Visual Object Tracking Using Matrix Sparse

Visual object tracking is an important area in the computer vision research direction. It has broad prospect for application in the fields such as military, transportation, financial, medical and entertainment, etc. After years of research, though a lot of tracking algorithms have been proposed, it is still a challenging task to develop an method to fully satisfy the increasing application requirements.Recently, cognitive scientists find that the human brain employing the so-called'sparse coding'strategy to represent the image with only few neurons. Sparse coding has many advantages such as saving energy, not sensitive to environmental changes and the reduction of storage requirements to human brain. The scientists of mathematics and computer vision have studied the concept of sparsity for a long time and have some theoretical results. Using "sparse coding" for visual object tracking which will be a worthy subject.In this paper we propose two robust visual tracking algorithms which employ the sparsity. First is the multi-cue target tracking algorithm based on sparse representation, and second is the multi-cue tracking algorithm based on low-rank matrix. First we cast the tracking problem as finding a sparse approximation in a template subspace using L 1 minimization and multiple features. We extract several features from the image and take candidate targets as a dictionary in the particle filter framework. We get the probability for each particle by L1 minimization and obtain final tracking result based on the weighted multiple features.The second algorithm has the same step as the first, The main framework is constructing a low-rank matrix for tracking. With multiple features, the low-rank matrix theory was introduced into the field of tracking.We conduct extensive experiments on standard video sequence. The experimental result shows that our proposed tracking algorithms are robust. The first algorithm is efficient for visual tracking and can handle the problems such as occlusion and scaling. The tracking algorithm using low rank matrix is a new tracking method, which is better than the first one. In the experiment, we have summed up the essence of sparse representation and low rank tracking. We make adequate comparison the sensitivity of the various features.In addition, this paper systematically introduces and summarizes the thoughts about the sparsity. After reading a mass of extensive technical, we did a careful study on its physiological basis and theoretical research and application meticulous summary. We introduce sparse representation theory and low rank matrix theory, comparing the advantages and disadvantages of various optimization algorithms. Considering the command of object tracking, we adopt L1 minimization and Augmented Lagrange Multiplier (minimization and ALM), which are validated to be very efficient on object tracking in our experiments.