概要

This paper describes the vision system which recognizes 3-D objects in real-time by modeling shape of objects and matching the generated models. We develop following methods to practically solve the indispensable problems of integration, like the estimation of sensor accuracy and real-time processing: 1) We generate hierarchical mesh models in real-time by reduction of computation of signed-distance which is necessary to apply Marching Cubes Algorithm and selecting the optimal resolution of models to be generated using Octree. 2) We efficiently match multiple objects of different size by applying Spin-image matching with selecting the resolution of generated models and the coarse-to-fine algorithm.