Integrated LiDAR-Based Localization Correction Using a Dedicated Support Sign for Autonomous Vehicles

Area of Science:

• Autonomous Driving Systems
• Robotics and Computer Vision
• Geospatial Navigation Technologies

Background:

• Global Navigation Satellite System (GNSS) and Inertial Navigation System (INS) based vehicle localization suffers from significant error accumulation in tunnels due to signal degradation.
• Lane-level decision-making and path-following stability are compromised by inaccurate localization in GNSS-denied environments.

Purpose of the Study:

• To propose a novel LiDAR-based localization support system using dedicated signs for enhanced accuracy in tunnels.
• To develop a point-cloud-registration algorithm for real-time correction of localization errors.

Main Methods:

• A dedicated localization support sign was designed and installed in a tunnel environment.
• A PointPillars-based detector identified the dedicated sign.
• Point cloud registration to a pre-built reference map estimated a rigid correction transform for online localization refinement.

Main Results:

• The proposed pipeline demonstrated a substantial reduction in localization error post-sign recognition compared to a GNSS/INS-only baseline.
• The dedicated sign provided more stable corrections than natural tunnel structures within the same registration pipeline.
• LiDAR-based localization effectively suppressed drift in GNSS-degraded tunnel sections.

Conclusions:

• The developed dedicated sign and LiDAR-based pipeline offer a robust solution for stable vehicle localization in tunnels.
• This approach significantly improves localization accuracy and stability, crucial for autonomous driving functions in challenging environments.
• The findings highlight the potential of targeted infrastructure and advanced sensor fusion for reliable navigation where GNSS is unreliable.