Semi-automatic 3D reconstruction of urban areas using epipolar geometry and template matching

Abstract

In this work we describe a novel technique for semi-automatic three-dimensional (3D) reconstruction of urban areas, from airborne stereo-pair images whose output is VRML or DXF. The main challenge is to compute the relevant information—building's height and volume, roof's description, and texture—algorithmically, because it is very time consuming and thus expensive to produce it manually for large urban areas. The algorithm requires some initial calibration input and is able to compute the above-mentioned building characteristics from the stereo pair and the availability of the 2D CAD and the digital elevation model of the same area, with no knowledge of the camera pose or its intrinsic parameters. To achieve this, we have used epipolar geometry, homography computation, automatic feature extraction and we have solved the feature correspondence problem in the stereo pair, by using template matching.