Surface and terrain reconstruction from very high resolution imagery: Where are we now?

3D Geographic Information and especially Digital Surface or Terrain Models are becoming essential data for many applications, from risk mapping (avalanches, volcanoes, floods, ..) to town and infrastructure planning with virtual and augmented reality. Automatic surface reconstruction from images with shape from stereo techniques has been a subject of constant attention within remote sensing, photogrammetric, image processing and computer vision communities since the 70's. These techniques had reached a limit when addressing scenes with poor textures, discontinuities and occlusions as encountered in urban areas with HR geospatial imagery. These limitations have faded lately with the rise of new technologies (e.g. high quality aerial digital cameras) and techniques (e.g multi-view processing) and consequently expectations of full (or nearly) automation at short term have become a trend again. Meanwhile, LASER scanning technologies, which provide a reliable description of the imaged scenes in the form of 3D cloud points, have achieved a certain degree of maturity.