FlexCloud: Direct, Modular Georeferencing and Drift-Correction of Point Cloud Maps

TL;DR

The paper addresses the challenge of turning SLAM-generated, locally referenced point cloud maps into globally georeferenced HD maps suitable for map-based localization. It introduces FlexCloud, a modular ROS 2 pipeline that (i) uses Keyframe Interpolation to fuse GNSS trajectories with SLAM odometry, (ii) applies Umeyama rigid alignment, and (iii) performs a 3D rubber-sheet transformation with a 3D Delaunay triangulation to correct long-term drift while preserving map structure. Key contributions include automatic GNSS-derived control-point interpolation, a scalable 3D rubber-sheet drift-correction mechanism, and an open-source implementation validated on Yas Marina Circuit and KITTI demonstrating sub-meter georeferencing accuracy under favorable GNSS conditions. The approach enables robust generation of globally referenced point cloud maps for HD-map creation and map-based localization in autonomous driving, with avenues for boundary-shape improvements and GNSS post-processing in future work.

Abstract

Current software stacks for real-world applications of autonomous driving leverage map information to ensure reliable localization, path planning, and motion prediction. An important field of research is the generation of point cloud maps, referring to the topic of simultaneous localization and mapping (SLAM). As most recent developments do not include global position data, the resulting point cloud maps suffer from internal distortion and missing georeferencing, preventing their use for map-based localization approaches. Therefore, we propose FlexCloud for an automatic georeferencing of point cloud maps created from SLAM. Our approach is designed to work modularly with different SLAM methods, utilizing only the generated local point cloud map and its odometry. Using the corresponding GNSS positions enables direct georeferencing without additional control points. By leveraging a 3D rubber-sheet transformation, we can correct distortions within the map caused by long-term drift while maintaining its structure. Our approach enables the creation of consistent, globally referenced point cloud maps from data collected by a mobile mapping system (MMS). The source code of our work is available at https://github.com/TUMFTM/FlexCloud.

Figures (10)

• Figure 1: Steps for HD map generation (extended from srinara_strategy_2023).
• Figure 2: Flowchart illustrating the creation of a global, georeferenced PCM using FlexCloud. The computations within the single modules are conducted with the global GNSS trajectory and the local odometry trajectory. The resulting transformations from the Rigid Alignment and the Rubber-Sheet Transformation are then applied to the PCM.
• Figure 3: Principle of Keyframe Interpolation. For interpolation of the global position corresponding to the odometry position at time $t_o$, four reference points on the global trajectory (blue points) are necessary. The interpolated global position $p_{g,t_0,inter}$ is on the resulting spline (green).
• Figure 4: Excerpt of a final PCM with color-coded point deformations by the Rubber-Sheet Transformation at the YMC. Gray points represent the PCM after Rigid Alignment with the orange mark being a selected CP subdividing the excerpt into tetrahedra.
• Figure 5: Deviation of the odometry to the GNSS trajectory (gray) after Rigid Alignment at the YMC.