Use of explicit knowledge and GIS data for the 3D evaluation of remote sensing images

The evaluation of 3D scenes observed from different sensors requires the co-registration of sensor images and the reconstruction of the 3D geometry. To solve both tasks the presented system exploits prior knowledge, represented explicitly by semantic nets, and uses a digital landscape model of a geographic information system (GIS) as a hint for the object location. This is shown for the detection of control points for image registration and the extraction of objects (roads, buildings) for 3D reconstruction. For real time visualization the 3D geometry is approximated by a polygon mesh with overlaid photo texture.