3D Reconstruction of Urban Environments Based on Fisheye Stereovision

In this paper, we show some results about 3D urban scenes reconstruction using a fisheye stereovision setup. We propose an analytical analysis of epipolar geometry of the system and an analytical description of tools to compute a 3D point cloud from matched pixels. The novelty is that we do not rectify the images and that we match points along 3D or 2D epipolar curves. The matching process is based on a global dynamic programming algorithm that we adapt to take into account continuous epipolar curve equation. We show 3D point cloud in the case of synthetic images.