Robot vision based on three-dimensional model

In order that robots recognize objects, the models of the objects are required. If there is not any constraint about an object scene, it is desirable that robot vision has the three-dimensional models of the things composing the scene. Since the preparation of three-dimensional models takes much time, here, the utilization of the geometrical models made by CAD is proposed. Besides, when the description of a scene and three-dimensional models are compared, to attempt the comparison with all attitudes of respective models is not efficient, therefore, stratum-wise comparison was proposed. As concrete examples, when input information is only in the direction of a part of surfaces such as lustrous bodies, when information can be obtained in all the directions of the surfaces of a body visible by a photometric stereo, and when complete three dimensional information is obtained by a distance-measuring instrument, the techniques of object recognition are described. In all cases, by carrying out the stratum-wise comparison based on three-dimensional models, the efficient and generalized object recognition was able to be achieved. (Kako, I.)