Multi-cue Visual Tracking Based on Sparse Representation

Abstract

Under dynamic and complex environment, the single feature methods usually can’t distinguish the target from background well, so that multiple features are considered in the paper. For each candidate, multiple features are extracted and conducted the sparse representation respectively, then observation probability is calculated by combinating reconstruction errors of multiple features in particle filter framework. Comparing with single feature method, the proposed method performed robust with better accuracy. And further experiments on some representative image sequences showed that the proposed method also performs well in complex scenarios, such as varying illumination, background clutter, and occlusion.