When “3D reconstruction is difficult or long feature trajectories are not available,” video stabilization is quite a challenge. This paper introduces a “robust, efficient, and streamable technique” for video stabilization that can handle those difficult cases.

The authors’ technique works on general videos. The idea is to smooth the trajectories that are integrated from the features. In this paper, the smoothing is represented as a global optimization problem by formulating some energy items. They use Bézier representation to guarantee the smoothness of each feature trajectory, which helps to reduce the complexity of the optimization problem. The optimized feature positions are computed and used to warp the video frames, forming a final result. Extensive experiments were conducted to evaluate the method, which show that the proposed technique “achieves high-quality camera motion on a variety of challenging videos” and outperforms other existing methods.

The main contribution of this paper is the presentation of a new method for video stabilization that accounts for how “the tracked features are spatially and temporally correlated.” One limitation of this method, as the authors point out in the paper, is that the algorithm “is not effective if there are no background features in some frames.”