A novel visual tracking with occlusion detection via sparse coefficient analysis

In recently years, visual tracking has achieved great development in the field of computer vision. But occlusion remains a challenging problem. Though sparse representation has been introduced into visual tracking, most of existing visual tracking methods based sparse representation treat the occlusion challenges as one of the special scenes simply, and did not make full use of sparse coefficient. In this paper, a novel occlusion detection via sparse analysis is proposed. We can judge whether the occlusion is happening and determine the definite occlusion area in current frame. And the detection result is introduced into the process of visual tracking in order to exclude the influence of occluding area of target object. In addition, we put forward a novel template update strategy. Both of these strategies collectively help the tracker to reduce the probability of drifting. Experimental results on a series of challenging image sequences demonstrate that the proposed visual tracking method achieves more favorable performance than other state-of-the-art tracking methods.