Visual tracking with multiple representative models based on sparse prototypes

Abstract

Online visual tracking plays a critical role in research and application of computer vision, and it is still a challenging task to alleviate the possibility of drift. In this paper, a robust visual tracker is proposed with multiple representative appearance models based on sparse prototypes. Benefitting from the representation with sparse prototypes, the multiple representative appearance models maintain representative and discriminative features of the target appearance. The multiple models are triggered to recognize the target in challenging cases with an effective strategy, which is demonstrated by the extensive experiments.