Feature-fusion based object tracking for robot platforms

Purpose - Object tracking has been a challenging problem of robot vision over the decades, which plays a key role in a wide spectrum of visual tracking-related applications such as surveillance, visual servoing, sensing and navigation in robotics, video compression. The purpose of this paper is to present a novel intensity, orientation codes and geometry (IOCG) histogram variant of the mean-shift algorithm for object tracking. Design/methodology/approach - Feature cues of intensity, orientation codes and geometric information are fused together to form an IOCG histogram in combination with a conventional mean-shift-based tracking algorithm. Findings - Experimental results demonstrate the effectiveness and efficiency of the proposed method. Not only do fusing orientation codes features allow the proposed algorithm to conduct tracking in a cluttered background, but partial occlusion is also solved in the tracker in that spatial information usually lost in a conventional histogram is compensated by the introduced geometric relations between tracked pixels and the center of a tracker template. Originality/value - The paper presents a novel vision tracking method for robots.