Font Size: a A A

Research On Human Abnormal Behavior Detection Method Based On Video Monitoring

Posted on:2021-11-19Degree:MasterType:Thesis
Country:ChinaCandidate:P Y JiangFull Text:PDF
GTID:2518306560996469Subject:Pattern Recognition and Intelligent Systems
Abstract/Summary:PDF Full Text Request
With the development of smart city and the concept of smart campus,more and more smart video surveillance is applied in various places.Compared with intelligent video surveillance,traditional video surveillance needs to waste a lot of manpower to watch the video,and the dangerous behaviors in some dangerous places can not be handled in time,which is easy to cause serious consequences.Therefore,this paper focuses on the research and analysis of pedestrian behavior detection methods in specific scenes based on pedestrian detection and tracking.In this paper,we mainly track and detect pedestrians in surveillance video,and judge whether there is abnormal behavior of pedestrians in specific occasions.Therefore,we need to study three aspects: moving object extraction,moving object tracking and abnormal behavior analysis.In the aspect of moving object extraction,this paper uses the improved vibe algorithm to extract the moving pedestrian in the video.Because the traditional vibe algorithm is prone to ghost and shadow when extracting moving objects,and the detection threshold of moving background cannot be changed according to the change of scene.Therefore,this paper combines the vibe algorithm and the average background method to eliminate the ghost problem in the initial video.At the same time,HSV color space is introduced to suppress the shadow caused by the light angle in the process of pedestrian movement.Finally,adaptive threshold is introduced to replace the fixed threshold in the traditional vibe background detection.In the aspect of moving target tracking,pedestrians may block each other during the process of moving,which leads to the failure of target tracking.Therefore,this paper chooses to combine the Kalman tracker and Hungarian optimal matching algorithm to predict the position of the next pedestrian frame by Kalman filter.In the next frame,the predicted position is matched by the Hungarian best matched pedestrian position,so as to complete the tracking of moving objects,so that the moving objects can still maintain good tracking performance under occlusion.In the aspect of detection,this paper mainly focuses on three kinds of abnormal behaviors: pedestrian fighting,regional invasion and pedestrian falling.In the detection of pedestrian fighting,the improved optical flow method and spatial information are used to detect whether there is fighting behavior.In the area intrusion detection,lawn position is extracted by color recognition,lawn boundary is extracted by Sobel edge detection,and intrusion behavior is detected by feature points.In the pedestrian fall detection,the acceleration and the degree of centroid deviation are used to determine whether the pedestrian falls.The improved method is simulated on the video set,and the results show that the improved algorithm can completely extract the moving pedestrian in the video monitoring,keep the continuous tracking of the pedestrian in the case of pedestrian occlusion,and detect a variety of abnormal behaviors defined in the video monitoring.
Keywords/Search Tags:intelligent video surveillance, dynamic background modeling, moving target tracking, abnormal behavior detection
PDF Full Text Request
Related items