Hybrid discriminative visual object tracking with confidence fusion for robotics applications

In this paper, we propose a hybrid visual tracking algorithm that combines two discriminative trackers. Discriminative trackers treat tracking as a classification problem, that is, they try to distinguish targets from backgrounds, and usually the trackers incorporate classifiers. The two trackers collectively determine the new object location via a process called confidence fusion. The two trackers are aimed to complement the ability of discrimination. To achieve this goal, one tracker extracts image features pixel by pixel, and the other extracts image features over several rectangular regions. In addition, the corresponding classifiers are trained using different learning algorithms. We not only model object tracking as a binary classification problem but also model it as a three-class classification problem. On-line learning algorithms are used to update the classifiers during tracking so that the trackers are adaptive to the variations of the appearance of the target. A set of rules including tracker switching and confidence fusion is devised to synthesize the two trackers. The experimental results show that our approach is competitive with other popular tracking algorithms.