When-to-Loop: Enhanced Loop Closure for LiDAR SLAM in Urban Environments Based on SCAN CONTEXT.

Area of Science:

• Robotics
• Computer Vision
• Autonomous Systems

Background:

• Global Navigation Satellite Systems (GNSSs) struggle with accuracy in urban canyons due to signal obstruction.
• Micro-Electro-Mechanical System (MEMS) Inertial Measurement Units (IMUs) offer autonomous navigation but suffer from error accumulation.
• LiDAR-based Simultaneous Localization and Mapping (SLAM) systems are used but face drift and error issues.

Purpose of the Study:

• To develop a novel loop closure detection method for LiDAR-based SLAM.
• To correct time-accumulated errors by identifying previously visited locations.
• To enhance the robustness and accuracy of autonomous navigation systems.

Main Methods:

• Leveraging vehicular drivable area and IMU trajectory for keyframe selection, identifying significant environmental changes.
• Extending the SCAN CONTEXT algorithm to incorporate overall point cloud distribution for robust loop closure constraints.
• Utilizing enhanced environmental feature extraction beyond simple height-based metrics.

Main Results:

• Achieved a 6% overall accuracy improvement on the KITTI dataset.
• Demonstrated a 17% accuracy improvement in local scenarios, showcasing enhanced robustness.
• Validated the effectiveness of the proposed method in correcting accumulated errors.

Conclusions:

• The novel loop closure detection method significantly improves LiDAR-based SLAM accuracy and robustness.
• Integrating drivable area and IMU data offers a more reliable approach to keyframe selection.
• The extended SCAN CONTEXT algorithm provides stronger loop closure constraints for autonomous navigation.