Scholarlink

<p>In large-scale environments, LiDAR-based Simultaneous Localization and Mapping (SLAM) often encounters loop closure failures, resulting in uncorrected global accumulated errors. Compared with visual SLAM, which benefits from a variety of feature detectors and descriptors, LiDAR SLAM has received relatively limited attention in leveraging structural information for scene representation. To address this limitation, this study proposes a global descriptor based on weighted ternary features, which exhibits both rotation and translation invariance. Firstly, geometrically constrained planar and edge points are extracted to enhance the discriminative power of features. Then, a local adaptive coordinate system is constructed using each feature point and its principal orientation, enabling the extraction of stable and physically interpretable weighted ternary features and the parameterized modeling of point-to-point geometric relationships. Finally, a weighting mechanism that integrates normal vector deviation and curvature difference is introduced to strengthen local structural consistency, along with a density-based weighting strategy to capture global-level representations. Experimental results on the public KITTI dataset demonstrate that the proposed method reduces the root mean square error (RMSE) by up to 32% compared to LeGO-LOAM and improves the similarity score by up to 12% over Scan Context.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Weighted Ternary Feature Based Loop Closure Detection Method for LiDAR SLAM

  • 期刊论文

  • 发表日期：

    2025年10月24日

查看全文
  • Dong Zhang,
  • Tao Xiang,
  • Hao Tang,
  • He Yan

摘要

In large-scale environments, LiDAR-based Simultaneous Localization and Mapping (SLAM) often encounters loop closure failures, resulting in uncorrected global accumulated errors. Compared with visual SLAM, which benefits from a variety of feature detectors and descriptors, LiDAR SLAM has received relatively limited attention in leveraging structural information for scene representation. To address this limitation, this study proposes a global descriptor based on weighted ternary features, which exhibits both rotation and translation invariance. Firstly, geometrically constrained planar and edge points are extracted to enhance the discriminative power of features. Then, a local adaptive coordinate system is constructed using each feature point and its principal orientation, enabling the extraction of stable and physically interpretable weighted ternary features and the parameterized modeling of point-to-point geometric relationships. Finally, a weighting mechanism that integrates normal vector deviation and curvature difference is introduced to strengthen local structural consistency, along with a density-based weighting strategy to capture global-level representations. Experimental results on the public KITTI dataset demonstrate that the proposed method reduces the root mean square error (RMSE) by up to 32% compared to LeGO-LOAM and improves the similarity score by up to 12% over Scan Context.