Multi-cue based tracking

Visual tracking is a central topic in computer vision. However, the accurate localization of target object in extreme conditions (such as occlusion, scaling, illumination change, and shape transformation) still remains a challenge. In this paper, we explore utilizing multi-cue information to ensure a robust tracking. Optical flow, color and depth clues are simultaneously incorporated in our framework. The optical flow can get a rough estimation of the target location. Then the part-based structure is adopted to establish the precise position, combining both color and depth statistics. In order to validate the robustness of the proposed method, we take four video sequences of different demanding situations and compare our method with five competitive ones representing state of the arts. Experiments prove the effectiveness of the proposed method.