RECONSTRUCTION OF SCAFFOLDING COMPONENTS FROM PHOTOGRAMMETRIC POINT CLOUDS OF A CONSTRUCTION SITE
RECONSTRUCTION OF SCAFFOLDING COMPONENTS FROM PHOTOGRAMMETRIC POINT CLOUDS OF A CONSTRUCTION SITE
复制标题
DOI：
10.5194/isprsannals-ii-3-w5-401-2015
复制
发表时间：
2015-08
期刊：
影响因子：
--
通讯作者：
Yusheng Xu;Yusheng Xu;J. He;J. He;S. Tuttas;Uwe Stilla
中科院分区：
文献类型：
--
作者：
Yusheng Xu;Yusheng Xu;J. He;J. He;S. Tuttas;Uwe Stilla
提出了一种数据驱动的从点云中检测脚手架构件的工作流程。确定属于脚手架构件的点，并将其从主要建筑结构中分离出来，重建两个基本元素，即脚板和管。工作流有四个主要的处理步骤。首先，对原始点云进行统计滤波和体素分割预处理。第二步，通过RANSAC提取建筑表面和脚手架的平面，然后根据它们的平行度和距离进行分组，从而分离出建筑立面。第三步，采用三维形状描述符FPFH和随机森林分类算法对建筑立面点数据进行分类，将其划分为属于不同元素的类。最后，利用线性拟合算法和SHOT形状描述符进行匹配，重构出管材和趾板的几何参数。结果表明，识别出这些目标的点，然后用柱面和长方体模型进行重建。最终结果表明，在被调查的立面中，超过60%的管道和近90%的脚板被重建，超过40%的重建物体被重建得很好。
This paper presents a data-driven workflow for the detection of scaffolding components from point clouds. The points belonging to the scaffolding components are identified and separated from the main building structures and two basic elements, namely the toeboard and the tube, are reconstructed. The workflow has four main processing steps. Firstly, the raw point clouds are preprocessed by statistical filtering and voxel girding. In the second step, the planar surfaces of the building surface and scaffoldings are extracted via RANSAC and then grouped by their parallelity and distance to separate the building facade. In the third step, the 3D shape descriptor FPFH and random forest classification algorithm are applied to classify the point data of building facades into classes belonging to different elements. Finally, by the use of linear fitting algorithm and matching using SHOT shape descriptor, the tubes and toeboards are reconstructed with their geometric parameters. It is shown that the points belonging to these objects are identified and then reconstructed with cylinder and cuboid models. The final results show that over 60% of the tubes and nearly 90% of the toeboards are reconstructed in the investigated facade, and more than 40% of the reconstructed objects are well rebuilt.