This work aims at detecting buildings in digital a­erial photographs. Here we model a set of buildings by a configuration of objects. We define a point process on the set of configurations, which splits into two parts : the first one is a prior model on the configurations which use interactions between objects, the second one is a data model which enforces the coherence with the image. Thus we have a posterior distribution whose maximum has to be found. In order to achieve this maximum, we use a MCMC simulation - a Metropolis-Hasting- s-Green algorithm - mixed with a simulated annealing. Then we test this method on both synthetic and real stereo-images.