Robust object tracking using a spatial pyramid heat kernel structural information representation

In this paper, we propose an object tracking framework based on a spatial pyramid heat kernel structural information representation. In the tracking framework, we take advantage of heat kernel structural information (HKSI) matrices to represent object appearance, because HKSI matrices perform well in characterizing the edge flow (or structural) information on the object appearance graph. To further capture the multi-level spatial layout information of the HKSI matrices, a spatial pyramid division strategy is adopted. Then, multi-scale HKSI subspace analysis is applied to each spatial pyramid grid at different levels. As a result, several grid-specific HKSI subspace models are obtained and updated by the incremental PCA algorithm. Based on the multi-scale grid-specific HKSI subspace models, we propose a tracking framework using a particle filter to propagate sample distributions over time. Theoretical analysis and experimental evaluations demonstrate the effectiveness of the proposed tracking framework.