Robot vision system capable of recognizing machine parts using multistage neural networks

A visual sense system for robots which can measure and recognize 3-D objects is proposed. This robot vision system is capable of recognizing machine parts moving on a line so that an automated transferring and assembly system with high flexibility and no dependence on external peripheral equipment can be realized. This system possesses two remarkable features. One is the measuring technique, which can effectively remove the false 3-D information due to mirror reflection using laser slit beams from two directions. The other is the technique of carrying out the recognition of polygonal surfaces by p-type Fourier descriptor, area and perimeter, and, further, the recognition of objects based on aspect graphs, using multistage neural networks. As compared with the conventional recognition method, the speed and accuracy of recognition rose a step