Speeded Up Visual Tracker with Adaptive Template Updating Method

Tracking an object with limited prior information regarding to its appearance is a challenging problem that attracts much attention. In this paper, we propose a speeded up visual tracker that is not only capable of long-term tracking but also of online tasks. The tracker treats object tracking as a binary classification problem between the object and background information. Usually, little information is available for training in real cases, which makes trackers with pre-defined distance metric to drift. To solve this problem, the proposed tracker adopts distance metric learning to update classifier after every frame for a more robust tracking result. We use dense SIFT feature to describe an object appearance and randomized principle component analysis (RPCA) to reduce the original feature space dimensionality. Additionally, a new partially-updated template library is proposed for a more robust tracking. The experiment results show that the proposed tracker performs preferable comparing to state-of-art trackers.