Automatically detecting geographical objects in high-resolution satellite images Détection automatique d’objets géographiques dans des images de satellite à haute résolution

Automatically detecting geographical objects such as bridges, roundabouts or road crossings on highresolution satellite images is useful for keeping up to date geographical databases, automatically locating the imaging satellite at low cost, and easily and quickly assessing the extent of damages in case of natural disasters such as flooding or earthquakes. In addition, it may help in content-based indexing of such satellite images. We present a set of techniques and methods to automatically detect bridges on this kind of images. These methods rely on radiometric features and neural networks to classify each pixel into several terrain types, and fixed rules to find bridges in this classification. They can be easily extended to other kinds of geographical objects, and integrated with existing techniques using geometric features.