Abstract
In order to recognize the 3-D objects in range images, the objects are first separated out from the background. The next step involves segmentation of the complex boundary surfaces of objects into primitive surface patches. The primitive surface patches are to be characterized so that the object surface can be represented by a combination of surfaces of known types. We here present a method for characterizing a surface as either planar or cylindrical or spherical. Any surface which does not fall into the category of the above surface types is represented by a combination of these primitive surface types.
Indexing terms: