TRIO: A CAD-based 3-D object recognition system in manufacturing.

摘要

Model-based 3-D object recognition is one of the central issues of research in the domain of computer vision in manufacturing systems. As tergrated CAD/CAM systems contribute to the integration of manufacturing systems, the idea of utilizing CAD as a geometric modeling tool for model-based recognition motivates the current research of building a CAD-based object recognition system. This research proposes a vision system, "TRIO" (The Recognition and Interpretation of Objects), which uses a hierarchical database of CAD and graph-based representations to achieve the goal of building a CAD-based object recognition system. Graph representation is chosen because it has strong; descriptive power to geometric features and relations. TRIO consists of three major modules: (1) CAD to graphs transformation and model base organization, (2) Image to graphs transformation, and (3) Matching graphs to generate valid object interpretations.;With the design of the reasoning hierarchy, the system is able to infer face identities with incomplete or distorted edge boundary. It then uses a constraint search technique to generate all the possible interpretations by solving a consistent labeling problem (CLP). The TRIO system is tested with successfully segmented real images with over 80% recognition success rate. This system is implemented using C language. The indexing tables in this system are implemented with GNU DabaBase Manager (GDBM) which is a set of C library routines for creating efficient data storage and retrieval mechanism. This runs on a Sun Sparc Unix-based machine with efficient run time.