Research On Abnormal Behavior Recognition Of Examination Room Based On Multi-features

At present,the infrastructure of electronic invigilator in China is gradually improving,which has laid the foundation for the construction of standardized examination room.However,the current examination invigilator verification method still adopts the traditional invigilator method,which combines human on-site inspection and video monitoring.This method makes invigilators work intensively,and the monitoring of examination room is prone to abnormal behavior.At the same time,the corresponding massive video data transmission,storage,post review pressure is huge.With the rapid progress of intelligent monitoring technology,intelligent monitoring can enhance the strength of invigilation,improve the efficiency of verification,relieve the pressure of video transmission,storage and review,and it has important practical significance to apply it to the examination room monitoring.In this thesis,based on the background of examination room environment,an intelligent monitoring method of examination room is realized,which can detect the number of examination room and identify the abnormal behavior of examinees(1)According to the behavior standard of examinee standard,the classification of examinee abnormal behavior and the corresponding judgment criterion are formulated.The actual demand of the examination room is analyzed,and the number of examinees and abnormal behavior identification are realized.(2)In view of the shortage of data sets in the background of the examination room,the data collection is carried out by classroom monitoring,and the single frame sequence data set is obtained.The standard data information classification is labeled by labelimg,and the data set of abnormal behavior in the test field is self-made.(3)For the number of people in the test room,YCb Cr color model,Haar feature and SSD target detection algorithm are used to detect the number of candidates.The experiments show that the SSD target detection algorithm has a good effect through the optimization and identification of internal parameters of the model.Finally,the number of people in the test field is detected by SSD target detection algorithm.(4)In the aspect of abnormal behavior recognition,two methods are used to recognize the abnormal behaviors of examinees.One is template matching method.The dynamic target detection method is used to extract abnormal behaviors,and the abnormal behavior template is established and the abnormal behavior recognition of test room based on template matching is realized.Another method is based on YOLO algorithm to recognize abnormal behavior in the test room.The experiment verifies that YOLO algorithm is superior to template matching method.The training model of YOLO algorithm is optimized by using frozen network,with training speed increased by 27.8%,and the average detection accuracy is improved by 2.2%.It can realize the recognition of abnormal behaviors of examinees in the test room,which has stronger universality.The accuracy and real-time performance meet the design requirements of the behavior recognition of the examination room.