Online exam cheating identification method, device, computer equipment and storage medium

Abstract

The present application relates to a method, device, computer equipment and storage medium for identifying cheating in online exams. The method comprises: obtaining a facial video of a candidate to be identified, sampling the facial video according to a preset sampling period, and obtaining no less than one single-frame image; performing single-frame face detection processing on the single-frame images in a time series, and obtaining the single-frame face coordinates corresponding to each single-frame image; based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, making a cheating determination on the candidate to be identified in each single-frame image, and generating a cheating determination result corresponding to each single-frame image; based on the cheating determination result corresponding to each single-frame image, determining whether the candidate to be identified is cheating. The use of this method can improve the efficiency of identifying cheating of online candidates.

Description

Technical Field

[0001] The present application relates to the field of computer vision technology, and in particular to an online examination cheating identification method, device, computer equipment and storage medium. Background Art

[0002] With the continuous development of educational informatization-related technologies, more and more traditional paper-based exams have been gradually moved online to reduce the cost of organizing exams. At the same time, paperless exams are also in line with national green environmental protection. The exams have moved from offline to online, and the exam papers have moved from paper to computers, mobile phones and other devices, which has brought new challenges to how to identify and prevent cheating in online exams and ensure exam fairness.

[0003] Existing online anti-cheating solutions mostly use video capture to capture pictures, or directly use videos to manually monitor and identify cheaters taking exams on smart devices. They are highly dependent on manual monitoring and judgment. Since the number of cheating videos that can be identified by humans in a unit of time is limited, the existing online exam cheating identification has the technical problem of low identification efficiency. Summary of the invention

[0004] Based on this, it is necessary to provide an online exam cheating identification method, device, computer equipment and storage medium that can improve the efficiency of online exam cheating identification in response to the above technical problems.

[0005] A method for identifying cheating in an online exam, the method comprising:

[0006] Obtain a facial video of the examinee to be identified, and perform sampling processing on the facial video according to a preset sampling period to obtain no less than one single-frame image;

[0007] Performing single-frame face detection processing on the single-frame images in time sequence to obtain single-frame face coordinates corresponding to each single-frame image;

[0008] Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a cheating determination is performed on the candidate to be identified in each single-frame image, and a cheating determination result corresponding to each single-frame image is generated;

[0009] Based on the cheating determination results corresponding to each single-frame image, it is determined whether the candidate to be identified is cheating.

[0010] In one embodiment, the method further includes: based on the single-frame face coordinates corresponding to each single-frame image, performing blink determination on the examinee to be identified, and obtaining a blink determination result;

[0011] Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, single-frame head posture determination is performed on the candidate to be identified to obtain the single-frame head posture determination result;

[0012] Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a single-frame binocular gaze angle determination is performed on the candidate to be identified, and a single-frame binocular gaze angle determination result is obtained;

[0013] Based on the blink determination results, the head posture determination results and the binocular gaze angle determination results of each single-frame image, cheating determination is performed on the candidate to be identified in each single-frame image to generate a cheating determination result corresponding to each single-frame image.

[0014] In one of the embodiments, the method further includes: calculating the eye aspect ratio of the candidate to be identified based on the single-frame face coordinates corresponding to each single-frame image;

[0015] If the eye length-to-width ratio of the candidate to be identified is lower than the preset eye length-to-width ratio threshold, outputting that the candidate to be identified is in a closed-eye state as the blink determination result;

[0016] If the eye length-to-width ratio of the candidate to be identified is greater than or equal to a preset eye length-to-width ratio threshold, the eye-opening state of the candidate to be identified is output as a blink determination result.

[0017] In one of the embodiments, the method further includes: processing the single-frame face coordinates based on a preset universal three-dimensional face model to obtain two-dimensional face coordinates corresponding to the single-frame face coordinates;

[0018] Obtaining the camera parameters of the camera that captures the facial video of the candidate to be identified; the camera parameters include the focal length and optical center of the camera;

[0019] The single-frame face coordinates, the two-dimensional face coordinates corresponding to the single-frame face coordinates, and the camera parameters are input into a preset rotation amount and offset calculation formula to obtain the rotation amount and offset as a single-frame head posture determination result.

[0020] In one embodiment, the method further includes: inputting each single frame image into a preset stacked hourglass network model for processing to obtain a plurality of eye key points;

[0021] Based on the eye key points and the single-frame face coordinates corresponding to each single-frame image, a single-frame binocular gaze angle of the candidate to be identified is determined, and the gaze angle is obtained as a determination result.

[0022] In one embodiment, the method further includes: judging whether the examinee to be identified is in an eye-opening state according to the blink determination result of each single frame image;

[0023] If the candidate to be identified is in a state of closing eyes, it is determined that the candidate to be identified in the single frame image is not looking at the screen;

[0024] If the examinee to be identified is in an open-eye state, the single-frame head posture determination result and the single-frame binocular gaze angle determination result are input into a preset single-frame cheating determination network model, and the cheating determination result corresponding to the single-frame image is output.

[0025] In one embodiment, the method further includes: arranging the cheating determination results of each single frame image in a time series to generate a cheating determination result sequence;

[0026] Obtaining a cheating rate of the cheating determination result sequence within a time window of a preset duration; the cheating rate is obtained by calculating the ratio of the number of times determined as cheating within the time window to the total number of times determined;

[0027] Based on the cheating rate in each time window and a preset cheating rate threshold, it is determined whether the candidate to be identified is cheating.

[0028] An online examination cheating identification device, the device comprising:

[0029] An image acquisition module is used to acquire a facial video of the examinee to be identified, and to perform sampling processing on the facial video according to a preset sampling period to acquire no less than one single-frame image;

[0030] A coordinate acquisition module, used to perform single-frame face detection processing on the single-frame images in time sequence to obtain the single-frame face coordinates corresponding to each single-frame image;

[0031] A single-frame determination module is used to determine cheating of the examinee to be identified in each single-frame image based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, and generate a cheating determination result corresponding to each single-frame image;

[0032] The comprehensive determination module is used to determine whether the candidate to be identified is cheating based on the cheating determination results corresponding to each single frame image.

[0033] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0034] Obtain a facial video of the examinee to be identified, and perform sampling processing on the facial video according to a preset sampling period to obtain no less than one single-frame image;

[0035] Performing single-frame face detection processing on the single-frame images in time sequence to obtain single-frame face coordinates corresponding to each single-frame image;

[0036] Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a cheating determination is performed on the candidate to be identified in each single-frame image, and a cheating determination result corresponding to each single-frame image is generated;

[0037] Based on the cheating determination results corresponding to each single-frame image, it is determined whether the candidate to be identified is cheating.

[0038] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0039] Obtain a facial video of the examinee to be identified, and perform sampling processing on the facial video according to a preset sampling period to obtain no less than one single-frame image;

[0040] Performing single-frame face detection processing on the single-frame images in time sequence to obtain single-frame face coordinates corresponding to each single-frame image;

[0041] Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a cheating determination is performed on the candidate to be identified in each single-frame image, and a cheating determination result corresponding to each single-frame image is generated;

[0042] Based on the cheating determination results corresponding to each single-frame image, it is determined whether the candidate to be identified is cheating.

[0043] The above-mentioned online examination cheating identification method, device, computer equipment and storage medium obtain a facial video of the candidate to be identified, sample the facial video according to a preset sampling period, obtain no less than one single-frame image, perform single-frame face detection processing on the single-frame images according to a time series, obtain the single-frame face coordinates corresponding to each single-frame image, and based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, perform cheating judgment on the candidate to be identified in each single-frame image, generate a cheating judgment result corresponding to each single-frame image, and finally, based on the cheating judgment result corresponding to each single-frame image, judge whether the candidate to be identified is cheating. By obtaining the cheating judgment result corresponding to each single-frame image to judge whether the candidate to be identified is cheating, the efficiency of identifying cheating of online candidates is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A diagram showing an application environment of an online examination cheating identification method in one embodiment;

[0045] Figure 2 A schematic diagram of a flow chart of a method for identifying cheating in online examinations in one embodiment;

[0046] Figure 3 A schematic diagram of single-frame face coordinates obtained in one embodiment;

[0047] Figure 4 Schematic diagram of a flow chart of a step of determining binocular gaze angle of a single frame in one embodiment;

[0048] Figure 5A schematic diagram of a flow chart of a single-frame image cheating determination step in one embodiment;

[0049] Figure 6 It is a structural block diagram of an online examination cheating identification device in one embodiment;

[0050] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0052] The online exam cheating identification method provided by this application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The terminal 102 and the server 104 can be used separately to perform the online exam cheating identification provided by this application. The terminal 102 and the server 104 can also be used to collaboratively perform the online exam cheating identification provided by this application. For example, the server 104 is used to obtain the facial video of the candidate to be identified, sample the facial video according to a preset sampling period, and obtain no less than one single-frame image; perform single-frame face detection processing on the single-frame images according to the time series, and obtain the single-frame face coordinates corresponding to each single-frame image; based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, perform cheating judgment on the candidate to be identified in each single-frame image, and generate a cheating judgment result corresponding to each single-frame image; based on the cheating judgment result corresponding to each single-frame image, judge whether the candidate to be identified is cheating.

[0053] The terminal 102 may be, but is not limited to, a terminal device capable of acquiring a facial video of the examinee to be identified, and the server 104 may be implemented as an independent server or a server cluster consisting of multiple servers.

[0054] In one embodiment, Figure 2 As shown, a method for identifying cheating in online examinations is provided. Figure 1 The terminal in is used as an example to illustrate, including the following steps:

[0055] Step 202, obtain the facial video of the candidate to be identified, sample the facial video according to a preset sampling period, and obtain no less than one single-frame image.

[0056] Specifically, the facial video of the candidate to be identified during the online exam is obtained in real time through the camera on the terminal, such as the camera on the mobile phone, the camera on the computer or other external cameras; the video resolution of the facial video of the candidate to be identified should be at least 720P. After obtaining the facial video, due to the redundancy of the video information, sampling frames can be extracted from the video sequence according to the actual situation and the carrying capacity of the machine, which can reduce the calculation amount of subsequent operations and achieve a balance between system performance and proctoring effect. The facial video is sampled according to the preset sampling period to obtain no less than one single frame image. Single-frame images can be extracted from the video stream at fixed intervals according to demand. For example, 2 frames / second can be used as the sampling rate of frame extraction, and it can also be adjusted according to specific circumstances.

[0057] Step 204 , performing single-frame face detection processing on the single-frame images in time sequence to obtain single-frame face coordinates corresponding to each single-frame image.

[0058] Specifically, single-frame face detection is performed on single-frame images in time series to obtain the single-frame face coordinates corresponding to each single-frame image. Single-frame face detection is used to determine the location of the facial area of ​​the candidate to be identified in the single-frame image, which is the prerequisite for subsequent processing steps. For the single-frame face detection method, the built-in face detection algorithm in the Python Dlib library can be used. Figure 3 FIG. 1 is a schematic diagram of a single-frame face coordinate obtained in one embodiment. Figure 3 As shown in the figure, facial key points are usually used to mark the positions of key parts of the face, including eyebrows, eyes, eyebrows, mouth and contours. To detect facial key points, you need to first determine the position of the face. After that, for the input face area image, the Python Dlib library can be used to obtain the two-dimensional position coordinates of 68 facial key points as the single-frame face coordinates corresponding to each single-frame image.

[0059] Step 206, based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, cheating determination is performed on the candidate to be identified in each single-frame image, and a cheating determination result corresponding to each single-frame image is generated.

[0060] Specifically, after obtaining each single-frame image and the single-frame face coordinates corresponding to each single-frame image, cheating determination is performed on the candidate to be identified in each single-frame image according to each single-frame image and the single-frame face coordinates corresponding to each single-frame image, and a cheating determination result corresponding to each single-frame image is generated. When cheating determination is performed on the candidate to be identified in each single-frame image, a single-frame cheating determination network model is generated through sample data training, and each feature extracted from each single-frame image and the single-frame face coordinates corresponding to each single-frame image is input into the single-frame cheating determination network model to obtain a cheating determination result corresponding to each single-frame image.

[0061] Step 208: determine whether the candidate to be identified is cheating based on the cheating determination results corresponding to each single-frame image.

[0062] Specifically, after obtaining the cheating determination results corresponding to each single-frame image, since the single-frame image only describes a local or a very small time segment in the exam, it is unfair to make a judgment based solely on the cheating determination results corresponding to the single-frame image, and it is easy to produce many false positives. The misjudgment rate will be greatly reduced by combining the cheating determination results corresponding to each single-frame image for comprehensive judgment. Therefore, by performing statistical analysis based on the cheating determination results corresponding to each single-frame image, it is determined whether the candidate to be identified is cheating by obtaining the cheating rate for a certain period of time. For example, a sliding time window is set, and the cheating rate of each single-frame image in the time window is obtained. If the cheating rate exceeds a certain threshold, it is determined that the candidate to be identified is cheating in the time window.

[0063] In the above-mentioned online examination cheating identification method, a facial video of the candidate to be identified is obtained, and the facial video is sampled according to a preset sampling period to obtain no less than one single-frame image, and single-frame face detection processing is performed on the single-frame images according to a time series to obtain the single-frame face coordinates corresponding to each single-frame image, and based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a cheating judgment is made on the candidate to be identified in each single-frame image, and a cheating judgment result corresponding to each single-frame image is generated. Finally, based on the cheating judgment result corresponding to each single-frame image, it is determined whether the candidate to be identified is cheating. By obtaining the cheating judgment result corresponding to the single-frame image to determine whether the candidate to be identified is cheating, the efficiency of identifying cheating of online candidates is improved.

[0064] In one embodiment, the cheating determination of the examinee to be identified in each single-frame image based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, and generating the cheating determination result corresponding to each single-frame image includes:

[0065] Based on the single-frame face coordinates corresponding to each single-frame image, the candidate to be identified is subjected to blink determination, and the blink determination result is obtained;

[0066] Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, single-frame head posture determination is performed on the candidate to be identified to obtain the single-frame head posture determination result;

[0067] Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a single-frame binocular gaze angle determination is performed on the candidate to be identified, and a single-frame binocular gaze angle determination result is obtained;

[0068] Based on the blink determination results, the head posture determination results and the binocular gaze angle determination results of each single-frame image, cheating determination is performed on the candidate to be identified in each single-frame image to generate a cheating determination result corresponding to each single-frame image.

[0069] Specifically, based on the single-frame face coordinates corresponding to each single-frame image, the candidate to be identified is judged to blink, and the blink judgment result is obtained; since the user will inevitably blink subconsciously during the test, when the user closes his eyes, the eyeball disappears, and the blink judgment result will be judged as closed eyes, and the single-frame image will be directly judged as not looking at the screen. Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, the candidate to be identified is judged to have a single-frame head posture judgment, and the single-frame head posture judgment result is obtained. The single-frame head posture estimation judges the direction angle and position of the head in a single-frame image relative to the camera. The single-frame head posture judgment problem can usually be converted into a Perspective-n-Point problem to solve. As long as there is a calibrated camera, the positions of n three-dimensional points on the object and the corresponding projection point positions onto the two-dimensional plane can obtain the posture of the object. Figure 4 FIG. 4 is a flow chart of a step of determining binocular gaze angle of a single frame in an embodiment. Figure 4 As shown, based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, the single-frame binocular gaze angle judgment is performed on the candidate to be identified, and the single-frame binocular gaze angle judgment result is obtained. The single-frame binocular gaze angle judgment is for the input eye picture, and the corresponding feature value can be fully extracted from it through the preset deep learning network. The entire network outputs 18 heat maps, corresponding to 18 eye key points of the orbital edge and pupil (wherein, 8 points are located around the orbit, 8 points are located around the pupil, 1 eyeball center and 1 pupil center). Based on these 18 eye key points and the eyeball model, the gaze angle of each eye can be calculated. Finally, according to the blink judgment result of each single-frame image, the single-frame head posture judgment result and the single-frame binocular gaze angle judgment result, the candidate to be identified in each single-frame image is judged to cheat, and the cheating judgment result corresponding to each single-frame image is generated.

[0070] In the present embodiment, based on the single-frame face coordinates corresponding to each single-frame image, a blink judgment is performed on the candidate to be identified to obtain a blink judgment result; based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a single-frame head posture judgment is performed on the candidate to be identified to obtain a single-frame head posture judgment result; based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a single-frame binocular gaze angle judgment is performed on the candidate to be identified to obtain a single-frame binocular gaze angle judgment result; finally, according to the blink judgment result, the single-frame head posture judgment result and the single-frame binocular gaze angle judgment result of each single-frame image, a cheating judgment is performed on the candidate to be identified in each single-frame image to generate a cheating judgment result corresponding to each single-frame image, thereby realizing the acquisition of the cheating judgment result corresponding to each single-frame image, and improving the accuracy of judging whether the candidate to be identified in each single-frame image is cheating.

[0071] In one embodiment, the blink determination of the examinee to be identified is performed based on the single-frame face coordinates corresponding to each single-frame image, and the blink determination result is obtained, including:

[0072] Based on the single-frame face coordinates corresponding to each single-frame image, the aspect ratio of the eyes of the candidate to be identified is calculated;

[0073] If the eye length-to-width ratio of the candidate to be identified is lower than the preset eye length-to-width ratio threshold, outputting that the candidate to be identified is in a closed-eye state as the blink determination result;

[0074] If the eye length-to-width ratio of the candidate to be identified is greater than or equal to a preset eye length-to-width ratio threshold, the eye-opening state of the candidate to be identified is output as a blink determination result.

[0075] Specifically, according to the single-frame face coordinates corresponding to each single-frame image, when judging the blinking of the candidate to be identified, we need to determine the position of both eyes through the positions of 68 facial key points, and calculate the eye aspect ratio of the candidate to be identified based on the single-frame face coordinates corresponding to each single-frame image, that is, the aspect ratio of the 6 feature points {P1, P2, P3, P4, P5, P6} around the eyes can be used to determine whether the candidate blinks, which is called EAR (Eye Aspect Ratio). The calculation formula of the eye aspect ratio can refer to the following formula:

[0076]

[0077] When the eye aspect ratio of the candidate to be identified is lower than the preset eye aspect ratio threshold, the candidate to be identified is output as the blink determination result, which indicates that the candidate is in a closed eye state; when the eye aspect ratio of the candidate to be identified is greater than or equal to the preset eye aspect ratio threshold, the candidate to be identified is output as the blink determination result, which indicates that the candidate is in an open eye state. When the user opens his eyes, the EAR will stabilize at a certain value, which is then determined as an open eye state; and when the user closes his eyes, the EAR will drop rapidly in a short period of time. When the EAR is less than the preset eye aspect ratio threshold (e.g., 0.15), it is determined as a closed eye state.

[0078] In the present embodiment, the eye aspect ratio of the candidate to be identified is calculated based on the single-frame face coordinates corresponding to each single-frame image. When the eye aspect ratio of the candidate to be identified is lower than the preset eye aspect ratio threshold, the output is that the candidate to be identified is in a closed-eye state as the blinking judgment result. When the eye aspect ratio of the candidate to be identified is greater than or equal to the preset eye aspect ratio threshold, the output is that the candidate to be identified is in an open-eye state as the blinking judgment result, thereby improving the accuracy of blinking judgment for the candidate to be identified.

[0079] In one embodiment, the single-frame head posture determination of the examinee to be identified is performed based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, and the single-frame head posture determination result is obtained, including:

[0080] Processing the single-frame face coordinates based on a preset universal three-dimensional face model to obtain two-dimensional face coordinates corresponding to the single-frame face coordinates;

[0081] Obtaining the camera parameters of the camera that captures the facial video of the candidate to be identified; the camera parameters include the focal length and optical center of the camera;

[0082] The single-frame face coordinates, the three-dimensional face coordinates corresponding to the single-frame face coordinates, and the camera parameters are input into a preset rotation amount and offset calculation formula to obtain the rotation amount and offset as a single-frame head posture determination result.

[0083] Specifically, the single-frame head posture determination of the candidate to be identified is used to determine the direction angle and position of the head of the candidate to be identified in the single-frame image relative to the camera, which is essentially a posture estimation problem. The posture estimation problem can usually be converted into a perspective-n-point problem. In this problem, as long as there is a calibrated camera, the position of n three-dimensional points on the object and the corresponding projection point position onto the two-dimensional plane can be used to obtain the posture of the object. The single-frame face coordinates are processed by a preset universal three-dimensional face model to obtain the two-dimensional face coordinates corresponding to the single-frame face coordinates; through the posture of the object relative to the camera, the three-dimensional point in the world coordinate can be converted into a three-dimensional point in the camera coordinate by rotation and translation. The position of the two-dimensional coordinate point in the single-frame image can use the coordinates of the 68 facial key points calculated in the previous text, and the three-dimensional coordinates corresponding to these 68 feature points can be obtained through a universal three-dimensional face model. Obtain the camera parameters of the camera that shoots the facial video of the candidate to be identified. The camera parameters include the focal length and optical center of the camera, which can be approximated by the image width and the center of the image in pixels, and assume that there is no radial distortion. Finally, the single-frame face coordinates, the three-dimensional face coordinates corresponding to the single-frame face coordinates, and the camera parameters are input into the preset rotation and offset calculation formula to obtain the rotation and offset as the single-frame head posture determination result. In OpenCV, the solvePnP function is provided to calculate the Perspective-n-Point problem. Among them, the solvePnP function needs to input the following parameters: objectPoints: three-dimensional coordinate points in the world coordinate system; imagePoints: points on the corresponding two-dimensional plane; cameraMatrix: camera parameters, including focal length and optical center in the image; distCoeffs: distortion coefficient. The solvePnP function outputs a rotation vector and a translation vector. Here, some two-dimensional plane points (nose tip, chin, left and right eyes, left and right corners of the mouth) in the face detection points obtained by DLib are input, and the corresponding rotation and displacement can be obtained by calculation. In this embodiment, the single-frame face coordinates, the two-dimensional face coordinates corresponding to the single-frame face coordinates, and the camera parameters are input into the preset rotation and offset calculation formula to obtain the rotation and offset as the single-frame head posture determination result.

[0084] In this embodiment, the single-frame face coordinates are processed based on a preset general three-dimensional face model to obtain the two-dimensional face coordinates corresponding to the single-frame face coordinates, and the camera parameters of the camera that shoots the facial video of the candidate to be identified are obtained. Finally, the single-frame face coordinates, the three-dimensional face coordinates corresponding to the single-frame face coordinates, and the camera parameters are input into the preset rotation and offset calculation formula to obtain the rotation and offset as the single-frame head posture determination result. By performing single-frame head posture determination based on the single-frame face coordinates, the three-dimensional face coordinates corresponding to the single-frame face coordinates, and the camera parameters, the rotation and offset are obtained as the single-frame head posture determination result, thereby improving the accuracy of single-frame head posture determination.

[0085] In one embodiment, the single-frame binocular gaze angle determination of the examinee to be identified is performed based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, and the single-frame binocular gaze angle determination result is obtained, including:

[0086] Input each single frame image into a preset stacked hourglass network model for processing to obtain multiple eye key points;

[0087] Based on the eye key points and the single-frame face coordinates corresponding to each single-frame image, a single-frame binocular gaze angle of the candidate to be identified is determined, and the gaze angle is obtained as a determination result.

[0088] Specifically, each single-frame image is input into a preset stacked hourglass network model for processing to obtain multiple eye key points, that is, the entire stacked hourglass network model outputs 18 heat maps, corresponding to the 18 eye key points of the orbital edge and the pupil (8 points are located around the orbit, 8 points are located around the pupil, 1 is the center of the eyeball and 1 is the center of the pupil). Based on these 18 eye key points and the eyeball model, the gaze angle of each eye can be calculated. Based on the eye key points and the single-frame facial coordinates corresponding to each single-frame image, a single-frame binocular gaze angle judgment is performed on the candidate to be identified, and the gaze angle is obtained as the judgment result. Gaze angle The calculation formula can refer to the following calculation formula:

[0089]

[0090]

[0091] The pupil center is (u i0 , v i0 ); the center of the eyeball is (u c , v c ); the eyeball radius is ruv; the values ​​in arccosine and arcsine are clipped to between -1 and 1 to prevent inverse trigonometric functions from being uncomputable.

[0092] In this embodiment, each single-frame image is input into a preset stacked hourglass network model for processing to obtain multiple eye key points, and a single-frame binocular gaze angle judgment is performed on the candidate to be identified based on the eye key points and the single-frame face coordinates corresponding to each single-frame image. The gaze angle is obtained as the judgment result, thereby realizing the acquisition of the gaze angle and improving the efficiency of obtaining the gaze angle.

[0093] In one embodiment, the cheating determination of the examinee to be identified in each single-frame image is performed according to the blink determination result, the head posture determination result and the binocular gaze angle determination result of each single-frame image, and the cheating determination result corresponding to each single-frame image is generated, including:

[0094] According to the blink determination results of each single frame image, it is determined whether the candidate to be identified is in an eye-opening state;

[0095] If the candidate to be identified is in a state of closing eyes, it is determined that the candidate to be identified in the single frame image is not looking at the screen;

[0096] If the examinee to be identified is in an open-eye state, the single-frame head posture determination result and the single-frame binocular gaze angle determination result are input into a preset single-frame cheating determination network model, and the cheating determination result corresponding to the single-frame image is output.

[0097] The preset single-frame cheating judgment network model is obtained by training the support vector machine (SVM) based on sample data; when training the SVM, the sample data extracted from the sample image is input into the initialized SVM, and the cheating judgment result of the single-frame image is output. When the training reaches the preset conditions, the training of the SVM is completed, and the single-frame cheating judgment network model is obtained. The sample data used for training is 9-dimensional feature data, including the average value of the gaze angle of the eyes of the person to be identified. The single-frame head posture determination result (r1, r2, r3); and the coordinates and length and width (x, y, w, h) of the center point of the face frame obtained based on the single-frame face coordinates.

[0098] Specifically, Figure 5 FIG. 1 is a flow chart of a single-frame image cheating determination step in an embodiment. Figure 5 As shown, first, according to the blink judgment result of each single-frame image, it is judged whether the candidate to be identified is in an open-eye state; when the candidate to be identified is in a closed-eye state, it is judged that the candidate to be identified in the single-frame image is not looking at the screen; when the candidate to be identified is in an open-eye state, the single-frame head posture judgment result and the single-frame binocular gaze angle judgment result are converted into the feature types required by the preset single-frame cheating judgment network model, including the average value of the gaze angle of the eyes of the person to be identified The single-frame head posture judgment result (r1, r2, r3); and the coordinates of the center point and length and width (x, y, w, h) of the face frame obtained based on the single-frame face coordinates, and the above feature data are input into the single-frame cheating judgment network model. Finally, the single-frame cheating judgment network model outputs the cheating judgment result corresponding to the single-frame image.

[0099] In the present embodiment, whether the examinee to be identified is in an open-eye state is determined based on the blink determination result of each single-frame image; when the examinee to be identified is in an open-eye state, the single-frame head posture determination result and the single-frame binocular gaze angle determination result are input into a preset single-frame cheating determination network model, and the cheating determination result corresponding to the single-frame image is output, thereby realizing the determination of the cheating result of the user to be identified in the single-frame image and improving the accuracy of cheating identification for the single-frame image.

[0100] In one embodiment, the step of determining whether the candidate to be identified is cheating based on the cheating determination results corresponding to each single frame image includes:

[0101] Arranging the cheating determination results of each single frame image in time series to generate a cheating determination result sequence;

[0102] Obtaining a cheating rate of the cheating determination result sequence within a time window of a preset duration; the cheating rate is obtained by calculating the ratio of the number of times determined as cheating within the time window to the total number of times determined;

[0103] Based on the cheating rate in each time window and a preset cheating rate threshold, it is determined whether the candidate to be identified is cheating.

[0104] Specifically, after obtaining the cheating determination results corresponding to each single-frame image, since the single-frame image only describes a local or a very small time segment in the exam, it is unfair to make a judgment based solely on the cheating determination results corresponding to the single-frame image, and it is easy to produce many false positives. The misjudgment rate will be greatly reduced by combining the cheating determination results corresponding to each single-frame image for comprehensive judgment. The cheating determination results of each single-frame image are arranged in time series to generate a cheating determination result sequence, and the cheating rate of the cheating determination result sequence within a time window of a preset time length is obtained. Based on the cheating rate of each time window and the preset cheating rate threshold, it is determined whether the candidate to be identified is cheating. For example, the time window can be set to 15 seconds; the cheating rate threshold can be set to 30%; the calculation formula of the cheating rate can refer to the following formula:

[0105]

[0106] Among them, cheat ratio is the cheating rate; cheat frames is the number of times that cheating is judged within the time window; ∑ framesis the total number of judgments.

[0107] When the cheating rate of each time window exceeds the preset cheating rate threshold, the candidate to be identified in the time window is determined to be cheating; when a preset number of time windows are identified as cheating by the candidates to be identified, the candidate to be identified is determined to be cheating, and the information with the video number, the relative time point where the cheating occurred, etc. can be output. Subsequently, this information can be used to automatically determine cheating or serve as input for manual secondary review.

[0108] In this embodiment, by arranging the cheating determination results of each single frame image according to a time series, a cheating determination result sequence is generated, and the cheating rate of the cheating determination result sequence within a time window of a preset duration is obtained. Finally, based on the cheating rate of each time window and a preset cheating rate threshold, it is determined whether the candidate to be identified is cheating, thereby realizing the determination of whether the candidate to be identified is cheating and improving the efficiency of identifying cheating of online candidates.

[0109] It should be understood that although Figure 2 , 4 The steps in the flowcharts of and 5 are shown in sequence as indicated by arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 2 , 4 At least part of the steps in 5 may include multiple steps or multiple stages. These steps or stages do not necessarily have to be executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily one by one, but can be executed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0110] In one embodiment, Figure 6 As shown, an online examination cheating identification device is provided, comprising: an image acquisition module 601, a coordinate acquisition module 602, a single frame determination module 603 and a comprehensive determination module 604, wherein:

[0111] The image acquisition module 601 is used to acquire the facial video of the examinee to be identified, and to perform sampling processing on the facial video according to a preset sampling period to acquire no less than one single-frame image.

[0112] The coordinate acquisition module 602 is used to perform single-frame face detection processing on the single-frame images in time sequence to obtain the single-frame face coordinates corresponding to each single-frame image.

[0113] The single-frame determination module 603 is used to perform cheating determination on the candidate to be identified in each single-frame image based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, and generate a cheating determination result corresponding to each single-frame image.

[0114] The comprehensive determination module 604 is used to determine whether the candidate to be identified is cheating based on the cheating determination results corresponding to each single frame image.

[0115] In one embodiment, the single-frame judgment module 603 is also used to: perform a blink judgment on the candidate to be identified based on the single-frame face coordinates corresponding to each single-frame image, and obtain a blink judgment result; perform a single-frame head posture judgment on the candidate to be identified based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, and obtain a single-frame head posture judgment result; perform a single-frame binocular gaze angle judgment on the candidate to be identified based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, and obtain a single-frame binocular gaze angle judgment result; perform a cheating judgment on the candidate to be identified in each single-frame image according to the blink judgment result, the single-frame head posture judgment result and the single-frame binocular gaze angle judgment result of each single-frame image, and generate a cheating judgment result corresponding to each single-frame image.

[0116] In one embodiment, the single-frame determination module 603 is further used to: calculate the eye aspect ratio of the candidate to be identified based on the single-frame face coordinates corresponding to each single-frame image; if the eye aspect ratio of the candidate to be identified is lower than a preset eye aspect ratio threshold, output that the candidate to be identified is in a closed-eye state as a blink determination result; if the eye aspect ratio of the candidate to be identified is greater than or equal to the preset eye aspect ratio threshold, output that the candidate to be identified is in an open-eye state as a blink determination result.

[0117] In one embodiment, the single-frame judgment module 603 is also used to: process the single-frame face coordinates based on a preset general three-dimensional face model to obtain the two-dimensional face coordinates corresponding to the single-frame face coordinates; obtain the camera parameters of the camera that shoots the facial video of the candidate to be identified; the camera parameters include the focal length and optical center of the camera; input the single-frame face coordinates, the two-dimensional face coordinates corresponding to the single-frame face coordinates, and the camera parameters into a preset rotation and offset calculation formula to obtain the rotation and offset as the single-frame head posture judgment result.

[0118] In one embodiment, the single-frame judgment module 603 is also used to: input each single-frame image into a preset stacked hourglass network model for processing to obtain multiple eye key points; perform single-frame binocular gaze angle judgment on the candidate to be identified based on the eye key points and the single-frame face coordinates corresponding to each single-frame image, and obtain the gaze angle as the judgment result.

[0119] In one embodiment, the single-frame judgment module 603 is also used to: judge whether the candidate to be identified is in an open-eye state based on the blink judgment result of each single-frame image; if the candidate to be identified is in a closed-eye state, judge that the candidate to be identified in the single-frame image is not looking at the screen; if the candidate to be identified is in an open-eye state, input the single-frame head posture judgment result and the single-frame binocular gaze angle judgment result into a preset single-frame cheating judgment network model, and output the cheating judgment result corresponding to the single-frame image.

[0120] In one embodiment, the comprehensive determination module 604 is further used to: arrange the cheating determination results of each single frame image in a time series to generate a cheating determination result sequence; obtain the cheating rate of the cheating determination result sequence within a time window of a preset duration; the cheating rate is obtained by calculating the ratio of the number of times judged as cheating within the time window to the total number of times of judgment; based on the cheating rate of each time window and a preset cheating rate threshold, determine whether the candidate to be identified is cheating.

[0121] The above-mentioned online examination cheating identification device obtains a facial video of the candidate to be identified, samples the facial video according to a preset sampling period, obtains no less than one single-frame image, performs single-frame face detection processing on the single-frame images according to a time series, obtains the single-frame face coordinates corresponding to each single-frame image, and makes a cheating determination on the candidate to be identified in each single-frame image based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, generates a cheating determination result corresponding to each single-frame image, and finally determines whether the candidate to be identified is cheating based on the cheating determination result corresponding to each single-frame image. By obtaining the cheating determination result corresponding to each single-frame image to determine whether the candidate to be identified is cheating, the efficiency of identifying cheating of online candidates is improved.

[0122] For the specific definition of the online exam cheating identification device, please refer to the definition of the online exam cheating identification method above, which will not be repeated here. Each module in the above-mentioned online exam cheating identification device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0123] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 7As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an online examination cheating identification method is implemented.

[0124] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0125] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0126] Obtain a facial video of the examinee to be identified, and perform sampling processing on the facial video according to a preset sampling period to obtain no less than one single-frame image;

[0127] Performing single-frame face detection processing on the single-frame images in time sequence to obtain single-frame face coordinates corresponding to each single-frame image;

[0128] Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a cheating determination is performed on the candidate to be identified in each single-frame image, and a cheating determination result corresponding to each single-frame image is generated;

[0129] Based on the cheating determination results corresponding to each single-frame image, it is determined whether the candidate to be identified is cheating.

[0130] In one embodiment, the processor also implements the following steps when executing the computer program: based on the single-frame face coordinates corresponding to each single-frame image, a blink judgment is performed on the candidate to be identified, and a blink judgment result is obtained; based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a single-frame head posture judgment is performed on the candidate to be identified, and a single-frame head posture judgment result is obtained; based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a single-frame binocular gaze angle judgment is performed on the candidate to be identified, and a single-frame binocular gaze angle judgment result is obtained; based on the blink judgment result, the single-frame head posture judgment result and the single-frame binocular gaze angle judgment result of each single-frame image, a cheating judgment is performed on the candidate to be identified in each single-frame image, and a cheating judgment result corresponding to each single-frame image is generated.

[0131] In one embodiment, when the processor executes the computer program, the following steps are also implemented: based on the single-frame facial coordinates corresponding to each single-frame image, the eye aspect ratio of the candidate to be identified is calculated; if the eye aspect ratio of the candidate to be identified is lower than a preset eye aspect ratio threshold, the candidate to be identified is in a closed-eye state as a blink judgment result, and if the eye aspect ratio of the candidate to be identified is greater than or equal to the preset eye aspect ratio threshold, the candidate to be identified is in an open-eye state as a blink judgment result.

[0132] In one embodiment, the processor also implements the following steps when executing the computer program: processing the single-frame face coordinates based on a preset general three-dimensional face model to obtain the two-dimensional face coordinates corresponding to the single-frame face coordinates; obtaining the camera parameters of the camera that shoots the facial video of the candidate to be identified; the camera parameters include the focal length and optical center of the camera; inputting the single-frame face coordinates, the two-dimensional face coordinates corresponding to the single-frame face coordinates, and the camera parameters into a preset rotation and offset calculation formula to obtain the rotation and offset as the single-frame head posture judgment result.

[0133] In one embodiment, when the processor executes the computer program, the following steps are also implemented: each single-frame image is input into a preset stacked hourglass network model for processing to obtain multiple eye key points; based on the eye key points and the single-frame face coordinates corresponding to each single-frame image, a single-frame binocular gaze angle judgment is performed on the candidate to be identified, and the gaze angle is obtained as a judgment result.

[0134] In one embodiment, when the processor executes the computer program, the following steps are also implemented: based on the blink judgment result of each single-frame image, it is determined whether the candidate to be identified is in an open-eye state; if the candidate to be identified is in a closed-eye state, it is determined that the candidate to be identified in the single-frame image is not looking at the screen; if the candidate to be identified is in an open-eye state, the single-frame head posture judgment result and the single-frame binocular gaze angle judgment result are input into a preset single-frame cheating judgment network model, and the cheating judgment result corresponding to the single-frame image is output.

[0135] In one embodiment, when the processor executes the computer program, the following steps are also implemented: arranging the cheating determination results of each single frame image in time series to generate a cheating determination result sequence; obtaining the cheating rate of the cheating determination result sequence within a time window of a preset duration; the cheating rate is obtained by calculating the ratio of the number of times judged as cheating within the time window to the total number of times judged; based on the cheating rate of each time window and a preset cheating rate threshold, determining whether the candidate to be identified is cheating.

[0136] The above-mentioned computer equipment obtains a facial video of the candidate to be identified, samples the facial video according to a preset sampling period, obtains no less than one single-frame image, performs single-frame face detection processing on the single-frame images according to a time series, obtains the single-frame face coordinates corresponding to each single-frame image, makes a cheating determination on the candidate to be identified in each single-frame image based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, generates a cheating determination result corresponding to each single-frame image, and finally determines whether the candidate to be identified is cheating based on the cheating determination result corresponding to each single-frame image. By obtaining the cheating determination result corresponding to each single-frame image to determine whether the candidate to be identified is cheating, the efficiency of identifying cheating of online candidates is improved.

[0137] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0138] Obtain a facial video of the examinee to be identified, and perform sampling processing on the facial video according to a preset sampling period to obtain no less than one single-frame image;

[0139] Performing single-frame face detection processing on the single-frame images in time sequence to obtain single-frame face coordinates corresponding to each single-frame image;

[0140] Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a cheating determination is performed on the candidate to be identified in each single-frame image, and a cheating determination result corresponding to each single-frame image is generated;

[0141] Based on the cheating determination results corresponding to each single-frame image, it is determined whether the candidate to be identified is cheating.

[0142] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: based on the single-frame face coordinates corresponding to each single-frame image, a blink judgment is performed on the candidate to be identified, and a blink judgment result is obtained; based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a single-frame head posture judgment is performed on the candidate to be identified, and a single-frame head posture judgment result is obtained; based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a single-frame binocular gaze angle judgment is performed on the candidate to be identified, and a single-frame binocular gaze angle judgment result is obtained; based on the blink judgment result, the single-frame head posture judgment result and the single-frame binocular gaze angle judgment result of each single-frame image, a cheating judgment is performed on the candidate to be identified in each single-frame image, and a cheating judgment result corresponding to each single-frame image is generated.

[0143] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: based on the single-frame face coordinates corresponding to each single-frame image, the eye aspect ratio of the candidate to be identified is calculated; if the eye aspect ratio of the candidate to be identified is lower than a preset eye aspect ratio threshold, the candidate to be identified is in a closed-eye state as a blink judgment result, and if the eye aspect ratio of the candidate to be identified is greater than or equal to the preset eye aspect ratio threshold, the candidate to be identified is in an open-eye state as a blink judgment result.

[0144] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: processing the single-frame face coordinates based on a preset general three-dimensional face model to obtain the two-dimensional face coordinates corresponding to the single-frame face coordinates; obtaining the camera parameters of the camera that shoots the facial video of the candidate to be identified; the camera parameters include the focal length and optical center of the camera; inputting the single-frame face coordinates, the two-dimensional face coordinates corresponding to the single-frame face coordinates, and the camera parameters into a preset rotation and offset calculation formula to obtain the rotation and offset as the single-frame head posture judgment result.

[0145] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: each single-frame image is input into a preset stacked hourglass network model for processing to obtain multiple eye key points; based on the eye key points and the single-frame face coordinates corresponding to each single-frame image, a single-frame binocular gaze angle judgment is performed on the candidate to be identified, and the gaze angle is obtained as a judgment result.

[0146] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: based on the blink judgment result of each single-frame image, determine whether the candidate to be identified is in an open-eye state; if the candidate to be identified is in a closed-eye state, determine that the candidate to be identified in the single-frame image is not looking at the screen; if the candidate to be identified is in an open-eye state, input the single-frame head posture judgment result and the single-frame binocular gaze angle judgment result into a preset single-frame cheating judgment network model, and output the cheating judgment result corresponding to the single-frame image.

[0147] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: arranging the cheating determination results of each single frame image in time series to generate a cheating determination result sequence; obtaining the cheating rate of the cheating determination result sequence within a time window of a preset duration; the cheating rate is obtained by calculating the ratio of the number of times judged as cheating within the time window to the total number of times judged; based on the cheating rate of each time window and a preset cheating rate threshold, determining whether the candidate to be identified is cheating.

[0148] The above-mentioned storage medium obtains the facial video of the candidate to be identified, samples the facial video according to a preset sampling period, obtains no less than one single-frame image, performs single-frame face detection processing on the single-frame images according to a time series, obtains the single-frame face coordinates corresponding to each single-frame image, and makes a cheating judgment on the candidate to be identified in each single-frame image based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, generates a cheating judgment result corresponding to each single-frame image, and finally judges whether the candidate to be identified is cheating based on the cheating judgment result corresponding to each single-frame image. By obtaining the cheating judgment result corresponding to each single-frame image to judge whether the candidate to be identified is cheating, the efficiency of identifying cheating of online candidates is improved.

[0149] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0150] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0151] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for identifying cheating in online examinations, characterized in that: The method comprises: Obtain a facial video of the examinee to be identified, and perform sampling processing on the facial video according to a preset sampling period to obtain no less than one single-frame image; Performing single-frame face detection processing on the single-frame images in time sequence to obtain single-frame face coordinates corresponding to each single-frame image; Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a cheating determination is performed on the candidate to be identified in each single-frame image, and a cheating determination result corresponding to each single-frame image is generated, including: Based on the single-frame face coordinates corresponding to each single-frame image, the examinee to be identified is subjected to blink determination, and the blink determination result is obtained, including: Based on the single-frame face coordinates corresponding to each single-frame image, the eye aspect ratio of the candidate to be identified is calculated; if the eye aspect ratio of the candidate to be identified is lower than the preset eye aspect ratio threshold, the candidate to be identified is in a closed-eye state as the blink determination result; if the eye aspect ratio of the candidate to be identified is greater than or equal to the preset eye aspect ratio threshold, the candidate to be identified is in an open-eye state as the blink determination result; Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, single-frame head posture determination is performed on the candidate to be identified to obtain the single-frame head posture determination result; Based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, a single-frame binocular gaze angle determination is performed on the candidate to be identified, and a single-frame binocular gaze angle determination result is obtained; Perform cheating judgment on the examinee to be identified in each single-frame image according to the blink judgment result, the single-frame head posture judgment result and the single-frame binocular gaze angle judgment result of each single-frame image, and generate cheating judgment results corresponding to each single-frame image, including: judging whether the examinee to be identified is in an open-eye state according to the blink judgment result of each single-frame image; if the examinee to be identified is in a closed-eye state, judging that the examinee to be identified in the single-frame image is not looking at the screen; if the examinee to be identified is in an open-eye state, inputting the single-frame head posture judgment result and the single-frame binocular gaze angle judgment result into a preset single-frame cheating judgment network model, and outputting the cheating judgment result corresponding to the single-frame image; Based on the cheating determination results corresponding to each single-frame image, determining whether the candidate to be identified is cheating includes: arranging the cheating determination results of each single-frame image in a time series to generate a cheating determination result sequence; obtaining a cheating rate of the cheating determination result sequence within a time window of a preset duration; obtaining the cheating rate by calculating the ratio of the number of times judged as cheating within the time window to the total number of times judged; and determining whether the candidate to be identified is cheating based on the cheating rate of each time window and a preset cheating rate threshold.

2. The method according to claim 1, characterized in that The single-frame head posture determination of the examinee to be identified is performed based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, and the single-frame head posture determination result is obtained, which includes: Processing the single-frame face coordinates based on a preset universal three-dimensional face model to obtain two-dimensional face coordinates corresponding to the single-frame face coordinates; Obtaining the camera parameters of the camera that captures the facial video of the candidate to be identified; the camera parameters include the focal length and optical center of the camera; The single-frame face coordinates, the two-dimensional face coordinates corresponding to the single-frame face coordinates, and the camera parameters are input into a preset rotation amount and offset calculation formula to obtain the rotation amount and offset as a single-frame head posture determination result.

3. The method according to claim 1, characterized in that The single-frame binocular gaze angle determination of the candidate to be identified is performed based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, and the single-frame binocular gaze angle determination result is obtained, which includes: Input each single frame image into a preset stacked hourglass network model for processing to obtain multiple eye key points; Based on the eye key points and the single-frame face coordinates corresponding to each single-frame image, a single-frame binocular gaze angle of the candidate to be identified is determined, and the gaze angle is obtained as a determination result.

4. An online examination cheating identification device, characterized in that: The device comprises: An image acquisition module is used to acquire a facial video of the examinee to be identified, and to perform sampling processing on the facial video according to a preset sampling period to acquire no less than one single-frame image; A coordinate acquisition module, used to perform single-frame face detection processing on the single-frame images in time sequence to obtain the single-frame face coordinates corresponding to each single-frame image; The single-frame judgment module is used to judge cheating of the candidate to be identified in each single-frame image based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, and generate cheating judgment results corresponding to each single-frame image, including: based on the single-frame face coordinates corresponding to each single-frame image, judge blinking of the candidate to be identified, and obtain blinking judgment results, including: based on the single-frame face coordinates corresponding to each single-frame image, calculate the eye aspect ratio of the candidate to be identified; if the eye aspect ratio of the candidate to be identified is lower than a preset eye aspect ratio threshold, output that the candidate to be identified is in a closed eye state as a blink judgment result; if the eye aspect ratio of the candidate to be identified is greater than or equal to the preset eye aspect ratio threshold, output that the candidate to be identified is in an open eye state as a blink judgment result; based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, judge single-frame head posture of the candidate to be identified, Obtain a single-frame head posture determination result; based on each single-frame image and the single-frame face coordinates corresponding to each single-frame image, perform a single-frame binocular gaze angle determination on the candidate to be identified, and obtain a single-frame binocular gaze angle determination result; perform a cheating determination on the candidate to be identified in each single-frame image according to the blink determination result of each single-frame image, the single-frame head posture determination result, and the single-frame binocular gaze angle determination result, and generate a cheating determination result corresponding to each single-frame image, including: judging whether the candidate to be identified is in an open-eye state according to the blink determination result of each single-frame image; if the candidate to be identified is in a closed-eye state, judging that the candidate to be identified in the single-frame image is not looking at the screen; if the candidate to be identified is in an open-eye state, inputting the single-frame head posture determination result and the single-frame binocular gaze angle determination result into a preset single-frame cheating determination network model, and outputting the cheating determination result corresponding to the single-frame image; The comprehensive determination module is used to determine whether the candidate to be identified is cheating based on the cheating determination results corresponding to each single-frame image, including: arranging the cheating determination results of each single-frame image in a time series to generate a cheating determination result sequence; obtaining the cheating rate of the cheating determination result sequence within a time window of a preset time length; the cheating rate is obtained by calculating the ratio of the number of times determined as cheating within the time window to the total number of determinations; based on the cheating rate of each time window and a preset cheating rate threshold, determining whether the candidate to be identified is cheating.

5. The device according to claim 4, characterized in that The single-frame determination module is also used to: process the single-frame face coordinates based on a preset general three-dimensional face model to obtain the two-dimensional face coordinates corresponding to the single-frame face coordinates; obtain the camera parameters of the camera that shoots the facial video of the candidate to be identified; the camera parameters include the focal length and optical center of the camera; input the single-frame face coordinates, the two-dimensional face coordinates corresponding to the single-frame face coordinates, and the camera parameters into a preset rotation and offset calculation formula to obtain the rotation and offset as the single-frame head posture determination result.

6. The device according to claim 4, characterized in that The single-frame judgment module is also used to: input each single-frame image into a preset stacked hourglass network model for processing to obtain multiple eye key points; perform single-frame binocular gaze angle judgment on the candidate to be identified based on the eye key points and the single-frame face coordinates corresponding to each single-frame image, and obtain the gaze angle as the judgment result.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images