Visual tracking by appearance modeling and sparse representation

Abstract

Appearance variation is a big challenge for object tracking. To deal with this problem, we propose a robust tracking method by online appearance modeling and sparse representation. In this method, we use the intensity matrix of image to represent the object, and learn a low dimensional subspace online to model the object appearance variations during tracking. Then applying the recent theory of sparse representation [1], we construct a likelihood function to measure the similarity between an object candidate and the learned appearance model. After that, tracking is led by the Bayesian inference framework, in which a particle filter is utilized to recursively estimate the object state over time. Theoretic analysis and experiments compared with state-of-the-art methods demonstrate the effectiveness of the proposed algorithm.