A fast video stabilization algorithm with unexpected motion prediction strategy

The vision systems mounted on mobile robots, which are moving on the rugged environment or enduring mechanical vibration, usually suffer from instable video during acquisition. In this paper, a fast video stabilization technique for robot application is presented. The proposed approach takes full advantage of intrinsic characteristics of the unwanted camera motion. There are four main steps involved in the proposed approach: feature extraction, motion estimation, motion compensation and undefined region mosaicing. We began with extracting the global features in 9 sub-regions of the current frame. The blocks with extracted features are matched in the reference frame from predicted start position. The computed motion consists of both expected and unexpected components. They are distinguished by motion smooth process and followed by motion compensation. As the result of frame shifting, some border regions are undefined. To solve this problem, inter-frames data is taken in consideration to fill those regions. A series of experiments with vivid data have been conducted. The results have revealed the effectiveness of the proposed technique.