Real-time Tracking Based on Three Images Motion Extraction

摘要

Using three successive images as a process group, not two images as usual, we present an efficient method to extract motion in a complicated environment. Up to now, for real-time motion detection, image subtraction combined with spatial information using two images as a process group is most widely used method. This method, however, not only extracts the true motion, but also produces a false motion caused by occluded background. In order to eliminate the false motion, we propose a new motion extraction technique based on three successive images. Using this method, we implement a real-time monocular tracking system, which robustly track a moving object (walking person) in complicated environments.