Detection And Tracking Of Moving Target In Dynamic Background

Various countries around the world appeared in public security incidents make securityissues particularly important. Now real-time monitoring for the public to maintain publicsafety is a major part of the intelligent video surveillance. Target detection and trackingtechnology is an important aspect in the field of computer vision. Currently target detectionand tracking technology has broad application prospects, such as monitoring of public security,the flow of traffic or people, industrial manufacturing production line, human-computerinteraction and other fields.This paper proposes the improved methods, the research works are as following:(1) Aiming at the shortage of Mean shift target tracking using single color characteristics,this paper proposes combination of color and texture feature. Texture feature is extracted ofLBP mode and Sobel operator. The tracking error caused by the color similarity ofbackground and the target is solved. Final position of target is based on the matching degreeof each block. The experimental results show this method can shorten the time withoutocclusion, and the tracking accuracy is improved when the target is partially occluded.(2) When the weights reduce the impact of the background, but also weaken the targetmodel at the same time in the Mean shift tracking algorithm. Considering this problem, thetarget and the candidate model and the probability of the pixel belonging to foreground isproposed to enhance the differences between the foreground and background. The proposedmethod suppresses the influence of the background pixels by this improvement. Projectionvector combined of the foreground probability map and Mean shift weights to enhancedifference between the foreground and background. The projection vectors are updated toadapt to the background changes. According to moments features of improved weight, theposition and size of the target are determined. The experimental results show this method canachieve the target tracking.(3) For target occlusion, tracking technology, blur arising from the high-speed motion andother issues, target tracking algorithm based on superpixel is proposed in this paper.Superpixels are used to establish confidence map of target, because of the advantages ofcomplete preservation of image boundary after segmentation. Updating the appearance modelensures the tracking process can adapt to the changes of target and background. Finally, thelocation of the target is found according to the confidence map. The experimental resultsshow that the proposed tracking algorithm can effectively find the target position in the caseof partial occlusion or background interference.