Ee368 Project Proposal Fast Video Stabilization for Hand-held Devices

Although modern hand-held devices, such as digital camcorders and smartphones, have enabled everyday users to capture high-quality video sequences, the quality of hand-held videos often suffers from unavoidable shaky camera motion. Video stabilization aims to remove this annoying hand-shaking motion from a video. Existing video stabilization algorithms [1, 2, 3, 4, 5] involve three major stages: camera motion estimation, motion smoothing, and image synthesis. By applying a 2-D affine or projective transformation, the 2-D video stabilization [1, 6] techniques can effectively remove small camera jitters and only require a low-complexity implementation. However, they fail to model parallax induced by a translational shift in a viewpoint. Li and Jin [4] considered a 3-D video stabilization approach by stabilizing the camera view along a 3-D trajectory path. Furthermore, dynamic content are preserved through shape-preserving and content-aware image warping. 3-D video stabilization generally produces a better visual result compared to traditional 2-D stabilization, but comes with an expensive computational cost.