A discontinuity detector for building extraction from digital elevation models by stochastic geometry

This work presents a framework to use stochastic geometry for auto-matic building extraction from different kinds of Digital Elevation Models (DEM). The goal is to extract some vectorial information from a DEM in order to ease precise 3 dimensional reconstruction. Using a spatial point process framework, we model cities as config-urations of unknown number of rectangles. An energy is defined, which takes into account both a low level information provided by the altimetry of the scene, and some geometric knowledge on the location of buildings in towns. The estimation is done by minimiz-ing an energy using simulated annealing. We present results on real data provided by IGN (French National Geographic Institute) con-sisting of laser and optical DEMs. 1