Online Ensemble of Exemplar-SVMs for Visual Tracking

Abstract

In this paper, we put forward a robust algorithm for visual tracking based on an ensemble of Exemplar-SVM classifiers. First of all, a simple yet effective Exemplar-SVM method originating from object detection is adapted for visual tracking, where the linear SVM classifier is trained using the tracked object as the exemplar and its surroundings as negatives. Secondly, we propose an online ensemble tracker, which integrates a set of Exemplar-SVMs and updates automatically online. Making good use of history information, the proposed algorithm achieves better discrimination of the object and its surrounding background. The experimental results prove that the proposed algorithm is robust and effective.