Fiducial marker-based Metric Initialization for Monocular SLAM

This paper proposes a novel approach for metric monocular SLAM Initialization using keypoints and a fiducial marker. SLAM approaches using one camera only can obtain the scale relevant to initial movement and the reference world origin must be attached to the initial frame. These make monocular SLAM approaches hard to achieve metric scale in localization and navigation problems or to provide true scale reference in augmented reality applications. In this paper, a fiducial marker is adopted at the initialization stage of SLAM. The relative and the absolute scale for 3D coordinates of some particular visual features would be calculated. The initial map would be adjusted by the ratio of these two values at the same time. Our experimental results show that feature-based monocular SLAM can acquire metric scale and the world origin can be coordinated to the marker plane after initialization. The scale outputs can reach centimeter accuracy with real-time performance in a normal computing unit.