Abstract:

Single target tracking is an important part of computer vision, and its robustness is restricted by target occlusion, illumination variation and target scale variation. To deal with these problems, we propose a visual tracking algorithm based on the analysis of the sparse error matrix in low rank projection. In order to overcome the effect of model drifting, target templates are updated dynamically with the similarity between target templates and candidate targets. In the framework of particle filter, the sparse error matrix of candidate targets is obtained by using the theory of robust principal component analysis and low rank projection, and the observation likelihood estimation of the next frame is achieved according to edge and smoothness information. Experimental results on multiple video sequences show that this algorithm has better robustness performance than that of the state-of-the-art tracker.

Key words: visual tracking, sparse error, particle filter, low rank projection, template update