Research On Pedestrian Abnormal Behavior Detection Method Based On Video

The continuous increase of population density puts forward higher requirements to the level of public security prevention in China.Only relying on manpower to monitor video surveillance cannot meet the needs of society.Using intelligent video surveillance technology to realize automatic detection of abnormal behavior can provide effective help for safeguarding public security and maintaining social order.Therefore,this paper studies how to effectively detect the abnormal behavior of pedestrians in video sequences.This paper mainly studies and improves the problems existing in moving target tracking and abnormal behavior detection.The main research work is as follows:In the aspect of moving target tracking,this paper proposes an improved target tracking algorithm based on optical flow feature points.On the one hand,Shi-Tomasi corner detection algorithm is introduced to solve the problem that traditional KLT tracking algorithm is sensitive to illumination and rotation changes.On the other hand,in order to solve the redundant information problem that the extracted feature points contain a large amount of background and shadow,the background feature points are filtered by counting the static state of the feature points and limiting the threshold.In addition,the shadow feature points were removed by using HSV color space.The improved target tracking algorithm has strong adaptability to the illumination change,and effectively filters out the feature points existing in the background and shadow,which improves the real-time performance of the algorithm.In this paper,the multi-target tracking algorithm is optimized by combining optical flow clustering and occlusion improvement in the scene of multi-target interactive motion.On the one hand,DBSCAN clustering algorithm is introduced to distinguish the targets with close distance and similar motion states.On the other hand,in order to solve the problem of target loss caused by local occlusion,the Kalman filter is introduced to predict the occlusion area.Experimental results show that the proposed algorithm has high accuracy,real-time performance and anti-jamming capability in multi-target tracking scenarios.In the aspect of abnormal behavior detection,this paper realizes the detection of four kinds of abnormal behaviors based on target tracking,which are regional intrusion behavior,wandering behavior in sensitive area,fighting behavior and following behavior.Firstly,regional intrusion detection is implemented by judging the position relationship between the feature points and the designated detection area on the rectangular frame of the tracking target.On this basis,wandering behavior detection is realized according to the distance displacement ratio and trajectory overlap in the region.In addition,by introducing the constraints of target distance and optical flow energy,the characteristics of optical flow direction were extracted and the SVM classifier was used to classify the behavior,and the fighting behavior detection was realized.Finally,according to the distance,speed and gait characteristics of the pedestrian target,combined with SVM classifier,the following behavior detection is realized.Simulation experiments are carried out on public database and self-collected video sets.Experimental results show that the proposed algorithm can achieve stable tracking of multiple targets in different indoor and outdoor scenes and under different lighting conditions.Also can realize a variety of specific abnormal behavior detection.