A method and apparatus for measuring accuracy of a gaze estimation algorithm

Abstract

The present application relates to the technical field of measuring the precision of gaze estimation algorithm, and proposes a method for measuring the precision of gaze estimation algorithm and a matching device thereof. The device comprises a three-degree-of-freedom motion controller for controlling a point laser emitter to move in the front-back direction opposite to the center of a projection plane, to rotate in the horizontal (yaw) direction, and to rotate in the vertical (pitch) direction. The device further comprises a point laser emitter for irradiating a gaze point on the projection plane, a projection plane for displaying the irradiation point of the point laser emitter, and a camera for capturing an eye image of a subject when the subject gazes at the gaze point. The method can satisfy the precision measurement of the gaze estimation algorithm in the positive and negative directions of pitch and yaw within any eye rotation angle range, can simultaneously satisfy the precision measurement of the gaze estimation algorithm for a larger gaze angle, and can satisfy the precision measurement requirements of the gaze estimation algorithm in different distance ranges.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gaze estimation, in particular to a method and device for measuring the accuracy of a gaze estimation algorithm. BACKGROUND

[0002] The movement of the human eye has three degrees of freedom, namely pitch, yaw and roll. Gaze estimation of the human eye is a technology for determining the rotation angles of the eyeballs in the above three directions and the gaze position of the human eye. Currently, those skilled in the art mainly focus on the movement of the human eye in the pitch and yaw directions.

[0003] Gaze estimation technology has important application value in the fields of advertising analysis, psychological research, medical diagnosis, etc. In human-computer interaction, eye gaze can replace traditional mouse clicking actions, making the interaction fast, interesting and natural. Thanks to the rapid development of neural networks and deep learning technology, those skilled in the art have proposed numerous eye gaze estimation algorithms. However, there is still a lack of a method for measuring the accuracy and precision of these developed gaze estimation algorithms in a large angular range and at different distances.

[0004] It is necessary to measure the accuracy of gaze estimation algorithms at different distances, which is also a result that must be measured in algorithm development. In the existing technology disclosed in the prior art, a patent with application publication number CN114646457A proposes a method and system for testing the accuracy of eye movement tracking, which includes: controlling an artificial eye to continuously rotate along the measurement direction with the center point as the starting measurement point and with a preset angle gradient value; collecting test data for each tracking angle by an eye movement tracking device and comparing it with the line-of-sight position detected by a preset eye movement algorithm to calculate the accuracy and precision of each tracking angle. However, the method in patent CN114646457A cannot satisfy the horizontal comparison of the accuracy and precision of gaze estimation algorithms in different distances, such as 30 cm, 50 cm, 80 cm, etc. Moreover, the method proposed in this patent is only applicable to artificial eyes and cannot be determined by the subject or mastered by the subject, which has great limitations and cannot be applied flexibly. The method proposed in this application avoids the above limitations and can measure the accuracy of gaze estimation algorithms at any angle in the positive and negative directions of pitch and yaw, and can be used in various test environments, such as different lighting, different head postures of the subject, real eyes and simulated artificial eyes, etc. SUMMARY

[0005] In view of the technical problems existing in the prior art, the present application provides a method for measuring the accuracy of a gaze estimation algorithm, and simultaneously provides a device for measuring the accuracy of a gaze estimation algorithm.

[0006] The technical scheme adopted by the present application to solve the above technical problems is:

[0007] A method for measuring the accuracy of a gaze estimation algorithm, comprising the following steps:

[0008] S201: System calibration, obtain system deviation pitch(0) and yaw(0);

[0009] S202: Set the number of data collection times N;

[0010] S203: The subject sets the preset angle β, yaw(i) in the pitch and yaw directions, and the motion controller controls the point laser emitter to rotate pitch(i) and yaw(i) degrees in the pitch and yaw directions for this preset angle;

[0011] S204: The point laser emitter irradiates the projection plane to generate the gaze point at the preset angle;

[0012] S205: The subject looks at the gaze point, the camera takes a picture of the subject's eye, and the gaze estimation algorithm takes the image as input to calculate the subject's eye rotation angle pitch(j), yaw(j) corresponding to the current gaze point;

[0013] S206: Calculate the deviation of the gaze estimation algorithm for the current preset angle;

[0014] S207: Determine whether the sampling number of the subject at the preset angle reaches N groups, if yes, execute S208; if not, return to S203 to continue collecting sample data;

[0015] S208: The true rotation angle measured by the system is taken as the true value, the average deviation mean(Δpitch) and mean(Δyaw) of the eye rotation angle detected by the gaze estimation algorithm of N groups of subjects from the true value are calculated as the measurement accuracy of the gaze estimation algorithm for the current preset angle of the subject; at the same time, the root mean square error RMSE(pitch) and RMSE(yaw) of the N groups of data from the true value are calculated as the accuracy of the measurement result of the gaze estimation algorithm for the preset angle of the subject.

[0016] Further, before measuring the accuracy of the gaze estimation algorithm each time using the method, the point laser emitter is in a zero state, i.e. the point laser emitter is perpendicular to the projection plane and has no deflection in the pitch and yaw directions. Before using the method each time, the subject needs to do system calibration, and the steps are:

[0017] S301: Set the calibration times M;

[0018] S302: In the zero-return state, the point laser emitter vertically irradiates the i-th initial fixation point at any position on the projection plane;

[0019] S303: The subject is located directly below the point laser emitter, and the gaze of the subject is fixed on the i-th initial fixation point. The camera captures the eye image of the subject at this time, and the gaze estimation algorithm takes this image as input to detect the i-th calibration result;

[0020] S304: Determine whether the calibration times have reached the expected set M times. If yes, stop calibration. If not, continue calibration;

[0021] S305: The gaze estimation algorithm takes M calibration images as input, and the average of the M yaw direction eye rotation angles detected is denoted as the system deviation yaw(0). The system obtains the pitch direction deviation pitch(0) after completing all calibration work, and the system deviation pitch(0) in the pitch direction at the i-th calibration i The calculation is as follows:

[0022]

[0023] The calculation method of pitch(0) and yaw(0) in the system calibration link is as follows:

[0024]

[0025]

[0026] Where h is the distance from the installation position of the point laser emitter to the subject's eye, d is the pre-fixed distance from the subject's eye to the projection plane, pitch i ′ is the pitch direction eye rotation angle value of the subject detected by the gaze estimation algorithm in the i-th calibration link, yaw(0) i is the yaw direction eye rotation angle value of the subject detected by the gaze estimation algorithm in the i-th calibration link.

[0027] Further: In the step of the subject setting a preset angle, the motion controller controlling the point laser emitter to rotate pitch(i) degrees and yaw(i) degrees in the pitch and yaw directions, the rotation angle of the point laser emitter is calculated as follows:

[0028] It is specified that the upward gaze of the subject's eye is the positive direction of pitch, and when the point laser emitter rotates pitch(i) degrees in the negative direction of pitch:

[0029]

[0030] When The point laser emitter rotates pitch(i) degrees in the positive direction of pitch:

[0031]

[0032] When the subject measures the accuracy of the gaze estimation algorithm in the negative direction of pitch, the point laser emitter rotates pitch(i) degrees in the negative direction of pitch:

[0033]

[0034] Wherein, β all refers to the preset angle value of the subject in the pitch direction, yaw(i) is the degree number of the point laser emitter rotating according to the preset angle value of the subject in the yaw direction, pitch(i) all refers to the rotation angle of the point laser emitter in the pitch direction when working, h all is the vertical distance from the installation position of the point laser emitter to the subject's eye, and d all is the vertical distance between the subject's eye and the projection plane.

[0035] Further, the distance d between the subject's eye and the projection plane is set by the subject according to the distance required for measuring the accuracy of the gaze estimation algorithm. Before using the method, the subject can independently set the distance between the eye and the projection plane, such as 30 cm, 50 cm, 70 cm, etc., that is, the accuracy of the gaze estimation algorithm developed by the subject in different distance ranges can be measured using the method.

[0036] Further: the preset angle of the subject is taken as the true value, and the deviation calculation method of the gaze estimation algorithm for the current preset angle of the subject is:

[0037] Δyaw=yaw(j)-yaw(i)-yaw(0)

[0038] Δpitch=pitch(j)-β-pitch(0)

[0039] Wherein, yaw(j) is the yaw direction detection result of the gaze estimation algorithm for the current preset angle of the subject, yaw(i) is the yaw direction rotation degree number of the eye preset by the subject, which is also the yaw direction rotation degree number of the point laser emitter; pitch(j) is the detection result of the gaze estimation algorithm for the preset rotation angle of the eye in the pitch direction of the subject, yaw(0) and pitch(0) are the system errors obtained after M times of calibration before sample collection, and β is the preset rotation angle in the pitch direction set by the subject.

[0040] Further: the average deviation calculation method of the eye rotation angle obtained by the N groups of subjects through the gaze estimation algorithm and the true rotation angle measured by the system is:

[0041]

[0042]

[0043] Further: the calculation method of the root mean square error RMSE (pitch) and RMSE (yaw) of the N groups of collected samples and true values is:

[0044]

[0045]

[0046] The preset angle of the subject is taken as the true value, and the multiple groups of collected gaze estimation algorithm detection results are taken as samples. The average deviation of the multiple samples from the true value is calculated as the accuracy measurement result of the gaze estimation algorithm, and the root mean square error of the multiple samples from the true value is calculated as the precision measurement result of the gaze estimation algorithm. Finally, the pitch and yaw direction detection precision of the gaze estimation algorithm for the preset eyeball rotation angle of the subject is obtained.

[0047] The device matched with the method for measuring the precision of the gaze estimation algorithm comprises: a 3-DOF motion controller controlling a point laser emitter to rotate in the pitch direction and the yaw direction and to move in the front-rear direction opposite to the center of the projection plane; the point laser emitter is set at a distance from the projection plane and irradiates a gaze point on the projection plane; the subject gazes at the gaze point on the projection plane through the naked eye at a position directly below the point laser emitter; a camera captures an eye image of the subject when the subject gazes at the gaze point; a gaze estimation algorithm takes the image as input to detect the rotation angle of the eyeball of the subject in the pitch and yaw directions; the rotation angle of the eyeball of the subject measured by the system is taken as the true value, and the average deviation and the root mean square error of the detection result of the gaze estimation algorithm from the true value are finally calculated as the precision measurement result of the gaze estimation algorithm.

[0048] Compared with the prior art, the application has the beneficial effects that the accuracy measurement of the gaze estimation algorithm in the positive and negative directions of pitch and yaw is met, the cumbersome calculation in the prior art is avoided, the time for measuring the accuracy of the gaze estimation algorithm is shorter, and the application is more suitable for practical application; the device and method for measuring the accuracy of the gaze estimation algorithm only need the simple cooperation of the subject to complete all the accuracy measurement work, the use scene has low requirements on the environment, is more universal, and can be used in various test environments, such as different illumination, different head postures of the subject, real naked eyes and simulated eyeball models; the method can meet the accuracy measurement of the gaze estimation algorithm in a larger gaze angle (such as 70°-90°), the subject can test the accuracy of the gaze estimation algorithm at different distances set by the subject by using the device and method, and the algorithm accuracy test needs of actual engineering project development are greatly met; the root mean square error between the gaze estimation algorithm detection result and the real rotation angle is used as the precision standard of the gaze estimation algorithm model, and in actual engineering measurement, the root mean square error can better reflect the precision of measurement. BRIEF DESCRIPTION OF DRAWINGS

[0049] FIG. 1 is a schematic diagram of the method proposed in the application in use; Figure 1 FIG. 1 is a schematic diagram of the method proposed in the application in use;

[0050] In the figure, 101 is a motion controller, 102 is a point laser emitter, 103 is a projection plane, 104 is a subject, 105 is a camera, the solid line indicates the laser irradiated by the point laser emitter, and the dotted line indicates the line of sight of the subject.

[0051] FIG. 2 is a flow chart of the whole system of the application; Figure 2 FIG. 2 is a flow chart of the whole system of the application;

[0052] FIG. 3 is a flow chart of the system calibration link of the application. Figure 3 FIG. 3 is a flow chart of the system calibration link of the application. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical scheme and advantages of the embodiments of the application more clear, the application will be further described in detail below with reference to the drawings and embodiments.

[0054] Embodiment 1:

[0055] Please refer to Figure 1The schematic diagram of the device for measuring the accuracy of the gaze estimation algorithm, the device comprises: (1) a three-degree-of-freedom motion controller, which controls the point laser emitter to rotate in the pitch direction and the yaw direction and to move in the front-back direction opposite to the center of the projection plane; (2) a point laser emitter, which is used to irradiate the gaze point on the projection plane; (3) a projection plane, which is used to display the irradiation point of the point laser emitter; and (4) a camera, which is used to capture the eye image of the subject when the subject gazes at the gaze point.

[0056] The device for measuring the accuracy of the gaze estimation algorithm has the following specific principles and characteristics:

[0057] (1) The three-degree-of-freedom motion controller is connected with the point laser emitter and controls the point laser emitter to move in the front-back direction opposite to the projection plane and to rotate in the vertical and horizontal directions to generate the gaze point at different angles.

[0058] (2) The point laser emitter generates a clear and visible gaze point when irradiating the projection plane, and the movement of the point laser emitter is controlled by the three-degree-of-freedom motion controller, and the rotation center of the point laser emitter is opposite to the center of the projection plane.

[0059] (3) The projection plane can clearly display the light point projected by the point laser emitter, and the size of the projection plane is calculated according to the following formula:

[0060] Length: 2d max *tanθ

[0061]

[0062] wherein d max is the maximum value of the distance range from the eye of the subject to the projection plane, which is also the maximum detection distance of the predetermined accuracy of the gaze estimation algorithm of the subject; θ is the maximum detection range of the eye rotation angle of the gaze estimation algorithm in the yaw direction, such as ±70°; is the maximum detection range of the eye rotation angle of the gaze estimation algorithm in the pitch direction, such as ±80°.

[0063] (4) The camera used in the method is installed between the subject and the projection plane, and the selection of the camera is based on the needs of the gaze estimation algorithm developed by the subject. The subject can use an RGB camera, of course, an IR camera or a Tof camera, etc. The selection of the camera is not limited in the application.

[0064] S201, please refer to Figure 2 and Figure 3 After the subject installs the complete set of devices of the application, the subject must follow Figure 3The system calibration link corrects the system angle error pitch(0) and yaw(0), and the calculation method is as follows:

[0065]

[0066]

[0067] Wherein, pitch(0) i and yaw(0) i are the system angle deviation in pitch and yaw directions at the i-th system calibration, pitch(0) i is calculated according to the following formula:

[0068]

[0069] In the formula, h is the distance from the point laser emitter installation position to the subject's eye, d is the pre-fixed distance from the subject's eye to the projection plane, pitch i is the value of the subject's eye rotation angle in the pitch direction detected by the gaze estimation algorithm in the i-th calibration link, and M is the number of calibrations.

[0070] In addition, it should be noted that when the subject uses the method proposed in the present application to detect the accuracy of the gaze estimation algorithm, the position must be directly below the point laser emitter.

[0071] S202, please refer to Figure 2 Each accuracy measurement of the predetermined eye rotation angle needs to repeat N experiments, and collect N sets of sufficient data. The data amount N can be set by the subject. If the number of collected times is less than N, the experiment needs to be continued. Only when the number of repeated experiments is large enough, the final calculation result of the accuracy of the gaze estimation algorithm can be true and accurate.

[0072] S203, after the subject fixes the distance d from the eye to the projection plane, the accuracy of the gaze estimation algorithm under the predetermined eye rotation angle (β and yaw(i)) in the pitch or yaw direction can be detected.

[0073] It should be noted that when measuring the accuracy of the yaw direction gaze estimation algorithm, the point laser emitter is directly rotated according to the yaw direction angle yaw(i) predetermined by the subject. When measuring the accuracy of the pitch direction gaze estimation algorithm, the rotation angle of the point laser emitter in the pitch direction is pitch(i) calculated after analysis and judgment, rather than the pitch direction angle β predetermined by the subject.

[0074] S204, the point laser emitter irradiates the projection plane to generate the gaze point under the predetermined angle.

[0075] S205, the subject's gaze looks at the fixation point, the camera shoots the subject's eyes, and the gaze estimation algorithm takes the image as input to calculate the subject's eye rotation angle pitch(j), yaw(j) corresponding to the current fixation point.

[0076] Please refer to Figure 2 S206-S208 gaze estimation algorithm accuracy measurement steps, the following yaw direction and pitch direction of the eye gaze accuracy measurement are introduced in detail, and the detailed calculation method is as follows:

[0077] When measuring the accuracy of yaw direction, the point laser emitter directly rotates yaw(i) degrees according to the preset angle of the subject under the control of the motion controller. At this time, the calculation method of the gaze estimation algorithm for the accuracy mean(Δyaw) and the accuracy RMSE(yaw) of the yaw direction eye rotation angle is:

[0078]

[0079]

[0080] Where yaw(j) is the yaw direction detection result of the gaze estimation algorithm for the subject's current preset angle, yaw(i) is the subject's preset eye rotation degree in the yaw direction, and also the point laser emitter's rotation degree in the yaw direction.

[0081] When measuring the accuracy of pitch direction, the subject's eye looking up is defined as the positive direction of pitch, and when the subject's preset eye rotation angle is pitch(i) degrees, the point laser emitter rotates pitch(i) degrees in the negative direction of pitch:

[0082]

[0083] When the point laser emitter rotates pitch(i) degrees in the positive direction of pitch:

[0084]

[0085] When the subject measures the accuracy of the gaze estimation algorithm in the negative direction of pitch, the point laser emitter rotates pitch(i) degrees in the negative direction of pitch:

[0086]

[0087] Wherein, βi all refer to the preset eye rotation angle value of the subject in the pitch direction, pitch(i) all refer to the rotation angle in the pitch direction when the point laser emitter works, hi all refer to the vertical distance from the installation position of the point laser emitter to the eye of the subject, di all refer to the vertical distance between the eye of the subject and the projection plane. At this time, the calculation method of the accuracy mean(Δpitch) and the precision RMSE(pitch) of the gaze estimation algorithm for the eye rotation angle in the pitch direction is as follows:

[0088]

[0089]

[0090] Wherein, pitch(j) is the detection result of the gaze estimation algorithm for the predetermined rotation angle of the eye of the subject in the pitch direction.

[0091] The above method is described below with a specific embodiment:

[0092] Taking an open-source eye gaze estimation algorithm on github as an example, the distance between the eye of the subject and the projection plane is 60 cm, the camera is fixed between the subject and the projection plane and the distance between the camera and the eye of the subject is 50 cm, and the point laser emitter is installed 30 cm above the eye of the subject. Before measuring the precision of the open-source gaze estimation algorithm using the method proposed in the present application, the subject obtains the system angle deviation through 20 calibrations. The subject is scheduled to measure the precision of the open-source algorithm for the eye rotation angle of -70°~70° in the yaw direction, and then the point laser emitter rotates the predetermined angle under the action of the motion controller and projects the gaze point on the projection plane. The subject looks at the gaze point on the projection plane, the camera captures the eye image of the subject when the gaze state is maintained and takes it as the input of the open-source algorithm, the open-source gaze estimation algorithm detects the input image, saves the detection result, and repeats the experiment for each predetermined angle. According to the obtained data, the precision detection results of the open-source gaze estimation algorithm for the predetermined angles of -70°, -60°, -50°…60°, 70° of the subject calculated by the method proposed in the present application are shown in the following table:

[0093] Predicted eye rotation angle / ° Measured algorithm accuracy / ° Measured algorithm precision / ° -70° 0.55715 0.12626 -60° 0.3827 0.1472 -50° 0.91605 0.7725 -40° 0.93029 0.74793 -30° 0.99206 0.83967 -20° 0.89847 0.71828 -10° 0.90317 0.68587 10° 0.56295 0.16044 20° 0.63264 0.28518 30° 0.69665 0.26704 40° 0.63842 0.3222 50° 0.75462 0.49656 60° 0.51351 0.35364 70° 0.98581 0.62675

[0094] The above only describes the preferred embodiments of the present application, and is not a limitation on the technical solutions of the present application. It should be noted that for those skilled in the art, further improvements and changes can be made without departing from the overall concept of the present application, and these improvements and changes should be covered within the protection scope of the present application.

Claims

1. A method of measuring accuracy of a gaze estimation algorithm, characterized by, The method comprises the following steps: S201: system calibration, obtain system deviation pitch(0) and yaw(0); S202: set the data acquisition times N; S203: The subject sets a preset angle in the pitch and yaw directions , yaw(i), the motion controller control point laser emitter rotates in the pitch and yaw directions by pitch(i) and yaw(i) degrees for this preset angle; S204: the point laser emitter irradiates the projection plane to generate the gaze point under the preset angle; S205: the subject looks at the gaze point, the camera takes the subject's eye, and the gaze estimation algorithm takes the image as input to calculate the subject's eye rotation angle pitch(j), yaw(j) corresponding to the current gaze point; S206: calculate the deviation of the gaze estimation algorithm for the current preset angle; S207: judge whether the sampling times of the subject under the preset angle reach N groups, if yes, execute S208; if not, return to S203 to continue collecting sample data; S208: The real rotation angle measured by the system as the true value, calculate the average deviation of the eye rotation angle detected by the gaze estimation algorithm of the N groups of subjects from the true value and , as the measurement accuracy of the gaze estimation algorithm for the preset angle of the current subject; at the same time, calculate the root mean square error of the N groups of data and the true value and as the accuracy of the measurement result of the gaze estimation algorithm for the preset angle of the subject; Distance between the subject's eye and the projection plane Set by the subject according to the accuracy of the measured gaze estimation algorithm distance requirements; The device for measuring the accuracy of the gaze estimation algorithm is used to measure the accuracy of the gaze estimation algorithm, and the device comprises: A three-degree-of-freedom motion controller connected with the point laser emitter, which controls the point laser emitter to move forward and backward in the direction opposite to the projection plane, and rotates in the vertical and horizontal directions to generate the gaze point under different angles; The point laser emitter irradiates the projection plane to generate a clear and visible gaze point, and its movement is controlled by the three-degree-of-freedom motion controller, and the rotation center of the point laser emitter is opposite to the center of the projection plane; The projection plane can clearly present the light point projected by the point laser emitter, and its size is calculated according to the following formula: Long: ; Wide: ; wherein, a maximum value of the distance range of the eye of the subject to the projection plane, which is also a maximum detection distance of the predetermined gaze estimation algorithm accuracy of the subject; a maximum detection range of the eye rotation angle of the gaze estimation algorithm in the yaw direction of the subject; a maximum detection range of the eye rotation angle of the gaze estimation algorithm in the pitch direction of the subject; The camera is installed between the subject and the projection plane.

2. The method of claim 1, wherein: The system calibration method comprises the following steps: S301: set the calibration times M; S302: when the zero state, the point laser emitter vertically irradiates the i-th initial gaze point at any position on the projection plane; S303: the subject is located directly below the point laser emitter, and looks at the i-th initial gaze point, the camera takes the image of the subject's eye at this time, and the gaze estimation algorithm takes the image as input to detect the i-th calibration result; S304: judge whether the calibration times have reached the expected set M times, if yes, stop calibration; if not, continue calibration; S305: The gaze estimation algorithm takes M calibration images as input, and detects the average of the M yaw direction eye rotation angles, denoted as the system bias yaw(0); the system bias pitch(0) in the pitch direction obtained after all calibration work is completed, and the system bias pitch(i) in the pitch direction at the i-th calibration This is calculated as follows: ； The calculation method of pitch(0) and yaw(0) in the system calibration link is: ； ； wherein, is a distance from a position of the point laser emitter to the eye of the subject, is a distance from the eye of the subject to the projection plane fixed in advance for the subject, is a value of the eye rotation angle of the subject in the pitch direction detected by the gaze estimation algorithm in the i-th calibration link, is a value of the eye rotation angle of the subject in the yaw direction detected by the gaze estimation algorithm in the i-th calibration link.

3. The method of claim 1, wherein: When measuring the yaw direction accuracy, the point laser emitter directly rotates yaw(i) degrees under the control of the motion controller according to the preset angle of the subject; when measuring the pitch direction accuracy, the rotation angle of the point laser emitter in the pitch direction is calculated according to the following method: The subject's eyes were positioned upwards in the positive direction of the pitch. At that time, the point laser emitter rotates in the negative pitch direction. Spend: ； When , the point laser emitter is rotated in the pitch positive direction degrees: ； When the subject measures the accuracy of the pitch negative gaze estimation algorithm, the dot laser emitter is rotated in the pitch negative direction degrees: ； wherein, both refer to a preset angle value of the subject in the pitch direction, yaw(i) is the degree number of the point laser emitter rotating at a preset angle value of the subject in the yaw direction, both refer to the rotation angle of the point laser emitter in the pitch direction when working, both are the vertical distance from the installation position of the point laser emitter to the eye of the subject, both are the distance from the eye of the subject to the projection plane fixed in advance by the subject.

4. The method of claim 1, wherein: The preset angle of the subject measured by the system is taken as the true value, and the deviation calculation method of the gaze estimation algorithm for the current preset angle of the subject is: ； ； wherein, is the yaw direction detection result of the gaze estimation algorithm for the subject's current preset angle, is the yaw direction rotation degree of the subject's preset eye, and is also the yaw direction rotation degree of the point laser emitter; is the pitch direction detection result of the gaze estimation algorithm for the subject's eye's predetermined rotation angle, and is the system error obtained after M calibrations before the sample is collected, is the pitch direction preset rotation angle set by the subject.

5. The method of claim 1, wherein: The average deviation calculation method of the eye rotation angle obtained by the gaze estimation algorithm of the N subjects and the true rotation angle measured by the system is: ； ； Further: the root mean square error of the N groups of collected samples and the true value and The calculation method is: ； 。 6. The method of claim 1, wherein: When the subject uses the device, he must be located directly below the installation position of the point laser emitter.

Images