Research On Video Stabilization Technology

With the rapid development of electronic technology, handheld imaging devices, such as mobile phones and digital cameras, have become much popular in people’s everyday life. However, when people are shooting video, the video quality is not always satisfactory because of the video jitter caused by hand shake or external interference without using specialized camera stand equipment. Thus, it is necessary to develop video stabilization technique to improve the quality of jittered video, which will provide users with comfortable video viewing experience.The research of this thesis mainly focuses on the development of video stabilization method, and several contributions are as follows:1. A real-time video stabilization algorithm based on hierarchical matching is proposed, which can well deal with the jittered video with small rotation and panning. By partitioning the current frame and the reference frame into multiple blocks, block matching with low computational complexity is first implemented to seek the affine transformation parameters between two frames. Furthermore, with these affine transformation parameters, the motion compensation for the current frame is carried out to obtain a stable video.2. For the jittered video caused by the pull-push of lens, we present a local feature matching based stabilization method. To remove the impact pf noise, RANSAC is used to improve the accuracy of SURF feature matching. With the refined pairs of features, the perspective projection transform is implemented to eliminate the impact of pull-push of lens, which will effectively improve the video quality.3. Considering the case of dramatic scene change, we present a novel video stabilization method. By using image mosaic technique, a wide-angle view map is first generated. Meanwhile, the optical flow tracking technique is adopted to obtain the actual camera motion path on the generated map. Furthermore, we can get a corresponding virtual camera motion path by smoothing processing on the actual camera motion path, based on which a synthesized stable video can be obtained.4. Based on the development platform of VS2010, a prototype system for video stabilization is built.