The Research On Moving Target Detection And Abnormal Behavior Analysis In Intelligent Video Surveillance System

Intelligent Video Surveillance(IVS) is a widely concerned subdiscipline in computer vision field in recent years. Machine vision technology and techniques related to image processing are used to interpret video signals in IVS, it aims to get a intelligent control of the surveillance system. What the IVS do can solve the problems of large amount of data processing, too long response time, the boundaries of human and so on that come from the traditional monitoring system and all these defects will result in poor efficiency, tedious work and some others.All the researches here are done for the demand of the applications of IVS. The study in this paper focuses on moving target detection and abnormal behavior analysis in video sequence. The specific work need to be done is to design a series of algorithms for moving target detection and tracking and on this basis an intelligent analysis on abnormal behaviors is available. There are the main works finished as follows:Section 1, Moving Target Detection. In order to overcome the failing in full foreground extracting of the traditional GMM-based moving target detection algorithm,a developed one is designed here. The developed algorithm combines GMM with frame-differ method together and it tackles well in full foreground detection. What’s more, a new moving target detection algorithm based of background modeling is primarily proposed. In the new algorithm, the frame-differ method helps detect the background parts of a video frame and all the background parts are gather together to establish a stable background model. Besides, an adaptive updating strategy inner is designed for real-time background update and helps to detect a satisfying foreground.Experimental results indicate that the algorithm proposed here are stable and work well.Section 2, Moving Target Tracking. In order to resolve the technical difficulties in moving target tracking under complex environment, a new moving target tracking algorithm based on perception hash is proposed. Perception hash is used to generate the hash code of images and hash code is used for marching. When tracking, the newalgorithm combines matching strategy with searching strategy together and a template evaluation function and update strategy is designed here to ensure a good adaptability against the occlusion and variation of the moving targets. In contrast, the tracking algorithm proposed here is lower time cost and more stable when it comes to noise,occlusion and variation of targets.Section 3, Abnormal Behavior Analysis. In view of the fact that an abnormal behavior is hard to define and easy to find out, a discriminated algorithm for abnormal behavior analysis based on target blobs and its motion path is adopted. The algorithm adopted tries to set different rules for each of the different abnormal behavior and once a rule is violated, related alarm will be triggered. The main works finished here include setting of region of interest, identification of moving targets and feature extraction. A bounding rectangle is introduced here to mark the moving target and the feature extracted inner include area of targets, barycentric coordinates, motion path and so on.Experimental results indicate that the algorithm adopted here is stable and works well.