Development of a 3D Vision Guided Robotic System Using Stereo Vision

摘要

As the flexibility becomes an important factor in robotrnapplications, there are strong demands from manufacturersrnfor a robust and accurate three-dimensional (3D) visionrnsystem in the robotic handling applications. However thernconventional 3D vision systems can be limitedly employedrnonly in the specific tasks due to restrictions caused by a fieldrnof view, number of key features and the shape of an object.rnIn this paper, we propose a new general purpose stereornvision based robotic system which can measure the 3D posernof the workpiece. In order to reduce the depth estimationrnerror, the pattern recognition method is used instead of thernstereo matching method to find the correspondence points inrnstereo images. The multiple pattern registration method isrnapplied to increase the reliability against illuminationrnchange. In addition the ellipse fitting algorithm is employedrnfor the objects of circular shape features so that thernperformance of the 3D position estimation is improved. Thern3D pose estimation is performed in a robot controller byrncomparing the geometrical relationship between therncoordinate of reference object and the coordinate ofrnincoming object measured by the stereo vision system.rnBased on the experimental results on the performance ofrnthe 3D position estimation, we validate that the proposedrnvision guided robotic system can be applied in the handlingrnprocess of an automotive plant.