Modeles stochastiques pour la reconstruction tridimensionnelle d'environnements urbains
l'Ecole des Mines de Paris, Paris, PhD Thesis, (2007)Abstract
This thesis tackles the problem of the 3D building reconstruction from very high resolution satellite images. The information provided by this kind of data are not accurate enough to allow an efficient use of the varied algorithms developped in an aerial framework. In this context, it is necessary to introduce strong prior knowledge related to the urban areas. The stochastic tools are especially well adapted to deal with this problem. A structural approach is proposed to address this topic. It consists in modeling a building through an assembling of basic urban structures which are extracted fom a library of 3D parametric models. First, the 2D supports of these structures are extracted from a Digital Elevation Model (DEM). The result is a quadrilateral layout of which the neighboring elements are connected. Then, the buildings are reconstructed by finding the optimal configuration of 3D models which are fixed onto the extracted supports. This configuration corresponds to the realization which maximizes a density. The last one measures the coherence between the realization and the DEM, and takes into account prior knowledge such as the assembling law of the structures. Finally, we discuss on the relevance of this approach by analysing the obtained results from satellite data (PLEIADES simulations). Experiments are also carried on from higher resolution aerial images.