Research On Behavior Recognition Of Examinees In Examination Room Based On Computer Vision

Cheating in the examination room refers to the fact that in the process of examination,the reference personnel obtain the answers through improper ways.The most important thing in the invigilation scene is to ensure the fairness and impartiality of the examination,so how to identify the cheating in the examination room in time is the key to the invigilation task.In the traditional manual invigilation mode,a large number of invigilators need to be assigned to maintain the order of the examination room.many examination places,long examination time and other factors bring great challenges to the physical strength and energy of the invigilators.The existing monitoring system has played a certain effect in the invigilation scene,which usually uses the monitoring specialist to view the situation of multiple examination rooms in real time,which can assist the work of the invigilator on the spot.At the same time,the monitoring system plays a more important role in storing examination video data.Although the existing monitoring system is effective,its effect is limited.First of all,it is difficult to take into account the overall situation by relying entirely on the eye scan of the background invigilators,and it is hard to avoid omission and misdetection when watching for a long time.Secondly,the monitoring system only stores data,cannot mark abnormal behavior,and can only be used as follow-up forensics.There is still a lot of room for improvement in real-time abnormal behavior detection.Therefore,in order to make full use of the existing monitoring system,this paper considers the application of computer vision technology to the behavior identification of candidates in the examination room to detect cheating in the examination room.Truly realize the real-time monitoring of the system,and bring more auxiliary functions to the task of invigilating the examination.Ensure the fairness and impartiality of the examination.Through systematic judgment and manual review of the form of parallel detection,the misjudgment of abnormal behavior is reduced to the maximum extent.The work of this paper is as follows:(1)The technical framework of human posture estimation based on Open Pose:this paper proposes an improved scheme of knowledge distillation and network structure optimization of feature extraction network.In the experiment based on COCO dataset,the optimized version of this paper improves 336.11% compared with the original method in CPU environment,while the performance of AP @ 0.5 decreases only 1.53%.(2)Behavior recognition algorithm based on Human body Node: The high-level semantic information such as motion direction,human limb and motion amplitude of the attached nodes are encoded into the image to enhance the feature representation ability of the behavior data.In the experiment based on the self-made behavior data set in this paper,the recognition accuracy of cheating in the examination room is achieved with 93.5%.(3)Object Detection algorithm based on YOLO:Aiming at the task of intelligent invigilation in this paper,a scheme of boundary box improvement and data enhancement is proposed.In the experiment based on Object365 dataset,the model recognition accuracy of this paper is higher than that of YOLO algorithm 7.7%,and the processing time of single image is reduced 51.2%.(4)Online intelligent invigilating system:An online intelligent invigilating system is designed and implemented to detect and record abnormal behavior in the examination room and make an alarm in time.The existing surveillance system is used as the video data input source,and the server uses the micro-service deployment of algorithm module to save computer resources and ensure the availability and efficiency of the system at the same time.Through the design of experiments on human posture estimation,behavior recognition and target detection and data analysis,the intelligent invigilating system designed and implemented in this paper can identify the current human behavior in video and detect abnormal targets in real time.It achieves the effect of assisting the invigilating task and helps to maintain the fairness and impartiality of the examination room.