Extraction of 3D Primitives from Stereopairs of Satellite Images for Automatic Reconstruction of Buildings

This study is a preliminary operation of a global project which goal is to interpret urban scenes from satellite images in order to recognize and classify roads and buildings. Actually, the research is focused on buildings. We plan to apply these algorithms on the forthcoming satellite data of PLEIADE HR which will follow SPOT5. The input data consists of a panchromatic stereopair of satellite images, a t a high resolution of 50-60 cm and a low Base to Height ratio B/H= 0.1,0.2. In this paper, we describe three algorithms involved in a global strategy for extracting reliable 3D primitives which is a key step and an important preliminary operation to make the generation of building hypothesis easier. All algorithms use photometric constraints a t every step to take advantage from low B/H and handle both views in a symmetric way. However, each algorithm can be used independently providing the specific primitives as input data.