In this work a new framework for generic scene interpretation is introduced. It is based on several statistical approaches ranging from a novel description of object shapes to their localization and classification in complex scenes. In the present context a scene is a static arrangement of several 3D rigid free-form objects represented by a large number of 3D-surface points and surface normals. The interpretation of a scene is the segmentation of the scene data into object-specific parts and classification of the identity of each part
«