Target Tracking based on Dynamic Model Updating and Result Fusing

• 1、College of Computer Science

Abstract：The result of single tracker can sometimes be unreliable in some situation like illumination variation, occlusion, object size change, etc. The ensemble approach can combine the advantages of difference models and overcome this limitation, improving the performance of visual tracking. In order to better ensemble the different trackers' result we propose a adaptively assemble method that can choice the weight of each track result by analyzing the fluctuation degree of response map. Moreover, we propose a adaptively update strategy to avoid the "drift" in tracking which result by unsuitable model update. In this paper, we expanded Staple using our methods and evaluate the performance on famous object tracking benchmark OTB. After experience, the proposed method can still outperforms effective in the situation of cluttered background, illumination variation, out-of-view, deformation, etc .

Keywords： object tracking result fusion model update