Object tracking based on improved MeanShift and SIFT

Traditional color histogram MeanShift (MS) algorithm only considered object´s color statistical information, and didn´t contain object´s space information, so when the object color was close to the background color, or the object´s illumination changed, the traditional MS algorithm easily caused object tracking inaccurately or lost. Aimmed at this issue, a novel tracking algorithm which fused improved MS and SIFT was proposed in this paper. Firstly, the improved MS algorithm got initial tracking results, which determined block method by the size of the lastest enclosing rectangle and determined its weight coefficient by the Bhattacharyya coefficient of each block. After obtained the tracking result with MS, then utilized SIFT to refine it. Finally, the proposed algorithm used linear weighted method to fuse the improved MS´s tracking result and SIFT´s. Experimental results show that the proposed method which fused improved MS and SIFT deals with occlusion, rotation and illumination change successfully.