关键词: 稀疏编码, 贝叶斯推理, 视频目标跟踪, l1-跟踪子

Abstract: The research on sparse coding has become one of the most important fields in the last years, and some researchers introduce it into the issue of video object tracking. Under the framework of Bayesian inference, l1-tracker deals with complex changes of objects in video scenes successfully, the tracking is robust. However, the computation cost of l1-tracker is too expensive to achieve real-time tracking. This paper proposes a fast l1-tracker, and improves three components, including construction of over-complete basis, process method of complex object changes, and update of tracking template. In the proposed algorithm, there is no need to update the object template for tracking. Experimental results show that, compared with original l1-tracker, the proposed algorithm has comparative accuracy and is faster.

Key words: sparse coding, Bayesian inference, video based object tracking, l1-tracking