A Fast Global Motion Estimation And Infrared Video Stabilization

With the rapid development of video technology, image stabilization is an important content and direction of the research on computer vision and image processing. Global motion estimation is a key part and the most important factor of video stabilization, so the performance of video stabilization is determined largely by the accuracy and efficiency of global motion estimation. Infrared imaging technology can be commonly used in the areas, such as infrared imaging guidance, supervisory system, target tracking and detection, etal, so the study on infrared video processing is a hot research direction.The primary process of video stabilization: global motion estimation, motion filtering, motion compensation. Global motion estimation on IR image sequence is the main content in this paper. The dithering parameters obtained by motion filtering are used to get stable video by the usage of motion compensation. In this thesis, a fast global motion estimation based on RANSAC is proposed. First, the infrared image should be preprocessing before the features got by multi-scale Harris corner detection algorithm, and a part of features on local motion could be excluded by corner threshold. Secondly, the principle of features matching is based on the similarity of the image block which center is corner coordinate, and the error matching points can be removed by the distance between them; Finally, the precise six-parameters are estimated by RANSAC algorithm with preview evaluation model, then the stable image is transformed from the current frame by using dithering vector which obtained by the usage of low-pass smoothing filter. This algorithm can be used to IR image sequence with scaling, rotation, translation and other movements to obtain high-quality, clear and stable infrared video.Finally, a variety of videos are stabilized by many existing algorithms in VC++ platform and MATLAB in order to make comparative analysis.