This paper presents a target tracking framework applicable to complex crowded scenes with random movements. A robust tracking algorithm using a local sparse appearance model associated with a multi-part color representation is proposed. Sparsity is achieved by solving an l1 regularized least squares problem.Candidates with the smallest projection error is taken as the tracking result. All candidates are drawn based on a density distribution in a Bayesian state inference framework. The target templates are dynamically updated to adapt appearance variation at the end of a tracking iteration. We test the approach on numerous videos including different type of very crowded scenes with serious occlusion and illumination variation. The proposed approach demonstrates excellent performance in comparison with previous methods.