Refinement of urban digital elevation models from very high resolution stereo satellite images

Digital elevation models (DEM) of high resolution and high quality are required for many applications like urban modeling, readiness for catastrophes or disaster assessment. A good source for the derivation of such DEMs from any place in the world are very high resolution (VHR) satellite stereo images as provided e.g. by Ikonos, QuickBird or WorldView. In this paper a method for the generation and refinement of urban high resolution DEMs from VHR imagery is presented and evaluated. Urban DEMs generated from very high resolution satellite imagery of ground sampling distances of about one meter are normally of resolutions of about three to ten meters. For the above mentioned applications of urban DEMs such results are often too coarse. In this paper an advanced method for the generation of dense digital elevation models is presented and discussed. The method is mainly based on dense stereo algorithms developed for computer vision applications. It is adapted and optimized to earth observation requirements. In the paper the DEM generation together with the additional refinement steps is presented and evaluated using very high resolution stereo imagery from Munich and Athens. The generated DEMs are compared to ground truth data where available and the quality and efficiency of the algorithms are analyzed and discussed.