Robust object tracking via constrained online dictionary learning

摘要

Robust object tracking has widespread applications in human motion analysis systems, but it is challenging due to various factors, such as occlusion, illumination variation, and complex backgrounds. In this paper, we present a novel tracking method on the basis of a constrained online dictionary learning algorithm. Some existing tracking methods cannot consider background effects and thus have weak discriminative ability. Moreover, some dictionary learning-based tracking methods directly collect target templates and background templates as positive and negative dictionaries, respectively. The main issue is that the dictionaries cannot effectively represent the target and background and handle appearance changes. Thus, a constrained online dictionary learning algorithm is proposed to obtain a discriminative dictionary, which can ensure that the proposed tracker has good discriminative ability in distinguishing targets from complex backgrounds. Experimental results show that the proposed algorithm performs favorably against other state-of-the-art methods in terms of accuracy and robustness.