Sparsity-constrained and dynamic group structured sparse coding for robust visual tracking

L1 tracker is one of the most effective methods in dealing with the occlusions for sparseness of coding coefficients of objects. However, the existing sparse coding algorithms do not use special sparse structure of coding coefficients in L1 tracker. In this paper, we propose a two-stage sparse coding algorithm for visual tracking based on constrained sparsity of target template coefficients and spatial continuity structure of trivial template coefficients with block coordinate optimization theory. At the first stage, the algorithm solves sparsity-constrained coding coefficients on target template set using orthogonal matching pursuit. At the second stage, the algorithm finds sparse coding coefficients with spatial continuity on trivial template set via dynamic group sparse coding. Robust visual tracking is achieved using the proposed sparse coding algorithm in particle filter framework. The experimental results demonstrate that the proposed tracking method has better robustness and higher precision than the state-of-the-art trackers. ©, 2015, Chinese Institute of Electronics. All right reserved.