Abstract
This paper aims to realize a vision system for robots which handle mechanical parts for assembly at production lines. The process to recognize mechanical parts consists of two portions. One is the off-line process as preceding process to recognize parts efficiently, the other is the on-line process to recognize parts by real time. In the off-line process, geometric models are made for mechanical parts and the characteristic of reflection of their surfaces are evaluated by experiments. The characteristic of reflection are represented by a reflection function. The reflection function used in this paper is based on Phong's equation and it is modified simpler to make the process easy. By using this knowlege, positions of highlight regions at various postures are estimated to make recognition algorithm of the on-line process. Highlight regions as features to recognize objects have following merits : (1) The process to extract them is easy and takes a short time. (2) Shapes of them don't change very much even if the postures of objects are changed. The recognition in the on-line process are made by matching extracted highlight regions with the estimated ones from parts model. By using this method positions and postures of mechanical parts put at random are identified precisely for a short time by slantwise fixed TV camera.
