Active hand-eye calibration via online accuracy-driven next-best-view selection

We propose a novel high-accuracy active hand-eye calibration approach. In our method, the robot movement and camera view selection are both driven by and targeting the improvement of calibration accuracy. During the calibration process, the data acquisition is guided by an online estimated discrete viewing quality field (DVQF), representing the calibration quality of different views in various 3D locations. The view quality is measured by how much it reduces the uncertainty of calibration results and increases the diversity of robot poses, contributing to the calibration precision. Based on DVQF, we select the next-best-view as the target moving pose for each time step. A fully automatic system is presented to perform the overall hand-eye calibration process without any human intervention. Numerous experiments are conducted both in real-world and simulated scenarios. The proposed algorithm outperforms other approaches and shows much superiority in accuracy and robustness.