摘要
Multiple image fragments have been used to represent the target for tracking in a video sequence. It is proved to be able to maintain spatial information of the target. In this paper, following the idea that represents the target with multiple image fragments, we propose a framework that can efficiently combine multiple spatially distributed fragment histograms for robust tracking. The framework ranks the importance of each fragment adaptively, which can increase the robustness to partial occlusions and pose variation. We derive a mean shift type algorithm for the framework that allows efficient target tracking with very low computational overhead. Extensive experiments on challenging real video sequences clearly demonstrate the benefits of our tracker.
- 出版日期2008
- 单位上海交通大学