Bayesian model identification: Application to building reconstruction in aerial imagery

In this paper, we deal with building reconstruction in stereoscopic aerial imagery. We present a statistical, and competitive approach to the segmentation of roofs in pre-segmented regions. This parametric method is based on a multi-plane model, interpreted as a Bayesian mixture model. The so-called augmentation of the model with indicator variables allows the using of Bayesian sampler algorithms to achieve both the estimation of the model's parameters and the segmentation of the selected region.