Robust Object Tracking Based On Joint Sparse Appearance Model

Visual tracking is a hot topic in computer vision research, and has made great progress in military and civilian applications. Despite the success, to achieve a reliable object tracking still needs to overcome many difficulties, including the appearance change, target partially occlusion, illumination variation, etc.This paper proposes a robust tracking method based on joint sparse appearance model which mainly includes a joint appearance model, an improved motion model and template updating strategy. In order to deal with the target appearance changes, this paper proposes a joint sparse appearance model by combining the sub-region partition method, the template structure rearrangement method and multi-scale partitioning method. In terms of motion model, this paper improves the traditional particle filter based motion model. On the one hand, this paper uses a twice search strategy to improve the efficiency of tracking, on the other hand gives each candidate particle a dynamic weight value to weigh the contribution degree of each particle. In addition, this paper also designs a dynamic dictionary building method and a flexible updating strategy, the target dictionary of each frame in the process of tracking is dynamically generated by a labeled template pool, this labeled template pool consists of many types of target template set, which can accurately depict the latest status of the target, and adopt a flexible mechanism to update the template set, this method can effectively avoid introducing error into template set and to prevent the accumulation of error.Both qualitative and quantitative evaluations on challenging image sequences demonstrate that the proposed tracking algorithm performs favorably against several state-of-the-art methods.