Research On Crowd Abnormal Behavior Detection Algorithm In Video Surveillance Scene

With the development of digital information,internet technology has evolved by leaps and bounds.In recent years,due to the government's emphasis on public safety,an increasing number of CCTV cameras have been installed in public places such as streets,subway stations and airports.Traditional video surveillance technology is gradually unable to handle the ever-increasing amount of massive video data.The intelligent video surveillance system which can automatically detect and identify abnormal behaviors has received widespread attention from the society.As the high-level processing in intelligent video surveillance algorithms,abnormal behavior detection mainly carries out real-time warning and detection of unusual behaviors in the monitoring scene to reduce the harm caused by unexpected abnormal events to society and citizens.This paper takes the medium and high density crowd as the research object,mainly focuses on the research and discussion on crowd abnormal behavior detection in video surveillance system.The main research contents and achievements of this paper are as follows:Firstly,the fundamental theory and methods of abnormal behavior detection are introduced in this paper.The fundamental principle of common methods for moving target detection is briefly introduced,and we analyze their advantages and disadvantages.Based on the abnormal behavior detection process,the definition of anomalous behavior,feature selection and motion pattern construction,and abnormal behavior detection mechanism are expounded.Secondly,when the 'escape event' occurs,pedestrians will escape in a disordered way with different velocities and directions to avoid the potential danger.Based on these characteristics,this paper proposes a Direction-Collectiveness model to detect escape behavior in crowd scenes.We use GMM to extract background images,and extract a set of trajectories from video sequences by using generalized Kanade-Lucas-Tomasi key point tracker(gKLT).Then,a Direction-Collectiveness model is built based on the randomness of velocity and orientation calculated from the trajectories to express the movement of the crowd.Threshold segmentation method is used to determine whether the crowd escape behavior occurs.Experimental results show that the proposed method can effectively detect the crowd escape behavior with low contrast,individual occlusion and complex background.However,when the escape behavior has an ordered motion pattern,it will be mistakenly detected as normal behavior.Finally,in order to solve the perspective deformation caused by the low shooting angle of surveillance equipment,a crowd abnormal behavior detection method based on multi-scale cell Direction-Collectiveness model is proposed.Considering that the target closer to the camera can provide more feature description information because of the larger area occupation in the frame,a multi-scale cell structure with increasing size from top to bottom is established.In order to reduce the computational complexity,feature extraction operation is only performed on active cell which contains sufficient foreground information.Because the Direction-Collectiveness feature has poor ability to characterize relatively disorderly normal behavior with small amplitude and relatively orderly abnormal behavior with large amplitude,we fusion velocity energy feature with Direction-Collectiveness feature.The anomaly inference mask based on GMM is used to detect the crowd abnormal behavior.Experimental results show that the proposed method can detect the crowd abnormal behavior accurately and robustly in the scene of low shooting angle,orderly escaping and disorderly normal walking.