Real-time video stabilization based on fast block matching and improved Kalman filter

Abstract

Mobile devices become widely used as their easily handling whereas unpleasant shaking effect comes along inevitably in the recorded sequences. This paper presents a robust real-time video stabilization algorithm applying fast block matching and an improved kalman filter. Key block picking and hierarchical searching improve the efficiency of the motion estimation section to make the method available for real-time application, meanwhile the proposed improved kalman filter outstands by preserving the initiative camera scanning while removing unwanted jitters properly at the same time adaptively. Theoretical analysis and practical experimental results demonstrate the effectiveness and robustness of the proposed method even if large scale of panning motion effect exists.