Multi-degree-of-freedom Mean-shift robust tracking algorithm based on SIFT

Mean-shift tracking plays an important role in computer vision applications due to its computational efficiency, ease of implementation and robustness, but fail to track when moving objects have changes in size or shape. To solve this problem, a multi-degree-of-freedom Mean-shift robust tracking algorithm based on SIFT is presented. At first, these obtained SIFT features in current frame are matched with the SIFT features of moving object in previous frame. Then affine transformation between possible object area in current frame and previous frame can be calculated. So bounding box parameters can be estimated through the obtained affine transformation model. The window width of kernel is also updated through the affine transformation. Experimental results show that the proposed algorithm has yielded marked improvement in accuracy of tracked bounding box.