Rapid Visual Tracking with Modified On-Line Boosting and Template Matching

Guibo Luo, Yuesheng Zhu, and Qing Zhang


Visual tracking, on-line boosting, template matching


On-line learning is increasingly popular in visual tracking, but the challenge that it faced is how to adapt the appearance changes and avoid the drifting or missing track. In this paper, a fast visual tracking algorithm is proposed to make the tracker more accurate and stable in the complex variations situations like occlusions, illuminations and shape deformations. In the proposed algorithm, a modified on-line boosting method is developed to make the tracker more adaptive to variable scene and a template matching model is used to constrain the training samples, so that the accumulating errors in self-update learning can be alleviated effectively. In addition, an optimization process is used to reduce the computational burden. Our experimental results have demonstrated that compared with other on-line tracking methods the target can be accurately tracked with lower drifting error in the complicated environments by using the proposed algorithm. Moreover, the new tracker runs at 60 frames per second, and is suitable for the real-time catching and tracking.

Important Links:

Go Back