Robust Visual Tracking with Combined Norm Regularized Sparse Coding and Adaptive Weighted Residual

In the visual tracking algorithm based on sparse representation,the objective function which used L0 penalty often resulted in NP-hard problem and which used L1 penalty was easy to cause biased estimation.This paper proposes a robust visual tracking method with combined norm regularized sparse coding under the Bayesian inference framework.First,we regularize the coding coefficient with the combined L0 and L1 penalty,and give different parameters values according to their contribution to enhance the robustness of visual tracking.Second,we introduce an adaptive weighted value which considered in the objective function to handle the occlusion,illumination and corrosion problems.Finally,since the minimization of objective function is a non-convex optimization ill-posed problem,we propose a generalized thresholding method under the accelerated proximal gradient method framework to obtain the optimal solution iteratively.Experiments on multiple challenging sequences demonstrate that our tracking method can be much more robust compared with other state-of-the-art methods.