Rapid Localization and Mapping Method Based on Adaptive Particle Filters.

Area of Science:

• Robotics
• Computer Vision
• Autonomous Systems

Background:

• Accurate localization and mapping are critical for autonomous vehicle operation.
• GPS-denied environments pose significant challenges for existing localization methods.

Purpose of the Study:

• To develop and validate a robust, GPS-independent localization and mapping architecture for autonomous vehicles.
• To enhance vehicle operation in challenging environments like urban canyons and tunnels.

Main Methods:

• Feature extraction from LiDAR scenes using K-means clustering to create local and global maps.
• An adaptive particle filter employing particle generation, motion update, and weighted selection based on map matching for localization.
• Data association between frames facilitated by concatenated local maps.

Main Results:

• The proposed method demonstrates effective localization without GPS reliance.
• Validation on Kitti and Pandaset datasets shows high performance across diverse environmental conditions.
• The approach achieves competitive speed and feature extraction representativeness compared to state-of-the-art techniques.

Conclusions:

• The developed localization and mapping system provides a reliable solution for autonomous vehicles in GPS-denied areas.
• The K-means and adaptive particle filter combination offers a robust and efficient method for real-time positioning.
• This work contributes to advancing the operational capabilities of autonomous vehicles in complex real-world scenarios.