A method for analyzing exam cheating

Abstract

The present disclosure provides a method for analyzing exam cheating. Among them, the method includes: obtaining multiple frames of images in a video stream and determining the paper offset direction of a first object in each frame of the multiple frames of images; judging whether there is a risk of paper exposure based on the paper coordinates and the hand coordinates of the first object in each frame of image; if there is a risk of paper exposure, determining a cheating suspect object associated with the first object based on the paper offset direction; determining an angle eigenvalue, an eye writing eigenvalue, and a paper result analysis eigenvalue corresponding to the cheating suspect object; determining a comprehensive score based on the angle eigenvalue, the eye writing eigenvalue, and the paper result analysis eigenvalue, and determining the degree of cheating suspicion of the cheating suspect object according to the comprehensive score. It realizes the use of intelligent video analysis technology, combined with the correlation analysis of the answer sheet results, improves the efficiency and accuracy of cheating analysis, reduces the burden of manual review, and improves the accuracy of cheating behavior recognition.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular, to a method for analyzing exam cheating. Background Art

[0002] In modern education, the fairness and effectiveness of exams are regarded as important criteria for measuring the quality of education and learning outcomes. To ensure the fairness of exams, traditional exam invigilation methods have been widely applied and supplemented by video surveillance systems.

[0003] Currently, the exam cheating analysis methods in related technologies record video data in the examination room in real time, providing intuitive monitoring images for invigilators. Invigilators remotely observe the exam status of candidates through video surveillance and promptly discover and handle suspicious behaviors.

[0004] However, the exam cheating analysis methods in related technologies have shown limitations, mainly manifested in the low efficiency and error-prone nature of manual review of video data, and at the same time, the lack of the ability to effectively associate video content with answer sheet results, making it difficult to accurately determine whether a candidate actually cheats. Summary of the Invention

[0005] The present disclosure provides a method for analyzing exam cheating.

[0006] According to a first aspect of the present disclosure, there is provided a method for analyzing exam cheating, the method comprising: obtaining multiple frames of images in a video stream and determining the paper offset direction of a first object in each frame of the multiple frames of images; based on the paper coordinates and the hand coordinates of the first object in each frame of image, determining whether there is an exposure risk of the paper; if there is an exposure risk of the paper, determining a cheating suspect object associated with the first object based on the paper offset direction; determining an angle feature value, a look-and-write feature value, and a paper result analysis feature value corresponding to the cheating suspect object; based on the angle feature value, the look-and-write feature value, and the paper result analysis feature value, determining a comprehensive score, and determining the degree of cheating suspicion of the cheating suspect object according to the comprehensive score.

[0007] In some embodiments of the present disclosure, obtaining multiple frames of images in a video stream and determining the paper offset direction of a first object in each frame of the multiple frames of images includes: respectively performing color segmentation and edge detection on each frame of image, extracting the paper region and the table region in each frame of image; determining the distance between the paper vertex coordinates of the paper region and the edge center coordinates of the table region; based on the distance and a preset safety threshold, determining a paper offset result; based on the paper offset result, determining the paper offset direction.

[0008] In some embodiments of the present disclosure, determining whether there is a risk of exposure of the test paper based on the test paper coordinates and the hand coordinates of the first object in each frame of image includes: determining the overlapping vertex coordinates of the overlapping area between the test paper area and the hand area based on the test paper vertex coordinates of the test paper area and the hand vertex coordinates of the hand area of the first object in each frame of image; if the overlapping vertex coordinates meet the preset overlapping conditions, determining the overlapping area of the overlapping area and the test paper area of the test paper area based on the overlapping vertex coordinates and the test paper vertex coordinates; if the ratio between the overlapping area and the test paper area is less than the preset proportion threshold, determining that there is a preliminary risk of exposure of the test paper; determining the overlapping ratio based on the hand vertex coordinates and the test paper vertex coordinates; if the overlapping ratio is greater than or equal to the preset overlapping ratio, determining that there is a risk of exposure of the test paper.

[0009] In some embodiments of the present disclosure, determining the cheating suspect object associated with the first object based on the test paper offset direction includes: determining the target analysis population corresponding to the test paper offset direction; using the head postures of each target analysis object in the target analysis population to determine the cheating suspect object in the target analysis population.

[0010] In some embodiments of the present disclosure, determining the target analysis population corresponding to the test paper offset direction includes: determining the associated population corresponding to the first object; according to the test paper offset direction and the position coordinates of each object in the associated population, determining the downward viewing angle of each object in the associated population looking at the first object; according to the downward viewing angle, determining the target analysis population in the associated population.

[0011] In some embodiments of the present disclosure, using the head postures of each target analysis object in the target analysis population to determine the cheating suspect object in the target analysis population includes: determining the head offset angle of each target analysis object according to the key point coordinates of each target analysis object; according to the test paper center point coordinates of each target analysis object and the position coordinates of the first object, determining the relative angle between the first object and each target analysis object; determining the angle ratio between the head offset angle and the relative angle of each target analysis object, and based on the angle ratio, determining the cheating suspect object in the target analysis population.

[0012] In some embodiments of the present disclosure, determining the angular feature value corresponding to the cheating suspect object includes: determining the horizontal deflection angle and the vertical pitch angle of the head of the first object, the seat included angle between the first object and the cheating suspect object, and the seat pitch angle between the head of the first object and the desktop of the cheating suspect object; determining the horizontal angle score based on the seat included angle between the first object and the cheating suspect object and the horizontal deflection angle of the head of the first object; determining the vertical angle score based on the vertical pitch angle of the head of the first object and the seat pitch angle between the head of the first object and the desktop of the cheating suspect object; determining the total angle score based on the horizontal angle score, the vertical angle score, and a preset angle weight; determining the average angle score based on the total angle score and the number of frames corresponding to the video stream; determining the angle standard deviation based on the total angle score and the average angle score; respectively comparing the angle standard deviation with a preset angle standard deviation threshold and the average angle score with a preset average angle score threshold, and obtaining the angular feature value based on the comparison results.

[0013] In some embodiments of the present disclosure, determining the eye writing feature value corresponding to the cheating suspect object includes: determining the residence time of the cheating suspect object in the writing area during a preset monitoring period, where the residence time in the writing area includes the gaze residence time and the pen tip residence time; determining the ratio between the residence time in the writing area and the preset monitoring period to obtain the residence time ratio in the writing area; comparing the residence time ratio in the writing area with a preset residence time ratio threshold, and determining the eye writing feature value according to the comparison result.

[0014] In some embodiments of the present disclosure, determining the test result analysis feature value corresponding to the cheating suspect object includes: obtaining the answering data of the first object and the cheating suspect object, where the answering data includes the answering answers and the answering question types; determining the question type similarity corresponding to each answering question type between the first object and the cheating suspect object based on the answering answers corresponding to each answering question type; determining the score variance corresponding to each answering question type based on the number of students and the student scores of each student under each answering question type; respectively determining the minimum score variance in the score variances corresponding to each answering question type, and using each minimum score variance as the exam difficulty coefficient corresponding to each answering question type; for each answering question type, determining the cheating suspect score corresponding to each answering question type based on the question type similarity and the exam difficulty coefficient; obtaining the test result analysis feature value based on a preset score weight and the cheating suspect scores corresponding to each answering question type.

[0015] In some embodiments of the present disclosure, based on the angle eigenvalue, the eye-writing eigenvalue, and the test paper result analysis eigenvalue, a comprehensive score is determined, and the cheating suspicion degree of the cheating suspect object is determined according to the comprehensive score, including: determining the comprehensive score based on the angle eigenvalue, the eye-writing eigenvalue, the test paper result analysis eigenvalue, and a preset eigenvalue weight; if the comprehensive score is greater than or equal to a preset comprehensive score threshold, determining that the cheating suspicion degree of the cheating suspect object is a high cheating suspicion; if the comprehensive score is less than the preset comprehensive score threshold, determining that the cheating suspicion degree of the cheating suspect object is a low cheating suspicion.

[0016] The exam cheating analysis method provided by the present disclosure obtains multiple frames of images in a video stream and determines the test paper offset direction of the first object in each frame of the multiple frames of images; based on the test paper coordinates and the hand coordinates of the first object in each frame of image, it is determined whether there is a risk of the test paper being exposed; if there is a risk of the test paper being exposed, then based on the test paper offset direction, a cheating suspect object associated with the first object is determined; the angle eigenvalue, the eye-writing eigenvalue, and the test paper result analysis eigenvalue corresponding to the cheating suspect object are determined; based on the angle eigenvalue, the eye-writing eigenvalue, and the test paper result analysis eigenvalue, a comprehensive score is determined, and the cheating suspicion degree of the cheating suspect object is determined according to the comprehensive score. It realizes the use of intelligent video analysis technology, combined with the correlation analysis of the answer sheet results, greatly improves the efficiency and accuracy of cheating analysis, not only reduces the burden of manual review, improves the accuracy of cheating behavior recognition, but also provides objective and sufficient evidence of cheating behavior for educational institutions, facilitating subsequent processing and investigation.

[0017] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present application, nor is it used to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings are used to better understand the solution and do not constitute a limitation to the present disclosure. Among them:

[0019] Figure 1 is a flowchart of a method for analyzing exam cheating provided by an embodiment of the present disclosure;

[0020] Figure 2 is a flowchart of another method for analyzing exam cheating provided by an embodiment of the present disclosure;

[0021] Figure 3 is a flowchart of another method for analyzing exam cheating provided by an embodiment of the present disclosure;

[0022] Figure 4 is a flowchart of another method for analyzing exam cheating provided by an embodiment of the present disclosure;

[0023] Figure 5 A flowchart of another method for analyzing exam cheating provided by an embodiment of the present disclosure;

[0024] Figure 6 A flowchart of another method for analyzing exam cheating provided by an embodiment of the present disclosure;

[0025] Figure 7 A flowchart of another method for analyzing exam cheating provided by an embodiment of the present disclosure;

[0026] Figure 8 A structural diagram of an exam cheating analysis device provided by an embodiment of the present disclosure;

[0027] Figure 9 A hardware structural diagram of an electronic device provided by an embodiment of the present disclosure. Detailed implementation manners

[0028] The following describes exemplary embodiments of the present disclosure with reference to the accompanying drawings. Various details of the embodiments of the present disclosure are included to assist understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted below.

[0029] In modern education, the fairness and effectiveness of exams are of crucial importance. Traditional exam invigilation methods, even with the assistance of video surveillance systems, still have many deficiencies, especially in the post-analysis of cheating behaviors. The video cheating analysis method of related technologies uses a video surveillance system as an auxiliary tool to record and remotely monitor the exam status in real time, enabling invigilators to detect and handle suspicious behaviors in a timely manner, and at the same time saving video recordings for subsequent review and analysis. However, with the development of technology and the continuous innovation of cheating methods, the main deficiencies of the video cheating analysis method of related technologies are:

[0030] 1. Low efficiency of manual review: The massive amount of video data makes the manual review workload huge, time-consuming and error-prone, and it is difficult to effectively detect hidden cheating behaviors.

[0031] 2. Lack of correlation analysis ability: In related technologies, usually only the video recordings can be viewed in isolation, and it is impossible to perform correlation analysis between the video content and the exam answers of candidates. For example, if a candidate frequently looks in a certain direction in the video, but it is difficult to determine whether there is a plagiarism behavior if it cannot be compared with the exam answers of the candidates in that direction.

[0032] 3. Insufficient post - analysis means: The related technologies lack effective tools and methods to comprehensively analyze videos and answer sheets after the exam to generate objective and comprehensive analysis results of cheating behaviors. This makes it difficult for educational institutions to effectively trace and handle cheating behaviors.

[0033] Therefore, the video cheating analysis methods of related technologies cannot fully utilize the role of video surveillance to ensure the fairness of the exam. This poses an integrity crisis for educational institutions and also affects the learning enthusiasm of students and the environment of fair competition.

[0034] To solve the problems in related technologies, the exam cheating analysis method proposed in this disclosure analyzes multiple frames of images in the video stream to accurately determine the offset direction and exposure risk of the test paper, and then locks the suspected cheating objects. At the same time, by combining the angle eigenvalue, the eye - writing eigenvalue, and the test - paper result analysis eigenvalue, it comprehensively evaluates the degree of suspected cheating, achieving efficient and accurate identification of cheating behaviors. This not only improves the intelligent level of exam invigilation but also effectively guarantees the fairness and impartiality of the exam.

[0035] Next, the exam cheating analysis method, device, electronic device, storage medium, and computer program product according to the embodiments of this disclosure will be described with reference to the accompanying drawings.

[0036] Figure 1 It is a flowchart of an exam cheating analysis method provided for the embodiments of this disclosure. As Figure 1 shown, the method includes:

[0037] Step 101: Obtain multiple frames of images in the video stream and determine the offset direction of the test paper of the first object in each frame of the multiple frames of images.

[0038] In some embodiments, this disclosure can capture the video stream through a camera and extract consecutive multiple frames of images from it. Specifically, after obtaining multiple frames of images, this disclosure can use image - processing technologies (such as color segmentation, edge detection, contour recognition, etc.) to accurately extract the test - paper area and the table area in each frame of the image.

[0039] After determining the test - paper area and the table area, this disclosure can also determine the offset direction of the test paper relative to the table by analyzing the coordinates of the test - paper area and the table area. Specifically, it is to compare the relative positions of the center points or specific points of the two groups of coordinates to determine whether the test paper is offset to the left, right, up, or down relative to the table.

[0040] Specifically, the present disclosure can calculate the distance between the vertex coordinates of the test paper area and the center coordinates of the edge of the table area, and this distance reflects the deviation between the test paper and the preset position of the table. Then, by comparing this distance with a preset safety threshold, an offset result can be obtained, that is, it can be determined whether the test paper is offset and the current offset direction. Finally, based on this offset result, it can be determined in which direction (such as upper left, lower right, etc.) the test paper is specifically offset.

[0041] Step 102: Based on the test paper coordinates and the first object's hand coordinates in each frame of the image, determine whether there is a risk of exposure of the test paper.

[0042] In some embodiments, when processing each frame of the video stream, the present disclosure can also determine whether there is a risk that the test paper is exposed by the hand of the first object (such as a candidate).

[0043] Specifically, the present disclosure can first determine the overlapping area between the two and its overlapping vertex coordinates according to the vertex coordinates of the test paper area and the vertex coordinates of the hand area. If the vertex coordinates of the overlapping area meet the preset overlapping conditions, it means that there is a possibility of contact or near contact between the test paper and the hand. Next, the present disclosure can calculate the ratio of the area of the overlapping area to the total area of the test paper. If this ratio is lower than the preset ratio threshold, it indicates that although there is an overlap, the area is not large. At this time, the present disclosure can determine that there is a preliminary risk of exposure of the test paper.

[0044] To more accurately evaluate the risk, the present disclosure can also directly calculate the overlapping ratio of the test paper according to the hand vertex coordinates and the test paper vertex coordinates. If the overlapping ratio reaches or exceeds the preset exposure threshold, the present disclosure can determine that there is indeed a risk of exposure of the test paper.

[0045] When it is determined that there is indeed a risk of exposure of the test paper, the present disclosure can also give a real-time exposure risk prompt to prompt the invigilator to take corresponding measures to prevent the leakage of the test paper or improper behavior, or can directly give an intelligent voice reminder to prompt the first object to adjust the posture to avoid the leakage of the test paper.

[0046] Step 103: If there is a risk of exposure of the test paper, determine the cheating suspect object associated with the first object based on the offset direction of the test paper.

[0047] In some embodiments, in the monitoring scenario, after the present disclosure determines that there is a risk of exposure of the test paper, the present disclosure can identify the potential line-of-sight intersection area or the path where the test paper may be spied on according to the offset direction of the test paper. Next, it will focus on the specific population located in this direction or path, that is, the associated population. The associated population can be candidates in the examination room who are relatively close to the first object. Due to their geographical location, they may see the test paper content that is partially or completely exposed due to the offset.

[0048] In order to screen out cheating suspect objects from the target analysis population, the present disclosure can first screen out the target analysis population in the associated population. The target analysis population refers to the population that can see the test paper of the first object. Then, observe and analyze the head postures of each target analysis object, such as the orientation of their heads, the specific direction of their gazes, and actions such as frequent head rotations. These behavioral characteristics can reveal the obvious attention or abnormal behaviors of the target analysis objects towards the test paper. In particular, if the head posture of a certain target analysis object significantly points towards the direction of the test paper, or their gazes stay in the exposed area of the test paper for a long time, then this target analysis object may be regarded as a cheating suspect object, and a cheating suspect object is an object with potential cheating behavior.

[0049] In addition, after determining the cheating suspect object, the present disclosure can also send a cheating suspicion notice to the invigilator in the examination room of the current cheating suspect object through remote control, so as to quickly remind the invigilator to focus on this cheating suspect object for more detailed observation and necessary intervention measures. In this way, not only can the invigilation efficiency be effectively improved, but also the fairness and impartiality of the examination can be maintained to the greatest extent, ensuring that each candidate can compete in a fair environment and show their true level.

[0050] Step 104: Determine the angle eigenvalue, eye writing eigenvalue, and test paper result analysis eigenvalue corresponding to the cheating suspect object.

[0051] In some embodiments, after determining the cheating suspect object, the present disclosure can, in combination with post-analysis or real-time monitoring analysis, determine potential cheating behaviors of the cheating suspect object in the examination environment. By analyzing multiple eigenvalues to determine whether the cheating suspect object has cheating behaviors. These eigenvalues mainly include the angle eigenvalue, the eye writing eigenvalue, and the test paper result analysis eigenvalue.

[0052] The angle eigenvalue is used to evaluate whether there are abnormal behaviors of the cheating suspect object in terms of the head and seat angles. The present disclosure can first collect the head and seat angle data of the first object (such as a reference student) and the cheating suspect object. This includes the horizontal deflection angle of the first object's head, the vertical pitch angle of the first object's head, as well as the seat included angle between the first object and the cheating suspect object and the seat pitch angle between the head and the cheating suspect object's desktop. Based on these data, calculate the horizontal angle score and the vertical angle score, and then combine the preset angle weights to determine the total angle score. By considering the number of frames of the video stream, the present disclosure can further calculate the average angle score and the angle standard deviation. Finally, compare the angle standard deviation and the average angle score with the preset threshold to determine the angle eigenvalue.

[0053] The eye-writing feature value is used to evaluate whether a suspected cheating object conveys information through eye or writing behavior, so as to determine whether there is cheating behavior. The present disclosure can monitor the residence time of the suspected cheating object in the writing area during a preset monitoring time period, including the gaze residence time and the nib residence time. By calculating the ratio between the residence time in the writing area and the preset monitoring time period, the residence time ratio in the writing area is obtained. This ratio is compared with a preset residence time ratio threshold to determine the eye-writing feature value.

[0054] The test paper result analysis feature value is used to be associated with the answer sheet of the exam, and evaluate the abnormal behavior of the suspected cheating object in terms of the exam results, so as to determine whether there is cheating behavior. The present disclosure can obtain the answering data of the first object and the suspected cheating object, including the answering answers and the question types of the answers. By comparing the answers of the two in each question type, the question type similarity is calculated. At the same time, based on the number of students and the scores of each student in each question type, the score variance corresponding to each question type is further determined, and the minimum value of the score variance is found as the exam difficulty coefficient. Combining the question type similarity and the exam difficulty coefficient, the cheating suspicion score corresponding to each question type is calculated. Finally, according to the preset score weight, the test paper result analysis feature value is obtained.

[0055] By comprehensively analyzing the angle feature value, the eye-writing feature value, and the test paper result analysis feature value, the present disclosure can more accurately determine whether a suspected cheating object has cheating behavior, providing strong technical support for exam invigilation.

[0056] Step 105: Based on the angle feature value, the eye-writing feature value, and the test paper result analysis feature value, determine a comprehensive score, and determine the degree of cheating suspicion of the suspected cheating object according to the comprehensive score.

[0057] In some embodiments, after detecting that a student may have cheating behavior (i.e., determining the suspected cheating object), the present disclosure can introduce a confidence score module to quantify the cheating risk of each student. This module will combine different weights and intervals based on the calculation results of multiple feature values, and finally obtain a comprehensive score to judge the degree of cheating suspicion of the student inferred above.

[0058] The present disclosure can set a feature weight for each feature value (angle feature value, eye-writing feature value, test paper result analysis feature value). These feature weights can be set based on the importance and reliability of each feature value in judging cheating behavior, and the specific values are not limited in the present disclosure. The present disclosure calculates the comprehensive score according to these feature weights and the corresponding feature values. The comprehensive score is a comprehensive index, which reflects the degree of abnormal behavior of the suspected cheating object in multiple aspects. The higher the comprehensive score, the higher the degree of cheating suspicion of the suspected cheating object.

[0059] Specifically, based on the angle eigenvalue, the eye-writing eigenvalue, and the test paper result analysis eigenvalue, a comprehensive score is determined, and the degree of cheating suspicion of the cheating suspect is determined according to the comprehensive score, including: determining the comprehensive score based on the angle eigenvalue, the eye-writing eigenvalue, the test paper result analysis eigenvalue, and the preset eigenvalue weights; if the comprehensive score is greater than or equal to the preset comprehensive score threshold, determining that the degree of cheating suspicion of the cheating suspect is high cheating suspicion; if the comprehensive score is less than the preset comprehensive score threshold, determining that the degree of cheating suspicion of the cheating suspect is low cheating suspicion.

[0060] In an alternative embodiment of the present disclosure, the comprehensive score can be calculated by the following formula:

[0061]

[0062] where is the angle eigenvalue, is the eye-writing eigenvalue, is the test paper result analysis eigenvalue, is the eigenvalue weight corresponding to the angle eigenvalue, is the eigenvalue weight corresponding to the eye-writing eigenvalue, is the eigenvalue weight corresponding to the test paper result analysis eigenvalue. If the comprehensive score ≥ the preset comprehensive score threshold , it indicates that the degree of cheating suspicion is relatively high, which is high cheating suspicion, and special attention needs to be paid to this cheating suspect; if the comprehensive score <the preset comprehensive score threshold , it indicates that the degree of cheating suspicion is relatively low, which is low cheating suspicion, and there is no need to pay special attention to this cheating suspect.

[0063] At the same time, in addition to determining the degree of cheating suspicion, the present disclosure can also generate a detailed cheating analysis report based on the obtained degree of cheating suspicion. The cheating analysis report can include each eigenvalue (angle eigenvalue, eye-writing eigenvalue, test paper result analysis eigenvalue) of the cheating suspect, the comprehensive score, and the final judgment result of the degree of cheating suspicion. The existence of the analysis report helps the invigilator quickly understand the situation of the cheating suspect and can also provide a strong basis for subsequent further investigation or handling.

[0064] It should be noted that the setting of the eigenvalue weights, the calculation method of the comprehensive score, and the determination of the comprehensive score threshold in the present disclosure can all be adjusted and optimized according to the actual situation.

[0065] In summary, the technical solution provided by the present disclosure obtains multiple frames of images in the video stream, and determines the paper offset direction of the first object in each frame of the multiple frames of images; based on the paper coordinates and the hand coordinates of the first object in each frame of the image, it determines whether there is a risk of the paper being exposed; if there is a risk of the paper being exposed, it determines the cheating suspect object associated with the first object based on the paper offset direction; determines the angle eigenvalue, the eye writing eigenvalue, and the paper result analysis eigenvalue corresponding to the cheating suspect object; based on the angle eigenvalue, the eye writing eigenvalue, and the paper result analysis eigenvalue, determines a comprehensive score, and determines the degree of cheating suspicion of the cheating suspect object according to the comprehensive score. It realizes the use of intelligent video analysis technology, combined with the correlation analysis of the answer sheet results, greatly improves the efficiency and accuracy of cheating analysis, not only reduces the burden of manual review, improves the accuracy of cheating behavior recognition, but also provides objective and sufficient evidence of cheating behavior for educational institutions, facilitating subsequent processing and investigation.

[0066] As a possible implementation, as Figure 2 shown in the flowchart of another method for analyzing exam cheating. On the basis of the above embodiments, the specific process of obtaining multiple frames of images in the video stream and determining the paper offset direction of the first object in each frame of the multiple frames of images includes the following steps:

[0067] Step 201: Perform color segmentation and edge detection on each frame of the image respectively, and extract the paper area and the table area in each frame of the image.

[0068] In some embodiments, when processing video frames or images, in order to extract the paper area and the table area, color segmentation and edge detection techniques can be applied respectively. For color segmentation of each frame of the image, first convert the image to the HSV color space, set the HSV threshold to identify and extract the color areas of the paper and the table, and then perform morphological operations to denoise and fill holes; subsequently, perform edge detection, use the Canny algorithm to detect the image edges, and extract the contours through the Suzuki-Abe algorithm, and finally accurately identify and extract the areas of the paper and the table.

[0069] The purpose of color segmentation is to identify and extract the color areas of the paper and the table from the image. The purpose of edge detection is to extract the edge contours of the paper and the table from the image.

[0070] In an alternative embodiment of the present disclosure, the present disclosure can first convert the color space of the image to the HSV color space. That is, use the cv2.cvtColor function of OpenCV to convert the image from the BGR color space to the HSV color space. The HSV color space is more suitable for color threshold processing because it separates the hue, saturation, and value, making color recognition more accurate.

[0071] Set the corresponding HSV thresholds (H ∈ [H_min, H_max], S ∈ [S_min, S_max], V ∈ [V_min, V_max]) according to the colors of the test paper and the table, so as to accurately identify the colors of the test paper and the table. Use the cv2.inRange function in OpenCV to create a binary mask according to the set HSV thresholds, so as to extract the regions within the specified color range. For example, Mask(x, y) = {1 if H(x, y) ∈ [H_min, H_max] and S(x, y) ∈ [S_min, S_max] and V(x, y) ∈ [V_min, V_max], otherwise 0}.

[0072] For the regions within the specified color range, use morphological operations (such as opening and closing operations) to remove noise and fill holes to obtain a cleaner contour. For example, FinalImage = Closing(Opening(I)), where the opening operation Opening(I) = (I ⊖ B) ⊕ B, and the closing operation Closing(I) = (I ⊕ B) ⊖ B. At the same time, use the binary mask to perform a bitwise AND operation with the original image to extract the color regions of the test paper and the table.

[0073] After obtaining the color regions of the test paper and the table, the cv2.cvtColor function in OpenCV can be used to convert the color image to a grayscale image first. Then use the cv2.GaussianBlur function in OpenCV to perform Gaussian blur processing on the grayscale image to reduce noise. Use the cv2.Canny function in OpenCV for Canny edge detection. And use the Suzuki-Abe algorithm (implemented by the cv2.findContours function in OpenCV) to extract the contours from the edge image, so as to accurately extract the test paper region and the table region.

[0074] Among them, before the present disclosure performs color segmentation and edge detection on each frame of image, the present disclosure needs to obtain multiple frames of images in the video stream according to the video, specifically including: reading the examination room monitoring video information and transmitting it to the background; using ffmpeg to execute the frame extraction command to obtain static pictures per second from the video stream; classifying and labeling the extracted pictures, and the pictures contain meta-information such as timestamps and student data. After that, use openpose to extract the pixel coordinates in the pictures, specifically including: setting the parameter configuration of OpenPose; specifying the called model, and selecting FACE_MODEL / BODY_25 / HAND_MODEL as the model for facial key point recognition; initializing the OpenPose instance, configuring and starting facial detection; loading the image to be processed; calling OpenPose to process the image to identify facial / body key points / hand key points; outputting the detected facial / body key points / hand key point data, and displaying the detection result image for subsequent cheating analysis.

[0075] Step 202: Determine the distance between the paper vertex coordinates of the paper area and the edge center coordinates of the table area.

[0076] In some embodiments, the present disclosure can obtain the paper vertex coordinates A(x1, y1), B(x2, y2), C(x3, y3), D(x4, y4) of the four corners of the paper and the edge center coordinates X(x5, y5), Y(x6, y6), Z(x7, y7), T(x8, y8) of the table area.

[0077] Calculate the distance distanceA between point A and the center X of the upper side of the table:

[0078]

[0079] Calculate the distance distanceC between point C and the center Y of the lower side of the table:

[0080]

[0081] Calculate the distance distanceB between point B and the center Z of the left side of the table:

[0082]

[0083] Calculate the distance distanceD between point D and the center T of the right side of the table:

[0084]

[0085] Through the above formula, the present disclosure can obtain the distances between each vertex of the test paper and the center of the corresponding edge of the table. These distances can reflect the positional relationship of the test paper on the table.

[0086] Step 203: Determine the test paper offset result based on the distances and a preset safety threshold.

[0087] In some embodiments, the present disclosure can set safety thresholds in four directions: Threshold_A, Threshold_C, Threshold_B, and Threshold_D. These safety thresholds can be adjusted according to the actual application scenario and the size ratio of the test paper and the table. For example, it can be set to 5% or 10% of the width of the test paper.

[0088] Specifically, for the vertical offset judgment:

[0089] If distanceA > Threshold_A and distanceA > distanceC + Offset_Vertical, it is considered that the test paper is offset upward, where Offset_Vertical is the allowable vertical offset tolerance.

[0090] If distanceC > Threshold_C and distanceC > distanceA + Offset_Vertical, it is considered that the test paper is offset downward.

[0091] For the horizontal offset judgment:

[0092] If distanceB > Threshold_B and distanceB > distanceD + Offset_Horizontal, it is considered that the test paper is offset to the left, where Offset_Horizontal is the allowable horizontal offset tolerance.

[0093] If distanceD > Threshold_D and distanceD > distanceB + Offset_Horizontal, it is considered that the test paper is offset to the right.

[0094] Through the above direction judgment, the present disclosure can determine whether the test paper offset result is a vertical offset or a horizontal offset, and whether the current offset direction of the test paper is to the left or the right.

[0095] Step 204: Determine the test paper offset direction based on the test paper offset result.

[0096] In some embodiments, after obtaining the paper offset result, the present disclosure can combine the offset results in the vertical and horizontal directions to comprehensively determine the offset direction of the paper. For example, if the conditions of "biased upward" and "biased leftward" are both met, it is considered that the offset direction of the paper is biased to the upper left.

[0097] Meanwhile, the present disclosure can also determine whether the paper undergoes continuous offset. Specifically, the present disclosure can record the timestamp of each detection. Set a time window t1 (e.g., 3 seconds). If the same-direction offset is detected continuously m1 times (e.g., 2 times) within this time window, it is considered that the paper continuously offsets in this direction.

[0098] For example, use a queue to store the offset directions of the most recent m1 times. Each time an offset is detected, add the offset direction to the queue. If all elements in the queue are the same, it is considered that the paper has undergone continuous offset.

[0099] In summary, the present disclosure performs color segmentation and edge detection on each frame of the image, accurately extracts the paper and table areas, calculates the distances between the paper vertices and the center of the table edge, combines the preset safety threshold to judge the paper offset situation, and then determines the offset direction, which can effectively monitor the position state of the paper on the table and improve the accuracy of cheating analysis.

[0100] As a possible implementation, as Figure 3 shown in the flowchart of another exam cheating analysis method, based on the paper coordinates and the hand coordinates of the first object in each frame of the image, the specific process of determining whether there is a risk of paper exposure includes the following steps:

[0101] Step 301: Based on the paper vertex coordinates of the paper area and the hand vertex coordinates of the hand area in each frame of the image, determine the overlapping vertex coordinates of the overlapping area between the paper area and the hand area.

[0102] In some embodiments, based on the vertex coordinates of the paper area and the hand vertex coordinates of the hand area in each frame of the image, the present disclosure can determine the overlapping part between these two areas and its overlapping vertex coordinates.

[0103] In an alternative embodiment of the present disclosure, assume that the paper vertex coordinates of the paper area are [(x1, y1), (x2, y2)], where (x1, y1) is the upper left corner and (x2, y2) is the lower right corner. At the same time, the hand vertex coordinates of the hand area are [(x9, y9), (x10, y10)], where (x9, y9) is the upper left corner and (x10, y10) is the lower right corner.

[0104] To determine whether there is a risk of paper exposure, that is, whether the hand obscures or touches the paper, the present disclosure needs to calculate the overlapping area.

[0105] The upper left coordinates of the overlapping area are composed of the larger x value and the smaller y value of the upper left coordinates of the two areas, that is, overlap_x1 = max(x1, x9) and overlap_y1 = min(y1, y9). Similarly, the lower right coordinates of the overlapping area are composed of the smaller x value and the larger y value of the lower right coordinates of the two areas, that is, overlap_x2 = min(x2, x10) and overlap_y2 = max(y2, y10).

[0106] Step 302: If the overlapping vertex coordinates meet the preset overlapping conditions, then based on the overlapping vertex coordinates and the paper vertex coordinates, determine the overlapping area of the overlapping area and the paper area of the paper area.

[0107] In some embodiments, when the present disclosure obtains the correct overlapping area coordinates [(overlap_x1, overlap_y1), (overlap_x2, overlap_y2)], the present disclosure can determine whether there is an overlap. That is, calculate the overlapping area and the paper area according to the vertex coordinates of the overlapping area, and further determine whether the preset overlapping conditions are met and calculate the proportion of the overlapping area in the paper area.

[0108] If overlap_x1 < overlap_x2 and overlap_y1 > overlap_y2, it is determined that the overlapping vertex coordinates meet the preset overlapping conditions, that is, the two rectangles overlap, otherwise there is no overlap.

[0109] If there is an overlap, the overlapping area overlap_area can be calculated by the following formula:

[0110]

[0111] where (overlap_x1, overlap_y1) are the upper left coordinates of the overlapping area, and (overlap_x2, overlap_y2) are the lower right coordinates of the overlapping area.

[0112] The paper area can be calculated by the following formula:

[0113]

[0114] where (x1, y1) are the upper left coordinates of the paper area, and (x2, y2) are the lower right coordinates of the paper area.

[0115] Step 303: If the ratio between the overlapping area and the paper area is less than the preset proportion threshold, it is determined that the paper has a preliminary exposure risk.

[0116] In some embodiments, the present disclosure may calculate the proportion of the overlapping area in the area of the test paper to determine whether there is a preliminary exposure risk for the test paper.

[0117] That is, the ratio shadow between the overlapping area and the area of the test paper can be calculated by the following formula:

[0118]

[0119] where overlap_area is the overlapping area and paper_area is the area of the test paper.

[0120] In the present disclosure, the preset proportion threshold can be adjusted according to the actual situation and is not limited in the embodiments of the present disclosure. Taking 30% as an example in the present disclosure, if the overlapping area (i.e., the occluded area) shadow < 30%, it is determined that the test paper may have an exposure risk, that is, there is a preliminary exposure risk.

[0121] Step 304: Determine the overlapping ratio based on the hand vertex coordinates and the test paper vertex coordinates.

[0122] In some embodiments, the present disclosure further determines the overlapping ratio to accurately determine whether there is an exposure risk.

[0123] The present disclosure may first calculate the distance between the hand and the edge of the test paper: Assuming that the left edge of the test paper is the exposure boundary, the present disclosure can calculate the distance between the left edge (x9) of the hand and the left edge (x1) of the test paper, that is, risk_distance = (x9 - x1). It should be noted that here it is assumed that the hand is on the left side of the test paper. If the hand is on the right side of the test paper, the calculation method needs to be adjusted accordingly.

[0124] The overlapping ratio represents the position of the hand relative to the width of the test paper, that is, whether the hand covers a part of the test paper. Since the width of the hand can be calculated by x10 - x9.

[0125] Therefore, the overlapping ratio can be expressed as risk_proportion = .

[0126] Step 305: If the overlapping ratio is greater than or equal to the preset overlapping ratio, it is determined that the test paper has an exposure risk.

[0127] In some embodiments, the preset overlapping ratio is a preset value used to determine whether the hand covers a sufficient area of the test paper to constitute an exposure risk. Taking the overlapping threshold as 50% in the present disclosure, that is, the hand covers half or more of the width of the test paper.

[0128] If the overlapping ratio is greater than or equal to 50%, it is determined that the test paper has an exposure risk.

[0129] Among them, in order to more accurately determine whether the test paper may be exposed, the present disclosure can combine the timestamp of the image and set a time window of t2 seconds. Within this time window, if the exposure risk is detected multiple times, it can be determined that the test paper is indeed exposed. For example, if the overlapping ratio is greater than or equal to 50%, an exposure risk event is recorded. Within the set time window, the number of exposure risk events is counted. If the number of exposure risk events exceeds a certain threshold (for example, m2 exposure risks are detected within t2 seconds), it is determined that the test paper is exposed.

[0130] In summary, based on the vertex coordinates of the test paper and the hand in each frame of the image, the present disclosure can accurately determine whether there is an exposure risk of the test paper, effectively monitor and warn of the exposure of the test paper, and ensure the fairness and security of the examination or assessment process.

[0131] As a possible implementation, as Figure 4 shown in the flowchart of another method for analyzing exam cheating, on the basis of the above embodiments, the specific process of determining the cheating suspect object associated with the first object based on the offset direction of the test paper includes the following steps:

[0132] Step 401, determine the target analysis population corresponding to the offset direction of the test paper.

[0133] In some embodiments, determining the target analysis population corresponding to the offset direction of the test paper includes: determining the associated population corresponding to the first object; according to the offset direction of the test paper and the position coordinates of each object in the associated population, determining the downward viewing angle of each object in the associated population looking at the first object; according to the downward viewing angle, determining the target analysis population in the associated population. The downward viewing angle is intended to simulate the perspective of other people in the examination room when observing the first object, so as to evaluate whether they can easily detect the offset of the test paper.

[0134] In an alternative embodiment of the present disclosure, in order to more precisely delimit the target analysis population at risk of cheating, the present disclosure may employ statistical methods such as the K-means clustering algorithm to first determine the associated population close to the position of the first object. This method can automatically divide students into several clusters according to the seat distribution in the examination room, and the positions of students within each cluster are relatively close. That is, the present disclosure first sets the number of clusters K (i.e., the number of clusters), and randomly selects K initial centroids (i.e., the center points of the clusters). Then, it calculates the distance between each student and each centroid, and assigns the student to the cluster where the centroid closest to it is located. Next, it recalculates the new centroid of each cluster based on the positions of the students within the cluster, and repeats the above assignment process until a convergence state is reached (i.e., the assignment of clusters and the positions of the centroids no longer change). After the clustering result converges, the present disclosure can obtain the cluster where the first object (such as candidate A) is located. Since the positions of the students within this cluster are relatively close, the downward viewing angles when they look at the first object may be smaller, and it is easier to notice the offset of the test paper. Therefore, we can regard the students within this cluster as the associated population that can see the test paper of the first object.

[0135] Among them, factors such as the selection of the K value and the selection of the initial centroid in the K-means clustering algorithm may affect the clustering result. In the embodiments of the present disclosure, no specific numerical values are restricted, and the actual situation shall prevail.

[0136] After obtaining the associated population, the present disclosure can further comprehensively evaluate whether other students in the examination room can see the offset of the test paper of the first object by combining the horizontal angle and the vertical angle. That is, the present disclosure can first define a reference vector, and based on the position coordinates of each object in the associated population pair, calculate the relative vector of each object in the associated population relative to the first object. Then, using the dot product formula and the inverse cosine function, it calculates the horizontal angle between these relative vectors and the reference vector. To evaluate the visibility in the vertical direction, the present disclosure can also calculate the downward viewing angle when each object in the associated population looks at the first object according to the height of the first object (obtained from the photo through proportional calculation) and the actual distance between the first object and each object in the associated population (also obtained through proportional calculation). Finally, according to the set downward viewing angle threshold, it can screen out those objects whose downward viewing angles are less than or equal to the preset threshold, and regard them as the target analysis population that can see the test paper of the first object. This method combines the calculation of the horizontal angle and the vertical angle, and can more precisely determine which students are most likely to observe and thus may be involved in cheating in the case of test paper offset.

[0137] Specifically, given the position coordinates of each student in the examination room and the position coordinates of the first object, the students are divided into k clusters by the k-means algorithm.

[0138] Suppose there are four students: (a1, b1), (a2, b2), (a3, b3), (a4, b4).

[0139] Randomly select k initial centroids: (a1, b1), (a2, b2).

[0140] Calculate the distance between the students and the centroids, and assign the students to the cluster closest to them. The calculation formula is as follows:

[0141]

[0142] Using the above formula, calculate the distance from each student to the two centroids, and assign each student to the cluster where the closest centroid is located.

[0143] By calculating the distance from each student to the centroid, the initial assignment result of the students can be obtained as follows:

[0144] Cluster 1 = (a1, b1), (a2, b2)

[0145] Cluster 2 = (a3, b3), (a4, b4).

[0146] Calculate the new centroid of each cluster: New centroid of Cluster 1 = ( , ); New centroid of Cluster 2 = ( , )

[0147] Repeat the calculation until convergence. Recalculate the distance from each student to the new centroid and reassign the students. After multiple iterations, the cluster assignment and the centroid no longer change, reaching a convergence state. Final clustering result: Cluster 1 = (a1, b1), (a2, b2); Cluster 2 = (a3, b3), (a4, b4). Obtain the cluster where the first object is located as the associated population.

[0148] After that, assume that the paper offset direction is to the right, and determine which objects in the associated population can see the paper of the first object.

[0149] Specifically, define the reference vector of the first object (Student A):

[0150] Suppose the coordinates of Student A are (a1, b1), and select the horizontal right direction (i.e., 0 degrees) as the reference vector, defined as V0 = (1, 0).

[0151] Calculate the vector of Student B (a certain object in the associated population) relative to Student B:

[0152] Assume the coordinates of student B are (a2, b2), and calculate the vector Vab from student A to student B as Vab = (a1 - a2, b1 - b2). Then use the dot product formula and the arccosine function to calculate the horizontal angle.

[0153] For example, use the dot product formula to calculate the angle between V0 and Vab : .

[0154] Then find the angle through the arccosine function : .

[0155] After that, calculate the student's height. Given the actual height of the table , the height of the table in the photo , and the height of the student in the photo (the vertical distance from the table surface to the student's head) .

[0156] Given the geometric ratio calculation formula .

[0157] According to the geometric ratio calculation formula, the height of student A can be calculated .

[0158] Similarly, the distance between student A and student B can be obtained through geometric ratio calculation.

[0159] Finally, through the distance between student A and student B and the height of student A , the downward viewing angle of student B looking at student A can be calculated , that is .

[0160] Step 402: Use the head postures of each target analysis object in the target analysis population to determine the cheating suspect objects in the target analysis population.

[0161] In some embodiments, using the head postures of each target analysis object in the target analysis population to determine the cheating suspect objects in the target analysis population includes: determining the head offset angle of each target analysis object according to the key point coordinates of each target analysis object; determining the relative angle between the first object and each target analysis object according to the center point coordinates of the test paper of each target analysis object and the position coordinates of the first object; determining the angle ratio between the head offset angle and the relative angle of each target analysis object, and based on the angle ratio, determining the cheating suspect objects in the target analysis population.

[0162] Specifically, the present disclosure can determine whether there is a cheating behavior by the viewing direction of the head posture of the target analysis object.

[0163] Among them, determining the head pose of the target analysis object specifically includes: using the face model of OpenPose to identify the facial key points of the target analysis object, and these facial key points will be used for subsequent pose estimation. To perform head pose estimation, the present disclosure can define a standard three-dimensional face model.

[0164] Define a standard three-dimensional face model, including the three-dimensional coordinates of feature points such as the tip of the nose (0, 0, 0), the outer corner of the left eye (-x, y, z), the outer corner of the right eye (x, y, z), the left corner of the mouth (-x', -y', z'), the right corner of the mouth (x', -y', z'), and the chin (0, -h, -d) (equivalent to an ideal facial model). These coordinates are usually provided by actual measurement or a standard template.

[0165] In an actual scenario, the mapping relationship between the coordinates in the two-dimensional image and the three-dimensional world coordinates depends on the internal parameters of the camera. The internal parameters of the camera include: the focal length and , usually based on the actual settings of the camera or the image resolution; the optical center , representing the center point position of the image.

[0166] Using the pinhole camera model, define the internal parameter matrix of the camera (This matrix is used to project the three-dimensional coordinates onto the two-dimensional image plane.): .

[0167] After that, use the PnP algorithm to calculate the rotation matrix and displacement matrix of the head according to the positions of the feature points in the two-dimensional image and the three-dimensional model points. Through the solvePnP function of OpenCV, input the known 3D model points, the corresponding 2D image points, the camera internal parameter matrix, and the distortion coefficients to obtain the rotation vector and displacement vector. Among them, PnP (Perspective-n-Point) is an algorithm for estimating the pose of a calibrated camera, given a set of 3D points and their 2D projections in the image. The goal of this algorithm is to calculate the rotation and translation of the camera relative to the world coordinate system, with six degrees of freedom (6 DOF), namely the rotation (roll, pitch, and yaw) and three-dimensional translation of the camera.

[0168] Through steps such as the conversion from the rotation vector to the rotation matrix and the conversion from the rotation matrix to the quaternion, obtain the pitch angle, yaw angle, and roll angle of the head. The specific method is as follows:

[0169] First, use the Rodrigues formula to convert the rotation vector r into a rotation matrix. The rotation vector is usually represented as a three-dimensional vector, indicating the angle of rotation around a certain unit vector.

[0170] The Rodrigues formula is as follows:

[0171] , where: R is the rotation matrix, is the rotation angle; is the cross product matrix of the rotation axis; M is the identity matrix.

[0172] Quaternion is a mathematical representation method that extends complex numbers. It is a four-dimensional number composed of four real numbers. Quaternions are mainly used in computer graphics, robotics, and physics. Especially in the representation and calculation of 3D rotations, quaternions have higher stability and efficiency than Euler angles and rotation matrices.

[0173] Conversion from the rotation matrix R to quaternion: q = (w, x, y, z).

[0174] Conversion from quaternion to rotation vector r: , where w is the scalar part of the quaternion, and (x, y, z) is the vector part of the quaternion. Among them, w, x, y, z are the components of the quaternion. The direction of the rotation vector r is given by the x, y, z components in the quaternion, and the rotation angle is determined by w.

[0175] In practical applications, such as in head motion analysis, the present disclosure can obtain the orientation of the head through the rotation matrix or quaternion. For example, the offsets of the head (such as up and down, left and right, rotation) can be described by calculating the rotation matrix or quaternion.

[0176] Pitch: Corresponding to the rotation around the horizontal axis (for example, the tilt of the head up and down). It can be calculated through the components of the quaternion or the rotation vector.

[0177] Yaw: Corresponding to the rotation around the vertical axis (for example, the rotation of the head left and right). It can also be calculated through the rotation vector or quaternion.

[0178] Roll: Corresponding to the rotation around the front and back axis (for example, the roll of the head). The calculation method is similar to that of the pitch angle.

[0179] Finally, the following values are obtained:

[0180] Pitch angle: [-90°, 90°]

[0181] Yaw angle: [-180°, 180°]

[0182] Roll angle: [-180°, 180°]

[0183] After that, the present disclosure can determine the relative angle between the target analysis object and the first object according to the seat distribution in the examination room:

[0184] Assume that the coordinates of the target analysis object are (a2, b2), and the coordinates of the center point of the test paper of Student B (i.e., the first object) are (a1, b1).

[0185] Use the cosine function to calculate the relative angle between two points :

[0186]

[0187] According to the yaw angle of the head of the target analysis object (indicating the rotation of the head left and right) and the relative angle , calculate the angle ratio .

[0188] The present disclosure can set a reasonable preset ratio threshold according to the actual situation to determine whether the target analysis object may be peeking at the test paper.

[0189] If the angle ratio is greater than or equal to the preset ratio threshold (or the direction of Yaw is consistent with and the difference is within a reasonable range), it is considered that the current target analysis object has a suspicion of cheating, and the current target analysis object is determined as a suspected cheating object.

[0190] At the same time, the present disclosure can also detect multiple cheating behaviors within a time window: combine the time stamp of the image and set a time window of n seconds. Within the time window, check whether the target analysis object is detected as peeking multiple times (that is, determine that the number of times the angle ratio of the target analysis object is greater than or equal to the preset ratio threshold within the time window is greater than or equal to the preset number of times). If detected multiple times, it is determined that the target analysis object may be cheating, and the target analysis object is a suspected cheating object.

[0191] In summary, the present disclosure can effectively screen out objects with suspected cheating by accurately identifying the target analysis population pointed to by the offset direction of the test paper and further using the head pose information of these students. It not only improves the pertinence of cheating detection but also ensures the accuracy and efficiency of detection.

[0192] As a possible implementation manner, as shown in Figure 5 the flowchart of another method for analyzing exam cheating, on the basis of the above embodiments, the specific process of determining the angle eigenvalue corresponding to the suspected cheating object includes the following steps:

[0193] Step 501: Determine the horizontal deflection angle of the head, the vertical pitch angle of the head of the first object, the seat angle between the first object and the suspected cheating object, and the seat pitch angle between the head of the first object and the desktop of the suspected cheating object.

[0194] In some embodiments, according to the above determination of the head offset angle and position coordinates, determine the horizontal deflection angle of the head of the first object and the vertical pitch angle of the head , the seat included angle between the first object and the object suspected of cheating and the seat pitch angle between the head of the first object and the desktop of the object suspected of cheating .

[0195] Step 502: Determine the horizontal angle score based on the seat included angle between the first object and the object suspected of cheating and the horizontal deflection angle of the head of the first object.

[0196] In some embodiments, based on the seat included angle between the first object and the object suspected of cheating and the horizontal deflection angle of the head of the first object, calculate the absolute difference of the horizontal deflection angle .

[0197] After that, through the following formula, calculate the horizontal deflection angle score, that is, the horizontal angle score:

[0198]

[0199] represents the horizontal angle tolerance, for example, 3°; represents the maximum horizontal angle difference, for example, 30°, for normalization.

[0200] Step 503: Determine the vertical angle score based on the vertical pitch angle of the head of the first object and the seat pitch angle between the head of the first object and the desktop of the object suspected of cheating.

[0201] In some embodiments, based on the vertical pitch angle of the head of the first object and the seat pitch angle between the head of the first object and the desktop of the object suspected of cheating, calculate the absolute difference of the vertical pitch angle .

[0202] After that, through the following formula, calculate the vertical pitch angle score, that is, the vertical angle score:

[0203]

[0204] represents the vertical angle tolerance; represents the maximum vertical angle difference, for normalization.

[0205] Step 504: Determine the total angle score based on the horizontal angle score, the vertical angle score and the preset angle weight.

[0206] In some embodiments, the present disclosure can calculate the total angle score of each frame of image based on the horizontal angle score, the vertical angle score and the preset angle weight of each frame of image by applying the following formula :

[0207]

[0208] Among them, is the horizontal angle score of each frame of image, is the preset angle weight corresponding to the horizontal angle score, is the vertical angle score of each frame of image, is the preset angle weight corresponding to the vertical angle score, , and the specific value of the weight can be set according to the scene requirements. For example, = 0.7, = 0.3.

[0209] Step 505: Determine the average angle score based on the total angle score and the number of frames corresponding to the video stream.

[0210] In some embodiments, the present disclosure can first determine the number of frames corresponding to multiple frames of images in the video stream, and then use the total angle score of each frame of image and the number of frames to calculate the average angle score of the current multiple frames of images , as shown in the following formula.

[0211]

[0212] Among them, N is the number of frames corresponding to multiple frames of images in the video stream (i.e., the total number of frames of the video segment in the current time period), is the time point corresponding to the th frame of image, is the total angle score corresponding to the i-th frame of image.

[0213] Step 506: Determine the angle standard deviation based on the total angle score and the average angle score.

[0214] In some embodiments, after obtaining the average angle score of the current multiple frames of images, the present disclosure can use the total angle score of each frame of image and the average angle score of the multiple frames of images to calculate the angle standard deviation of the current multiple frames of images (i.e., the current time period) . As shown in the following formula:

[0215]

[0216] Among them, N is the number of frames corresponding to multiple frames of images in the video stream (i.e., the total number of frames of the video segment in the current time period), is the time point corresponding to the th frame of image, is the total angle score corresponding to the i-th frame of image, is the average angle score of the current multiple frames of images.

[0217] Step 507: Compare the angular standard deviation with a preset angular standard deviation threshold and the angular average score with a preset angular average score threshold respectively. Based on the comparison results, obtain the angular feature value.

[0218] In some embodiments, after obtaining the angular standard deviation, the present disclosure can use the angular standard deviation and the angular average score to determine the angular feature value.

[0219] Specifically, the present disclosure can compare the angular standard deviation with a preset angular standard deviation threshold and the angular average score with a preset angular average score threshold respectively. If the angular average score is greater than or equal to the preset angular average score threshold (for example, 0.8), and the angular standard deviation is less than or equal to the preset angular standard deviation threshold (for example, 0.3), it indicates that the cheating possibility of the current cheating suspect object is relatively high in most time slices and the behavior patterns are relatively consistent. At this time, determine that the angular feature value is 1, otherwise the angular feature value is 0.

[0220] In summary, the present disclosure accurately calculates the head rotation angle of the first object, the seat angle with the cheating suspect object, and the seat pitch angle. The system can carefully analyze the visual attention of the cheating suspect object to the first object. Combining the scores of the horizontal angle and the vertical angle, and the preset weights, the system generates the total angle score, and further calculates the angular average score and the angular standard deviation, so as to comprehensively evaluate the line-of-sight dynamics of the cheating suspect object. Finally, by comparing the calculated angular standard deviation and average score with the preset thresholds respectively, the system can accurately determine and output the angular feature value reflecting the line-of-sight characteristics of the first object, providing a strong basis for identifying potential cheating behaviors.

[0221] As a possible implementation manner, as Figure 6 shown in the flowchart of another method for analyzing exam cheating, on the basis of the above embodiments, the specific process of determining the eye-writing feature value corresponding to the cheating suspect object includes the following steps:

[0222] Step 601: Determine the residence time of the cheating suspect object in the writing area during the preset monitoring time period. The residence time in the writing area includes the gaze residence time and the pen tip residence time.

[0223] In some embodiments, the time of staying in the writing area on the test paper refers to the time that the eyes or pen tip of the suspected cheating object stays in the writing area of the test paper. When the present disclosure determines the time of staying in the writing area of the suspected cheating object within the preset monitoring time period, the present disclosure comprehensively considers the eye fixation time and the pen tip staying time, and these two indicators jointly reflect the activities of the suspected object in the writing area. To achieve this goal, the present disclosure can adopt an eye tracking device and supporting software to accurately track and record relevant data.

[0224] Step 602: Determine the ratio between the time of staying in the writing area and the preset monitoring time period to obtain the time proportion of staying in the writing area.

[0225] In some embodiments, the present disclosure defines the time proportion of staying in the writing area (Writing_ratio), which represents the percentage of the time that the eyes or pen tip of the suspected cheating object stays in the writing area in the entire monitoring time period. Among them, the preset monitoring time period of the present disclosure can be set according to actual needs, and the present disclosure takes 30 minutes as an example.

[0226] Step 603: Compare the time proportion of staying in the writing area and the preset time proportion threshold, and determine the eye-writing characteristic value according to the comparison result.

[0227] In some embodiments, the present disclosure presets a time proportion threshold (T_writing) as a reference standard for judging whether to stay away from the writing area for a long time.

[0228] The present disclosure compares the calculated time proportion of staying in the writing area with the preset time proportion threshold. According to the comparison result, the eye-writing characteristic value is determined. The eye-writing characteristic value can intuitively reflect the degree of attention of the suspected object to the writing area during the monitoring time period, thus providing an important basis for the subsequent judgment of cheating behavior.

[0229] Specifically, if the time proportion of staying in the writing area (Writing_ratio) is greater than or equal to the preset time proportion threshold (T_writing), the eye-writing characteristic value is determined to be 1, indicating that the suspected cheating object is writing normally most of the time; if the time proportion of staying in the writing area (Writing_ratio) is less than the preset time proportion threshold (T_writing), the eye-writing characteristic value is determined to be 0, indicating that the suspected cheating object may have potential cheating behaviors, such as not looking at or writing on the test paper for a long time.

[0230] Among them, the present disclosure can reasonably set and adjust the time proportion threshold according to historical data and experience to improve the accuracy and efficiency of cheating behavior recognition, and the present disclosure takes 70% as an example.

[0231] In summary, through precise monitoring technology, the present disclosure can accurately calculate the residence time of a cheating suspect in the writing area during a preset monitoring period, which covers the residence time of their eyes and pen tip in this area. Further, we compare this residence time with the total monitoring time to obtain the proportion of the residence time in the writing area. This proportion intuitively reflects the degree of focus of the suspect on the writing area. Subsequently, we compare the obtained proportion with a preset threshold of the residence time proportion. Based on the comparison result, we can determine and generate an eye-writing feature value. This feature value serves as an important indicator for judging whether a cheating suspect is focused on the writing activity and provides strong data support for subsequent identification of cheating behaviors.

[0232] As a possible implementation, as Figure 7 shown in the flowchart of another method for analyzing exam cheating. Based on the above embodiments, the specific process of determining the test result analysis feature value corresponding to a cheating suspect includes the following steps:

[0233] Step 701: Obtain the answering data of the first object and the cheating suspect, where the answering data includes the answering answers and question types.

[0234] In some embodiments, the present disclosure can collect the answering answers and question types of each question of the cheating suspect and the first object. At the same time, the present disclosure can also collect the answering data of other students in the same class, which is mainly used for subsequent calculation of the option probabilities of multiple-choice questions.

[0235] After obtaining the answering data, the present disclosure can determine the question types of each question, including but not limited to multiple-choice questions, true-false questions, multiple-choice questions, fill-in-the-blank questions, and essay questions.

[0236] Among them, after obtaining the answering data, the present disclosure can also perform text preprocessing on text-based question types (fill-in-the-blank questions and essay questions), including converting the answers of fill-in-the-blank questions and essay questions to lowercase to eliminate the influence of case differences on text similarity calculation; removing punctuation marks from the answers to make the text cleaner and facilitate subsequent processing; performing word segmentation on the answers to split continuous text into meaningful lexical units; and replacing the words in the segmented results with synonyms or near-synonyms to further improve the accuracy of text similarity. Optionally, the present disclosure can also remove stop words (such as common but meaningless words like "of" and "already") from the answers to reduce noise.

[0237] In addition, for each multiple-choice question type, the present disclosure counts the number of times each option j is selected in the class. Divide the number of times each option is selected by the total number of students in the class to obtain the probability P(j) of that option being selected.

[0238] Step 702: Based on the answer for each question type, determine the similarity of question types corresponding to each question type between the first object and the object suspected of cheating.

[0239] In some embodiments, after obtaining the answer data, the present disclosure may calculate the relative student question type similarity. That is, the present disclosure may obtain the answer sets corresponding to each question type of the first object and the object suspected of cheating, and then based on the answer set corresponding to each question type of the first object and the answer set corresponding to each question type of the object suspected of cheating, determine the similarity coefficient between the first object and the object suspected of cheating for each question type. Finally, based on the similarity coefficient, determine the similarity of question types corresponding to each question type between the first object and the object suspected of cheating.

[0240] Among them, the similarity coefficient and the similarity of question types can be calculated by the following formula:

[0241]

[0242]

[0243] is the similarity coefficient between the first object and the object suspected of cheating for each question type; is the similarity of question types corresponding to each question type between the first object and the object suspected of cheating; A is the answer set of the first object; B is the answer set of the object suspected of cheating; ∣A∩B∣ represents the size of the intersection of A and B (i.e., the number of answers that are the same for both objects), and ∣A∪B∣ represents the size of the union of A and B (i.e., the number of answers filled in by at least one object).

[0244] In an alternative embodiment of the present disclosure, taking the multiple-choice question type as an example, the present disclosure may obtain the option order of a certain multiple-choice question type of an object (assumed to be student A) and the object suspected of cheating (assumed to be student B) (for multiple-choice questions, it is required that the selected option sets are exactly the same), determine the option sets of student A and student B based on the option order respectively, and then calculate the similarity of question types corresponding to the multiple-choice question type between student A and student B based on the option sets of student A and student B.

[0245] That is, the option set of student A ; the option set of student B

[0246] The similarity of question types corresponding to the multiple-choice question type is calculated by the following formula:

[0247]

[0248]

[0249] Among them, is the similarity coefficient corresponding to the multiple-choice question type; is the question type similarity corresponding to the multiple-choice question type; A is the set of option answers of student A; B is the set of option answers of student B; ∣A∩B∣ represents the number of options selected by both students, and ∣A∪B∣ represents the number of options selected by at least one student.

[0250] Taking the text question type as an example, the present disclosure can regard the answers of student A and student B as text sets. For example, student A may give a text answer set A containing multiple keywords or phrases, and student B gives another text answer set B.

[0251] The question type similarity corresponding to the multiple-choice question type is calculated by the following formula:

[0252]

[0253]

[0254] wherein, is the similarity coefficient corresponding to the text question type; is the question type similarity corresponding to the text question type; A is the text answer set of student A; B is the text answer set of student B; ∣A∩B∣ represents the number of keywords or phrases common to both answer sets, and ∣A∪B∣ represents the number of all non-repeating keywords or phrases in both answer sets.

[0255] It should be noted that the value of the relative coefficient of the present disclosure is between 0 and 1: when = 0, it means that there is no intersection between the two answer sets, that is, student A and student B did not fill in any identical options. When = 1, it means that the two answer sets are exactly the same, that is, student A and student B filled in exactly the same answers.

[0256] Step 703: Based on the number of students and the student scores of each student in each question type, determine the score variance corresponding to each question type.

[0257] In some embodiments, the present disclosure can calculate the score variance corresponding to each question type based on the number of students and the scores of each student in each question type. The score variance reflects the degree of dispersion of the students' scores in this question type.

[0258] wherein, the score variance can be calculated by the following formula:

[0259]

[0260] is the student score of the i-th student in a certain question type for answering questions; is the mean (average value) of the student scores in a certain question type for answering questions; is the number of students in the question type for answering questions (the total number of students).

[0261] Step 704: Determine the minimum value of the score variance corresponding to each question type for answering questions respectively, and use each minimum value of the score variance as the exam difficulty coefficient corresponding to each question type for answering questions.

[0262] In some embodiments, using the minimum value of the score variance as the difficulty coefficient actually reflects the degree of consistency of the students' scores in this question type. The smaller the variance, the closer or more consistent the students' performance in this question type for answering questions, which means that the difficulty of this question type for answering questions is moderate or the question design is relatively reasonable. Therefore, the present disclosure determines the minimum value minF of the score variance corresponding to each question type for answering questions, and uses this minimum value as the exam difficulty coefficient of this question type for answering questions. The smaller the difficulty coefficient, the easier this question type for answering questions is.

[0263] Step 705: For each question type for answering questions, determine the cheating suspicion score corresponding to each question type for answering questions based on the question type similarity and the exam difficulty coefficient.

[0264] In some embodiments, the present disclosure can calculate the cheating suspicion score corresponding to each question type for answering questions by combining the question type similarity and the exam difficulty coefficient for each question type for answering questions.

[0265] Specifically, the cheating suspicion score can be obtained through the following formula;

[0266]

[0267]

[0268] where q is the comprehensive exam difficulty coefficient, n is the number of students, minF is the exam difficulty coefficient corresponding to each question type for answering questions, is the question type similarity corresponding to each question type for answering questions between the first object and the cheating suspicion object; is the preset cheating suspicion score weight.

[0269] Step 706: Obtain the test paper result analysis eigenvalue based on the preset score weight and the cheating suspicion score corresponding to each question type for answering questions.

[0270] In some embodiments, after obtaining the cheating suspicion score corresponding to each question type for answering questions, the present disclosure can perform weighted fusion on the cheating suspicion score corresponding to each question type for answering questions through the following formula to obtain the test paper result analysis eigenvalue.

[0271] Taking the number of question types in this disclosure as 3 as an example, three cheating suspicion scores are obtained , , .

[0272]

[0273] Among them, , , are the cheating suspicion scores corresponding to each question type respectively, , , are the preset score weights.

[0274] In summary, this disclosure calculates the similarity between the first object and the cheating suspicion object in each question type by comprehensively comparing the answering data of the first object and the cheating suspicion object, including the answers and question types, and combines the variance of the scores of the student group in each question type to determine the exam difficulty coefficient, and then generates cheating suspicion scores for each question type. Finally, based on the preset score weights, the result analysis eigenvalue of the test paper is summarized to provide objective technical support for the answer sheet, which can accurately evaluate the cheating possibility of the cheating suspicion object and improve the accuracy and efficiency of cheating detection.

[0275] Corresponding to the above exam cheating analysis method, the present invention also proposes an exam cheating analysis device. Since the device embodiment of the present invention corresponds to the above method embodiment, the details not disclosed in the device embodiment can be referred to the above method embodiment, and will not be elaborated in the present invention.

[0276] Figure 8 is a schematic structural diagram of an exam cheating analysis device provided by an embodiment of this disclosure. As Figure 8 shown, the device includes:

[0277] The offset determination unit 810 is configured to obtain multiple frames of images in the video stream and determine the paper offset direction of the first object in each frame of the multiple frames of images;

[0278] The exposure monitoring unit 820 is configured to determine whether there is a risk of paper exposure based on the paper coordinates and the hand coordinates of the first object in each frame of image;

[0279] The object monitoring unit 830 is configured to, if there is a risk of paper exposure, determine a cheating suspicion object associated with the first object based on the paper offset direction;

[0280] The comprehensive analysis unit 840 is configured to determine the angle eigenvalue, the eye writing eigenvalue, and the result analysis eigenvalue of the test paper corresponding to the cheating suspicion object;

[0281] A scoring unit 850 is configured to determine a comprehensive score based on an angle feature value, an eye-writing feature value, and an exam paper result analysis feature value, and determine the degree of cheating suspicion of a cheating suspect based on the comprehensive score.

[0282] In some embodiments of the present disclosure, an offset determination unit 810 is configured to: perform color segmentation and edge detection on each frame of image respectively, extract the exam paper area and the table area in each frame of image; determine the distance between the vertex coordinates of the exam paper in the exam paper area and the edge center coordinates of the table area; determine an exam paper offset result based on the distance and a preset safety threshold; and determine the exam paper offset direction based on the exam paper offset result.

[0283] In some embodiments of the present disclosure, an exposure monitoring unit 820 is configured to: determine the overlapping vertex coordinates of the overlapping area between the exam paper area and the hand area of the first object based on the vertex coordinates of the exam paper in the exam paper area and the vertex coordinates of the hand area of the first object in each frame of image; if the overlapping vertex coordinates meet a preset overlapping condition, determine the overlapping area of the overlapping area and the exam paper area of the exam paper area based on the overlapping vertex coordinates and the vertex coordinates of the exam paper; if the ratio between the overlapping area and the exam paper area is less than a preset proportion threshold, determine that there is a preliminary exposure risk for the exam paper; determine an overlapping ratio based on the vertex coordinates of the hand and the vertex coordinates of the exam paper; and if the overlapping ratio is greater than or equal to a preset overlapping ratio, determine that there is an exposure risk for the exam paper.

[0284] In some embodiments of the present disclosure, an object monitoring unit 830 is configured to: determine a target analysis population corresponding to the exam paper offset direction; and use the head postures of each target analysis object in the target analysis population to determine a cheating suspect in the target analysis population.

[0285] In some embodiments of the present disclosure, an object monitoring unit 830 is configured to: determine an associated population corresponding to the first object; determine the downward viewing angle of each object in the associated population looking at the first object according to the exam paper offset direction and the position coordinates of each object in the associated population; and determine the target analysis population in the associated population according to the downward viewing angle.

[0286] In some embodiments of the present disclosure, an object monitoring unit 830 is configured to: determine the head offset angle of each target analysis object according to the key point coordinates of each target analysis object; determine the relative angle between the first object and each target analysis object according to the exam paper center point coordinates of each target analysis object and the position coordinates of the first object; determine the angle ratio between the head offset angle and the relative angle of each target analysis object, and determine a cheating suspect in the target analysis population based on the angle ratio.

[0287] In some embodiments of the present disclosure, the comprehensive analysis unit 840 is configured to: determine the horizontal deflection angle and the vertical pitch angle of the head of the first object, the seat included angle between the first object and the cheating suspect object, and the seat pitch angle between the head of the first object and the desktop of the cheating suspect object; determine the horizontal angle score based on the seat included angle between the first object and the cheating suspect object and the horizontal deflection angle of the head of the first object; determine the vertical angle score based on the vertical pitch angle of the head of the first object and the seat pitch angle between the head of the first object and the desktop of the cheating suspect object; determine the total angle score based on the horizontal angle score, the vertical angle score, and the preset angle weight; determine the average angle score based on the total angle score and the number of frames corresponding to the video stream; determine the angle standard deviation based on the total angle score and the average angle score; respectively compare the angle standard deviation with the preset angle standard deviation threshold and the average angle score with the preset average angle score threshold, and obtain the angle feature value based on the comparison result.

[0288] In some embodiments of the present disclosure, the comprehensive analysis unit 840 is configured to: determine the stay time of the cheating suspect object in the writing area during the preset monitoring period, where the stay time in the writing area includes the gaze stay time and the nib stay time; determine the ratio between the stay time in the writing area and the preset monitoring period to obtain the stay time ratio in the writing area; compare the stay time ratio in the writing area with the preset stay time ratio threshold, and determine the eye writing feature value according to the comparison result.

[0289] In some embodiments of the present disclosure, the comprehensive analysis unit 840 is configured to: obtain the answer sheet data of the first object and the cheating suspect object, where the answer sheet data includes the answer to the question and the question type; determine the question type similarity corresponding to each question type between the first object and the cheating suspect object based on the answer to each question type; determine the score variance corresponding to each question type based on the number of students and the student scores of each student under each question type; respectively determine the minimum score variance in the score variance corresponding to each question type, and use each minimum score variance as the exam difficulty coefficient corresponding to each question type; for each question type, determine the cheating suspect score corresponding to each question type based on the question type similarity and the exam difficulty coefficient; obtain the test result analysis feature value based on the preset score weight and the cheating suspect score corresponding to each question type.

[0290] In some embodiments of the present disclosure, the evaluation unit 850 is configured to: determine the comprehensive score based on the angle feature value, the eye writing feature value, the test result analysis feature value, and the preset feature weight; if the comprehensive score is greater than or equal to the preset comprehensive score threshold, determine that the cheating suspect degree of the cheating suspect object is high cheating suspect; if the comprehensive score is less than the preset comprehensive score threshold, determine that the cheating suspect degree of the cheating suspect object is low cheating suspect.

[0291] It should be noted that the foregoing explanations of the method embodiments also apply to the devices in this embodiment. The principles are the same and will not be further limited in this embodiment.

[0292] Based on the above method as Figures 1 to 7 shown, correspondingly, this embodiment also provides a computer program product, including a computer program which, when executed by a processor, implements the above method as Figures 1 to 7 shown.

[0293] Based on the above method as Figures 1 to 7 shown, correspondingly, this embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the above method as Figures 1 to 7 shown.

[0294] Based on such an understanding, the technical solution of this application can be embodied in the form of a software product. This software product can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods in various implementation scenarios of this application.

[0295] As Figure 9 shown, the following is a schematic diagram of the hardware structure of an electronic device according to the present invention, including:

[0296] At least one processor 901; and,

[0297] A memory 902 communicatively connected to at least one of the processors 901; wherein,

[0298] The memory 902 stores instructions executable by at least one of the processors. The instructions are executed by at least one of the processors so that at least one of the processors can execute the exam cheating analysis method as described above.

[0299] Figure 9 Taking one processor 901 as an example in

[0300] The electronic device may further include: an input device 903 and a display device 904.

[0301] The processor 901, the memory 902, the input device 903, and the display device 904 may be connected through a bus or other means. In the figure, connection through a bus is taken as an example.

[0302] The memory 902, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the review content generation method in the embodiments of the present application. For example, Figures 1 to 7 the method flow shown. The processor 901 executes various functional applications and data processing by running the non-volatile software programs, instructions, and modules stored in the memory 902, that is, implements the exam cheating analysis method in the above embodiments.

[0303] The memory 902 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the review content generation method, etc. In addition, the memory 902 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 902 may optionally include a memory remotely set relative to the processor 901, and these remote memories can be connected to the device executing the review content generation method through a network. Examples of the above network include but are not limited to the Internet, enterprise intranet, local area network, mobile communication network, and combinations thereof.

[0304] The input device 903 can receive user clicks input and generate signal inputs related to user settings and function controls of the review content generation method. The display device 904 may include a display screen and other display devices.

[0305] When the one or more modules are stored in the memory 902 and run by the one or more processors 901, the exam cheating analysis method in any of the above method embodiments is executed.

[0306] Optionally, the above physical device may further include a user interface, a network interface, a camera, a radio frequency (RF) circuit, sensors, an audio circuit, a WI-FI module, etc. The user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), etc. Optionally, the user interface may further include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), etc.

[0307] Those skilled in the art can understand that the above physical device structure provided in this embodiment does not limit the physical device, and it may include more or fewer components, or combine certain components, or have different component arrangements.

[0308] The storage medium may also include an operating system and a network communication module. The operating system is a program for managing the hardware and software resources of the above-mentioned entity device, and supports the operation of information processing programs and other software and / or programs. The network communication module is used to implement communication between components inside the storage medium, as well as communication with other hardware and software in the information processing entity device.

[0309] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform, or can also be implemented by hardware. By applying the solution of this embodiment, compared with the current existing technologies, in this embodiment, multiple frames of images in a video stream are acquired, and the paper offset direction of the first object in each frame of the multiple frames of images is determined; based on the paper coordinates and the hand coordinates of the first object in each frame of image, it is judged whether there is a risk of the paper being exposed; if there is a risk of the paper being exposed, the cheating suspect object associated with the first object is determined based on the paper offset direction; the angle feature value, the eye writing feature value, and the paper result analysis feature value corresponding to the cheating suspect object are determined; based on the angle feature value, the eye writing feature value, and the paper result analysis feature value, a comprehensive score is determined, and the degree of cheating suspicion of the cheating suspect object is determined according to the comprehensive score. It realizes the use of intelligent video analysis technology, combined with the correlation analysis of the answer sheet results, to greatly improve the efficiency and accuracy of cheating analysis, not only reducing the burden of manual review, improving the accuracy of cheating behavior recognition, but also providing objective and sufficient evidence of cheating behavior for educational institutions, facilitating subsequent processing and investigation.

[0310] It should be noted that in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover a non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0311] The above are only specific embodiments of the present application, enabling those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments described herein, but rather will conform to the broadest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for analyzing cheating in an examination, characterized in that: The method comprises: Acquire multiple frames of images in a video stream, extract the test paper area and the table area in each frame of the image using image processing technology, and determine the test paper offset direction of the first object in each frame of the image in the multiple frames of the image by analyzing the coordinates of the test paper area and the table area, wherein the test paper offset direction is the offset direction of the test paper relative to the table; Based on the test paper coordinates and the hand coordinates of the first subject in each frame of the image, determining whether the test paper has an exposure risk; If the test paper has an exposure risk, determine the target analysis population corresponding to the deviation direction of the test paper, wherein the target analysis population refers to the population that can see the test paper of the first subject; and determine the suspected cheating subjects in the target analysis population by using the head posture of each target analysis subject in the target analysis population; Determine the angle feature value, eye expression and writing feature value, and test paper result analysis feature value corresponding to the suspected cheater, wherein the angle feature value is used to evaluate whether the suspected cheater has abnormal behavior in terms of head and seat angles; the eye expression and writing feature value is used to evaluate whether the suspected cheater transmits information through eye expression or writing behavior, thereby determining whether there is cheating behavior; the test paper result analysis feature value is used to associate with the answer test paper to evaluate the abnormal behavior of the suspected cheater in terms of answer results, thereby determining whether there is cheating behavior; Determine a comprehensive score based on the angle feature value, the eye contact and writing feature value, and the test paper result analysis feature value, and determine the degree of cheating suspicion of the suspected cheater according to the comprehensive score; Wherein, determining the angle feature value corresponding to the suspected cheating object includes: Determine, according to the head offset angle and position coordinates of the first subject, the horizontal deflection angle and vertical pitch angle of the head of the first subject, the seat angle between the first subject and the suspected cheater, and the seat pitch angle between the head of the first subject and the tabletop of the suspected cheater; Determine an absolute difference in horizontal deflection angles based on a seat angle between the first subject and the suspected cheater and a horizontal deflection angle of the head of the first subject, and determine a horizontal angle score based on the absolute difference in horizontal deflection angles; Determine an absolute difference between a vertical angle pitch angle based on a vertical pitch angle of the first subject's head and a seat pitch angle between the first subject's head and the tabletop of the suspected cheater, and determine a vertical angle score based on the absolute difference between the vertical angle pitch angle; The horizontal angle score is determined based on the absolute difference of the horizontal deflection angle and the vertical angle score is determined based on the absolute difference of the vertical angle pitch angle according to the following formula: Wherein, when determining the horizontal angle score based on the absolute difference of the horizontal deflection angle, represents the horizontal angle score, represents the horizontal angle tolerance, Indicates the maximum difference in horizontal angle, represents the absolute difference in horizontal deflection angle, , represents the seat angle, represents the horizontal deflection angle of the head, When determining the vertical angle score based on the absolute difference between the vertical angle and the pitch angle, represents the vertical angle score, Indicates the vertical angle tolerance; Indicates the maximum vertical angle difference, Indicates the absolute difference between the vertical angle and the pitch angle. , represents the vertical pitch angle of the head, represents the seat pitch angle; Based on the horizontal angle score, the vertical angle score and the preset angle weight, the total angle score is determined according to the following formula: in, Represents the total score of the angle Represents the horizontal angle score of each frame image, Indicates the preset angle weight corresponding to the horizontal angle score, Represents the vertical angle score of each frame image, Indicates the preset angle weight corresponding to the vertical angle score, ; Based on the total angle score and the number of frames corresponding to the video stream, the average angle score is determined according to the following formula: in, represents the average score of the angle, N represents the number of frames corresponding to the multiple frames in the video stream, Indicates The time point corresponding to the frame image, is the total angle score corresponding to the i-th frame image; Based on the total angle score and the average angle score, the angle standard deviation is determined according to the following formula: in, represents the angle standard deviation, N represents the number of frames corresponding to the multiple frames in the video stream, Indicates the The time point corresponding to the frame image, represents the total angle score corresponding to the i-th frame image, represents the average score of the angle; The angle standard deviation is compared with a preset angle standard deviation threshold, and the angle average score is compared with a preset angle average score threshold. If the comparison result meets a preset condition, the angle feature value is determined to be 1. The preset condition means that the angle average score is greater than or equal to the preset angle average score threshold, and the angle standard deviation is less than or equal to the preset angle standard deviation threshold. If the comparison result does not meet the preset condition, the angle feature value is determined to be 0.

2. The method according to claim 1, characterized in that The step of acquiring multiple frames of images in a video stream and determining the test paper offset direction of the first object in each frame of the multiple frames of images includes: Performing color segmentation and edge detection on each frame of the image respectively, and extracting the test paper area and the table area in each frame of the image; Determine the distance between the test paper vertex coordinates of the test paper area and the edge center coordinates of the table area; Based on the distance and a preset safety threshold, determining a test paper deviation result; Based on the test paper offset result, determine the test paper offset direction.

3. The method according to claim 1, characterized in that The step of judging whether the test paper has an exposure risk based on the test paper coordinates and the hand coordinates of the first object in each frame of the image includes: Based on the test paper vertex coordinates of the test paper area in each frame of the image and the hand vertex coordinates of the hand area of ​​the first object, determining the overlapping vertex coordinates of the overlapping area between the test paper area and the hand area; If the overlapping vertex coordinates meet the preset overlapping conditions, determining the overlapping area of ​​the overlapping region and the test paper area of ​​the test paper region based on the overlapping vertex coordinates and the test paper vertex coordinates; If the ratio between the overlapping area and the test paper area is less than a preset ratio threshold, it is determined that the test paper has a preliminary exposure risk; Determine an overlap ratio based on the hand vertex coordinates and the test paper vertex coordinates; If the overlap ratio is greater than or equal to a preset overlap ratio, it is determined that there is an exposure risk in the test paper.

4. The method according to claim 1, characterized in that The target analysis population corresponding to the test paper deviation direction is determined to include: Determine the associated population corresponding to the first object; Determine, according to the test paper offset direction and the position coordinates of each object in the associated group, a bird's-eye view angle of each object in the associated group looking at the first object; According to the overhead viewing angle, a target analysis population in the associated population is determined.

5. The method according to claim 1, characterized in that The step of using the head posture of each target analysis object in the target analysis population to determine the suspected cheating object in the target analysis population includes: Determining the head offset angle of each target analysis object according to the key point coordinates of each target analysis object; Determine the relative angle between the first object and each target analysis object according to the test paper center point coordinates of each target analysis object and the position coordinates of the first object; The angle ratio between the head deviation angle of each target analysis object and the relative angle is determined, and based on the angle ratio, the suspected cheating object in the target analysis population is determined.

6. The method according to claim 1, characterized in that Determining the eye contact writing feature value corresponding to the suspected cheater includes: Determine the time that the suspected cheater stays in the writing area within a preset monitoring time period, where the time that the suspected cheater stays in the writing area includes the time that the eyes stay and the time that the pen tip stays; Determine the ratio between the residence time in the writing area and the preset monitoring time period to obtain the residence time ratio in the writing area; The dwell time ratio of the writing area is compared with a preset dwell time ratio threshold, and the eye contact writing feature value is determined according to the comparison result.

7. The method according to claim 1, characterized in that Determining the test paper result analysis feature value corresponding to the suspected cheater includes: Acquire answer data of the first subject and the suspected cheater, the answer data including answer and question type; Determining the question type similarity between the first object and the suspected cheating object for each question type based on the answer to each question type; Based on the number of students and the scores of each student under each question type, determine the score variance corresponding to each question type; Determine the minimum score variance among the score variances corresponding to each answering question type, and use each score variance minimum value as the test difficulty coefficient corresponding to each answering question type; For each question type, based on the question type similarity and the test difficulty coefficient, determine the cheating suspicion score corresponding to each question type; Based on the preset score weights and the cheating suspicion scores corresponding to each question type, the test paper result analysis feature values ​​are obtained.

8. The method according to claim 1, characterized in that The step of determining a comprehensive score based on the angle feature value, the eye contact and writing feature value, and the test paper result analysis feature value, and determining the degree of cheating suspicion of the cheating suspect according to the comprehensive score, includes: Determine a comprehensive score based on the angle feature value, the eye contact and writing feature value, the test paper result analysis feature value, and a preset feature weight; If the comprehensive score is greater than or equal to a preset comprehensive score threshold, the cheating suspicion level of the suspected cheating object is determined to be high cheating suspicion; If the comprehensive score is less than the preset comprehensive score threshold, it is determined that the cheating suspicion level of the cheating suspect is low cheating suspicion.

Images