Research On Object Tracking Algorithm Based On Sparse Representation

Object tracking technology, an important research subject in computer vision technology, involving image processing, pattern recognition and other subjects, is widely used in video monitoring, military and medical field, transportation, machine vision,etc. The main purpose of object tracking is to extract and distinguish the object interesting, so as to obtain its position, velocity and other information. Because of the existence of all kinds of interference factors in real life, such as occlusion, object rotation, changing illumination and scale, video tracking algorithm has to be faced with a great challenge on its accuracy and robustness.In this paper, we review and summarize some popular tracking algorithms and carry out a detailed research on object tracking algorithm based on sparse representation. The main parts of this paper is as follows:(1) Analyze the problems of object tracking algorithm. After that, propose an object tracking algorithm based on sparse representation. With this algorithm, construct a discriminative model by making the subtraction of sorted grouping, then distinguish the target object and background according to the likelihood of the candidates. At the same time, computational complexity is reduced by utilizing 2norm models to solve the coefficient vector of signal’s sparse representation.(2) Study the template updating method in the tracking algorithm. By combining incremental principal component analysis and sparse representation, the method used in template updating can effectively solve the problem of target drifting and scale changes.(3) Analyze the discriminative and generative model based on sparse representation. This paper uses logistic regression function to select features and construct the sparse representation of signal. After that, use sparsity concentration index to select samples which can be used to construct discriminative model. At the same time, calculate the equal weighted sum of coefficient vectors of sparse representation and use reconstructed error of partial image block to handle the problem of partial occlusion. Lastly, generation model is built by convergence feature which is obtained through alignment pooling methods.(4) Analyze target tracking algorithm which is based on single discriminative model and generative model. By combing discriminative and generative model, propose a tracking algorithm using hybrid model based on sparse representation. Simulation experiments demonstrate our algorithm’s high accuracy and robustness against occlusion, rotation, motion blur and scale variation.