Research On Human Abnormal Behavior Detection Algorithm Under Security Monitoring

With the increasement of people’s sense of security,how to use a wide range of video surveillance equipment to detect and warn human abnormal behaviors such as falling,wandering and fighting in time and effectively has become a research hotspot.The traditional monitoring system adopts human duty,which is not only time-consuming and laborious,but also has the problem of missing judgment and misjudgment.Human abnormal behavior detection technology founded on image processing theory and video analysis technology can timely identify and warn human abnormal behavior in the monitoring scene,which has important research significance.Human abnormal behavior detection includes three steps: target detection,target tracking and abnormal behavior judgment.The main contribution of this paper is as follows:(1)Aiming at the problem of incomplete detection results in target detection,a moving target detection algorithm based on mean background and three frame difference method was proposed.Firstly,the mean method was used for background modeling,and then the background template was continuously updated by three frame difference method.Finally,the background subtraction method was used to extract the binary image of moving target.Experiments showed that this method improves the detection accuracy compared with the traditional target detection algorithm.(2)Aiming at the problem of tracking failure due to occlusion in target tracking,a pedestrian tracking algorithm founded on multi-feature fusion was established.In the process of moving target feature extraction,the gradient feature,color feature and texture feature are fused to obtain a 350 dimensional feature vector;Kalman filter was used to predict the target motion trend;In template matching,TSS method was used to complete target matching.Experiments show that the tracking success rate on the data set CASIR was 94.9% and the detection speed was 16.3f/s.(3)According to the different characteristics of abnormal behavior,a specific abnormal behavior detection algorithm was proposed.In the process of fighting behavior detection,according to the characteristics of large amplitude and chaotic direction of limb interaction between moving targets,Lucas Kanade optical flow method was used to solve the optical flow feature vector of image frame,and the entropy of improved optical flow direction histogram based on amplitude weighting was calculated;In the process of falling behavior detection,according to the characteristics of low center of gravity of moving target and long falling time,a method to calculate the change rate of moving human target centroid,the width height ratio of external rectangular frame and the falling time was proposed;In the process of wandering behavior detection,according to the characteristics of more direction conversion times of motion trajectory and longer residence time,Freeman chain code was used to record the motion trajectory,count the distance angle,observe the change of motion direction,and calculate the time when the human body is in the monitoring scene.The comprehensive experimental result showed that the algorithm can detect the three abnormal behaviors of fighting,falling and wandering efficiently and accurately,and complete the timely early warning work.The experimental results showed that the improved human abnormal behavior detection algorithm can realize the efficient detection and timely early warning of fighting,falling and wandering.