Study On The Methods Of Real-Time Digital Video Stabilization Based On FAST Feature Match

Video captured under nature conditions have undesired motion effects commonly due to the irregular movement of camera carrier or camera. Video jitter was known as all kinds of undesired motion between video frames, such as translation, rotation, scaling. In recent years, Digital image processing technology like face detection, object tracking and object reorganization develop rapidly. But the video jitter will badly affect their processing results. As the development of image process technology and computer technology, digital video stabilization technology is becoming more and more important.The paper is working on the following subjects:(1) The basic frame and mathematical model of digital video stabilization have been first introduced in this paper, and classical algorithms of video stabilization are analyzed. Pros and cons of different kinds of motion vector estimation are summarized later. The motion compensation algorithm-Kalman filter is described in detail in the end of this part.(2) A real-time video stabilization method base on FAST feature match is proposed for videos with jitter of translation and rotation. First of all, extract the FAST feature points with locations at referenced frame and detect the corresponding points at current frame with Pyramidal Lucas-Kanade optical flow. Secondly, remove the mismatched points with forward-backward matching method and RANSAC algorithm. Then, the rigid transformation matrix between two continuous frames is estimated by lease square. At last, calculating scanning motion vector by Kalman filter, and compensate the frame by nearest interpolation. The given method is applied to the real-time video stabilization for videos resolution of856*480successfully. The subjective evaluation of the effect of video stabilization is made and the comparison of the peak signal noise ratio (PSNR) of the original and stabilized videos is also made. The14sets of test video data include7sets of videos captured by author and7sets of classical videos for video stabilization test from Internet.(3)To applied the video stabilization method to mobile operating system. The real-time video stabilization applications based on the method given above are designed and implemented on Android and iOS platform. Videos with resolution of320*240can be stabilized by designed Android and iOS application real-timely. The subjective evaluation of the effect of video stabilization is made and the comparison of the peak signal noise ratio (PSNR) of the original and stabilized videos is also made. The experiments are made on scanning shooting, object tracking shooting, fixed shooting and random shooting videos. The experimental result show excellent performance of robust and precision on the basis of real-time. For videos with large rotation vibration and high-frequency jitter, the proposed method performs well.