Correlation Tracking with Correcting and Adaptive Update Strategy

Abstract

Object tracking plays an important role in the research field of computer vision. Correlation filter (CF) based tracking algorithms have shown remarkable performance recently. However, there are two problems: (1) the online model is prone to drift due to the fixed coefficient update strategy; (2) a tracking error is susceptible to lead the failure of the following tracking task due to the absence of a correcting strategy. To deal with these limitations, we proposed a new correlation tracking filter that includes an adaptive update strategy and a correcting strategy. The adaptive update strategy is based on the confident degree of the tracking result, which can minimize the effect of image noise. And the correcting strategy is based on a four-level classifier that can enhance the error correcting ability. Based on these two strategies, the proposed CCAS not only can improve the accuracy of the correlation tracking, but also can build a detector with strong error correcting ability to handle a variety of challenges. Experiments show the proposed method has the effective correcting ability and can resist model drift. It is noted that the proposed algorithm not only outperforms other state-of-art algorithms but also runs enough fast for real time application.