Line-preserving video stitching for asymmetric cameras

Abstract

The narrow FOV on mobile phones is becoming unsatisfying with the development of video live broadcast. Fast and stable video stitching algorithm is urgently needed with the advent of multi-device bracket and other equipment. In this paper, we propose a novel stitching method to stitch multiple synchronized video streams which are captured from different fixed cameras. This procedure can be divided into two stages: initial template construction and seam updating. In the stage of initialization, we align the extracted background with proposed line-preserving alignment, and save the mesh for future stitching. Especially, scale estimation allows the algorithm to be applied to videos from asymmetric cameras of different resolutions and scales. In the seam updating stage, we propose a last-frame-guided seam updating method to find a smooth transition on spatial-temporal volumes. Experimental results demonstrate that our method can efficiently stitch videos without ghosting and artifacts.