基于树冠体素特征的森林机载-地基激光点云无控配准

A Mark-Free Registration Method for Forest Airborne and Terrestrial Laser Scanning Point Clouds Fusion Based on Canopy Voxel Features

摘要:

高精度森林机载-地基点云配准能够极大提高点云数据完整性，为森林生态参数反演提供良好的数据支撑。针对森林环境特征稀疏下的机载-地基点云配准问题，提出了一种基于树冠体素特征的机载-地基激光点云数据无控配准方法。通过包围盒位置方向初始化、树高-频数的树冠点云滤波、树冠点云体素化、基于滑动模板匹配的特征提取以及旋转矩阵求解等步骤实现机载-地基点云配准。通过不同森林结构复杂度的4组数据对所提出的方法进行了实验验证，获得的平均距离差异分别为0.245m、0.238m、0.184m以及0.020m。与最近邻迭代法、人工配准及二者组合方法这3种方法进行对比实验，结果表明，本文提出的无控配准方法在精度和稳定性上均优于这3种方法。

Abstract:

Accurate registration for forest airborne laser scanning （ALS） point clouds and terrestrial laser scanning （TLS） point clouds improves data integrity， which can better serve for forest ecological parameter retrieval. A marker-free registration method is proposed to fuse ALS point clouds and TLS point clouds based on canopy voxel features. The method achieves ALS-TLS point clouds registration with the following steps： bounding box orientation， density-height-based canopy point clouds filtering， canopy point clouds voxelization， tie feature extraction with sliding voxel templates， and transformation matrix calculation. The proposed method was validated by four datasets with different forest structure complexities， which shows that its average differences are 0.245m， 0.238m， 0.184m， and 0.020m respectively. The comparisons indicates that the proposed method outperforms its counterparts including iterative closest point algorithm， artificial matching， and the combination of the two.