3D scene reconstruction from 2D images has been one of the major issues in computer vision for years. And how to find the matching points from the stereo pairs, which is one of the most important steps in 3D reconstruction, is also a basic problem in computer vision. So far, Although fruitful achievements have been obtained in the field, most of them has been focused on linear low-distortion cameras that can be modeled by the pinhole mapping. In recent years, the fisheye lens has received increasing attention and use due to its extremely wide field of view. However, little has been published on stereo reconstruction of fisheye images thanks to the fisheye images several distortion. This thesis is focused on 3D reconstruction from multiple fisheye cameras. The main work can be summarized as: 1) A new feature matching method specially designed for calibrated fisheye cameras is proposed and validated by extensive experiments. 2) Base on the self-developed four fisheye cameras stereo vision system, the main issues in 3D reconstruction from multiple fisheye cameras system, such as system calibrating, feature matching, 3D reconstruction, object segmentation and so on are investigated. Our preliminary experimental results show that our fisheye cameras system not only has a very large view, but it can also achieve relatively accurate 3D reconstruction results.
修改评论