Sparse Representation and Particle Filter Based Online Object Tracking Algorithm

Changes of object appearance and occlusion always lead to tracking performance degradation or drift problem during tracking. To solve these problems, a sparse representation and particles filter based online tracking algorithm was proposed which used pyramid histogram of oriented gradients (PHOG) to describe the object template. In the framework, the candidate template can be represented by object templates and trivial templates sparsely, and then L1 minimization was exploited to find the optimal solution. To ensure the accuracy of tracking, the object function was divided into two parts to model the occlusion part and non-occlusion part separately. Both qualitative and quantitative evaluations on challenging image sequences demonstrate that the proposed algorithm performs favorably against several state-of-the-art methods.