Constraint motion filtering for video stabilization

Video stabilization objective is to remove 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 (jitter) caused by unwanted camera motions. The corrective motion, whose magnitude is often limited by the system, is calculated from the observed raw motion by a procedure called motion filtering. In this paper we propose a novel motion filtering approach that takes into consideration the existence of a practical system constraint with respect to the amount of corrective motion that can be applied on each video frame. The proposed filtering procedure extends the Kalman filtering method by incorporating the system constraint in an optimal manner. The experimental results reveal that the proposed approach improves the stabilization performance in the presence of a system constraint, performing significantly better than a trivial incorporation of the system constraint into the stabilization algorithm.