ロボットビジョンによる重なり合った物体の個別凸多面体モデルの自動生成

抄録

In order to obtain the information of the environment for the robot to work, it may be the effective way to work on the environment through the robot's action. In accordance with this idea, a method dealing with the automatic three-dimension partitioning of overlapped objects has been proposed by using the robot vision system which was proposed and constructed by authors. As compared with other researches about partitioning until now, our method can obtain the three-dimension partitioning planes among unknown overlapped objects and this makes it possible to generate the each geometric model of those objects. Within this method, some active techniques such as using laser spotter scanning to obtain the appropriate visual points for investigating the partitioning condition of overlapped objects have been developed. This technique has been proved to be effective to obtain the threedimension partitioning plane of those overlapped objects. Experiments dealing with the several cases of overlapped objects' partitioning make clear the validity of this method. And with these obtained 3-D partitioning plane, the each geometric model of the overlapped objects has been generated automatically.