处理地形点云后，如何将PyVista的UnstructuredGrid转存为STL？

问题描述 #

使用PyVista处理3D扫描的.xyz点云数据，生成网格后导出STL用于Blender。单个连续区域的处理流程正常，但处理含多个不连续小区域的点云时，delaunay_2d()会生成连接不同区域的大面。尝试通过面面积阈值删除大面后，网格类型从可导出STL的PolyData变为UnstructuredGrid，无法直接导出，希望省去后续用MeshLab处理的步骤。

连续区域处理代码：

import numpy as np
import pyvista as pv

xyz_path = r"Path_to_file.xyz"
stl_path = r"Path_to_file.stl"

data = np.loadtxt(xyz_path)
cloud = pv.PolyData(data)
surf = cloud.delaunay_2d()
surf.plot(show_edges=True)
surf.save(stl_path)

不连续区域尝试的过滤代码（导致类型变为UnstructuredGrid）：

import numpy as np
import pyvista as pv

xyz_path = r"Path_to_file.xyz"
stl_path = r"Path_to_file.stl"

data = np.loadtxt(xyz_path)
cloud = pv.PolyData(data)
surf = cloud.delaunay_2d()

cell_sizes = surf.compute_cell_sizes(length=False, area=True, volume=False)
areas = cell_sizes['Area']
max_area_threshold = 100  # 自定义面积阈值

valid_cells = areas < max_area_threshold
filtered_mesh = surf.extract_cells(valid_cells)
filtered_mesh.plot(show_edges=True, scalars='Area')

# 原代码错误保存了未过滤的surf，且filtered_mesh类型不支持直接导出STL

解决方案 #

方法1：将UnstructuredGrid转换为PolyData #

过滤后的UnstructuredGrid若仅包含三角形面，可直接转换为PolyData，支持STL导出：

import numpy as np
import pyvista as pv

xyz_path = r"Path_to_file.xyz"
stl_path = r"Path_to_file.stl"

data = np.loadtxt(xyz_path)
cloud = pv.PolyData(data)
surf = cloud.delaunay_2d()

cell_sizes = surf.compute_cell_sizes(length=False, area=True, volume=False)
areas = cell_sizes['Area']
max_area_threshold = 100

valid_cells = areas < max_area_threshold
filtered_mesh = surf.extract_cells(valid_cells)

# 转换为PolyData（若网格仅含三角形面，直接转换即可）
polydata_mesh = pv.PolyData(filtered_mesh)
# 或用extract_surface确保是纯表面网格
# polydata_mesh = filtered_mesh.extract_surface()

polydata_mesh.plot(show_edges=True, scalars='Area')
polydata_mesh.save(stl_path)

方法2：使用threshold过滤，保持PolyData类型 #

改用threshold方法基于单元面积过滤，该方法处理PolyData时会保留原类型：

import numpy as np
import pyvista as pv

xyz_path = r"Path_to_file.xyz"
stl_path = r"Path_to_file.stl"

data = np.loadtxt(xyz_path)
cloud = pv.PolyData(data)
surf = cloud.delaunay_2d()

# 直接在原网格上计算单元面积（inplace=True修改原对象）
surf.compute_cell_sizes(length=False, area=True, volume=False, inplace=True)
# 过滤面积小于阈值的单元，返回仍为PolyData
filtered_mesh = surf.threshold(scalars='Area', upper=100)

filtered_mesh.plot(show_edges=True, scalars='Area')
filtered_mesh.save(stl_path)

方法3：先聚类点云，分区域生成网格 #

针对多不连续区域，先对点云聚类拆分，再分别生成网格后合并，从根源避免跨区域大面：

import numpy as np
import pyvista as pv

xyz_path = r"Path_to_file.xyz"
stl_path = r"Path_to_file.stl"

data = np.loadtxt(xyz_path)
cloud = pv.PolyData(data)

# 聚类点云，tolerance需根据点云密度调整（单位与点云坐标一致）
clustered = cloud.cluster_points(tolerance=5.0)
labels = clustered['ClusterId']
unique_labels = np.unique(labels)

# 逐个处理每个独立区域
mesh_list = []
for label in unique_labels:
    # 提取当前区域的点
    region_points = clustered.extract_points(labels == label)
    # 生成该区域的网格
    region_mesh = region_points.delaunay_2d()
    mesh_list.append(region_mesh)

# 合并所有区域的网格为单个PolyData
combined_mesh = pv.MultiBlock(mesh_list).combine()
combined_mesh.save(stl_path)

内容的提问来源于stack exchange，提问作者Melch