Semi-Automatic 3D Reconstruction of Urban Areas Using Epipolar Geometry and Template Matching

This paper describes a new technique for obtaining the volumetric description of large urban areas. This technique uses automatic procedures coupled with some interactive procedures for the three-dimensional (3D) reconstruction of urban areas from airborne stereo-pair images whose output is virtual reality modeling language (VRML) or Drawing Exchange Format (DXF). The main challenge is to compute the relevant information, such as building's height and volume, roof's description, and texture, algorithmically in a time- and cost-efficient manner. The algorithm requires some initial calibration input and is able to compute building characteristics using the stereo pair of aerial images and two-dimensional computer-aided design and digital elevation models of the same area. No knowledge of the camera pose or its intrinsic parameters is needed. The authors use epipolar geometry, homography computation and automatic feature extraction in this approach. They have solved the feature correspondence problem in the stereo pair by using template matching. Directions for future work are discussed.