Research On Video Tracking Based On Sparse Decomposition

Object tracking is a popular research area in Computer vision. Now, it is extensively used in intelligent monitoring, intelligent transportation system, finance, biomedicine,computer interaction and so on. Many years later, many researchers propose different tracking algorithms. But for tracking applications, it is still an interesting research in computer vision.Now, sparse representation has been applied to the computer interaction and got many achievements. Specially, the representation model, which is made up by the sparse representation, not only can overcome the influence of noise and cover, but also conquer the light variation in many challenging environments. On this basis, it is a challenging research topic by using sparse representation to track object.Our paper proposes three different improved algorithms, which are based on sparse representation. Through reading many literatures, we focus on the dictionary update, we propose three object tracking algorithms. Based on the paper [22], we use the local overlapped block of object and background to build the dictionary. And we update the dictionary by using the sparseness of candidates’sparse coefficients. Besides, we use the candidates as the object dictionary in which the aim is to reduce the complication of LI computation. Lastly, we update the dictionary through the using of online dictionary learning (ORDL). For the three improved algorithms, we compare with other algorithms and our results can overcome many difficulties very well.In this paper, we use many standard test videos to do our experiments. The experimental results explain that the three improved algorithms are better to against the occlusion, light variation, and blurred object. Meanwhile, our paper analyzes the feasibility of dictionary update in algorithm3and compares the quantitative and qualitative evaluation in same test video by using different algorithms.