ロボットビジョンによる物体の凸多面体モデルの自動生成

抄録

For an intelligent robot, determining efficiently the data of shape and posture of the objects in the real world environment will increase greatly the adaptation capacity for the unknown environment. With respect to these unknown 3-D objects, a robot vision system is proposed which is capable of constructing automatically their geometric models which include the information of both the shape and the posture or position of the objects. By means of this system, a “Visual Points Selecting Cone Intersection Method” is proposed which can select appropriate camera viewing point and direction automatically as investigating the partial shape of the object. With this method, experiments was performed which make clear that this method can get the model of object more accurately and effectively.