Research On Video Stabilization Algorithm Based On Feature Tracking

With the rise of self-media,recording life with mobile smart devices has become a mainstream way.Although some mobile phones appearing on the market today have anti-shake functions,in the process of shooting short videos on handheld mobile devices,due to the interference of the external environment,the video shot still has a phenomenon of jitter.As a result,video stabilization technology has emerged.The purpose of video stabilization technology is mainly to eliminate or reduce video jitter and generate stable video.The difficulty is that the generated image stabilization video frame contains black borders and the original jitter video has large foreground motion.For the black border problem,the current cropping method is used to remove it,but it will lead to the lack of video information and reduced resolution.The problem is usually solved based on the background recognition method,but it will make the image stabilization time longer.This article focuses on image stabilization of different types of jitter videos captured on handheld mobile devices,especially for original jitter videos with foreground motion.The main work is as follows:(1)A video stabilization algorithm for multi-grid path motion is proposed.The feature points of the video frames are extracted through the SIFT algorithm,the feature points are tracked using the KLT algorithm,the affine transformation matrix between adjacent frames is estimated using the RANSAC algorithm,the video frames are divided into uniform grids,and the transformation matrix of the adjacent video frames.It is applied to a uniform grid to calculate the original motion trajectory of the camera,and then to optimize and smooth multiple grid paths by minimizing the energy function.Finally,the relationship between the original camera path and the smooth camera path is used to calculate the compensation matrix between adjacent frames,and the compensation matrix is used to geometrically transform each frame to obtain a stable video.Experiments show that this method has good image stabilization effect.However,through cropping rate,it is found that if there is foreground motion in the original jitter video,the image stabilization difficulty will become greater,resulting in the lack of information content of the final image stabilization video.(2)Aiming at six types of original jittery video with foreground motion captured by handheld mobile devices,a video stabilization algorithm combining optical flow method and Kalman filtering is proposed.First,pre-stabilize the jitter video by optical flow method to reduce the video jitter component,which reduces the amount of subsequent calculation,and then use the idea of feature tracking in the motion estimation section to perform Shi-Tomasi corner detection on the pre-stabilized video frame generated by it.And track the corners,estimate the affine transformation matrix between adjacent frames,from the affine transformation matrix between adjacent frames,solve the homography transformation matrix between pre-stabilized video frames,calculate the original camera path,and then pass The Kalman filter optimizes the camera path to obtain a smooth camera path,and finally finds the amount of compensation for each frame of the video,and geometrically transforms the video frames one by one,thereby obtaining a stable video output.Experiments show that the final image stabilization effect of this method is better.