This paper presents an alternative to segmentation of point clouds tailored specifically to large-scale steel buildings. Typical segmentation approaches process the 3D point data directly and focus on large blocky structures such as concrete; these are not generalizable to smaller, more complex geometries found in steel elements. The method takes advantage of image processing techniques by utilizing 2D “slices” of the point cloud, rather than the original 3D point cloud. Centroids of targeted structural cross sections are extracted from these slices using 2D convolution as a template-matching operation, and then projected back to 3D. From this, member centroidal axes are extracted using a custom linear region growing algorithm to create a 1D beam line model, including connections. Experimental results from scans of four prefabricated steel buildings indicate that the method is robust to variations in framing systems, clutter (e.g., obstructions, nonstructural elements), and point cloud sparsity.

中文翻译：

---

从钢结构建筑的点云中自动提取结构梁线和连接

本文提出了一种专门为大型钢结构建筑量身定制的点云分割的替代方法。典型的分割方法直接处理 3D 点数据，并专注于大型块状结构，如混凝土；这些不能推广到钢元素中更小、更复杂的几何形状。该方法通过利用点云的 2D“切片”而不是原始 3D 点云来利用图像处理技术。使用 2D 卷积作为模板匹配操作从这些切片中提取目标结构横截面的质心，然后投影回 3D。由此，使用自定义线性区域增长算法提取成员质心轴，以创建一维光束线模型，包括连接。