High-Accuracy Calibration of a Visual Motion Measurement System for Planar 3-DOF Robots Using Gaussian Process

Abstract

Visual motion measurement systems (VMMSs) are widely used in robotics for pose estimation. A common characteristic of VMMSs is that the measurement repeatability is high, but the absolute measurement accuracy is relatively low, once a commercial lens with a large field of view is used. To improve the measurement accuracy of the VMMS for pose tracking of planar 3-degree-of-freedom (3-DOF) robots, a coarse-to-fine calibration method is presented in this paper. In this paper, after the VMMS is constructed, a degenerated perspective-n-point (DPnP) algorithm for pose estimation of 3-DOF robots is introduced. Then, an analytical calibration technique is implemented to obtain the camera intrinsic parameters required by the DPnP algorithm. Subsequently, a fine calibration step based on Gaussian process is developed to compensate the estimated pose of the DPnP algorithm. To investigate the effectiveness and performance of the proposed method, a series of the simulations and experiments are carried out on a planar 3-DOF robot. The results demonstrate that the proposed calibration method is quite robust and can improve the measurement accuracy up to 90.47%. Specifically, in a measurement field of view of 200 mm $\times200$ mm, the absolute errors of the 3-DOF pose obtained from the VMMS are reduced to below 0.03 mm and 0.016°.