Robust Visual Tracking Based on Adaptive Simultaneous Sparse Representation

Sparse representation has already been successfully applied to the visual tracking, but it also faces with unstable factors. In this paper, we propose an adaptive tracking algorithm based on simultaneous sparse representation which considers the Gaussian noise and Laplace noise. The sparse model is chosen adaptively according to the energy of Laplace noise. The proposed algorithm can better solve the problem of targets such as occlusion, pose change, illumination variation and background clutter, and the algorithm has better robustness. Secondly, the template update method based on the subspace learning and unsupervised learning (K-means) is given to respond the target state in a timely and effective manner and to avoid the template update too fast and introduce significant error. Then, the LASSO algorithm is used to further improve the model. Finally, comparing the current nine state-of-the-art tracking algorithms with the proposed algorithm in this paper, the experimental results illustrate that the proposed algorithm have the better performance in terms of robustness, accuracy and real time.