Practical and accurate calibration of RGB-D cameras using spheres
Résumé
RGB-Depth (or RGB-D) cameras are increasingly being adopted in robotic and vision applications, including mobile robot localization and mapping, gesture recognition, and at-home healthcare monitoring. As with any other sensor, calibrating RGB-D cameras is needed to increase their sensing accuracy, especially since the manufacturer’s calibration parameters might change between models. In this paper, we present a novel RGB-D camera-calibration algorithm for the estimation of the full set of intrinsic and extrinsic parameters. Our method is easy to use, can be utilized with any arrangement of RGB and depth sensors, and only requires that a spherical object (e.g., a basketball) is moved in front of the camera for a few seconds. Our image-processing pipeline automatically and robustly detects the moving calibration object while rejecting noise and outliers in the image data. Our calibration method uses all the frames of the detected sphere and leverages novel analytical results on the multi-view projection of spheres to accurately estimate all the calibration parameters. Extensive numerical simulations and real-world experiments have been conducted to validate our algorithm and compare its performance with that of other state-of-the-art calibration methods. An RGB-D Calibration Toolbox for MATLAB is also made freely available for the scientific community.