Video stitching based on iterative hashing and dynamic seam-line with local context

Abstract

Video stitching fuses multi-camera videos with differing centers of projection into a single panoramic video. Image registration and video fusion is the key of video stitching. In this paper, a robust and real-time video stitching method based on iterative hashing and best dynamic seam-line with local context model is proposed, which can eliminate ghosting and ensure real-time in video stitching. Firstly, the iterative hashing algorithm is proposed to improve the speed and precision of image registration. Image feature points are matched by constructing a multi-table, extracting candidates and refining candidates with iteration, which enhances the local sensitivity of hashing and speeds up the process of feature points matching. Secondly, the method of finding the best seam-line dynamically and blending with local context is proposed to improve the quality of video fusion. The proposed video fusion method is able to eliminate ghosting and illumination variation during video stitching. In addition, the speed of video stitching can be enhanced. Experimental results on several scenes show the efficiency and effectiveness of the proposed video stitching method.