A system and method are disclosed, that solve for rotational updates in 360 videos by removing camera shakes, while preserving user intended motions. The method uses a constrained nonlinear optimization approach in quaternion space. At first, optimal 3D camera rotations are computed between key frames. 3D camera rotations between consecutive frames are then computed. The first, second, and third derivatives of the resulting camera path are minimized, to stabilize the camera orientation path. The computation strives to find a smooth path, while also limiting its deviation from the original path. The system keeps the orientations close to the original, for example, even when the videographer takes a turn. Each frame is then warped to the stabilized path, which results in a smoother video. The rotational camera updates may be applied to the input stream at source or added as metadata. The technology may influence standards by making rotational updates metadata a component of 360 videos.
KEYWORDS: 360 degree video, camera rotation, removing camera shake, computing camera rotation
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Arora, Chetan and Kwatra, Vivek, "Intent Preserving 360 Video Stabilization Using Constrained Optimization", Technical Disclosure Commons, (May 03, 2018)