ABSTRACT The objective of this study was to evaluate the effects of the three dimensional (3D) point cloud density derived from Unmanned Aerial Vehicle (UAV) photogrammetry (using Structure from Motion (SfM) and Multi-View Stereopsis (MVS) techniques), the interpolation method for generating a digital terrain model (DTM), and the resolution (grid size (GS)) of the derived DTM on the accuracy of estimated heights in small areas, where a very accurate high spatial resolution is required. A UAV-photogrammetry project was carried out on 13 m × 13 m bare soil with a rotatory wing UAV at 10 m flight altitude (equivalent ground sample distance = 0.4 cm), and the 3D point cloud was derived. A stratified random sample (200 points in each square metre) was extracted and from the rest of the cloud, 15 stratified random samples representing 1, 2, 3, 4, 5, 10, 15, 20, 30, 40, 50, 60, 70, 80, and 90% were extracted. Five replications of each percentage were extracted to analyse the effect of cloud density on DTM accuracy. For each of these 15 × 5 = 75 samples, DTMs were derived using four different interpolation methods (Inverse Distance Weighted (IDW), Multiquadric Radial Basis Function (MRBF), Kriging (KR), and Triangulation with Linear Interpolation (TLI)) and 15 DTM GS values (20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0.67, 0.50, and 0.40 cm). Then, 75 × 4 × 15 = 4500 DTMs were analysed. The results showed an optimal GS value for each interpolation method and each density (most of the cases were equal to 1 cm) for which the Root Mean Square Error (RMSE) was the minimum. IDW was the interpolator that yielded the best accuracies for all combinations of densities and GS. Its RMSE when considering the raw cloud was 1.054 cm and increased by 3% when a point cloud with 80% extracted from the raw cloud was used to generate the DTM. When the point cloud included 40% of the raw cloud, RMSE increased by 5%. For densities lower than 15%, RMSE increased exponentially (45% for 1% of raw cloud). The GS minimizing RMSE for densities of 20% or higher was 1 cm, which represents 2.5 times the ground sample distance of the pictures used for developing the photogrammetry project.

中文翻译：

---

微地形级点云密度、插值方法和网格大小对导出的数字地形模型精度的影响

摘要 本研究的目的是评估从无人驾驶飞行器 (UAV) 摄影测量（使用运动结构 (SfM) 和多视立体视 (MVS) 技术）得出的三维 (3D) 点云密度的影响，用于生成数字地形模型 (DTM) 的插值方法，以及派生 DTM 的分辨率（网格大小 (GS)）对小区域中估计高度的准确性，其中需要非常准确的高空间分辨率。使用旋翼无人机在10 m飞行高度（等效地面采样距离= 0.4 cm）在13 m×13 m裸土上进行无人机摄影测量项目，并导出3D点云。抽取了分层随机样本（每平方米 200 个点），并从云的其余部分提取了 15 个分层随机样本，分别代表 1、2、3、4、5、10、提取了 15、20、30、40、50、60、70、80 和 90%。提取每个百分比的五次重复以分析云密度对 DTM 精度的影响。对于这 15 × 5 = 75 个样本中的每一个，使用四种不同的插值方法（反距离加权 (IDW)、多二次径向基函数 (MRBF)、克里金法 (KR) 和带线性插值的三角测量 (TLI)）推导出 DTM 和15 个 DTM GS 值（20、15、10、9、8、7、6、5、4、3、2、1、0.67、0.50 和 0.40 厘米）。然后，分析了 75 × 4 × 15 = 4500 个 DTM。结果显示了每种插值方法和每种密度（大多数情况下等于 1 cm）的最佳 GS 值，其中均方根误差 (RMSE) 最小。IDW 是对所有密度和 GS 组合产生最佳精度的插值器。考虑到原始云时，其 RMSE 为 1.054 cm，当使用从原始云中提取 80% 的点云生成 DTM 时，其 RMSE 增加了 3%。当点云包含原始云的 40% 时，RMSE 增加了 5%。对于低于 15% 的密度，RMSE 呈指数增长（原始云的 1% 为 45%）。密度为 20% 或更高的 GS 最小 RMSE 为 1 厘米，这表示用于开发摄影测量项目的图片地面样本距离的 2.5 倍。