Digital Terrain Models in Dense Urban Areas

In this paper we consider the problem of DTM extraction in dense urban areas. To this aim, we need valuable and reliable data in the tiny open areas among buildings, and a suitable algorithm to reconstruct the terrain discarding these same structures. So, we compare first LIDAR and aerial photogrammetry by evaluating the relative accuracy of the three-dimensional reconstructed surface in the small open areas in the town center. Then, we characterize the digital terrain model (DTM) of the whole town using a filtering and building detection approach. It comes out, as expected, that the best filter width depends on the terrain and built structure characteristics, and we show that, after training on test areas comprising all possible combination, it is possible to obtain highly precision DTMs filling the built areas with surrounding terrain without significantly affecting the open areas.