Illegal Object Detection For Online Invigilation

In the context of global informatization,online education conforms to the trend of the times and is vigorously developing in a more convenient and intelligent direction..Combining artificial intelligence with online invigilation to improve anti-cheating measures more accurately and efficiently has also become an urgent problem to be solved in the implementation of online education.The phenomenon of cheating using mobile phones,books,headsets and other items in the examination has never been solved.Adding the detection of illegal items in the invigilation link of the examination system will provide a new solution to the problem of cheating by examinees,and is of great significance for maintaining the fairness of the examination and evaluating the real strength of the examinees.This paper takes the invigilation link of the online examination system as the research background,starting from the basic framework of the YOLOV4 object detection model,and is committed to improving the detection effect of the object detection model.The work content is as follows:First of all,this article focuses on the lack of detection and identification of illegal items such as mobile phones,books and headsets in the online invigilation examination,and analyzes the theoretical knowledge of algorithms and corresponding evaluation models involved in the illegal object detection for online invigilation.Performance indicators with test results.The demand analysis and framework design of the invigilation link of the online examination system are carried out,and then the special training data set and test data set are made for online invigilation by collecting and labeling pictures.Secondly,in order to solve the problem that the YOLOV4 objection detection model used for online invigilation is too large,the detection speed and accuracy are reduced,and it is difficult to meet the real-time requirements.The original YOLOV4 backbone network has been light-weighted.A network structure model of M-YOLOV4 is proposed.The ablation experiment was conducted on the online proctored data set to verify the effectiveness of the method.Finally,this paper aims at the missed detection problem of the single-frame detector M-YOLOV4 used for online invigilation.Combining minimum delay detection and MYOLOV4,this paper proposes a minimum delay and multi-frame object detection model.The model is compared with the original single-frame model on the data set for online invigilation,and the effectiveness of the method is analyzed from both qualitative and quantitative aspects.