Have I Seen This Place Before? A Fast and Robust Loop Detection and Correction Method for 3D Lidar SLAM

Area of Science:

• Robotics
• Computer Vision
• Geospatial Data Processing

Background:

• Accurate and efficient loop detection and correction are crucial for Simultaneous Localization and Mapping (SLAM) systems.
• Existing methods often struggle with significant variations in point cloud orientation and computational efficiency.

Purpose of the Study:

• To develop a robust and efficient system for loop detection and correction using lidar data.
• To improve the precision-recall trade-off in loop candidate identification.
• To enhance the accuracy of pose estimation through effective loop closure.

Main Methods:

• A hybrid approach combining a global point cloud matcher with a novel registration algorithm for loop candidate detection.
• GPU acceleration for the global point cloud matcher, achieving a speed-up factor of 2⁻4.
• Development of a new loop correction algorithm to refine pose estimates.

Main Results:

• The combined approach demonstrates superior loop detection reliability, achieving up to a 7% increase in precision at nearly 100% recall.
• The novel registration method effectively handles point clouds with large orientation deviations, improving efficiency.
• The loop correction algorithm reduces average and median pose error by a factor of 2 in seconds.

Conclusions:

• The proposed system offers a significant advancement in lidar-based SLAM, enhancing both detection accuracy and correction efficiency.
• The method provides a practical solution for real-time applications requiring precise mapping.
• This work contributes to more reliable and accurate autonomous navigation systems.