The Research On Video-based Online Examination Cheating Behavior Detection Method

With the rapid development of computer technology and Internet technology,online exams have gradually become a new test method widely adopted by test organizations.The test mode has been more and more widely recognized and developed.However,online examinations and field examinations are faced with various kinds of cheating and even more forms of cheating.Therefore,in order to ensure the fairness of online exams,effective online exam cheating detection techniques are required to supervise candidates.By studying various methods of detecting cheating in online exams at home and abroad,the advantages and shortcomings of the online exams were summarized.According to the form of cheating in the online exam,this study proposed a method based on the combination of head posture,screen fixation point estimation and mouth state recognition to detect online cheating behavior.The main research contents included the following parts:(1)In this research,MTCNN-based transfer learning method was used for face detection and face alignment,which effectively solved the problem of detector failure under dark light or face deflection conditions in traditional face detection algorithm.At the same time,29 specific facial feature points were obtained.(2)In the head pose estimation part,this research adopted the multi-loss function head pose estimation algorithm based on feature fusion,and obtained the strong features of the face image by combining the features of the FC layer of the MTCNN network and the ResNet50 network.The multi-loss function was introduced to estimate the head pose.This method refined the estimation result of head pose classification using only the cross entropy loss function,which could improve the accuracy of head pose and avoid the influence of utilizing facial 2D feature points and 3D face coordinates using Posit algorithm(3)In the estimation part of the screen fixation point,the facial feature point coordinates obtained by combining face alignment and the head deflection angle obtained by head pose estimation were taken as original features.After feature construction and feature screening,cascade XGBoost classification was adopted.And the method of regression model was used to estimate the gaze point.The experimental results showed that the screen gaze estimation method effectively improved the accuracy of the gaze area and reduced the error value of gaze point estimation.In the mouth state recognition part,in order to simplify the calculation and improve the processing speed,the mouth-like roundness calculation was performed by the image processing technology,thereby identifying the open and closed state of the mouth.(4)In the detection stage of online exam cheating behavior,this research expounded the judgment basis of candidates' cheating behaviors in various parts of head pose,screen fixation point and mouth state,and conducted simulation experiments on the overall cheating detection process.The experimental results show that the detection method of online exam cheating behavior in this research is comprehensive and accurate.