Study Of Human Abnormal Behavior Detection Algorithm In Surveillance Video

With the improvement of living standard and the development of science and technology,people's demand for social public safety is becoming more and more obvious.Simple camera surveillance has not satisfied people's demand for safety.Therefore,the design of an intelligent surveillance system which can quickly find useful information from massive surveillance video is the development trend of the surveillance system.As a part of the intelligent surveillance system,the human abnormal behavior detection algorithm has gradually attracted the attention of everyone.Therefore,it is necessary to study the human abnormal behavior detection algorithm in surveillance video.Based on machine vision theory,image processing theory and video analysis technology,three key technologies of human abnormal behavior detection in surveillance video are studied in this paper: moving object detection,moving object tracking and abnormal behavior detection.The main contents of this paper can be summarized as follows:(1)A moving target detection algorithm based on mean background and three frame difference is studied.Based on the research of several commonly used object detection algorithms,aiming at the incomplete detection results and the "empty" problem in the process of target detection while using common target detection algorithms,the background extraction algorithm is considered to be combined with the three frame difference algorithm.A target detection algorithm based on mean background and three frame difference is obtained by adding the background extracted by the improved mean background to the three frame difference process.Finally,the common target detection algorithm and the algorithm proposed in this paper are evaluated qualitatively and quantitatively.The results show that the accuracy and integrity of the algorithm proposed in this paper are superior to those of the other four commonly used target detectionalgorithms,and the detection speed of the algorithm is fast.(2)Study on video target tracking algorithm based on spatio-temporal context and Kalman filtering.Through the simulation of the common target tracking algorithms,the advantages and disadvantages of these algorithms are analyzed.Aiming at the problem of tracking failure in common target tracking algorithm when the target is occluded,In this paper,the combination of spatio temporal context tracking and Kalman filtering is considered.When the target is seriously occluded,the Kalman prediction value is used to update the target position,and eliminate the influence of occlusion on target tracking.And the improved tracking algorithm is used as a tracker to track multiple targets at the same time and then the task of video multi-target tracking has been completed.Finally,the common target tracking algorithms and algorithm proposed in this paper are evaluated qualitatively and quantitatively.The results show that the proposed algorithm can overcome the tracking failure problem when the target is seriously occluded,and can achieve real-time tracking.(3)Study on human abnormal behavior detection algorithm.First of all,the abnormal behavior is divided into single person abnormal behavior and the two person interactive abnormal behavior.Single person's abnormal behavior includes crouching,fainting,and so on.The two people's interactive abnormal behavior includes fighting,robbing,etc.Aiming at the characteristics of single person's abnormal behavior,the target area compactness and target centroid speed are extracted to detect abnormal human behavior;In view of the characteristics of two person interactive abnormal behavior,the improved histogram of optical flow direction feature which based on the effective optical flow field in the target area is extracted,and the entropy of the optical flow direction histogram is used to detect the human abnormal behavior.By introducing the abnormal detection accuracy and abnormal detection error rate,the proposed algorithm and traditional abnormal behavior detection algorithm based on area optical flow feature are compared and analyzed.The results show that compared with the traditional abnormal behavior detection algorithm the proposed algorithm has better detection effect,can detect abnormal behaviors in surveillance video effectively,and has a low detection error rate.