A novel video stabilization method with global motion vector prediction and border region recovery

This paper presents a novel approach of vision stabilization, potentially applied to mobile robot or unmanned vehicle, which is inspired by the characteristics of human eye. Vision information processing is a key technology of intelligent mobile platform. However, in some condition, such as the vision module mount on the vehicle moving on the rugged ground, the system could not acquire stable vision, which influence the post-process of image information and make the observer uncomfortable. With video stabilization technology, the unwanted vibrations or shakes of video sequences is smoothed at real time speed. The algorithm proposed is mainly based on digital image stabilization technology, which divides the whole vision in 9 sub-area averagely, then processes searching and matching strategy by macro block with size of 16 ×16 or 8×8 pixels. Both the start point for searching and motion vector are predicted by the motion trend. After the motion compensation procedure, the video is stabilized. During the procedure above, the edge area of the video is lost. In this paper, a recovery method inspired by human vision is discussed, which consider the duration of vision factors. With the evaluating a sequence of unstable images acquired by vision system on the mobile robot, we can draw the conclusion that the effectiveness and validation of the method aforementioned is implemented.