Robotic hand–eye calibration utilizing limited geometric features object

Abstract

Hand–eye calibration is essential for intelligent robots to accurately perceive their environment, primarily focused on determining the transformation matrix between the robot flange coordinate system and the 3D sensor coordinate system. However, current robot hand–eye calibration methods heavily depend on costly specialized calibration objects, such as calibration boards and spheres, which complicate the calibration process and hinder the robot’s ability to perform self-calibration at any time and in any location. To address this issue, this paper proposes a novel robot hand–eye calibration method that utilizes the reconstruction of common objects with limited geometric features. Specifically, a point cloud feature description method that integrates eigenvalue entropy is introduced to extract feature points from multi-pose point clouds of these objects. Subsequently, a registration strategy based on the random sampling consensus of partitioned point clouds is employed for the coarse registration of the point cloud, estimating the initial hand–eye relationship, followed by iterative optimization through fine registration to determine precise hand–eye parameters. Extensive experimental results demonstrate that the proposed method offers a simple and efficient calibration process, eliminates reliance on specialized calibration objects, and achieves calibration accuracy comparable to that of high-precision calibration boards, thereby showcasing the advantages of the proposed approach.