A new approach to video stabilization with iterative smoothing

A new approach to video stabilization is proposed in this paper. Firstly, SURF (Speeded-Up Robust Features) is employed to extract feature points from input frames. Then, the local feature matching strategy is presented to speed up the global motion estimation procedure. Secondly, a weak Gaussian smoother is constructed to smooth the camera motion iteratively. In order to make the weak smoother adapt to different videos, the acceleration of feature trajectory is calculated to control the iteration of smoothing. Finally, a series of comparisons with other methods, such as spline interpolation, Kalman filter, are carried out to demonstrate the well performance of the proposed approach, which also depend less on the selection of parameters.