Research On Object Tracking Algorithm Based On Sparse Representation

Visual tracking is a comprehensive interdisciplinary subject which integrates image processing, pattern recognition, image database, computer graphics, and robot vision and so on. As the key technology of the computer vision, it has been widely used in military, industrial, civilian and other fields.When appearance variation of object, partial occlusion or illumination change in object images occurs, most existing tracking approaches cannot meet the requirement of the robustness and accuracy. Used the visual property of sparse representation, this paper introduced the sparse representation theory into particle filter framework to reduce the influence of target appearance change, and proposed a robust visual tracking method based on sparse representation. The following is done in this paper:Firstly, this paper introduced and summarized the sparse decomposition algorithms. The paper divided all those algorithms into two categories as greedy algorithm and convex relaxation algorithm. In this paper,Orthogonal matching pursuit (OMP) and block orthogonal matching pursuit(BOMP) which belong to greedy algorithm are mainly studied. Basis pursuit(BP) which is a convex relaxation algorithm is introduced to compare with OMP and BOMP from running time, reconstruction error and reconstruction quality. Simulation results show that BOMP is faster than OMP and BP in running time, BP has the best reconstruction quality, BOMP comes second. In running time and reconstruction performance, BOMP achieves a balance. So BOMP is more suitable for solving sparse representation model.Secondly, when appearance variation of object, partial occlusion or illumination change in object images occurs, most existing tracking approaches fail to track the target effectively. To deal with the problem, this paper proposed a robust video tracking method based on appearance modeling and sparse representation. The proposed method exploits two-dimensional principal component analysis (2DPCA) with sparse representation theory for constructing appearance model. Then tracking is achieved by Bayesian inference framework, in which a particle filter is applied to evaluate the target state sequentially over time. In addition, to make the observation model more robust, the incremental learning algorithm is used to update the template set. Both qualitative and quantitative evaluations on four publicly available benchmark video sequences demonstrate that the proposed visual tracking algorithm performs better than several state-of-the-art algorithms.