Title : Visual Object Tracking Using Patch-based Sparse Representation
Abstract :
Sparse representation has been applied to various challenging problems, and it can effectively deal with fastidious artifacts in computer vision such as occlusion, corruption, and illumination changes. In this paper, we propose a patch-based sparse representation model to the robust visual tracking. The patch-based approach can handle the dynamic appearance changes of a tracking target. We model the dynamic appearance of the object by a combination of local rigid patches. The local parts of the target are tracked separately, and the tracked position is estimated by a mean-shift based spatial voting. The integration of the sparse representation and a voting based approach make visual tracking robust to occlusion and illumination variation. We present the experimental result with some challenging video sequences and compared with a state-of-art tracker. We show the proposed tracking system successfully handle the occlusion, noise, scale variance, illumination change, and appearance change of the target.
Abstract :
Sparse representation has been applied to various challenging problems, and it can effectively deal with fastidious artifacts in computer vision such as occlusion, corruption, and illumination changes. In this paper, we propose a patch-based sparse representation model to the robust visual tracking. The patch-based approach can handle the dynamic appearance changes of a tracking target. We model the dynamic appearance of the object by a combination of local rigid patches. The local parts of the target are tracked separately, and the tracked position is estimated by a mean-shift based spatial voting. The integration of the sparse representation and a voting based approach make visual tracking robust to occlusion and illumination variation. We present the experimental result with some challenging video sequences and compared with a state-of-art tracker. We show the proposed tracking system successfully handle the occlusion, noise, scale variance, illumination change, and appearance change of the target.