Correlation Filter Tracking via Global and Adaptive Local Parts

Abstract

Owing to the excellent performance and high efficiency, correlation filters have attracted attention in visual tracking recently. However, they often lead to tracking failures because of the high sensitivity to occlusion. Part-based tracking methods can deal with partial occlusions to some extent, but the ability of these methods to cope with other challenges, such as scale variations, or low resolution, is not robust enough. In this work, we propose a novel correlation filter-based tracking approach via global and adaptive local parts to address this issue. Specifically, the object is divided adaptively according to its appearance at first. For local parts tracking, coarse position and scale factor are then obtained by employing the spatial-temporal information of reliable parts. Finally, global tracking is performed with the rough position to localize the object more accurately. Experiments illustrate that the proposed tracker outperforms several state-of-the-art methods.