Object tracking is one of the key techniques in computer vision. Present algorithms are mainly implemented in static platforms. In this paper, we propose a novel technique for real-time object tracking in videos captured by cameras on moving platforms. First, we rule out feature points that have optical flows inconsistent with those of background. Second, optical flows on the rest of the feature points are utilized to estimate the global motion of the camera. Finally, the kinematic function of particle filtering is modified by the global motion of the camera, together with color-space histogram as appearance model, to achieve robustness in unstable video sequences. The proposed algorithm is tested on several video sequences, compared to mean-shift algorithm and traditional particle filtering tracking, it shows promising real-time tracking performance. Experiments demonstrate that our algorithm can track moving object robustly in videos captured by moving cameras.
