A Simple and Generic Method for Real-time Accurate Hand-eye Calibration Using a 3D Vision Sensor

Abstract

Hand-eye calibration is the first step of any vision system for robot manipulators. With 3D vision sensors becoming popular recently, how they can be used for real-time accurate hand-eye calibration is awaiting to be studied. In this paper, we propose a generic method for both eye-to-hand and eye-in-hand calibration using a 3D sensor. By exploiting a given feature point and the flange center point, we can formulate the two hand-eye calibration problems as one simple form: p <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">'</sup> =Rp+t. Herein, the flange center point is found by fitting the different locations of the feature point to a sphere surface. To do the sphere fitting and handle the above equation, we present a closed-form solution and two iterative solutions by least square method and random sample consensus algorithm. As traditional calibration marker is unnecessary, we can eliminate large measurement errors caused by the pose estimation of the marker. The performance of our method is simulated by two steps of calibration and validated with autonomous experiments. This work can be done instantaneously after the robot finishes the movements, which represents a simple and real-time step towards accurate hand-eye calibration in robotic applications.