OpenStreetMap-based LiDAR Global Localization in Urban Environment without a Prior LiDAR Map

Younghun Cho, Giseop Kim, Sangmin Lee, Jee-Hwan Ryu

Abstract

Using publicly accessible maps, we propose a novel vehicle localization method that can be applied without using prior light detection and ranging (LiDAR) maps. Our method generates OSM descriptors by calculating the distances to buildings from a location in OpenStreetMap at a regular angle, and LiDAR descriptors by calculating the shortest distances to building points from the current location at a regular angle. Comparing the OSM descriptors and LiDAR descriptors yields a highly accurate vehicle localization result. Compared to methods that use prior LiDAR maps, our method presents two main advantages: (1) vehicle localization is not limited to only places with previously acquired LiDAR maps, and (2) our method is comparable to LiDAR map-based methods, and especially outperforms the other methods with respect to the top one candidate at KITTI dataset sequence 00.

Introduction

For autonomous vehicles to navigate safely, precise global localization is essential. Traditionally, this "kidnapped robot" problem is solved by matching real-time sensor data against a dense, pre-built 3D LiDAR map. However, creating and maintaining these maps is a significant logistical challenge, as environments change and mapping every road is computationally and financially expensive.

This research explores an alternative: Can we localize a vehicle globally using only free, globally accessible 2D maps like OpenStreetMap (OSM)? By leveraging building footprints as stable geometric anchors, the proposed method eliminates the need for prior LiDAR maps while achieving competitive accuracy.

Methodology

The core of the approach is the creation of a lightweight, 1D descriptor that represents the "skyline" of surrounding buildings from both the map and the sensor's perspective.

1. Descriptor Generation

  • OSM Descriptors: The system interpolates candidate points along roads in OSM. At each point, virtual rays are cast in 360 degrees to calculate the distance to the nearest building footprint, resulting in a 1D distance array.
  • LiDAR Descriptors: Live LiDAR scans are processed through a semantic segmentation network (RangeNet++) to isolate building points. The shortest distance to these points is calculated at regular angular intervals to create a 1D array matching the OSM format.

2. Two-Stage Matching Pipeline

To handle the unknown orientation of the vehicle and the vast search space of the map, the system uses a two-stage process:

  • Stage 1 (Fast Retrieval): The 1D descriptors are transformed into rotation-invariant "keys." By comparing these keys, the system quickly identifies the top 200 candidate locations from the OSM database.
  • Stage 2 (Fine Matching): For the top candidates, the original 1D LiDAR descriptor is rotated through 360 degrees and compared against the OSM descriptors to determine the final position and heading.
The proposed localization pipeline: OSM data and LiDAR scans are processed into descriptors and keys for a two-stage global matching process.

Experimental Results

The method was evaluated on the KITTI Odometry and KITTI-360 datasets, using standard OSM layers for road and building information.

Accuracy and Robustness

On KITTI sequence 00, the method achieved a 48.34% Top-1 accuracy, outperforming the deep-learning-based PointNetVLAD (47.90%), which requires a prior LiDAR map. The system proved particularly robust at intersections, where the rotation-invariant keys allowed for successful localization even when approaching from novel directions.

Comparison of localization results: (a) the proposed OSM-based method achieves results comparable to (b) traditional methods that rely on pre-built LiDAR maps.

Computational Efficiency

The two-stage pipeline is highly efficient, processing frames in approximately 0.23 seconds on a standard Intel i7 CPU. This efficiency stems from the rotation-invariant keys, which allow the algorithm to discard 98% of the search space before performing expensive rotational alignments.

Conclusion

This work demonstrates that buildings serve as effective geometric anchors for bridging the gap between 3D sensors and 2D maps. By decoupling rotation from translation during the search process, global localization becomes feasible without the overhead of dense prior maps. Future research will focus on improving robustness in building-less environments and handling occlusions from non-building structures like dense vegetation or fences.

Summary updated

Abstract

Using publicly accessible maps, we propose a novel vehicle localization method that can be applied without using prior light detection and ranging (LiDAR) maps. Our method generates OSM descriptors by calculating the distances to buildings from a location in OpenStreetMap at a regular angle, and LiDAR descriptors by calculating the shortest distances to building points from the current location at a regular angle. Comparing the OSM descriptors and LiDAR descriptors yields a highly accurate vehicle localization result. Compared to methods that use prior LiDAR maps, our method presents two main advantages: (1) vehicle localization is not limited to only places with previously acquired LiDAR maps, and (2) our method is comparable to LiDAR map-based methods, and especially outperforms the other methods with respect to the top one candidate at KITTI dataset sequence 00.
BibTeX
@article{cho2022openstreetmap,
  author = {Younghun Cho and Giseop Kim and Sangmin Lee and Jee-Hwan Ryu},
  title = {{OpenStreetMap-based LiDAR Global Localization in Urban Environment without a Prior LiDAR Map}},
  journal = {IEEE Robotics Autom. Lett.},
  year = {2022},
  url = {https://openreview.net/forum?id=6DK7i94t5AG},
}
Figure