A Fast LiDAR Place Recognition Descriptor Based on Density Classification and Multi-Modal Fusion Place Recognition Strategy

Abstract

Place recognition has become increasingly important for navigation and localization in diverse environments, including urban areas, large-scale buildings, and mixed indoor-outdoor spaces. However, this task remains challenging due to sensor limitations and environmental appearance changes. To address these challenges, we propose Density Classification Scan Context (DCSC), a novel LiDAR descriptor that enhances place recognition efficiency without compromising accuracy. Additionally, we introduce Aw-Fusion, an adaptive weighting-based Visual-LiDAR fusion method for robust place recognition across varying environments. Aw-Fusion dynamically adjusts the contributions of visual and LiDAR features in the global descriptor, improving adaptability in different scenarios. This approach is suitable for applications such as robotic navigation, autonomous vehicles, and loop closure detection. Extensive experiments on the KITTI and UrbanNav datasets, as well as real-world platforms, validate the effectiveness of our method. Results demonstrate that DCSC significantly improves place recognition efficiency, while Aw-Fusion enhances system robustness in diverse environments.