Multiple cameras visual servoing used for large scale 3D positioning

Abstract

In this paper, a multiple cameras visual servoing system is presented to correct the pose of robot end-effector in large scale 3D positioning in robot manufacturing system. A 3D CAD laser projector is used to provide four target points on the projection screens for four cameras respectively. Eye-in-hand architecture is used to perform visual servoing and the image based visual servoing (IBVS) is adapted as the control strategy. This paper is focused on detecting the four target points by four cameras respectively. Theoretically, the image Jacobian matrix doesn't hold in this configuration because the target points are floating in base frame along the laser beam during the visual servoing process. But in a step of iteration, the target points on the projection screens can be referred as fixed points which are also called local desired target points. The image features of local desired target points in image planes of cameras are the desired features in this iteration. By observing the four target points on projection screens, the IBVS controller can drive the end-effector of robot to its local desired position in each step of iteration and ultimately to its calibration position. In this configuration, the simple constant depths are adopted for obtaining the image Jacobian matrix during visual servoing. A simulation was carried out to verify the effectiveness of the proposed method.