A video stabilization algorithm based on affine SIFT

Electronic image stabilization (EIS) is an important video enhancement technology which aims at removing annoying random jitter from videos. At present, the commonly EIS algorithm is to filter the feature trajectory, such as Kalman filter and Median filter. However, the filtered trajectories of these methods are greatly deviated from the original trajectory, and the images often lose large information after being stabilized. This paper uses affine SIFT (ASIFT) feature matching method to get the best estimating the affine matrix, then Gaussian low-pass filtering of the original path can compensate for the motion of the smoothed path, then the jitter frame is stabilized. Compared with Kalman filter, the experiments show that Gaussian filter better retains the camera active motion, stabilized video sequences lost less pixel information.