Chao Jia (The University of Texas at Austin), “Video Stabilization and Rectification for Handheld Cameras” (2014)

Chao Jia (The University of Texas at Austin), “Video Stabilization and Rectification for Handheld Cameras”, Advisor: Prof. Brian L. Evans (2014)

Video acquired by smart phones and other handheld cameras suffers from interframe jitter due to user hand shake and walking, and from rolling shutter effects due to fast relative motion between the camera and scene. In this dissertation, the author proposed to fuse gyroscope and visual data for stabilization of interfame jitter and rectification of rolling shutter artifacts. Digital video stabilization and rectification estimated camera motion, regenerated camera motion, and synthesized new frames. For estimating camera motion, the author proposed an online camera-gyroscope calibration method while a user is capturing video. The method used an implicit extended Kalman filter and a rolling shutter camera model. Camera motion estimates were smoothed by exploiting the manifold structure of 3D rotation matrices. Offline motion smoothing was formulated as a geodesic-convex regression problem and solved by manifold optimization methods. Online motion smoothing was achieved by recursive filtering or sequential probabilistic estimation. All proposed methods were constrained so that no black borders intrude into the stabilized frames.

For details, please contact the author or visit the thesis page.