: Object tracking is an important task within the field of computer vision, which still attracts many researchers’ attention because of unsolved problems. The aimoftracking is to locate a moving object or several ones in thevideoframesbased on theknowledge oftheir locationsinthe firstframe. It is wildly used in many applications, such as video surveillance, human-computerinterface, intelligentrobots, militaryand etc. The meanshifttrackingalgorithmis oneof the most populartrackingalgorithms. It is a local search algorithm based on color histogram matching. Bhattacharyya coefficient is used as a similarity function to measure the similarity between the object model and the object candidate. This tracker have some advantages such asreal-timecapability, robustness andsimpleimplementation. However, it is very sensitive to influence of background information, illumination changes and occlusions. Therefore, the applicability of tracker is limited in many real world complex conditions. In orderto deal with theseproblems, inthisthesis, we firstextract the suboptimalcolors of targets regions by using theK-means clusteringalgorithm and then we use the corrected background weighted histogram (CBWH) method to reduce the interference of background intarget localization. To deal with thepartialand completeocclusion, the Forward-Backward theoryandthe Kalman filter areused. Experimental results on various videos verify that the proposed method has betterperformancethan the other recent methods. Keywords: mean shift tracking, colorhistogram, K-means clusteringalgorithm, Kalman filter
