Robust method of video stabilization

Video stabilization objective is to remove the unwanted motion fluctuations from video data. Typically, this is achieved by applying a certain amount of corrective motion displacement onto each video frame, such that to cancel the effect of high frequency fluctuations caused by unwanted camera motions. In this paper we present a robust video stabilization system comprising three operations: motion estimation, motion filtering, and motion correction. A motion estimator robust to moving objects in the scene is designed in order to identify the camera motion component that follows to be stabilized by the system. A critical role in the quality of the stabilization is then played by the ability of the system to correctly distinguish between the unwanted and the intended motion of the camera. To do this we develop a procedure that extends the Kalman filtering algorithm by incorporating the practical system constraints with respect to the amount of the corrective motion that can be applied on each video frame. The experimental results show the ability of the proposed algorithm to reduce the unwanted motion fluctuations and, at the same time, to follow the user intentional motion. Robustness with respect to large moving objects in the scene, as well as the ability of the proposed method to stabilize in the presence of motion constraints are demonstrated through a series of experiments and comparisons.