Research On Abnormal Behavior Detection In Examination Room Based On OpenPose

With the maturity of software and hardware technologies,video surveillance technology is commonly used in examination rooms to record every move of each student in real time.It is effective to maintain the order of examination and ensured the fairness of the test by using video surveillance.However,the current examination room monitoring system is still consume a lot of time and effort to deal with the video,and some abnormal behaviors will be omitted at the same time.In that situation,the valuable video resources will be wasted,and the abnormal behaviors at the room can’t be prevented in time.With the development of artificial intelligence technology,it is possible to detect the students’ behavior in the frame image of the the video.In order to solve this problem,this paper proposes a abnormal behavior detection system based on OpenPose.This model can detect bone key points of multiple candidates in the examination room,and extract useful feature vectors to train the model based on the key point information,which can detect abnormal behaviors such as stretching head,stretching hand and standing up.This method can effectively reduce the interference caused by light in the environment and be used in different scenes,it will improve the efficiency of video detection of abnormal behavior greatly.The main work of this paper is as follows:(1)Screen the detection area.Inter-frame difference method and the adaptive threshold YCb Cr skin color detection method are used to detect the frame images and general areas where abnormal behaviors may occur in the video stream,which improves the detection efficiency.(2)Extract feature vectors.The OpenPose model is used to extract the coordinate information of the corresponding bone key points in the data set and the generated bone map,which can provide data support for subsequent training.(3)Training related models.The key point coordinate information of the skeleton is extracted to useful feature vectors as a data set,and a fully connected neural network is used for training;The generated skeleton diagram is used as a data set,and use the convolutional neural network to train the model,then compare the training result.(4)Design the system interface.Design a system interface for the trained model,visualize the examination room data,which can help the invigilator to analyze the result,discriminate the behavior of each student,so as to maintain the order of the examination room.This system collects a large number of data sets of various actions from multiple angles and targets to train the model,which improves the robustness of the abnormal behavior detection system effectively,and uses skin color detection and motion detection as pre-detection methods.The detected frame images improve the detection efficiency,and can detect variety of cheating behaviors well in the examination room.