5 Selection of Roof Hypotheses 4 Hierarchical Grouping and Matching of Features

A method for detection and description of rectilinear buildings from two or more aerial intensity images is described. The multiple images may be taken at different times under differing imaging conditions. Detection and description of buildings under these conditions is a difficult task. It is not possible to compute a dense 3-D map just by intensity correlation. Instead, we need to match features and organize the features into coherent structures. We describe a hierarchical perceptual grouping and matching technique for this. The grouping and matching are performed at several levels, each higher level having more information to make its decisions. The grouping process gives rise to 3D hypotheses for roofs of buildings which are then further examined for evidence of expected vertical walls, and shadows on the ground. The output of the system is model of 3D building hypotheses. Experimental results are presented.