Terrain Dependent Correspondence Search for 3D Urban Modeling Using Multiple High-Resolution Satellite Images

This paper presents a methodology for generating semi-automated 3D models of urban area from multiple high- resolution satellite images. Adoption of Rational Function Models (RFM's) during recent years as a replacement for satellite's rigorous sensor model by many commercial imagery providers makes 3D information extraction fast and reliable for most of end-users. This paper presents an application of RFM in generating 3D terrain models, i.e. bear-ground models void of any vegetation and man-made objects, as well as 3D surface modeling (surface models with man-made objects). A fast one- dimensional terrain-dependent object-based search technique for identifying correspondences in multiple images followed by a refined 2D search to obtain an optimal correspondence match is introduced. Corresponding objects are then used in conjunction with the RFM to generate 3D model of the area. Example of correspondence matching and a 3D model generated for a selected area of Yokohama, Japan using QuickBird stereo images is presented.