Sparse Coding And Counting For Robust Visual Tracking

Visual tracking plays an important role in computer vision and has many applications such as video surveillance, robotics, motion analysis and human computer interaction. Even though various algorithms have come out, it is still a challenge problem due to complex object motion, heavy occlusion, illumination change and background clutter. In this paper, we propose a novel sparse coding and counting method for visual tracking to address the aforementioned problem.In subspace representation, the object reconstruction will still have some errors when the object under occlusion or background clutter, but the sparsity constraint enables the tracker to effectively handle difficult challenge. As we know, Bayesian framework has been successfully applied to select variables by enforcing appropriate priors. Laplace priors were used to enforce sparsity in sparse linear model, however, they may cause over-penalization. To further enforce sparsity and reduce over-penalization, each coefficient is assigned with a Bernoulli variable. Therefore, a novel sparse coding and counting model interpreted from a Bayesian perspective by carrying maximum a posteriori (MAP) is proposed, which turns out to be a combination of Lo and L\norm to regularize the linear coefficients of incrementally updated linear basis. To achieve realtime processing, we propose a fast and efficient numerical algorithm for solving the proposed model. Although it is an NP-hard problem, the proposed accelerated proximal gradient (APG) approach is guaranteed to converge to a solution quickly. At the same time, we proposed a dictionary reinitialization method to deal with tracking drift, which can update the dictionary when error exceeds the threshold and track again. Besides, we provide a closed solution of com-bining Lq and L] regularized representation to obtain better sparsity. To evaluate the performance of our tracking algorithm, we perform our algorithm on challenging vedio sequences where the target object undergo large changes in heavy occlusion, fast motion, illumination, pose and s-cale, cluttered background. Experimental results on challenging video sequences demonstrate that the proposed method achieves state-of-the-art results both in accuracy and speed.