This paper describes the method and the last implemented modules within the frame of an ongoing study in the field of 3D indoor / outdoor scenes reconstruction, but in man-made environments. It deals more specifically with a cooperative segmentation methodology, very much application driven, and designed from a rough geometrical modelisation of such scenes, as sets of polyhedra.
First of all, principles of the segmentation methodology are given. Then, information transfers between both kinds of feature extractors: region (relating homogeneous properties), and edge (relating local difference properties) are pointed out. They allow an easy merge of the different types of data. Later, to conclude with the advantages of this method, compared to other classical ones are underlined.