Robust Visual Tracking with Incremental Subspace Learning Sparse Model

摘要

Sparse representation based trackers have achieved impressive tracking performance in recent years, the utilization of trivial templates could help to improve the trackers' performance when partial occlusion occurs. In this paper, we propose a novel incremental subspace learning sparse model for robust visual tracking. The proposed model collaboratively exploits the advantages of both sparse representation and the incremental subspace learning by modeling reconstruction errors caused by sparse representation and the eigen subspace representation simultaneously. We also propose a customized APG method for solving the optimization solution. In addition, a robust observation likelihood metric is proposed. Both qualitative and quantitative evaluations over challenging sequences demonstrate that our tracker performs favorably against several state-of-the-art trackers. Furthermore, we indicate the drawbacks of our tracker and analyze the underlying problem.