Building detection from disparity of edges

The availability of metric or sub-metric satellite images like Ikonos or Quickbird has enabled the fast and automatic detection of man-made structures for applications such as risk or damage assessment and change detection. In particular, the use of stereo pairs reveals the third dimension so distinctive for buildings and responsible for artefacts like shadow and occlusions complicating traditional 2D approaches. Unfortunately, areas to be studied are quite large and algorithms for height estimation are computationally demanding. We propose a simple and fast approach for building detection which computes the disparity at pixels of sufficient gradient and highlights linear segments so representative of building outlines. We have applied it to an Ikonos pair aiming at detecting large buildings in the context of security within the GMOSS Network of Excellence.