Geo-Registration of Terrestrial LiDAR Point Clouds with Satellite Images without GNSS

TL;DR

This paper tackles geo-registration of terrestrial LiDAR point clouds to satellite imagery in GNSS-denied urban environments, addressing the need for accurate, city-scale 3D reconstruction without prior localization. It introduces a topology-driven framework that first extracts road skeletons from LiDAR and maps, then performs a hierarchical alignment: a global rigid transformation based on skeleton intersections followed by a non-rigid RBF warp and terrain-aware elevation correction using SRTM data. Key contributions include (i) a cross-modal skeleton-based matching paradigm, (ii) a four-stage pipeline integrating semantic road segmentation, skeleton extraction, rigid and non-rigid alignment, and elevation correction, and (iii) quantitative evaluation on KITTI and a new Perth dataset demonstrating substantial planimetric and elevation improvements without GNSS input. The approach enables reliable urban geo-referencing for 3D city modeling and autonomous driving, offering a practical path toward GNSS-denied georeferencing and data fusion across LiDAR and map sources.

Abstract

Accurate geo-registration of LiDAR point clouds remains a significant challenge in urban environments where Global Navigation Satellite System (GNSS) signals are denied or degraded. Existing methods typically rely on real-time GNSS and Inertial Measurement Unit (IMU) data, which require pre-calibration and assume stable signals. However, this assumption often fails in dense cities, resulting in localization errors. To address this, we propose a structured geo-registration method that accurately aligns LiDAR point clouds with satellite images, enabling frame-wise geo-registration and city-scale 3D reconstruction without prior localization. Our method uses a pre-trained Point Transformer to segment road points, then extracts road skeletons and intersections from the point cloud and the satellite image. Global alignment is achieved through rigid transformation using corresponding intersection points, followed by local non-rigid refinement with radial basis function (RBF) interpolation. Elevation discrepancies are corrected using terrain data from the Shuttle Radar Topography Mission (SRTM). To evaluate geo-registration accuracy, we measure the absolute distances between the roads extracted from the two modalities. Our method is validated on the KITTI benchmark and a newly collected dataset of Perth, Western Australia. On KITTI, our method achieves a mean planimetric alignment error of 0.69m, representing 50% improvement over the raw KITTI data. On Perth dataset, it achieves a mean planimetric error of 2.17m from GNSS values extracted from Google Maps, corresponding to 57.4% improvement over rigid alignment. Elevation correlation improved by 30.5% (KITTI) and 55.8% (Perth). A demonstration video is available at: https://youtu.be/0wkACAB-O6E.

Figures (6)

• Figure 1: 1) Point cloud processing: road segmentation, outlier removal, and 2D projection to derive the road skeleton. 2) Map processing: tile identification, loading, merging, cropping, road network extraction, and elevation sampling to generate skeleton and height maps. 3) Rigid alignment: distance preprocessing, transform estimation, and similarity scoring to establish initial point cloud–map correspondence. 4) Non-rigid alignment and elevation matching: point pair identification, RBF interpolation, warping, and ground point extraction to produce the final geo-registered point cloud.
• Figure 2: Alignment results for selected KITTI sequences (00, 02, 05, 08, 09). Magenta curves show geo-registered point cloud trajectories overlaid on reference map tiles, illustrating correspondence with the urban road network.
• Figure 3: Frame-level registration results for selected frames in each KITTI sequence. Green dots: estimated trajectory points after geo-registration. Blue dots: manually labeled ground truth. Yellow dots: GPS-based reference. LiDAR points (red) are projected on the underlying map imagery. The results confirm accurate local alignment under diverse urban conditions.
• Figure 4: Point-pair distance distribution analysis across five KITTI sequences. Histograms compare distances between KITTI GNSS data (yellow) and our point cloud method using rigid alignment (blue) and non-rigid alignment (green), with fitted normal distributions and std ($\sigma$) for each approach.
• Figure 5: Cross-sequence validation of elevation correction accuracy. Comparison of the proposed point cloud-based method with KITTI GNSS ground truth across sequences 00, 02, 05, 08, and 09. The left plot shows index-based correlations between corrected and ground truth elevations, and the right plot presents error probability distributions with fitted std ($\sigma$). The bottom-left subfigure visualizes corrected elevations in the XY-plane for sequence 09.