Multi-Scale Triplet Descriptors for Global LiDAR Localization with Maximum Clique-Based Enhancement

This paper presents an accurate and robust global localization system by matching a single LiDAR scan against a global map. To enhance global pose estimation accuracy in environments with sparse semantic information, we first introduce a triplet descriptor based on multi-scale edge structures. By segmenting edge lengths with multiple thresholds, the method constructs triangular structures at different scales, enabling the extraction of hierarchical vertex descriptors that better enhance discriminability and maximize the use of limited information. To support the proposed descriptor structure, we further design a dynamic maximal clique enhancement strategy that enhances inlier selection accuracy in sparse semantic scenes while avoiding redundant information in semantically rich environments, maintaining computational efficiency. Experimental results on public datasets demonstrate that our proposed method outperforms existing state-of-the-art approaches in terms of both descriptor discriminability and pose estimation accuracy. © 2025 IEEE.