A Robust Algorithm for Estimating Digital Terrain Models from Digital Surface Models in Dense Urban Areas

This paper describes an algorithm in order to derive DTMs (Digital Terrain Models) from correlation DSMs (Digital Surface Models) and above-ground (buildings and vegetation) masks in dense urban areas. Among all the methods found in literature, the Elastic Grid method shows a good capability to reconstruct the topographic surface. This method consists in interpolating height values under above-ground masks by minimizing an energy. Nevertheless, this method is ill-adapted to outliers in input data (above-ground points out of above-ground masks). The main contribution of our study is the use of a method based on robust statistics in order to reject outliers from calculation so that the final DTM fits the “true” topographic surface for the best. For that purpose, the initial Elastic Grid has been noticeably changed. The results of the new method for 2 test sites with a pixel ground size of 20 cm (the first one is relatively flat and the second one is hilly) show the quality of the final DTM and the robustness of our method. Tests have been carried out with lower resolution DSMs and without any mask and show the feasability of extending the method to a more general context.