Research On Object Tracking Algorithm Based On Color Image Sequence

The method of target tracking in color image sequence is a hot topic ofcomputer vision’s research, it is of great use in the monitoring of military，traffic，security and other fields. Mean Shift algorithm is widely used in the presentmainstream of algorithms，because it has many remarkable advantages as its simplestructure, good real-time system, the remarkable tracking effect and so on. Thispaper mainly studied the algorithm of moving target detection and trackingalgorithm. Considering the environment of the object motion, in this paper wepropose a new method for target detection and tracking. In this algorithm, Backsubtraction algorithm based on statistical average is used to detect target. Besides, totrack the moving target, a newly tracking algorithm is proposed with Camshift andKalman filter. Between the target detection algorithm and target tracking algorithm,we propose a target feature extraction algorithm to connect them.This paper firstly introduces the typical video tracking system architecture andthe treatment method which is based on this architecture. Then we introduce a targetdetection algorithm which is based on back subtraction. In this algorithm, firstly theimage is pre-processed with gauss filter. Then we extract the binary image of themoving target with the back subtraction algorithm. After that，through the analysisof the connected region, the geometrical features and moving trajectory of themoving target are extracted from the binary image. Finally, according to thegeometric parameters, we can get the color images of the targets, which are used asfeature template in the target tracking algorithm. And then we give a detailedintroduction of the mathematical theory of Mean Shift algorithm, put forward thekernel density estimation, color density function, the Mean Shift vector andBhattacharyya coefficient,. Besides，the Mean shift target tracking algorithm isalso introduced. According to color image extracted from the target detectionalgorithm and MeanShift algorithm, was proposed a new algorithm to track themoving target. This algorithm is based on Camshift and Kalman filter. To verifiedthe performance of the method, we processed three videos with the algorithm at theend of the paper.Mean Shift’ algorithm has the features of good real-time system and good Robustness when it faces the partial occlusion and deformation of objects. But whenit faces the fast moving object，the similar color of background and the object andthe partial occlusion of objects.，the algorithm often failure，which can make theobject lost. Besides，search window of traditional Mean Shift algorithm is fixed， itcannot have the function of adaptive adjustment according to the switching size ofobject，so it can make the tracking lost. To solve these problems，this paper choosesthe adaptive search window of Camshift algorithm to deal with the problem aboutthe switching size of object and introduced the Kalman filter to predict the locationof the moving object，so it resolves the problem about too fast moving of the object.Proved by the experiment，the algorithm of this paper preferably finishes the task oftarget tracking in complex circumstance.