Visual Tracking via Locality-constrained Affine Subspace Coding

The dictionary plays an important role in the visual tracking whose sample is encoded by a dictionary. However, the dictionary atom is a representative point (visual word) in the previous tracking algorithms. Representative point neglects the geometric manifold structure which the point surrounded by immediately, which leads to limited ability of characterizing the distribution of features on the data manifold. In this paper, we propose a novel visual tracking algorithm with Locality-constrained Affine Subspace Coding (LASC) which explicitly explores the geometric structure of the manifold where these representative points lie. Unlike other encoding methods, LASC formulates the dictionary atom as the subspace attached to the representative point. To further improve the discrimination and diversity of the dictionary, we introduce the weighted spatial information to the representative point. We conduct experiments on challenging benchmarks and the experimental results demonstrate the effectiveness of our method.