KNOWLEDGE-BASED 3D VISION SYSTEM USING A MODELING WITH CLASS CONCEPT.

A three-dimensional (3-D) vision system using a modeling with class concept has been constructed. It is based on the object-oriented programming paradigm, which is effective for implementing a complex system incrementally. The models, which work as references in the understanding process, are represented as hierarchical compositions of plane clusters, planes and line segments. For each hierarchical level, the models are defined, going from a coarse description to a fine one, utilizing the inheritance function of the object-oriented programming. This model description allows an efficient coarse-to-fine understanding of a 3-D object from an input image. The system also has a function to supplement incomplete edge-lines caused by an illumination condition or occlusion.