Scholarlink

<p>Accurate forest mapping is essential for understanding and managing forest ecosystems. In recent years, Light Detection and Ranging (LiDAR) technology has become a powerful tool for forest monitoring. However, registering UAV-LiDAR strips remains challenging in GNSS-degraded forest environments due to sparse ground features, canopy occlusion, and limited availability of reliable control points. Existing registration methods often rely on surface-based or feature-plane matching, which can be suboptimal in structurally heterogeneous forests. To address this gap, we present a robust and target-less registration framework for UAVLS data that leverages tree localization as structural anchors. Each detected tree is encoded into a graph using three spatial descriptors: (a) planimetric distance, (b) vertical distance, and (c) height difference relative to neighboring trees. Initial correspondences between trees across overlapping strips are established using the Hungarian algorithm, which maximizes a similarity function derived from these descriptors. Subsequently, a three-dimensional rigid transformation (rotation and translation) is estimated via Particle Swarm Optimization (PSO) to refine the spatial alignment. The method was validated on eight circular forest plots (30 m in diameter) derived from two overlapping UAVLS strips. The root mean square error (RMSE) of residual distances between registered tree pairs ranged from ±17.6 cm to ±27.4 cm for poplar plots and from ±17.5 cm to ±24.3 cm for dawn blackwood plots. Furthermore, the proposed method improved the tree matching ratio by 17%–27% compared to a baseline approach, demonstrating higher alignment accuracy. By integrating graph-based correspondence with swarm-based optimization, this study contributes a scalable, accurate, and GNSS-independent registration solution tailored to the complexities of forest environments. The framework has strong potential to support UAV-LiDAR-based forest mapping, monitoring, and inventory tasks in both research and operational settings.</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

Target-less registration of UAV-LiDAR point clouds based on graph matching of tree locations in forest environments

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月15日

查看全文
  • Reda Fekry,
  • Eslam Ali,
  • Abubakar Sani-Mohammed,
  • Mohamed Abdallah,
  • Doha Amr

摘要

Accurate forest mapping is essential for understanding and managing forest ecosystems. In recent years, Light Detection and Ranging (LiDAR) technology has become a powerful tool for forest monitoring. However, registering UAV-LiDAR strips remains challenging in GNSS-degraded forest environments due to sparse ground features, canopy occlusion, and limited availability of reliable control points. Existing registration methods often rely on surface-based or feature-plane matching, which can be suboptimal in structurally heterogeneous forests. To address this gap, we present a robust and target-less registration framework for UAVLS data that leverages tree localization as structural anchors. Each detected tree is encoded into a graph using three spatial descriptors: (a) planimetric distance, (b) vertical distance, and (c) height difference relative to neighboring trees. Initial correspondences between trees across overlapping strips are established using the Hungarian algorithm, which maximizes a similarity function derived from these descriptors. Subsequently, a three-dimensional rigid transformation (rotation and translation) is estimated via Particle Swarm Optimization (PSO) to refine the spatial alignment. The method was validated on eight circular forest plots (30 m in diameter) derived from two overlapping UAVLS strips. The root mean square error (RMSE) of residual distances between registered tree pairs ranged from ±17.6 cm to ±27.4 cm for poplar plots and from ±17.5 cm to ±24.3 cm for dawn blackwood plots. Furthermore, the proposed method improved the tree matching ratio by 17%–27% compared to a baseline approach, demonstrating higher alignment accuracy. By integrating graph-based correspondence with swarm-based optimization, this study contributes a scalable, accurate, and GNSS-independent registration solution tailored to the complexities of forest environments. The framework has strong potential to support UAV-LiDAR-based forest mapping, monitoring, and inventory tasks in both research and operational settings.