Branch-Pipe: Improving Graph Skeletonization around Branch Points in 3D Point Clouds
Branch-Pipe: Improving Graph Skeletonization around Branch Points in 3D Point Clouds
复制标题
DOI：
10.3390/rs13193802
复制
发表时间：
2021-09
期刊：
影响因子：
--
通讯作者：
Illia Ziamtsov;Kian Faizi;Saket Navlakha
中科院分区：
文献类型：
--
作者：
Illia Ziamtsov;Kian Faizi;Saket Navlakha
现代植物表型分析需要对噪声和缺失数据具有鲁棒性的工具，同时能够有效地处理大量植物。在这里，我们研究了从3D点云中对植物结构的重建，这对于许多下游任务至关重要，包括植物形状，形态和分支角度的分析。具体而言，我们开发了一种算法，以提高在分支点（叉）的几何特性的圆柱体周围的分支点。我们在一组不同的番茄和烟草植物的高分辨率3D点云上测试了这种算法，这些点云生长在五种环境中，跨越多个发育时间点。与现有的3D重建方法相比，我们的方法有效，更准确地估计分支角度，即使在有噪声，丢失或非均匀采样数据的区域。我们的方法也适用于无机数据集，如工业管道或包含复杂圆柱形网络的城市场景的扫描。
Modern plant phenotyping requires tools that are robust to noise and missing data, while being able to efficiently process large numbers of plants. Here, we studied the skeletonization of plant architectures from 3D point clouds, which is critical for many downstream tasks, including analyses of plant shape, morphology, and branching angles. Specifically, we developed an algorithm to improve skeletonization at branch points (forks) by leveraging the geometric properties of cylinders around branch points. We tested this algorithm on a diverse set of high-resolution 3D point clouds of tomato and tobacco plants, grown in five environments and across multiple developmental timepoints. Compared to existing methods for 3D skeletonization, our method efficiently and more accurately estimated branching angles even in areas with noisy, missing, or non-uniformly sampled data. Our method is also applicable to inorganic datasets, such as scans of industrial pipes or urban scenes containing networks of complex cylindrical shapes.