Development of a robot vision system for measuring 3D pose of large object using virtual plane algorithm

Abstract

In this paper, we present a new robot vision solution to determine the accurate 3D position and orientation of large object using the virtual plane algorithm which uses arbitrary feature points of the 2D image and known CAD information. A typical example of industrial processes for measuring 3D pose of large object is the interior sealing of a car body in automotive painting shop. In this process, the trajectory of the painting robot should be adjusted according to real position and orientation of the car body placed on the conveyor. The laser vision sensor and the stereo vision system are conventional solutions for this process, but they have disadvantages such as narrow field of view, high cost, and maintenance problem. To overcome these problems, we propose a metrological method to measure 3D position and orientation of the large object using multiple sets of single camera located around recognizable feature points of the large object. Each camera measures local 3D pose information of the object based on the virtual plane algorithm. Accurate 3D pose information of the large object can be determined by some matrix calculation using known geometric relation between local coordinates in the large object. Through experiments on a commercial workpiece, we validate that the proposed system effectively measures the 3D pose of the workpiece with small errors sufficient to the robotic sealing applications.