Developments and experimental results of K-B system for workpieces recognition

A knowledge-based (KB) system for the recognition of partially occluded two-dimensional workpieces is described. After the image acquisition phase with a TV camera, the generalized Hough transform is used to extract elementary edge patterns and to derive a symbolic description. Hypothesis generation and test is the strategy utilized in the recognition process, on the basis of matchings between models and data. In particular, the system contains new modules devoted to automatic model acquisition and search-function generation for every model primitive. Results obtained confirm the validity of the proposed KB approach. The system has been tested on several images and has proved robust in terms of noise and occlusions