LiDAR-OSM-Based Vehicle Localization in GPS-Denied Environments by Using Constrained Particle Filter

Area of Science:

• Robotics
• Computer Vision
• Geographic Information Systems

Background:

• Accurate vehicle localization is critical for autonomous driving systems.
• Existing methods often rely on complex learning models or extensive labeled data.

Purpose of the Study:

• To develop a novel, deterministic approach for cross-modal vehicle localization.
• To improve localization accuracy by integrating LiDAR data with OpenStreetMaps (OSM) information.

Main Methods:

• A constrained particle filter framework was utilized.
• LiDAR point clouds were fused with geometric constraints derived from OpenStreetMaps (OSM).
• The approach does not involve any learning components or require labeled data.

Main Results:

• The proposed method achieved accurate vehicle pose tracking.
• A mean position error of less than 3 meters was recorded on the KITTI dataset.
• The integration of OSM provided geometrical constraints that enhanced localization accuracy.

Conclusions:

• The novel approach significantly improves vehicle localization accuracy for automated driving.
• This method offers state-of-the-art performance compared to existing OSM or satellite map-based techniques.
• The deterministic nature and lack of learning requirements make it a practical solution.