基于三维点云的结构表面裂缝提取方法  

Research on Structural Surface Crack Extraction Method Based on 3D Point Cloud

作　　者：吴玉龙[1] 张昊宇 张忠举 WU Yuong;ZHANG Haoyu;ZHANG Zhongju(Kunshan Construction Engineering Quality Inspection Center,Kunshan Jiangsu 215337,China;Nanjing University of Science and Technology,Nanjing Jiangsu 210094,China;PowerChina Sinohydro Bureau 10 Co.,Ltd.,Chengdu Sichuan 611830,China)

机构地区：[1]昆山市建设工程质量检测中心,江苏昆山215337 [2]南京理工大学,江苏南京210094 [3]中国水利水电第十工程局有限公司,四川成都611830

出　　处：《工程质量》2024年第11期73-79,共7页Construction Quality

摘　　要：裂缝识别是构建裂缝表面三维点云的基础,而获取高质量的点云是提取裂缝边缘信息的关键环节。针对传统裂缝图像无法关注裂缝所在混凝土结构全貌,提出一种基于三维点云的裂缝去噪和投影方法,利用改进的U-net 网络识别并标注图像中的裂缝区域;采用实景三维重建方法获取混凝土表面三维模型点云;通过基于空间分布的去噪算法(SOR),并结合点云RGB自适应阈值,完成裂缝兴趣区域提取和噪声消除。解决了较长贯穿裂缝的识别与点云提取问题,所提方法在裂缝宽度检测时绝对误差在范围内,具有一定工程应用价值。Crack identification is the basis for constructing 3D point clouds on crack surfaces,and obtaining high-quality point clouds is the key link for extracting crack edge information.Aiming at the traditional crack images which cannot focus on the whole concrete structure where the cracks are located,a crack denoising and projection method based on 3D point cloud is proposed,using the improved U-net network to identify and label the crack regions in the images;a real-world 3D reconstruction method is used to obtain the 3D model point cloud of the concrete surface;through the spatial distribution-based denoising algorithm(SOR),and combined with the point cloud RGB adaptive thresholding,the crack region of interest extraction and noise elimination.The problem of identification and point cloud extraction of longer penetration cracks is solved,and the proposed method has certain engineering application value as the absolute error in crack width detection is within±0.2mm.

关 键 词：裂缝提取 裂缝提取方法 三维点云 裂缝 深度学习 

分 类 号：TU755.7[建筑科学—建筑技术科学]