Real-time video stabilization based on vibration compensation by using feature block

Based on feature-block (FB) basis, this paper presents a real-time video stabilization method by using motion compensation for repairing unstable frames captured by a vibrating camera. The basic strategy is first to check if the frame is vibrant and then select the feature block to calculate the vibration vector for compensating the vibration area. The feature block, which is selected from nine fixed observation blocks (FOBs), has certain features and hence can avoid interference of any moving-object intrusion. With the vibration vector, the vibration area can be compensated in order to adjust the vibrating frames to be stable. Experimental results show that the proposed algorithm can effectively stabilize the video frames with moderate vibration (smaller than 20 pixels) in real-time. Besides, a comparison with four reported methods demonstrates that the proposed video stabilization method can cope with the problem of moving-object interference and has less computation time. From the comparison, it also manifests the attractiveness of the proposed approach, when compared with other reported methods.