Building semantic descriptions from boundary data

A method for extracting 3D shape primitives of a polyhedral object from the boundary face and edge data is presented. These primitives which provide semantic information as well as geometric information are useful in CAD/CAM applications and in model-based vision. A graph-based hypothesis generate-eliminate approach is used to identify and classify 3D shape primitives of the object. Several techniques, such as node unification and virtual links, are introduced to solve the problems associated with the interaction between the primitive features. Results from a system implemented in LISP are presented