Video Stabilization by Multi-Trajectory Mapping and Smoothing

To remove visually annoying vibrations caused by jittery camera from video is a very challenging issue. This paper proposed a video stabilization approach based on multi-trajectory mapping and smoothing. First, the estimated affine parameters in consecutive frames are mapped to a number of trajectories. Then, the trajectories are smoothed by the Bezier curve fitting and an improved B-spline curve fitting respectively. Finally, a set of new affine parameters are estimated from the smoothed trajectories. With the smoothed affine parameters, the video is reconstructed by frame wrapping, while the undefined regions after frame wrapping are filled by the corresponding pixels in neighboring frames. The effectiveness and robustness of the proposed method have been demonstrated by comparison experiments