3D rotational video stabilization using manifold optimization

We present a novel video stabilization method for cell phone cameras. Our video stabilization is based on a pure 3D rotation motion model, which can better capture the motion of the camera compared with 2D models. 3D camera rotation can be reliably captured by a gyroscope as commonly found on a smart phone or tablet. In this paper we directly smooth the sequence of camera rotation matrices for the video frames. Our contributions are (1) a smoothness metric for a sequence of 3D rotation matrices based on geodesic distance on a non-linear manifold, and (2) an efficient global motion smoothing algorithm using manifold optimization. Our smoothness metric better exploits the manifold structure of sequences of rotation matrices. Experimental results show that our video stabilization method outperforms state-of-the-art methods by generating more stable and visually pleasant videos.