Online examination anti-cheating method and system based on AI fusion sight tracking

Abstract

The invention provides an on-line examination anti-cheating method and system based on AI fusion sight tracking, and relates to the technical field of on-line education safety, and the method comprises the steps: extracting a head motion feature sequence of an examinee, calculating the coordinates and duration of a sight fixation point, analyzing the fixation behavior of a functional region, and recognizing a periodic jump mode in a fixation migration feature map. The gazing mode mutation point is identified by combining autoregression spectrum analysis and wavelet analysis, so that the cheating behavior is accurately identified, the safety and fairness of the online examination are effectively improved, and the misjudgment rate is reduced.

Description

Technical Field

[0001] The present invention relates to the field of online education security technology, and in particular to an online exam anti-cheating method and system based on AI-integrated eye tracking. Background Art

[0002] With the rapid development of online education, online exams have become a crucial method for educational assessment. However, cheating in online exams has also increased, posing serious challenges to educational equity and exam quality. Traditional online exam anti-cheating systems rely primarily on camera monitoring, screen locking, and behavior logging, but these methods have numerous limitations in practical application.

[0003] Currently, online exam cheating prevention technologies primarily include camera-based facial recognition and behavior monitoring, screen sharing and locking technologies, and AI-based abnormal behavior detection. These technologies monitor examinees' behavioral patterns in an attempt to detect cheating. With the development of artificial intelligence (AI), combining AI with eye tracking technology has provided a new approach to online exam cheating prevention.

[0004] Existing online exam anti-cheating technologies have the following defects and deficiencies: Traditional surveillance systems struggle to effectively detect subtle cheating behaviors, especially those achieved through minute eye movements and gaze shifts. Test-takers can control their heads to remain still and view illegal information solely through eye movements, which poses a challenge to conventional camera surveillance systems.

[0005] Existing technologies lack the ability to deeply analyze examinees' gaze patterns, making it difficult to accurately distinguish between normal reading and browsing behavior and abnormal gaze patterns that involve searching for external answers. This is especially true in multi-functional exam interfaces, where gaze patterns in different areas have different meanings, making it difficult for existing systems to effectively analyze these weights.

[0006] There is a lack of accurate modeling and analysis of the temporal characteristics of gaze behavior. Cheating behaviors often manifest as abnormal gaze migration patterns and periodic gaze behavior, which require advanced time series analysis methods to identify. Existing systems mostly use simple threshold judgments and are unable to capture these complex spatiotemporal patterns, resulting in high false positives or serious missed detections. Summary of the Invention

[0007] The embodiments of the present invention provide an online exam anti-cheating method and system based on AI-fused eye tracking, which can solve the problems in the prior art.

[0008] A first aspect of an embodiment of the present invention provides an online exam cheating prevention method based on AI-integrated eye tracking, comprising: The camera of the examinee's terminal device is used to extract the real-time motion trajectory of the examinee's head in three-dimensional space and establish a head motion feature sequence; The screen display content of the examinee's terminal device is divided into multiple functional areas, and a corresponding gaze weight coefficient is set for each functional area; when the actual gaze duration or number of gazes in a functional area exceeds the area switching frequency threshold, the corresponding gaze action is marked as abnormal gaze; Calculating the real-time coordinate position and gaze duration of the examinee's gaze point on the screen through gaze tracking, and combining the head movement feature sequence and the gaze weight coefficient to form gaze behavior feature data; Extract the transfer sequence of the examinee's gaze point between different functional areas, calculate the time interval and spatial distance of the gaze point transfer, and generate a gaze migration feature map; when a periodic jump pattern is detected in the gaze migration feature map, and the time interval of the gaze point transfer shows abnormal periodicity, and the spatial jump speed of the gaze point exceeds the gaze point jump speed threshold, the corresponding period is marked as a suspicious period; The gaze behavior feature data within the suspicious period is extracted, and the spectrum energy distribution features of the gaze point trajectory are extracted using autoregressive spectrum analysis. Wavelet analysis is combined to identify gaze pattern mutation points. When the spectrum energy distribution of the spectrum energy distribution features is abnormal and there are gaze pattern mutation points, accompanied by abnormal gaze, it is determined to be cheating behavior.

[0009] The real-time coordinate position and gaze duration of the examinee's gaze point on the screen are calculated by gaze tracking, and the gaze behavior feature data are formed by combining the head movement feature sequence and the gaze weight coefficient, including: Obtaining the test-taker's left eye sight vector and right eye sight vector through a sight tracking device, and determining the coordinates of the initial sight gaze point according to the intersection of the left eye sight vector and the right eye sight vector with the screen plane; Based on the head motion feature sequence, a head yaw angle rotation matrix, a head pitch angle rotation matrix, and a head roll angle rotation matrix are calculated and multiplied to obtain a head motion compensation matrix, a matrix operation is performed on the initial sight gaze point coordinates and the head motion compensation matrix to obtain compensated gaze point coordinates; a difference operation is performed on the compensated gaze point coordinates at the current moment and the compensated gaze point coordinates at the previous moment to obtain a gaze point movement speed; Continuously collecting a preset number of frames of the compensated gaze point coordinates to calculate spatial discreteness, and when the spatial discreteness is less than a gaze determination threshold, recording the gaze start and end times to generate a gaze duration; The compensated gaze point coordinates, the gaze point moving speed, the gaze duration and the gaze weight coefficient are combined into a gaze behavior feature vector.

[0010] Extract the transfer sequence of the examinee's gaze point between different functional areas, calculate the time interval and spatial distance of the gaze point transfer, and generate the gaze migration feature map including: Establishing a regional index matrix using multiple functional areas on the screen of the examinee's terminal device; Performing density peak clustering processing on the compensated gaze point coordinates within a preset time window to obtain the gaze point coordinates after density peak clustering, performing distance weight matching on the gaze point coordinates after density peak clustering with the region index matrix to obtain the functional region identifier corresponding to each gaze point coordinate; recording the functional region identifiers in chronological order to generate a gaze region transfer sequence; Counting the time points at which two adjacent functional area identifiers are different in the fixation area transfer sequence, marking the corresponding fixation point information as a migration event, and extracting the starting functional area identifier and the target functional area identifier of the migration event; The spatial jump distance is calculated according to the starting gaze point coordinates and the target gaze point coordinates of the migration event, the migration time interval is calculated according to the starting time and the ending time of the migration event, the starting functional area identifier, the target functional area identifier, the spatial jump distance and the migration time interval are combined to form a migration feature vector, the statistical features of the migration feature vector are calculated, and a gaze migration feature map is generated.

[0011] Performing density peak clustering processing on the compensated gaze point coordinates within a preset time window to obtain the gaze point coordinates after density peak clustering, performing distance weight matching on the gaze point coordinates after density peak clustering with the region index matrix, and obtaining the functional region identifier corresponding to each gaze point coordinate includes: Obtaining a compensated gaze point coordinate sequence within a preset time window, calculating the Euclidean distance between all gaze point pairs in the gaze point coordinate sequence, and calculating a local density value of each gaze point based on the Euclidean distance; Calculate the decision value based on the local density value of each fixation point and the minimum Euclidean distance between each fixation point and other fixation points, and select a preset number of fixation points with the largest decision values ​​as density peak points; Establish a gaze point connection graph, establish a connection between each non-density peak point and the gaze point closest to it in space, and calculate the weight value of each connection based on the local density value and the Euclidean distance; based on the connection graph, start from each non-density peak point and move along the connection with the largest weight value until reaching the density peak point, and divide the non-density peak point into the cluster where the corresponding density peak point is located; use the local density value of the gaze point as the weight, calculate the weighted average of all gaze point coordinates in each cluster, and obtain the gaze point coordinates after density peak clustering; Calculate the vertical distance from each density peak point in the gaze point coordinates after the density peak clustering to the boundary of the functional area, divide the vertical distance by the smoothing coefficient and take the negative exponent to obtain the distance weight value of each functional area corresponding to the density peak point, and determine the functional area identification of all gaze points in the cluster where the density peak point is located based on the maximum distance weight value.

[0012] When a periodic jump pattern is detected in the gaze migration feature map, and the time interval of the gaze point transfer presents an abnormal periodicity, and the gaze point spatial jump speed exceeds the gaze point jump speed threshold, the corresponding period is marked as a suspicious period, including: Extracting a jump pattern feature sequence from the gaze migration feature map, calculating the autocorrelation coefficient of the jump pattern feature sequence at different time delays, and marking the current period as a periodic jump period when the maximum autocorrelation coefficient is greater than a jump periodicity determination threshold; For the periodic jump period, extract the time interval sequence of adjacent fixation points and perform fast Fourier transform to obtain spectrum data. Calculate the ratio of the energy value of the main frequency component to the total energy value in the spectrum data to obtain the main frequency energy proportion. Calculate the ratio of the standard deviation of the time interval sequence to the mean to obtain the coefficient of variation. When the main frequency energy proportion is greater than the period intensity threshold and the coefficient of variation is less than the eye movement fluctuation threshold, mark the current period as an abnormal periodic period. For the abnormal period, the ratio of the Euclidean distance between adjacent fixation points and the corresponding time interval is calculated to obtain the fixation point spatial jump speed. When the fixation point spatial jump speed is greater than a preset fixation point jump speed threshold, the current fixation point is marked as a suspicious fixation point. The number and distribution of the suspicious gaze points in the abnormal period are counted, and when the number of the suspicious gaze points is greater than a density threshold and the distribution interval is less than an aggregation threshold, the current abnormal period is determined as a suspicious period.

[0013] Extract the gaze behavior feature data during the suspicious period, use autoregressive spectrum analysis to extract the spectral energy distribution characteristics of the gaze point trajectory, and combine wavelet analysis to identify gaze pattern mutation points. When the spectral energy distribution characteristics of the spectral energy distribution characteristics are abnormal and there are gaze pattern mutation points, accompanied by abnormal gaze, it is determined to be cheating behavior including: Extracting the spatial coordinate sequence of the gaze point from the gaze behavior feature data within the suspicious period; Performing autoregressive spectrum analysis on the spatial coordinate sequence to obtain power spectrum density data, dividing the power spectrum density data into multiple frequency bands according to frequency, and calculating the ratio of the power spectrum density value of each frequency band to the total power spectrum density value to obtain spectrum energy distribution characteristics; performing a continuous wavelet transform on the spatial coordinate sequence to obtain multi-scale wavelet coefficients, performing empirical mode decomposition on the multi-scale wavelet coefficients to obtain a plurality of intrinsic mode functions, constructing an analytical signal for each intrinsic mode function, calculating an instantaneous amplitude and an instantaneous phase based on the analytical signal, and marking a gaze pattern mutation point based on a rate of change of the instantaneous amplitude and the instantaneous phase at adjacent moments; When the component features of the spectral energy distribution features deviate from the feature subspace of the normal gaze pattern, and there is a mutation point of the gaze pattern, and the current gaze action is marked as abnormal gaze, the behavior in the current suspicious period is determined to be cheating behavior.

[0014] A second aspect of an embodiment of the present invention provides an online exam anti-cheating system based on AI-integrated eye tracking, comprising: The first unit is used to extract the real-time motion trajectory of the examinee's head in three-dimensional space through the camera of the examinee's terminal device to establish a head motion feature sequence; The second unit is used to divide the screen display content of the examinee's terminal device into multiple functional areas and set a corresponding gaze weight coefficient for each functional area; when the actual gaze duration or number of gazes in a functional area exceeds the area switching frequency threshold, the corresponding gaze action is marked as abnormal; The third unit is used to calculate the real-time coordinate position and gaze duration of the examinee's gaze point on the screen through gaze tracking, and combine the head movement feature sequence and the gaze weight coefficient to form gaze behavior feature data; The fourth unit is used to extract the transfer sequence of the examinee's gaze point between different functional areas, calculate the time interval and spatial distance of the gaze point transfer, and generate a gaze migration feature map. When a periodic jump pattern is detected in the gaze migration feature map, and the time interval of the gaze point transfer shows abnormal periodicity, and the spatial jump speed of the gaze point exceeds the gaze point jump speed threshold, the corresponding time period is marked as a suspicious period. The fifth unit is used to extract the gaze behavior feature data during the suspicious period, use autoregressive spectrum analysis to extract the spectral energy distribution characteristics of the gaze point trajectory, and combine wavelet analysis to identify gaze pattern mutation points. When the spectral energy distribution of the spectral energy distribution characteristics is abnormal and there is a gaze pattern mutation point, accompanied by abnormal gaze, it is determined to be cheating behavior.

[0015] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0016] According to a fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0017] The beneficial effects of this application are as follows: The present invention provides an online exam anti-cheating method based on AI-fused gaze tracking. By extracting the examinee's head movement trajectory and gaze features, combined with functional area weights and gaze migration map analysis, it accurately identifies abnormal gaze behavior and suspicious time periods, effectively solving the problem that traditional anti-cheating technology cannot accurately locate cheating behavior, and improving the accuracy of cheating behavior detection.

[0018] The present invention introduces autoregressive spectrum analysis and wavelet analysis technology to conduct in-depth analysis of the gaze behavior characteristic data during suspicious periods. Through the dual verification of spectral energy distribution characteristics and gaze pattern mutation points, the false positive rate is significantly reduced, the incorrect labeling of normal candidates is avoided, and the fairness and reliability of the anti-cheating system are enhanced.

[0019] By analyzing the sequence of gaze point transfers between different functional areas and combining the time interval and spatial distance calculation of gaze point transfers, the present invention can effectively identify periodic jumping patterns and abnormal gaze behaviors. It has a good ability to identify advanced cheating methods such as transmitting information through line of sight, and significantly improves the integrity and fairness of the online examination environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of an online exam anti-cheating method based on AI-integrated eye tracking according to an embodiment of the present invention; Figure 2 Schematic diagram for comparative analysis of head motion compensation effects; Figure 3 Schematic diagram for comparative analysis of spectrum energy distribution. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0022] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0023] Figure 1 This is a flow chart of an online exam anti-cheating method based on AI-integrated eye tracking according to an embodiment of the present invention. Figure 1 As shown, the method includes: The camera of the examinee's terminal device is used to extract the real-time motion trajectory of the examinee's head in three-dimensional space and establish a head motion feature sequence; The screen display content of the examinee's terminal device is divided into multiple functional areas, and a corresponding gaze weight coefficient is set for each functional area; when the actual gaze duration or number of gazes in a functional area exceeds the area switching frequency threshold, the corresponding gaze action is marked as abnormal gaze; Calculating the real-time coordinate position and gaze duration of the examinee's gaze point on the screen through gaze tracking, and combining the head movement feature sequence and the gaze weight coefficient to form gaze behavior feature data; Extract the transfer sequence of the examinee's gaze point between different functional areas, calculate the time interval and spatial distance of the gaze point transfer, and generate a gaze migration feature map; when a periodic jump pattern is detected in the gaze migration feature map, and the time interval of the gaze point transfer shows abnormal periodicity, and the spatial jump speed of the gaze point exceeds the gaze point jump speed threshold, the corresponding period is marked as a suspicious period; The gaze behavior feature data within the suspicious period is extracted, and the spectrum energy distribution features of the gaze point trajectory are extracted using autoregressive spectrum analysis. Wavelet analysis is combined to identify gaze pattern mutation points. When the spectrum energy distribution of the spectrum energy distribution features is abnormal and there are gaze pattern mutation points, accompanied by abnormal gaze, it is determined to be cheating behavior.

[0024] In an optional embodiment, the gaze behavior feature data is formed by calculating the real-time coordinate position and gaze duration of the examinee's gaze point on the screen through gaze tracking, and combining the head movement feature sequence and the gaze weight coefficient. Obtaining the test-taker's left eye sight vector and right eye sight vector through a sight tracking device, and determining the coordinates of the initial sight gaze point according to the intersection of the left eye sight vector and the right eye sight vector with the screen plane; Based on the head motion feature sequence, a head yaw angle rotation matrix, a head pitch angle rotation matrix, and a head roll angle rotation matrix are calculated and multiplied to obtain a head motion compensation matrix, a matrix operation is performed on the initial sight gaze point coordinates and the head motion compensation matrix to obtain compensated gaze point coordinates; a difference operation is performed on the compensated gaze point coordinates at the current moment and the compensated gaze point coordinates at the previous moment to obtain a gaze point movement speed; Continuously collecting a preset number of frames of the compensated gaze point coordinates to calculate spatial discreteness, and when the spatial discreteness is less than a gaze determination threshold, recording the gaze start and end times to generate a gaze duration; The compensated gaze point coordinates, the gaze point moving speed, the gaze duration and the gaze weight coefficient are combined into a gaze behavior feature vector.

[0025] This method uses a gaze tracking device to obtain the examinee's left and right eye gaze vectors. This device can be a camera-based eye tracker, which works by capturing the relationship between the corneal reflection light point and the pupil center position to calculate the gaze direction. For example, if the examinee's left eye gaze vector is captured as (0.3, 0.2, -1.0) and the right eye gaze vector is captured as (0.25, 0.22, -1.0), the initial gaze point coordinates are determined based on the intersection of these two vectors with the screen plane. Assuming the screen plane equation is z = -50, by solving the intersection of the gaze vector and the plane, the initial gaze point coordinates are (15.0, 10.0, -50.0).

[0026] The head motion feature sequence contains time-series data for the head's yaw, pitch, and roll angles. For example, when a head yaw angle of 5 degrees, a pitch angle of 3 degrees, and a roll angle of 1 degree are detected, the rotation matrices corresponding to these three angles are calculated. The yaw rotation matrix accounts for the effects of left and right head rotation, the pitch rotation matrix accounts for the effects of up and down nodding, and the roll rotation matrix accounts for the effects of left and right head tilt. Multiplying these three rotation matrices together yields the head motion compensation matrix. In practice, the calculated head motion compensation matrix is ​​assumed to be a 3×3 matrix with values ​​of [[0.996, 0.087, 0.005], [-0.087, 0.995, 0.052], [-0.001, -0.053, 0.998]].

[0027] The initial gaze point coordinates are represented as a column vector and multiplied by the head motion compensation matrix to obtain the compensated coordinates. For example, multiplying the initial coordinates (15.0, 10.0, -50.0) by the compensation matrix yields the compensated gaze point coordinates (16.4, 9.5, -50.2). This compensation mechanism effectively eliminates the effects of slight head movements on gaze tracking accuracy.

[0028] To calculate gaze velocity, the current compensated gaze coordinates are subtracted from the previous compensated gaze coordinates. Assuming the previous compensated gaze coordinates were (16.2, 9.6, -50.2) and the current coordinates are (16.4, 9.5, -50.2), the gaze velocity is (0.2, -0.1, 0.0), indicating a movement of 0.2 units in the x-direction, -0.1 units in the y-direction, and no movement in the z-direction.

[0029] Typically, the preset frame count is set to 10-30 frames, and the sampling frequency is set to 60Hz or 90Hz. The compensated gaze point coordinates are continuously collected to calculate spatial dispersion. Spatial dispersion is calculated by taking the standard deviation of the gaze point coordinates across these frames. For example, if 20 frames of data are continuously collected and the calculated standard deviation of the x-coordinate is 0.8 and the y-coordinate standard deviation is 0.6, then the spatial dispersion can be expressed as the square root of the sum of the squares of 0.8 and 0.6, or 1.0. When the spatial dispersion is less than the gaze determination threshold, the examinee is considered to be gazing at a specific point. The gaze determination threshold is typically set to 1.5-2.0 degrees of visual angle, approximately equivalent to a range of 15-20 pixels on the screen. When the spatial dispersion is detected to be below the threshold of 1.5 for 20 consecutive frames, the start of the gaze is recorded, for example, at frame 100. When the spatial dispersion exceeds the threshold again, the end of the gaze is recorded, for example, at frame 150. Calculated at a 60Hz sampling rate, the duration of the gaze is (150-100) / 60 = 0.833 seconds.

[0030] The compensated fixation point coordinates, fixation point movement speed, fixation duration, and fixation weight coefficient are combined to form a fixation behavior feature vector. The fixation weight coefficient reflects the importance of fixations in different areas and is usually preset based on the area division of the exam content. For example, the question stem area has a weight of 0.4, the options area has a weight of 0.3, the chart area has a weight of 0.2, and other areas have a weight of 0.1. The final fixation behavior feature vector can be expressed as: [(16.4, 9.5, -50.2), (0.2, -0.1, 0.0), 0.833, 0.3], corresponding to the compensated fixation point coordinates, fixation point movement speed, fixation duration, and fixation weight coefficient, respectively.

[0031] This gaze behavior feature vector can be used to subsequently analyze a student's reading patterns, cognitive load, and problem-solving strategies, providing data support for exam anti-cheating systems and learning behavior analysis. For example, by analyzing the distribution and duration of gaze points in different areas, it can be determined whether a student exhibits suspicious behaviors such as frequent looking around or communicating. By comparing the degree of alignment between a student's gaze sequence and a standard problem-solving path, it can be used to assess the student's problem-solving efficiency and thinking style.

[0032] Figure 2 The figure shows a comparative analysis of the effects of head motion compensation. The figure demonstrates the significant advantages of the present invention over traditional methods, verified by 1000 frames of continuous test data. As can be seen from the figure, the Pupil Labs method exhibits significant fluctuations in gaze point coordinate deviation, ranging from 2.1 to 22.1 pixels, with an average deviation of approximately 13.8 pixels, demonstrating significant instability. In particular, significant deviations exceeding 20 pixels are observed around frames 350, 500, and 900. While the Tobii Pro method shows some improvement over the Pupil Labs method, the deviation still fluctuates between 1.8 and 16.8 pixels, with an average deviation of approximately 10.2 pixels, and significant tracking errors persist under large head movements. The present invention, through its innovative three-axis rotation matrix head motion compensation algorithm, controls gaze point coordinate deviation within a very small range of 0.8 to 3.6 pixels, with an average deviation of only 2.3 pixels. This represents an 83.3% reduction compared to traditional methods and a 77.5% reduction compared to the Tobii Pro method. The compensation effect curve of the present invention shows excellent stability and consistency, and can maintain high-precision tracking even when the head moves with a large amplitude, fully demonstrating the technical superiority of the three-dimensional rotation compensation mechanism based on yaw angle, pitch angle and roll angle, and providing more reliable basic data for eye tracking for online exam anti-cheating systems.

[0033] In an optional embodiment, extracting the transfer sequence of the examinee's gaze point between different functional areas, calculating the time interval and spatial distance of the gaze point transfer, and generating the gaze migration feature map includes: Establishing a regional index matrix using multiple functional areas on the screen of the examinee's terminal device; Performing density peak clustering processing on the compensated gaze point coordinates within a preset time window to obtain the gaze point coordinates after density peak clustering, performing distance weight matching on the gaze point coordinates after density peak clustering with the region index matrix to obtain the functional region identifier corresponding to each gaze point coordinate; recording the functional region identifiers in chronological order to generate a gaze region transfer sequence; Counting the time points at which two adjacent functional area identifiers are different in the fixation area transfer sequence, marking the corresponding fixation point information as a migration event, and extracting the starting functional area identifier and the target functional area identifier of the migration event; The spatial jump distance is calculated according to the starting gaze point coordinates and the target gaze point coordinates of the migration event, the migration time interval is calculated according to the starting time and the ending time of the migration event, the starting functional area identifier, the target functional area identifier, the spatial jump distance and the migration time interval are combined to form a migration feature vector, the statistical features of the migration feature vector are calculated, and a gaze migration feature map is generated.

[0034] The screen can be divided into functional areas, including the question area, answer area, toolbar area, timer area, and submit button area. Each functional area corresponds to a unique area identifier. For example, the question area is identified as "A," the answer area is identified as "B," the toolbar area is identified as "C," the timer area is identified as "D," and the submit button area is identified as "E." The area index matrix contains the coordinate range of each functional area and the corresponding area identifier. For example, the coordinate range of the question area is (100, 100) for the upper left corner and (500, 400) for the lower right corner, and the area identifier is "A."

[0035] The distance between each fixation point and other fixations is calculated, and the local density of each fixation point and its minimum distance to points with higher density are determined. These two metrics are used to identify density peaks. For example, if there are 10 fixations within a time window, density peak clustering will result in three cluster centers with coordinates (315, 235), (420, 310), and (250, 180).

[0036] Based on distance-weighted matching, the functional region identifier corresponding to each fixation point is determined. Distance-weighted matching considers the distance between the fixation point and the center point of each functional region. The closer the distance, the greater the weight. For example, fixation point (315, 235) is closest to the center of the question area and is therefore matched to region identifier "A"; fixation point (420, 310) is matched to region identifier "B"; and fixation point (250, 180) is matched to region identifier "C".

[0037] The sequence of gaze area transfer within a period of time is "AABBBCAAD", which means that the examinee's gaze point starts from the question area, moves to the answer area, then to the toolbar area, back to the question area, and finally to the timing area.

[0038] In the above sequence, the transition points from "A" to "B", from "B" to "C", from "C" to "A", and from "A" to "D" are marked as migration events. The starting functional region identifier and the target functional region identifier of each migration event are extracted. For example, the starting functional region of the first migration event is "A" and the target functional region is "B".

[0039] The starting fixation point coordinates are (315, 235), the target fixation point coordinates are (420, 310), and the spatial jump distance is the Euclidean distance between the two points, which is approximately 137.7 pixels. The start time is 10.5 seconds, the end time is 10.8 seconds, and the time interval is 0.3 seconds.

[0040] The migration feature vector is composed of the starting functional area identifier, target functional area identifier, spatial jump distance, and migration interval. For example, the feature vector for the first migration event is {"A", "B", 137.7, 0.3}. Statistical features are calculated for all migration feature vectors, including the frequency of migrations between different functional areas, the average spatial jump distance, and the average migration interval. For example, there were five migrations from area "A" to area "B", with an average spatial jump distance of 130.5 pixels and an average migration interval of 0.28 seconds.

[0041] A gaze migration feature map can be generated in matrix form, with rows and columns representing the starting and target functional areas, respectively. Matrix elements contain the frequency of migrations between corresponding areas, the average spatial jump distance, and the average migration interval. For example, the migration feature from area "A" to area "B" is {5 times, 130.5 pixels, 0.28 seconds}. This gaze migration feature map visualizes the examinee's gaze patterns, reflecting their problem-solving strategies and cognitive processes, and provides a basis for performance assessment.

[0042] Through the above method, we can comprehensively analyze the examinees' gaze migration behavior during the exam, identify the gaze switching patterns between different functional areas, and provide objective data support for understanding the examinees' problem-solving ideas and cognitive processes.

[0043] In an optional embodiment, density peak clustering is performed on the compensated gaze point coordinates within a preset time window to obtain the gaze point coordinates after density peak clustering, and distance weight matching is performed on the gaze point coordinates after density peak clustering with the region index matrix to obtain the functional region identifier corresponding to each gaze point coordinate, including: Obtaining a compensated gaze point coordinate sequence within a preset time window, calculating the Euclidean distance between all gaze point pairs in the gaze point coordinate sequence, and calculating a local density value of each gaze point based on the Euclidean distance; Calculate the decision value based on the local density value of each fixation point and the minimum Euclidean distance between each fixation point and other fixation points, and select a preset number of fixation points with the largest decision values ​​as density peak points; Establish a gaze point connection graph, establish a connection between each non-density peak point and the gaze point closest to it in space, and calculate the weight value of each connection based on the local density value and the Euclidean distance; based on the connection graph, start from each non-density peak point and move along the connection with the largest weight value until reaching the density peak point, and divide the non-density peak point into the cluster where the corresponding density peak point is located; use the local density value of the gaze point as the weight, calculate the weighted average of all gaze point coordinates in each cluster, and obtain the gaze point coordinates after density peak clustering; Calculate the vertical distance from each density peak point in the gaze point coordinates after the density peak clustering to the boundary of the functional area, divide the vertical distance by the smoothing coefficient and take the negative exponent to obtain the distance weight value of each functional area corresponding to the density peak point, and determine the functional area identification of all gaze points in the cluster where the density peak point is located based on the maximum distance weight value.

[0044] Obtain the compensated gaze point coordinate sequence within a preset time window. This time window can be set to 1000 milliseconds, and this time window contains multiple gaze point coordinates, such as coordinate points P1 (300, 250), P2 (320, 260), P3 (310, 255), P4 (500, 450), P5 (515, 445), etc.

[0045] For the obtained gaze point coordinate sequence, calculate the Euclidean distance between all gaze point pairs in the sequence. For example, the Euclidean distance between points P1 and P2 is 22.36, the Euclidean distance between points P1 and P3 is 11.18, and so on.

[0046] After calculating all Euclidean distances, the local density value of each fixation point is calculated based on these distances. The local density value reflects the degree of aggregation of fixations around a certain fixation point. Set a cutoff distance dc, for example, dc = 25 pixels. For any two points Pi and Pj, if the Euclidean distance between them is less than dc, they are considered to be neighbors. The local density value ρi of point Pi is defined as the number of points whose distance to Pi is less than dc. For example, if there are 3 points (including P2 and P3) around P1 whose distance is less than dc, then the local density value ρ1 of P1 is 3.

[0047] For each point Pi, find the point Pj with the smallest distance from Pi among the points with higher local density than Pi. Record this smallest distance as δi. If Pi has the highest local density, δi is set to the maximum distance in the sequence. The decision value γi = ρi × δi represents the probability that the point will be the density peak. For example, if P1 has a local density of ρ1 = 3, and among the points with higher density than P1, the smallest distance between P3 and P1 is 11.18, then the decision value for P1 is γ1 = 3 × 11.18 = 33.54.

[0048] When selecting a preset number of fixations with the largest decision values ​​as density peak points, the preset number can be set according to the application scenario. For example, the three points with the largest decision values ​​are selected as density peak points. Assuming that P1, P4, and P5 have the largest decision values ​​after calculation, they are selected as density peak points.

[0049] Each non-density peak point is connected to its closest spatial distance to the fixation point to generate a fixation point connection map. For example, if non-density peak point P2 has a distance of 22.36 from P1 and 14.14 from P3, and its distances to P4 and P5 are both greater than 30, then P2 is connected to P3. Similarly, P3 is connected to P1 because P1 is the closest point to P3.

[0050] The connection weight can be defined as the product of the local density values ​​of the two endpoints divided by the square of the Euclidean distance between them. For example, the weight of the connection between P2 and P3 is (ρ2×ρ3) / (d 2 P2P3), assuming ρ2=2, ρ3=3, dP2P3=14.14, then the weight value is (2×3) / (14.14 2 )=0.03.

[0051] Starting from P2, if the connection weight between P2 and P3 is the largest, move to P3; if the connection weight between P3 and P1 is the largest and P1 is the density peak point, then P2 and P3 are both divided into the cluster where P1 is located.

[0052] In clustering, the local density of the fixation point is used as the weight to calculate the weighted average of all fixation point coordinates, resulting in the density peak clustered fixation point coordinates. For example, assuming cluster P1 contains P1, P2, and P3, whose local density values ​​are 3, 2, and 3, respectively, the weighted average coordinates of this cluster are [(300×3+320×2+310×3) / (3+2+3), (250×3+260×2+255×3) / (3+2+3)] = [309.38, 254.38].

[0053] Assume that there are three functional areas on the interface: Area A (rectangle, upper left corner coordinates (250, 200), lower right corner (350, 300)), Area B (rectangle, upper left corner coordinates (450, 400), lower right corner (550, 500)), and Area C (rectangle, upper left corner coordinates (600, 300), lower right corner (700, 400)). For the weighted average coordinate of cluster P1 (309.38, 254.38), its distance to the boundary of Area A is 0 (because the point is within Area A), the distance to the boundary of Area B is approximately 149.65, and the distance to the boundary of Area C is approximately 290.62.

[0054] Set the smoothing coefficient to 50. For cluster P1, the distance weight value of the corresponding region A is e -(0 / 50) =1, the distance weight value corresponding to area B is e -(149.65 / 50) =0.05, the distance weight value corresponding to area C is e -(290.62 / 50) =0.003. Based on the maximum distance weight value, the functional area to which all fixations in cluster P1 belong is identified as area A.

[0055] Similarly, performing the same calculation on the clusters formed by P4 and P5 will result in the result that they belong to area B. In this way, the gaze points can be effectively mapped to the corresponding functional areas, improving the accuracy of eye tracking in human-computer interaction.

[0056] In an optional embodiment, when a periodic jump pattern is detected in the gaze migration feature map, and the time interval of the gaze point transfer presents an abnormal periodicity, and the gaze point spatial jump speed exceeds the gaze point jump speed threshold, marking the corresponding period as a suspicious period includes: Extracting a jump pattern feature sequence from the gaze migration feature map, calculating the autocorrelation coefficient of the jump pattern feature sequence at different time delays, and marking the current period as a periodic jump period when the maximum autocorrelation coefficient is greater than a jump periodicity determination threshold; For the periodic jump period, extract the time interval sequence of adjacent fixation points and perform fast Fourier transform to obtain spectrum data. Calculate the ratio of the energy value of the main frequency component to the total energy value in the spectrum data to obtain the main frequency energy proportion. Calculate the ratio of the standard deviation of the time interval sequence to the mean to obtain the coefficient of variation. When the main frequency energy proportion is greater than the period intensity threshold and the coefficient of variation is less than the eye movement fluctuation threshold, mark the current period as an abnormal periodic period. For the abnormal period, the ratio of the Euclidean distance between adjacent fixation points and the corresponding time interval is calculated to obtain the fixation point spatial jump speed. When the fixation point spatial jump speed is greater than a preset fixation point jump speed threshold, the current fixation point is marked as a suspicious fixation point. The number and distribution of the suspicious gaze points in the abnormal period are counted, and when the number of the suspicious gaze points is greater than a density threshold and the distribution interval is less than an aggregation threshold, the current abnormal period is determined as a suspicious period.

[0057] To detect periodic jump patterns in the gaze migration feature map, a jump pattern feature sequence is extracted, consisting of the Euclidean distance values ​​between adjacent fixations. Taking an eye tracking device with a 60Hz sampling rate as an example, the distance between each pair of adjacent fixations in 30 seconds of eye movement data is calculated, forming a jump pattern feature sequence containing approximately 120-180 elements. To detect periodicity in the sequence, the autocorrelation coefficient of the sequence at different time delays is calculated. Specifically, the maximum delay is set to half the sequence length, and the autocorrelation coefficient is calculated for each delay value between 0 and the maximum delay value. When the maximum autocorrelation coefficient exceeds the preset jump periodicity threshold of 0.65, the current period is marked as a periodic jump period. For example, in one test, the maximum autocorrelation coefficient of a fixation sequence was detected to be 0.72, occurring at a delay value of 12, indicating that a similar jump pattern occurs approximately every 12 sampling points in the sequence, thus being marked as a periodic jump period.

[0058] Extract the time interval sequence between adjacent gaze points, and perform fast Fourier transform on the sequence to obtain spectrum data. In practical applications, for a time interval sequence of length 128, an FFT algorithm based on a power of 2 is used for transformation to obtain spectrum data consisting of 128 complex values. Calculate the ratio of the energy value of the main frequency component to the total energy value in the spectrum data to obtain the main frequency energy ratio. At the same time, calculate the ratio of the standard deviation of the time interval sequence to the mean to obtain the coefficient of variation. When the main frequency energy ratio is greater than the preset period intensity threshold of 0.35, and the coefficient of variation is less than the preset eye movement fluctuation threshold of 0.25, the current period is marked as an abnormal period period. In one instance, it was detected that the main frequency energy ratio of a certain segment of eye movement data was 0.42, and the coefficient of variation was 0.18, which met the characteristics of abnormal periodicity and was therefore marked as an abnormal period period.

[0059] For data marked as abnormal periodic periods, the spatial jump speed of the gaze point is further calculated, that is, the ratio of the Euclidean distance between adjacent gaze points to the corresponding time interval. In normal human eye movement, the spatial jump speed of the gaze point usually does not exceed a certain range. The gaze point jump speed threshold is set to 500 pixels / second. When the calculated spatial jump speed of the gaze point exceeds this threshold, the current gaze point is marked as a suspicious gaze point. For example, during a certain abnormal periodic period, the Euclidean distance between two adjacent gaze points is detected to be 300 pixels, the corresponding time interval is 0.5 seconds, and the calculated spatial jump speed is 600 pixels / second, which exceeds the preset threshold, so the gaze point is marked as a suspicious gaze point.

[0060] To identify suspicious periods, the number and distribution of suspicious gaze points within the abnormal period are counted. A density threshold of 10 / 30 seconds and an aggregation threshold of 3 seconds are set. When the number of suspicious gaze points within an abnormal period exceeds the density threshold, and the time interval between adjacent suspicious gaze points is less than the aggregation threshold, the current abnormal period is identified as suspicious. In practice, 15 suspicious gaze points were detected within a 30-second abnormal period, with an average time interval of 2.1 seconds. This meets both the density and aggregation criteria, and the period is therefore ultimately identified as suspicious.

[0061] The above method can effectively identify abnormal patterns in eye movement data, particularly suspicious periods that exhibit unnatural periodicity and unusual spatial gaze jump speeds. These suspicious periods are often associated with abnormal human eye movement behavior, such as deceptive behavior using eye movement simulation programs or robots. In practical applications, marking these suspicious periods provides a basis for subsequent security verification and behavioral analysis.

[0062] In an optional embodiment, the gaze behavior feature data within the suspicious period is extracted, the spectral energy distribution features of the gaze point trajectory are extracted using autoregressive spectrum analysis, and wavelet analysis is combined to identify gaze pattern mutation points. When the spectral energy distribution of the spectral energy distribution features is abnormal and there is a gaze pattern mutation point, accompanied by abnormal gaze, the cheating behavior is determined to be: Extracting the spatial coordinate sequence of the gaze point from the gaze behavior feature data within the suspicious period; Performing autoregressive spectrum analysis on the spatial coordinate sequence to obtain power spectrum density data, dividing the power spectrum density data into multiple frequency bands according to frequency, and calculating the ratio of the power spectrum density value of each frequency band to the total power spectrum density value to obtain spectrum energy distribution characteristics; performing a continuous wavelet transform on the spatial coordinate sequence to obtain multi-scale wavelet coefficients, performing empirical mode decomposition on the multi-scale wavelet coefficients to obtain a plurality of intrinsic mode functions, constructing an analytical signal for each intrinsic mode function, calculating an instantaneous amplitude and an instantaneous phase based on the analytical signal, and marking a gaze pattern mutation point based on a rate of change of the instantaneous amplitude and the instantaneous phase at adjacent moments; When the component features of the spectral energy distribution features deviate from the feature subspace of the normal gaze pattern, and there is a mutation point of the gaze pattern, and the current gaze action is marked as abnormal gaze, the behavior in the current suspicious period is determined to be cheating behavior.

[0063] In the gaze behavior feature data extracted during the suspicious period, the spatial coordinate sequence of the examinee's gaze point is obtained from the eye tracking device. This sequence contains the examinee's eye movement trajectory during the test. The spatial coordinate sequence is usually in the form of a time series {(x1,y1), (x2,y2), ..., (x n ,y n )}, where (x t ,y t ) represents the two-dimensional coordinates of the gaze point on the screen at time point t.

[0064] Autoregressive spectral analysis estimates the power spectral density of a sequence by building an autoregressive model. To implement this, select an appropriate autoregressive model order p (e.g., p=10) and estimate the model parameters using the Yule-Walker equation. Substitute the estimated parameters into the autoregressive spectral density function to calculate the power spectral density data. For example, analyzing the gaze trajectory of a test-taker over a one-minute period reveals the distribution of power spectral density values ​​within the 0-50 Hz frequency range.

[0065] The 0-50Hz frequency range can be divided into five frequency bands: 0-10Hz, 10-20Hz, 20-30Hz, 30-40Hz, and 40-50Hz. For each frequency band, the ratio of the power spectral density value within that band to the total power spectral density value is calculated to obtain the spectral energy distribution characteristics. For example, during normal reading, the energy proportion of the 0-10Hz band is approximately 65%, the 10-20Hz band is approximately 20%, the 20-30Hz band is approximately 10%, the 30-40Hz band is approximately 3%, and the 40-50Hz band is approximately 2%. Cheating behavior, on the other hand, is manifested by an abnormal increase in high-frequency components, such as an energy proportion exceeding 10% in the 30-40Hz band.

[0066] Perform a continuous wavelet transform on the same spatial coordinate sequence, using the Morlet wavelet as the mother wavelet function, and calculate the wavelet coefficients at multiple scales. For example, select 8 scales for analysis and obtain a wavelet coefficient matrix W(a,t) on the time-scale plane, where a represents scale and t represents time.

[0067] The signal is decomposed into multiple intrinsic mode functions (IMFs) and a residual term through iterative filtering. For a gaze trajectory, 4-6 IMFs are typically obtained. Each IMF represents the oscillation mode of the original signal at a different characteristic time scale.

[0068] The analytical signal is converted to complex form by performing a Hilbert transform on the IMF. The analytical signal consists of a real part (the original IMF) and an imaginary part (the Hilbert transform of the original IMF). The instantaneous amplitude and phase can be calculated from the analytical signal. The instantaneous amplitude is the square root of the modulus of the analytical signal and represents the energy of the signal at each moment. The instantaneous phase is the inverse tangent of the real and imaginary parts of the analytical signal and represents the phase angle of the signal at each moment.

[0069] When the relative rate of change of the instantaneous amplitude exceeds a preset threshold (e.g., 50%) or the rate of change of the instantaneous phase exceeds a preset threshold (e.g., π / 4 per second), that moment is marked as a gaze pattern mutation point. In practical applications, a threshold for the number of mutation points within a time window (e.g., 0.5 seconds) can be set. A valid mutation point is identified when the number of mutation points exceeds the threshold. For example, in a case of cheating by a student, three valid mutation points were detected within one second when the student switched from normal reading to viewing a hidden cheating device.

[0070] Principal component analysis is used to establish a characteristic subspace for normal gaze patterns. The first k principal components (e.g., k = 3) are used to represent the main variations in normal gaze patterns. The distance from the current spectral energy distribution feature to this characteristic subspace is calculated. If the distance exceeds a threshold (e.g., three standard deviations of the mean distance), the spectral energy distribution is considered abnormal.

[0071] Based on the above judgment results, if three conditions are simultaneously met: the spectral energy distribution characteristics deviate from the characteristic subspace of the normal gaze pattern (for example, the reconstruction error exceeds 0.25), there are gaze pattern mutation points (for example, the number of mutation points exceeds 5 within 10 seconds), and the current gaze action is marked as abnormal (for example, the gaze duration exceeds the 95th percentile of the normal distribution or the gaze point falls within the prohibited area), the behavior during the current suspicious period will be judged as cheating. The judgment result, including the timestamp, abnormal feature value, and judgment confidence level, will be recorded for further review by the invigilator.

[0072] Figure 3This is a schematic diagram of spectrum energy distribution comparison analysis, which shows the comparative effects of spectrum energy distribution of different spectrum analysis methods when identifying cheating behavior, clearly demonstrating the significant technical advantages of the present invention over traditional methods. As can be seen from the figure, in the normal gaze mode, the spectral energy is mainly concentrated in the low-frequency band, with the 0-10Hz band accounting for 65% of the energy, which is consistent with the physiological characteristics of natural human gaze. The traditional FFT method shows that the energy in the 0-10Hz band is 58%, and the 10-20Hz band is 23%. Although it can detect certain spectral changes, its sensitivity is limited. The detection results of the Welch power spectrum method are slightly improved, with the energy in the 0-10Hz band being 61%, but its ability to identify high-frequency anomalies is still insufficient. The present invention successfully identifies the spectral characteristic anomalies of cheating behavior through the autoregressive spectrum analysis algorithm. The detection results show that the energy in the 0-10Hz band is significantly reduced to 42%, while the energy in the high-frequency bands of 20-30Hz, 30-40Hz, and 40-50Hz reaches 18%, 8%, and 4%, respectively, which are increases of 80%, 167%, and 100% compared to the normal gaze mode. This abnormal increase in high-frequency energy precisely reflects the abnormal behavior patterns of cheating, such as rapid eye movements and frequent switching of gaze targets. The present invention can accurately capture these subtle spectral changes and, by quantitatively analyzing the redistribution of spectral energy in each frequency band, provides a reliable technical basis for accurately determining cheating behavior, fully verifying the innovation and practicality of autoregressive spectral analysis in the field of eye movement anomaly detection.

[0073] An embodiment of the present invention provides an online exam anti-cheating system based on AI-integrated eye tracking, the system comprising: The first unit is used to extract the real-time motion trajectory of the examinee's head in three-dimensional space through the camera of the examinee's terminal device to establish a head motion feature sequence; The second unit is used to divide the screen display content of the examinee's terminal device into multiple functional areas and set a corresponding gaze weight coefficient for each functional area; when the actual gaze duration or number of gazes in a functional area exceeds the area switching frequency threshold, the corresponding gaze action is marked as abnormal; The third unit is used to calculate the real-time coordinate position and gaze duration of the examinee's gaze point on the screen through gaze tracking, and combine the head movement feature sequence and the gaze weight coefficient to form gaze behavior feature data; The fourth unit is used to extract the transfer sequence of the examinee's gaze point between different functional areas, calculate the time interval and spatial distance of the gaze point transfer, and generate a gaze migration feature map. When a periodic jump pattern is detected in the gaze migration feature map, and the time interval of the gaze point transfer shows abnormal periodicity, and the spatial jump speed of the gaze point exceeds the gaze point jump speed threshold, the corresponding time period is marked as a suspicious period. The fifth unit is used to extract the gaze behavior feature data during the suspicious period, use autoregressive spectrum analysis to extract the spectral energy distribution characteristics of the gaze point trajectory, and combine wavelet analysis to identify gaze pattern mutation points. When the spectral energy distribution of the spectral energy distribution characteristics is abnormal and there is a gaze pattern mutation point, accompanied by abnormal gaze, it is determined to be cheating behavior.

[0074] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0075] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0076] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An online exam anti-cheating method based on AI-integrated eye tracking, characterized by: include: The camera of the examinee's terminal device is used to extract the real-time motion trajectory of the examinee's head in three-dimensional space and establish a head motion feature sequence; The screen display content of the examinee's terminal device is divided into multiple functional areas, and a corresponding gaze weight coefficient is set for each functional area; when the actual gaze duration or number of gazes in a functional area exceeds the area switching frequency threshold, the corresponding gaze action is marked as abnormal gaze; Calculating the real-time coordinate position and gaze duration of the examinee's gaze point on the screen through gaze tracking, and combining the head movement feature sequence and the gaze weight coefficient to form gaze behavior feature data; Extract the transfer sequence of the examinee's gaze point between different functional areas, calculate the time interval and spatial distance of the gaze point transfer, and generate the gaze migration feature map; When a periodic jump pattern is detected in the gaze migration feature map, and the time interval of the gaze point transfer presents an abnormal periodicity, and the gaze point spatial jump speed exceeds the gaze point jump speed threshold, the corresponding period is marked as a suspicious period; The gaze behavior feature data within the suspicious period is extracted, and the spectrum energy distribution features of the gaze point trajectory are extracted using autoregressive spectrum analysis. Wavelet analysis is combined to identify gaze pattern mutation points. When the spectrum energy distribution of the spectrum energy distribution features is abnormal and there are gaze pattern mutation points, accompanied by abnormal gaze, it is determined to be cheating behavior.

2. The method according to claim 1, characterized in that The real-time coordinate position and gaze duration of the examinee's gaze point on the screen are calculated by gaze tracking, and the gaze behavior feature data are formed by combining the head movement feature sequence and the gaze weight coefficient, including: Obtaining the test-taker's left eye sight vector and right eye sight vector through a sight tracking device, and determining the coordinates of the initial sight gaze point according to the intersection of the left eye sight vector and the right eye sight vector with the screen plane; Based on the head motion feature sequence, a head yaw angle rotation matrix, a head pitch angle rotation matrix, and a head roll angle rotation matrix are calculated and multiplied to obtain a head motion compensation matrix, a matrix operation is performed on the initial sight gaze point coordinates and the head motion compensation matrix to obtain compensated gaze point coordinates; a difference operation is performed on the compensated gaze point coordinates at the current moment and the compensated gaze point coordinates at the previous moment to obtain a gaze point movement speed; Continuously collecting a preset number of frames of the compensated gaze point coordinates to calculate spatial discreteness, and when the spatial discreteness is less than a gaze determination threshold, recording the gaze start and end times to generate a gaze duration; The compensated gaze point coordinates, the gaze point moving speed, the gaze duration and the gaze weight coefficient are combined into a gaze behavior feature vector.

3. The method according to claim 1, characterized in that Extract the transfer sequence of the examinee's gaze point between different functional areas, calculate the time interval and spatial distance of the gaze point transfer, and generate the gaze migration feature map including: Establishing a regional index matrix using multiple functional areas on the screen of the examinee's terminal device; Performing density peak clustering processing on the compensated gaze point coordinates within a preset time window to obtain the gaze point coordinates after density peak clustering, performing distance weight matching on the gaze point coordinates after density peak clustering with the region index matrix to obtain the functional region identifier corresponding to each gaze point coordinate; recording the functional region identifiers in chronological order to generate a gaze region transfer sequence; Counting the time points at which two adjacent functional area identifiers are different in the fixation area transfer sequence, marking the corresponding fixation point information as a migration event, and extracting the starting functional area identifier and the target functional area identifier of the migration event; The spatial jump distance is calculated according to the starting gaze point coordinates and the target gaze point coordinates of the migration event, the migration time interval is calculated according to the starting time and the ending time of the migration event, the starting functional area identifier, the target functional area identifier, the spatial jump distance and the migration time interval are combined to form a migration feature vector, the statistical features of the migration feature vector are calculated, and a gaze migration feature map is generated.

4. The method according to claim 3, characterized in that Performing density peak clustering processing on the compensated gaze point coordinates within a preset time window to obtain the gaze point coordinates after density peak clustering, performing distance weight matching on the gaze point coordinates after density peak clustering with the region index matrix, and obtaining the functional region identifier corresponding to each gaze point coordinate includes: Obtaining a compensated gaze point coordinate sequence within a preset time window, calculating the Euclidean distance between all gaze point pairs in the gaze point coordinate sequence, and calculating a local density value of each gaze point based on the Euclidean distance; Calculate the decision value based on the local density value of each fixation point and the minimum Euclidean distance between each fixation point and other fixation points, and select a preset number of fixation points with the largest decision values ​​as density peak points; Establish a gaze point connection graph, establish a connection between each non-density peak point and the gaze point closest to it in space, and calculate the weight value of each connection based on the local density value and the Euclidean distance; based on the connection graph, start from each non-density peak point and move along the connection with the largest weight value until reaching the density peak point, and divide the non-density peak point into the cluster where the corresponding density peak point is located; use the local density value of the gaze point as the weight, calculate the weighted average of all gaze point coordinates in each cluster, and obtain the gaze point coordinates after density peak clustering; Calculate the vertical distance from each density peak point in the gaze point coordinates after the density peak clustering to the boundary of the functional area, divide the vertical distance by the smoothing coefficient and take the negative exponent to obtain the distance weight value of each functional area corresponding to the density peak point, and determine the functional area identification of all gaze points in the cluster where the density peak point is located based on the maximum distance weight value.

5. The method according to claim 1, wherein When a periodic jump pattern is detected in the gaze migration feature map, and the time interval of the gaze point transfer presents an abnormal periodicity, and the gaze point spatial jump speed exceeds the gaze point jump speed threshold, the corresponding period is marked as a suspicious period, including: Extracting a jump pattern feature sequence from the gaze migration feature map, calculating the autocorrelation coefficient of the jump pattern feature sequence at different time delays, and marking the current period as a periodic jump period when the maximum autocorrelation coefficient is greater than a jump periodicity determination threshold; For the periodic jump period, extract the time interval sequence of adjacent fixation points and perform fast Fourier transform to obtain spectrum data. Calculate the ratio of the energy value of the main frequency component to the total energy value in the spectrum data to obtain the main frequency energy proportion. Calculate the ratio of the standard deviation of the time interval sequence to the mean to obtain the coefficient of variation. When the main frequency energy proportion is greater than the period intensity threshold and the coefficient of variation is less than the eye movement fluctuation threshold, mark the current period as an abnormal periodic period. For the abnormal period, the ratio of the Euclidean distance between adjacent fixation points and the corresponding time interval is calculated to obtain the fixation point spatial jump speed. When the fixation point spatial jump speed is greater than a preset fixation point jump speed threshold, the current fixation point is marked as a suspicious fixation point. The number and distribution of the suspicious gaze points in the abnormal period are counted, and when the number of the suspicious gaze points is greater than a density threshold and the distribution interval is less than an aggregation threshold, the current abnormal period is determined as a suspicious period.

6. The method according to claim 1, characterized in that Extract the gaze behavior feature data during the suspicious period, use autoregressive spectrum analysis to extract the spectral energy distribution characteristics of the gaze point trajectory, and combine wavelet analysis to identify gaze pattern mutation points. When the spectral energy distribution characteristics of the spectral energy distribution characteristics are abnormal and there are gaze pattern mutation points, accompanied by abnormal gaze, it is determined to be cheating behavior including: Extracting the spatial coordinate sequence of the gaze point from the gaze behavior feature data within the suspicious period; Performing autoregressive spectrum analysis on the spatial coordinate sequence to obtain power spectrum density data, dividing the power spectrum density data into multiple frequency bands according to frequency, and calculating the ratio of the power spectrum density value of each frequency band to the total power spectrum density value to obtain spectrum energy distribution characteristics; performing a continuous wavelet transform on the spatial coordinate sequence to obtain multi-scale wavelet coefficients, performing empirical mode decomposition on the multi-scale wavelet coefficients to obtain a plurality of intrinsic mode functions, constructing an analytical signal for each intrinsic mode function, calculating an instantaneous amplitude and an instantaneous phase based on the analytical signal, and marking a gaze pattern mutation point based on a rate of change of the instantaneous amplitude and the instantaneous phase at adjacent moments; When the component features of the spectral energy distribution features deviate from the feature subspace of the normal gaze pattern, and there is a mutation point of the gaze pattern, and the current gaze action is marked as abnormal gaze, the behavior in the current suspicious period is determined to be cheating behavior.

7. An online exam anti-cheating system based on AI-integrated eye tracking, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to extract the real-time motion trajectory of the examinee's head in three-dimensional space through the camera of the examinee's terminal device to establish a head motion feature sequence; The second unit is used to divide the screen display content of the examinee's terminal device into multiple functional areas and set a corresponding gaze weight coefficient for each functional area; when the actual gaze duration or number of gazes in a functional area exceeds the area switching frequency threshold, the corresponding gaze action is marked as abnormal; The third unit is used to calculate the real-time coordinate position and gaze duration of the examinee's gaze point on the screen through gaze tracking, and combine the head movement feature sequence and the gaze weight coefficient to form gaze behavior feature data; The fourth unit is used to extract the transfer sequence of the examinee's gaze point between different functional areas, calculate the time interval and spatial distance of the gaze point transfer, and generate the gaze migration feature map; When a periodic jump pattern is detected in the gaze migration feature map, and the time interval of the gaze point transfer presents an abnormal periodicity, and the gaze point spatial jump speed exceeds the gaze point jump speed threshold, the corresponding period is marked as a suspicious period; The fifth unit is used to extract the gaze behavior feature data during the suspicious period, use autoregressive spectrum analysis to extract the spectral energy distribution characteristics of the gaze point trajectory, and combine wavelet analysis to identify gaze pattern mutation points. When the spectral energy distribution of the spectral energy distribution characteristics is abnormal and there is a gaze pattern mutation point, accompanied by abnormal gaze, it is determined to be cheating behavior.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images