Research On The Technology Of Video Stabilization In The Motion State

With the development of information technology and popularization of camera equipments, video has become an important way for people to get information resources due to its rich content, visual expression and easy to obtain, storage and transmission. However, when the camera equipment is in the state of motion, the changing shooting posture and the mechanical vibration and other reasons will make video image instable and blurred, leading to a serious impact on the use of video information. Therefore, the jitter of the video obtained in the motion state is an urgent problem to be solved.The paper focused on the video stabilization in the motion state, especially the links of motion estimation and the reference frame selection strategy. In the motion estimation module, the paper firstly detected the ORB feature points. To solve the problem of local aggregation, the distance constraint method was used to control equable distribution and a moderate number of the extracted feature points. Secondly, the paper selected LSH approximate nearest neighbor search algorithm for feature point matching. Based on the characteristics of R-BRIEF descriptors, LSH used Hamming distance for descriptor similarity measurement. And in view of the problem of error matching and doesn't one-to-one mapping between the left and right matching,this paper used the constraint of the characteristic value and the Hamming distance ratio and the cross filtering to remove the matching point pair which with the poor reliability. Then RANSAC algorithm was adopted to eliminate the error matched points and remove the influence of local motion. To solve the low inliers accuracy of RANSAC in different depths of field and fuzzy image, the paper used PROSAC algorithm to improve inliers and matching accuracy by sorting points and choosing those points which with high quality. In the reference frame selection strategy module, in order to solve the problem of frame jump caused by the traditional selection strategy, a reference frame update strategy based on the cumulative transformation is presented in this paper. By this method, the cumulative global motion parameter of the current frame with respect to the first frame is obtained, and the compensation standard is unified. Aiming at the problem of cumulative error in the process, the change of the reference frame is adjusted by setting the matching threshold, and the number of parameters is reduced. In addition, the paper also studies the motion filter and motion compensation. We separated the global motion parameters into intention motion and random jitter through Kalman filtering, and finally compensate the jitter through bilinear interpolation method. After the steps above, we could finally get the smooth and stable output video.Finally, we created a set of electronic image stabilization system through VS2010 and used a large number of videos to test the effect of stabilization. The experimental results showed that the algorithm can quickly and effectively realize the electronic image stabilization in the motion state.