As a fundamental research topic of computer vision, background subtraction technology can be implemented in many applications. This problem becomes even more challenging once the videos are obtained with a moving camera. To solve the challenging task, we analyze the principal motion of pixels for subtracting background in videos obtained from freely moving cameras. In particular, the optical flow is captured for the representation of motion for pixels. Moreover, assuming that the background motion should be the principal part, Robust Principal Components Analysis (RPCA) is utilized for subtracting the background of videos obtained from freely moving cameras, in which both angle and magnitude of the optical flow are utilized for the analysis of motion. Finally, super-pixels are utilized to compensate for the defects produced by inaccuracies in optical flow, where a double threshold strategy is proposed to integrate the foreground results captured from the angle as well as the magnitude. Experiments based on several videos from standard benchmark datasets illustrate that the proposed method achieves promising performance compared to the state-of-the-art.
展开▼
