Study On The Digital Video Stabilization Algorithm Based On Projection

Digital video captured under nature conditions have undesired motion effects commonly and inevitably. Video jitter was known as all kinds of undesired motion between digital image sequences, such as translation, rotation, scaling. The unwanted positional fluctuations of the video sequences influence the visual quality and impede the accuracy of intelligent video surveillance system. As the development of the electronic technology, the digital stabilization system has replaced the traditional mechanical stabilization and optical stabilization systems. An effective digital stabilization system plays a vital role in enhancing the visual quality of digital videos.This paper concluded the current situation and analyzed technical problems to be solved in the digital stabilization field on the basis of reading a number of literatures at home and abroad. Given the research above, the paper is working on the following subjects.(1) The current theory and technology of digital stabilization has been analyzed and contrasted in this paper. The first part tends to summarize characteristics of video jitters, and establish mathematical models according to different types of jitters. The next part introduces and compares kinds of algorithms of translational motion vector estimation, rotational motion vector estimation and some popular motion compensation algorithms respectively. A new compensation component restriction method is proposed in the end of this part.(2) A real-time video stabilization method is proposed to compensate simple translational motion. Firstly, five local motion vectors is estimated by five areas gray projection, secondly, robust global motion vector is selected from the five local motion vectors. Then, calculating jitter components by Kalman filter and to correct the current frame by compensation components.(3) A fast video stabilization method with good precision and robustness is proposed to compensate the combination of translational and rotational motion. The method adopts two steps to estimate motion vector, the first step is to compute the translational motion vector and registing the center of image rotation. The second step is to compute the rotational motion vector by ring projection. Finally, correcting the current frame by Kalman filter and interpolation. Compared with other state-of-the-art algorithms at home or abroad. the experimental results of our methods show that the proposed methods both provide excellent performance at real-time, robustness and precision.