A fast video stabilization method is presented, which consists of sub-image phase correlation based global motion estimation, Kalman filtering based motion smoothing and motion modification based compensation. Global motion is decided using phase correlation in four sub-images. Then, the motion vectors are accumulated to be Kalman filtered for smoothing. The ordinal motion compensation is applied to each frame with modification to prevent error propagation. Experimental results show that this stabilization system can remove unwanted translational jitter of video sequences and follow intentional scan at real-time speed.