Research And Design Of Abnormal Behavior Detection System For Candidates Based On Video Analysis

Cheating in examinations is a frequent occurrence and has become an important factor in the fairness of examinations.Due to factors such as invigilation,examination patrol and review of video surveillance work,there are often problems of missed or mis-detected irregularities such as cheating by candidates,and with the increase in online examinations,the problem has become more of a concern.Research into methods and systems for the rapid and accurate identification and detection of examination cheating,using emerging technologies and facilities,offers a way to address this problem.Using offline examinations as the research object and surveillance video as the basic facility and data source,this theis focuses on the automatic detection algorithm of common abnormal behaviours of candidates in the examination hall,such as looking around,passing notes and peeking down,in order to establish a corresponding automatic monitoring system to assist examiners in detecting and judging abnormal behaviours of candidates in a timely manner.For the problem of accurate implementation of candidate abnormal behaviour detection,the YOLOv5 model is chosen as the base abnormal behaviour detection model.Combining the small magnitude of the candidate’s abnormal behaviour,the presence of an obscured target and the small size of the target image,the CBAM module and Ghost module were incorporated into the YOLOv5 model to improve the accuracy of detection.Due to the lack of available samples,video data from actual examinations were collected to obtain raw video and image data,in which the three cheating behaviours of looking around,peeking down and passing notes were sampled,and Label Img software was used to label the samples to construct a sample dataset of candidates’ abnormal behaviours to support algorithm learning and testing.The ablative experiments on the sample dataset show that the improved YOLOv5 model has a high detection rate for small movements and maintains a detection efficiency that is largely consistent with the YOLOv5 model,which is of good practicality.Based on the improved YOLOv5 examinee abnormal behaviour detection algorithm,this theis uses the Python programming language and the Py Qt5 framework to develop an examinee abnormal behaviour detection system,completing tasks such as video recording,abnormal behaviour detection,real-time capture and abnormal behaviour video analysis.Tests on actual examination videos show that the detection of abnormal behaviour can be achieved more accurately.