Moving Target Detectiong And Tracking In Intelligent Video Surveillance

Intelligent Video Surveillance is a rising application direction in computer vision field and anadvanced topic of concern. With the rapid developing of networking and digital videotechnologies, monitoring technology is striding forward intelligence and networking. Thefunctions of monitoring system are increasingly powerful, but the system needs people to observethe screen and analyze different conditions round the clock, so the work is heavy and tired.Intelligent video surveillance use computer vision and video analysis algorithm to automaticallyanalyze the image sequence captured by camera, and moving targets detection and tracking is themost basic of the key technology in intelligent video surveillance. They are the foundation ofkinds of follow-up advanced processing, such as target classification, behavior analysis, incidentdetection, activity identification, video image compression and semantic indexing. They are alsothe key to make the video monitoring system automatic, intelligent and real-time.This paper analyzes the present and more commonly used detection algorithm, and throughthe analysis of the advantages and disadvantages of their own, we put forward an improvedalgorithm based on background difference algorithm: we use mixed Gaussian backgroundmodeling to extract background, avoiding the condition that when moving object occurs in thefirst frames, we can’t extract the valid background using background difference method; when weget the valid background we replace mixed Gaussian background modeling with thresholdadaptive and background update method, and the new method is simple, practical, and processedrapidly and make up for the shortage of heavy calculation and bad practical when using mixedGaussian background modeling; and the introduction of the frame differential method solve theproblem of the environment changes, lighting changes when using background difference method,and at the same time improves the stability of the monitoring system.In the motion tracking field, according to the different tracking method, this paper introducesthree kinds of algorithm: tracking method based on the partial differential equations, trackingmethod based on mean shift and tracking method based on filtering theory. We analyze mean shiftalgorithm in detail, and implement a tracking window adaptive mean shift algorithm——CamShifttracking algorithm, and according to the problems existing in the experimental results, we introduceKalman filter and particle filter, and analyze the characteristics and implementation principle of thetwo algorithms and the advantages and disadvantages in tracking process. Kalman filter can only beapplied in linear and gaussian system, and cannot obtain the ideal effect using in nonlinear ornon-gaussian system. And the particle filter algorithm is complex, and needs heavy calculation. It cannot satisfy the requirement of real-time when using in complex environment. Based on theresearch above, we proposed a tracking algorithm based on kalman particle filter. The basic idea isthat: On the basic of the particle filter, we use kalman filter of consider the latest observationinformation. The sampling strategy will effectively lead to the big area of the likehood function, socan substantially reduce the needed number of particles. Finally through a large number of specificexperiments, mainly solving the scale change and keep-out problems, we propose an improvedkalman particle tracking algorithm. And the improved algorithm can well realize target tracking andhas good real-time performance.