The recognition and localization of 3-D natural quadric objects based on an optical proximity sensor system mounted on a robot end-effector is addressed. The goal is to achieve the tasks with a minimum number of probings as well as a minimum amount of surface information (orientation and distance). A natural quadric object is represented by a surface description vector (SDV) distribution graph and a hierarchical table. A series of probings allows the construction of a hierarchical table and a SDV distribution graph for the test object, and the derivation of possible interpretations on the test object. For an efficient selection of the correct interpretation, all possible interpretations are overlapped to form a multiple interpretation image (MII), from which an optimal probing plane is determined. It intersects the MI so that the intersecting surfaces of the MII can provide the maximum discrimination. An optimal probing plane is defined by selecting two vectors on that plane: the sensor orientation (the optimal beam orientation) and the direction of the sensor move (the optimal probing direction). Simulation results are shown.
展开▼
