360-degree Video Stitching for Dual-fisheye Lens Cameras Based On Rigid Moving Least Squares

摘要

Dual-fisheye lens cameras are becoming popular for 360-degree video capture,especially for User-generated content (UGC), since they are affordable andportable. Images generated by the dual-fisheye cameras have limited overlap andhence require non-conventional stitching techniques to produce high-quality360x180-degree panoramas. This paper introduces a novel method to align theseimages using interpolation grids based on rigid moving least squares.Furthermore, jitter is the critical issue arising when one applies theimage-based stitching algorithms to video. It stems from the unconstrainedmovement of stitching boundary from one frame to another. Therefore, we alsopropose a new algorithm to maintain the temporal coherence of stitchingboundary to provide jitter-free 360-degree videos. Results show that the methodproposed in this paper can produce higher quality stitched images and videosthan prior work.