Seam Reconstruct: Dynamic scene stitching with Large exposure difference

摘要

Panoramic stitching of static and dynamic scenes is a very important and challenging research area. For static scenes, number of approaches has been proposed so far. The result produced by these approaches provides great similarity between resultant panorama and input images with almost zero seam visibility. However for dynamic scenes, we have found that existing approaches are unable to overcome the challenges introduced due to simultaneous presence of (A) Large exposure difference and (B) Moving Objects. In this paper we propose a Seam-Reconstruction technique which overcomes these limitations. Seam-Reconstruction technique is a two step approach. The first step resolves the position of moving objects and prevents ghosting due to parallax by building a reference panorama with the help of optimal seam evaluation technique while the second step removes the remaining differences along the seam with the help of Poisson's equation. Experimental result depicts that we were able to stitch dynamic scene containing large exposure difference maintaining the quality as well as photometric and geometric similarity.