then training can be performed using the collected training data. After the eye tracking model has been fully trained, the eye tracking model can then be employed in the eye tracking system to predict eye tracking data in real time.

[0121] Previous reference Figures 3 to 11 The methods and schemes described represent a first aspect of the present disclosure. Figure 4 The described system 400 represents the second aspect of the present disclosure. The system 400 (or the processing circuit system 430 of the system 400) can, for example, be configured to perform the method of any embodiment of the first aspect described above. The system 400 can, for example, be configured to perform the method described above with reference to Figure 3 The method 300 described or previously referenced Figure 6 Method 600 is described.

[0122] The system 400 may, for example, include a processing circuit system 430 (or one or more processors 431) and one or more memories 432, wherein the one or more memories 432 contain instructions that can be executed by the processing circuit system 430 (or one or more processors 431), whereby the system 400 can be operated to perform the method of any embodiment of the first aspect disclosed herein.

[0123] As previously referenced Figure 4 As described, system 400 does not necessarily include Figure 4 All components shown in .

[0124] A third aspect of the present disclosure is represented by an embodiment of a non-transitory computer-readable storage medium 432 having stored thereon instructions that, when executed by the system 400 (or by the processing circuitry 430 of the system 400), cause the system 400 to perform the method of any of the embodiments of the foregoing first aspect (such as those described above with reference to FIG. Figure 3 The method 300 described, or the method previously referenced Figure 6 Method 600 described).

[0125] As previously referenced Figure 4 As depicted, storage medium 432 is not necessarily included in system 400 .

[0126] Those skilled in the art realize that the suggested methods given in this disclosure are by no means limited to the aforementioned preferred embodiments. On the contrary, many modifications and variations are possible within the scope of the appended claims. For example, the aforementioned reference Figures 1 to 11 The described embodiments may be combined to form further embodiments. Further, it will be understood that Figure 4 The system 400 shown in FIG. 4 is intended to be exemplary only, and other systems may also perform the aforementioned Figures 1 to 11 It will also be understood that reference Figure 3 、 Figure 6 、 Figure 8 、 Figure 9 、 Figure 10 and Figure 11 The method steps described do not necessarily have to be performed in the exact order shown in the figures.

[0127] It will be understood that the processing circuit system 430 (or one or more processors) may include a combination of one or more of the following: a microprocessor, a controller, a microcontroller, a central processing unit, a digital signal processor, an application specific integrated circuit, a field programmable gate array, or any other suitable computing device, resource, or combination of hardware, software and / or encoded logic that is operable to provide computer functionality alone or in combination with other computer components (such as memory or storage media).

[0128] It will also be understood that the memory or storage medium 432 (or computer-readable medium) may include any form of volatile or non-volatile computer-readable memory, including but not limited to persistent storage, solid-state memory, remotely mounted memory, magnetic media, optical media, random access memory (RAM), read-only memory (ROM), mass storage media (e.g., a hard disk), removable storage media (e.g., a flash drive, a compact disk (CD), or a digital video disk (DVD)), and / or any other volatile or non-volatile, non-transitory device-readable and / or computer-executable memory device that stores information, data, and / or instructions that may be used by a processor or processing circuit system.

[0129] In addition, from a study of the drawings, the present disclosure and the appended claims, those skilled in the art can understand and implement various changes to the disclosed embodiments when practicing the claimed invention. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude plurality. In the claims, the word "or" is not to be interpreted as an exclusive or (sometimes referred to as "XOR"). On the contrary, unless otherwise indicated, expressions such as "A or B" cover all cases "A but not B", "B but not A" and "A and B". The mere fact that certain measures are listed in different dependent claims does not indicate that a combination of these measures cannot be used to advantage. Any figure marks in the claims should not be interpreted as limiting the scope.

Claims

1. A method (300) for training an eye tracking model (710), wherein: The eye tracking model is adapted for estimating eye tracking data based on sensor data (709) from a first eye tracking sensor (411) of a new eye tracking system (410), wherein the first eye tracking sensor (411) is arranged for monitoring an eye (100), the method comprising: Receiving (301) sensor data obtained by the first eye tracking sensor at a certain moment; receiving (302) reference eye tracking data generated for approximately the same moment in time by an old eye tracking system (420) comprising a second eye tracking sensor (421) arranged to monitor the eye, wherein the reference eye tracking data is generated by the old eye tracking system based on sensor data obtained by the second eye tracking sensor at the approximately same moment in time, the approximately same moment in time being different from the certain moment in time and having a slight deviation relative to the certain moment in time; and The eye tracking model is trained (303) based on the sensor data obtained by the first eye tracking sensor at the certain moment in time and the generated reference eye tracking data for the approximately same moment in time.

2. The method of claim 1, further comprising: At said certain moment in time, sensor data is obtained using (601) said first eye tracking sensor.

3. The method according to any one of the preceding claims, further comprising: The old eye tracking system is used (602) to generate the reference eye tracking data for the substantially same moment in time, the substantially same moment in time being different from the certain moment in time and having a slight deviation relative to the certain moment in time.

4. The method according to claim 1, in, The eye tracking data predicted by the eye tracking model indicates a predicted gaze point (508) of an eye (100), and wherein the generated reference eye tracking data indicates a reference gaze point (505) of the eye; and / or wherein the eye tracking data predicted by the eye tracking model indicates a predicted line of sight of the eye (701), and wherein the generated reference eye tracking data indicates a reference line of sight of the eye (704); and / or wherein the eye tracking data predicted by the eye tracking model indicates a predicted position of the eye in space (702), and wherein the generated reference eye tracking data indicates a reference position of the eye in space (705).

5. The method according to claim 1, wherein Training the eye tracking model includes: predicting (801) eye tracking data for the certain moment in time using the eye tracking model and the sensor data obtained by the first eye tracking sensor at the certain moment in time; applying (802) an objective function to at least the eye tracking data predicted by the eye tracking model for the certain instant in time and the generated reference eye tracking data for the approximately same instant in time; and The eye tracking model is updated (803).

6. The method according to claim 5, wherein: Applying the objective function to at least the eye tracking data predicted by the eye tracking model for the certain moment in time and the generated reference eye tracking data for the approximately same moment in time comprises: forming a distance (509) between a predicted gaze point (508) indicated by the predicted eye-tracking data for the certain moment in time and a reference gaze point (505) indicated by the generated reference eye-tracking data; and / or forming a deviation (707) between a predicted line of sight (701) indicated by the predicted eye-tracking data for the certain moment and a reference line of sight (704) indicated by the generated reference eye-tracking data; and / or A distance (708) is formed between a predicted eye position (702) indicated by the predicted eye tracking data for the certain moment in time and a reference eye position (705) indicated by the generated reference eye tracking data.

7. The method of claim 1, wherein: The first eye tracking sensor is an imaging device, and / or wherein the second eye tracking sensor is an imaging device.

8. The method of claim 1, wherein: The old eye tracking system includes a light emitting device (422, 423, 424) that outputs light (428, 429) within a certain wavelength range to illuminate the eye (100), the second eye tracking sensor provides sensor data based on the light within the wavelength range, and the first eye tracking sensor is provided with a filter (415) for suppressing light within the wavelength range.

9. The method of claim 1, wherein: Training the eye tracking model includes: predicting (901) eye tracking data for the certain moment in time using the eye tracking model and the sensor data obtained by the first eye tracking sensor at the certain moment in time; and In response to a deviation (509, 707, 708) between the eye tracking data predicted by the eye tracking model for the certain moment and the generated reference eye tracking data for the approximately same moment exceeding a threshold, the eye tracking model is trained (903) based on the eye tracking data predicted by the eye tracking model for the certain moment and the generated reference eye tracking data for the approximately same moment.

10. The method of claim 1, further comprising: using the old eye tracking system (1001) to detect a specific triggering action of the eye (100), The specific triggering actions include: Staring; and / or Saccades; and / or Smooth follow-up; Wherein, receiving the sensor data obtained by the first eye tracking sensor at the certain moment and / or training the eye tracking model are performed in response to detecting the specific trigger action of the eye.

11. The method of claim 1, wherein: The eye tracking model is one of several eye tracking models, the several eye tracking models being based on any one of a machine learning eye tracking model, an artificial neural network, or a convolutional neural network, the several eye tracking models being associated with corresponding potential users, the method comprising: Detecting (1101) the presence of a user; selecting (1102) an eye tracking model associated with the user; and The selected eye tracking model is trained (1103) based on the sensor data obtained by the first eye tracking sensor at the certain moment in time and the generated reference eye tracking data for the approximately same moment in time.

12. The method of claim 1, comprising: receiving sensor data acquired by the first eye tracking sensor at a series of time instants; receiving reference eye tracking data generated by the old eye tracking system for the series of time instants, wherein the reference eye tracking data for the series of time instants was generated by the old eye tracking system based on sensor data obtained by the second eye tracking sensor for the series of time instants; as well as The eye tracking model is trained based on the sensor data obtained by the first eye tracking sensor for the series of moments and the generated reference eye tracking data for the series of moments, and / or the sensor data obtained by the first eye tracking sensor for the series of moments and the generated reference eye tracking data for the series of moments are stored.

13. A system (400) for training an eye tracking model (710), wherein: The eye tracking model is adapted for estimating eye tracking data based on sensor data (709) from a first eye tracking sensor (411) of a new eye tracking system (410), wherein the first eye tracking sensor (411) is arranged to monitor an eye (100), the system comprising processing circuitry (430) configured to: receiving sensor data obtained by the first eye tracking sensor at a certain moment; Characterized in that the processing circuit system (430) is further configured to: receiving reference eye tracking data generated for approximately the same instant in time by an old eye tracking system (420) comprising a second eye tracking sensor (421) arranged to monitor the eye, wherein the reference eye tracking data is generated by the old eye tracking system based on sensor data obtained by the second eye tracking sensor at the approximately same instant in time, the approximately same instant in time being different from the certain instant in time and having a slight deviation relative to the certain instant in time; and The eye tracking model is trained based on the sensor data obtained by the first eye tracking sensor at the certain moment in time and the generated reference eye tracking data for the approximately same moment in time.

14. The system of claim 13, further comprising the first eye tracking sensor, the processing circuitry being further configured to: Sensor data is obtained using the first eye tracking sensor at the certain moment in time.

15. The system of any one of claims 13 to 14, further comprising the legacy eye tracking system, the processing circuitry being further configured to: The reference eye tracking data is generated for the substantially same instant in time using the old eye tracking system.

16. The system of claim 13, wherein: The first eye tracking sensor is an imaging device, and / or wherein the second eye tracking sensor is an imaging device.

17. The system of claim 13, wherein: The old eye tracking system includes a light emitting device (422, 423, 424) configured to output light (428, 429) within a certain wavelength range to illuminate the eye (100), the second eye tracking sensor configured to provide sensor data based on the light within the wavelength range, and the first eye tracking sensor provided with a filter (415) for suppressing light within the wavelength range.

18. The system of claim 13, wherein: The processing circuitry is configured to train the eye tracking model by at least: predicting eye tracking data for the certain moment in time using the eye tracking model and the sensor data obtained by the first eye tracking sensor at the certain moment in time; and In response to a deviation (509, 707, 708) between the eye tracking data predicted by the eye tracking model for the certain moment and the generated reference eye tracking data for the approximately same moment exceeding a threshold, the eye tracking model is trained based on the eye tracking data predicted by the eye tracking model for the certain moment and the generated reference eye tracking data for the approximately same moment.

19. The system of claim 13, wherein: The processing circuitry is further configured to: Using said old eye tracking system to detect a specific triggering movement of the eye (100), The specific triggering actions include: Staring; and / or Saccades; and / or Smooth follow-up; The processing circuit system is configured to receive the sensor data obtained by the first eye tracking sensor at the certain moment and / or train the eye tracking model in response to detecting the specific trigger action of the eye.

20. A non-transitory computer-readable storage medium (432) storing instructions for training an eye tracking model (710), wherein: The eye tracking model is adapted to predict eye tracking data based on sensor data (709) from a first eye tracking sensor (411) of a new eye tracking system (410), wherein the first eye tracking sensor (411) is arranged to monitor an eye (100), wherein the instructions, when executed by the system (400), cause the system to: receiving sensor data obtained by the first eye tracking sensor at a certain moment; Characterized in that, when the instructions are executed by the system (400), the system further causes the system to perform the following operations: receiving reference eye tracking data generated for approximately the same instant in time by an old eye tracking system (420) comprising a second eye tracking sensor (421) arranged to monitor the eye, wherein the reference eye tracking data is generated by the old eye tracking system based on sensor data obtained by the second eye tracking sensor at the approximately same instant in time, the approximately same instant in time being different from the certain instant in time and having a slight deviation relative to the certain instant in time; and The eye tracking model is trained based on the sensor data obtained by the first eye tracking sensor at the certain moment in time and the generated reference eye tracking data for the approximately same moment in time.

Images