A LiDAR-Based Indoor Localization Method Using Skeleton Indexing and sLink3D

LiDAR-based localization systems can achieve centimeter-level accuracy. However, global pose ambiguity remains a significant challenge in highly self-similar indoor environments, such as long corridors. This challenge is exacerbated during the initial localization stage without any prior pose information or after task interruption. In these cases, the absence of an initial pose estimate or a continuous trajectory prior makes global localization particularly challenging. To address this issue, a global LiDAR-based indoor localization method using skeleton indexing and sLink3D (Structured Link3D) descriptors is proposed to enhance localization performance by exploiting indoor structural elements and their spatial connectivity. Semantic structures, including wall corners as well as door and window features, are first extracted from point cloud data. A skeleton graph is then constructed to index the spatial connectivity of semantic structures across the global map. Local feature subsets are derived from skeleton-indexed partitions with an adaptive step size, which helps maintain neighborhood consistency during retrieval and matching. Finally, within each local feature subset, the sLink3D feature descriptor is constructed to encode both geometric and semantic information for robust matching and global localization. Experimental results on indoor data sequences with long corridors and similar layouts demonstrate that the proposed method significantly improves global localization performance and pose estimation accuracy compared with representative global localization methods.