Robust Visual Tracking Via Part-Based Template Matching with Low-Rank Regulation

Abstract

This paper presents a simple yet effective visual tracking method to attack the challenge when the target object undergoes partial or even full occlusion. First, a fixed number of image patches are sampled as the template set around current object location. In the detection stage, candidate image patches are sampled as the candidate set around the object location in the previous frame. Second, both the template set and candidate set patches are divided into sub-regions and features can be efficiently extracted via random projections. The confidence score for a specific candidate patch is computed through compressive features’ low-rank regulation with the template set patches. The lowest confidence score in the current frame indicates the new object location. The encouraging experimental results show that our proposed method outperforms several state-of-the-art algorithms, especially when the target object suffers partial or even full occlusion.