Panoramic Mosaic Based On Precise Camera Tracking

This paper primarily study the methods of panoramic mosaic based on accurate camera tracking, which mainly apply to video shot taken by a moving camera. Panorama mosaic is one of the hot topics of image based rendering(IBR) since 90s of the last century, but panoramic mosaic is usually constructed from images taken from a fixed position. It is a difficult problem to compose a panoramic image from images taken from a camera with obvious movement.This paper proposes a novel method to mosaic multi-view panorama. We construct a quasi-perspective picture viewing from one point depending on precise camera tracking technology. Moreover, we propose a novel approach to camera tracking based on path constraint. This approach can deal with the long video sequences with distinct accumulated error, in which path and space are twisted. Our algorithm refine the key frames and gold points based on path constraint in order to eliminate the accumulated error. Meanwhile, recover the structure 3D points and camera parameters. The implemented examples demonstrate very precise structure and motion recovery, and prove the efficiency and robustness of the proposed method.Finally, we give an introduction to the method which is used in panorama mosaic. Moreover, we study two kinds of panoramas respectively: one is motionless viewpoint panorama, another is multi-viewpoint panorama. We analyze the process of the panorama mosaic in detail. For multi-view panorama, we not only display our mosaic process and result, but also discuss the limitations and future work.