A Scene Feature Based Eye-in-Hand Calibration Method for Industrial Robot

The industrial robot eye-in-hand calibration has been a hot research topic for a long time. When the RGB-D camera is mounted on the end-effector of the robot, the pose of object can be obtained by the camera, then the pose transformation from camera coordinate system to robot coordinate system is realized by the hand-eye transformation matrix. Traditional marker-based calibration methods are so complex to operate and time-costing that are not suitable for actual application in industrial field. In this paper, with the aim of reducing the calibration complexity and time, a novel eye-in-hand calibration method based on scene feature is proposed. At First, the image sequence obtained by small motion of camera is processed by ORB feature extraction and Bundle Adjustment (BA), and the camera intrinsic parameters are solved. Then, ORB feature and Perspective-n-Point (PnP) is used to solve pose transformation between multi-view images, followed by BA optimization. Finally, the hand-eye transformation matrix is obtained by solving the corresponding relations between pose transformations of multi-view images and pose transformations of the robot. Since our method needs no man-made marker, it is thought as an auto calibration method and can be done many times during the operation process. Experiments show that compared with existing calibration methods, our method is simple, fast with high accuracy and more suitable for industrial application.