Multi-sensor extrinsic calibration with the Adam optimizer

Abstract

In this paper, we propose the application of the Adam optimizer to extrinsic calibration of the multi -sensory system. Our robot is equipped with three RGB-D cameras. The first camera is attached to the wrist of the arm, the second camera is mounted in the robot's head, and the third camera is attached to the mobile base. Additionally, the pose of the wrist camera changes with respect to the robot frame and depends on the configuration of the robotic arm. The proposed method finds all relative transformations between cameras in a single optimization procedure. We compare the proposed application of Adam method with black-box evolutionary algorithm, Levenberg-Marquardt optimization, and graph-based optimization. We also evaluated three cost functions to verify the influence of various parameterization methods of the SO (3) rotation on the calibration results.