Improving object visual tracking performance by scene occluder estimation for video surveillance

In this paper, we propose an approach for improving the object tracking accuracy in video surveillance scenarios by estimating and compensating the occlusion introduced by static scene objects. Specifically, the scene occluder map is first estimated by analyzing the gradient of a normalized cumulative motion map from the frames of the first several minutes of a surveillance video. Then, a scene occlusion compensation approach for Mean Shift tracking is proposed to improve the object tracking accuracy by using the estimated scene occluder map. Experimental results on two public data sets demonstrate the effectiveness of the proposed approach.