Robust Object Tracking Via Weight-Based Local Sparse Appearance Model

Zhiyong Li,Dongming Wang,Ke Nai,Tong Shen,Ying Zeng

DOI: https://doi.org/10.1109/FSKD.2016.7603234

2016-01-01

Abstract:Developing an effective and efficient appearance model for robust visual tracking is difficult because of various interfering factors, such as postural change, occlusion, and rapid motion. More and more visual tracking methods tend to exploit local sparse appearance model to deal with the above problems. Since in the local sparse appearance model, all individual patch together to form a complete target which mean each patch can describe some extent of the target appearance information. In this work, we propose a simple yet effective tracking method to exploit the important patches through weights operating which can be tracked effectively through the entire tracking sequences. Moreover, in order to further improves the robust of tracking method, temporal information was used in our tracker. Finally, we incorporate weight-based local sparse appearance model and temporal information into the Bayesian inference tracking framework. Experimental result on 2015 PAMI tracking benchmark dataset shows that the propose method achieves approving performance than alternatives reported in the recent literature.