Robust Visual Tracking with Improved Subspace Representation Model

Abstract

This paper is under in-depth investigation due to suspicion of possible plagiarism on a high similarity index

In this paper, we propose a robust visual tracking with an improved subspace representation model. Different from traditional subspace representation model, we use sparse representation, but not the collaborative representation to reconstruct the observation samples, which can avoid the redundant object features in subspace effectively. Moreover, to reject the outliers in the process of tracking, we also propose the combination of sparse box templates and Laplacian residual. To solve the minimization problem of object representation efficiently, a fast numerical algorithm that accelerated proximal gradient (APG) approach is proposed for the Lagrangian function. Finally, experimental results on several challenging video sequences show better performance than LSST and many state-of-the-art trackers.