Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Beyond the Vehicle: Cooperative Localization by Fusing Point Clouds for GPS-Challenged Urban Scenarios

Kuo-Yi Chao, Ralph Rasshofer, Alois Christian Knoll

TL;DR

The paper tackles reliable vehicle localization in GPS-challenged urban scenarios by merging point-cloud data from vehicles and intersection infrastructure. It presents a cooperative, four-stage pipeline that fuses multi-modal sensor data, performs online SLAM, and registers ego point clouds against a shared reference map using ICP with RANSAC. Fast Point Feature Histograms provide robust local features for registration, enabling accurate alignment despite urban multipath and occlusions. CARLA-based experiments demonstrate that the fused system achieves millimeter-level positional accuracy in valid registrations, far surpassing GPS-only and SLAM-alone baselines, highlighting the practical potential of V2X-enabled cooperative localization for autonomous driving.

Abstract

Accurate vehicle localization is a critical challenge in urban environments where GPS signals are often unreliable. This paper presents a cooperative multi-sensor and multi-modal localization approach to address this issue by fusing data from vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) systems. Our approach integrates cooperative data with a point cloud registration-based simultaneous localization and mapping (SLAM) algorithm. The system processes point clouds generated from diverse sensor modalities, including vehicle-mounted LiDAR and stereo cameras, as well as sensors deployed at intersections. By leveraging shared data from infrastructure, our method significantly improves localization accuracy and robustness in complex, GPS-noisy urban scenarios.

Beyond the Vehicle: Cooperative Localization by Fusing Point Clouds for GPS-Challenged Urban Scenarios

TL;DR

The paper tackles reliable vehicle localization in GPS-challenged urban scenarios by merging point-cloud data from vehicles and intersection infrastructure. It presents a cooperative, four-stage pipeline that fuses multi-modal sensor data, performs online SLAM, and registers ego point clouds against a shared reference map using ICP with RANSAC. Fast Point Feature Histograms provide robust local features for registration, enabling accurate alignment despite urban multipath and occlusions. CARLA-based experiments demonstrate that the fused system achieves millimeter-level positional accuracy in valid registrations, far surpassing GPS-only and SLAM-alone baselines, highlighting the practical potential of V2X-enabled cooperative localization for autonomous driving.

Abstract

Accurate vehicle localization is a critical challenge in urban environments where GPS signals are often unreliable. This paper presents a cooperative multi-sensor and multi-modal localization approach to address this issue by fusing data from vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) systems. Our approach integrates cooperative data with a point cloud registration-based simultaneous localization and mapping (SLAM) algorithm. The system processes point clouds generated from diverse sensor modalities, including vehicle-mounted LiDAR and stereo cameras, as well as sensors deployed at intersections. By leveraging shared data from infrastructure, our method significantly improves localization accuracy and robustness in complex, GPS-noisy urban scenarios.
Paper Structure (10 sections, 2 figures, 1 table)

This paper contains 10 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Methodology
  3. Sources of GPS Error
  4. Online SLAM
  5. Fast Point Feature Histograms
  6. Point Cloud Registration
  7. RANSAC
  8. Experiments and Results
  9. Conclusion
  10. Acknowledgments.

Figures (2)

  • Figure 1: Pipeline for sensor fusion to correct noisy GPS data. The pipeline contains 4 steps to correct its localization by using external sensor information from other agents: sensor fusion, registration, trajectory estimation, and finally correction.
  • Figure 2: Simulation screenshots CARLA from each scenario at a single intersection, displaying the ego vehicle (orange path), two agent vehicles (blue paths/dots), and four infrastructure sensors (aqua dots) at corners for traffic monitoring. Sim 0: ego crosses parallel opposing paths with one agent; Sim 1: agents stationary; Sim 2: ego crosses perpendicular paths with one agent; Sim 3: ego and one agent execute left turn.