DTD: Density Triangle Descriptor for 3D LiDAR Loop Closure Detection.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Simultaneous Localization and Mapping (SLAM) systems require loop closure detection for long-term consistency.
• Existing LiDAR-based loop closure methods struggle with performance in complex environments due to reliance on limited geometric features.
• A need exists for robust and efficient loop closure techniques in SLAM.

Purpose of the Study:

• To introduce a novel loop closure detection method, the Density Triangle Descriptor (DTD).
• To improve the robustness and accuracy of loop closure detection in SLAM systems.
• To enhance computational efficiency compared to existing approaches.

Main Methods:

• Extraction of keypoints from LiDAR-derived density images.
• Construction of a rotation- and translation-invariant triangle-based global descriptor.
• Modeling local neighborhoods with Gaussian distributions and deriving entropy-based local descriptors.
• Utilizing hash indexing, entropy-based verification, and singular value decomposition for loop closure detection and pose estimation.

Main Results:

• The proposed DTD significantly improves average F1 max score (18.30%) and EP (20.08%) compared to STD.
• DTD achieves a 50.57% improvement in computational efficiency.
• The method demonstrates robust performance on solid-state LiDAR with non-repetitive scanning patterns.

Conclusions:

• The Density Triangle Descriptor (DTD) offers a robust and efficient solution for loop closure detection in SLAM.
• DTD's invariance properties and entropy-based local descriptors enhance performance in complex environments.
• The method shows broad applicability and generalization capabilities for various LiDAR systems.