基于三维重建的结构表面裂缝检测方法
A Structural Surface Crack Detection Method Based on 3D Reconstruction
张昊宇 1丁勇 1李登华2
作者信息
- 1. 南京理工大学理学院,南京 210094
- 2. 南京水利科学研究院,南京 210029;水利部水库大坝安全重点实验室,南京 210029
- 折叠
摘要
针对传统基于数字图像处理的裂缝监测方法缺乏对裂缝区域参数集成化表达,提出一种基于三维重建的结构表面裂缝检测方法.该方法在提高裂缝边缘识别精度的同时,实现了裂缝的区域集成及参数计算.使用二维裂缝映射到三维模型的方法完成裂缝的定位.根据建筑物的立面高度和结构特点,规划摄影路径、采集影像数据、构建针对建筑裂缝的影像数据集;结合SfM重建技术建立裂缝所在外立面区域三维模型,根据相机光心到裂缝所在拟合平面的距离,实现无测距设备条件下物距信息的获取;利用级联神经网络完成结构表面裂缝的识别与分割;利用实景三维模型获取像素解析度完成裂缝参数的计算.试验结果表明,所提方法可准确识别建筑外立面裂缝,同时计算裂缝宽度和长度,最终完成裂缝集成与定位.裂缝宽度相对误差小于1.5％,具有良好的实用价值.
Abstract
In view of the lack of integrated expression of crack region parameters in traditional digital image processing-based crack monitoring methods,the paper proposed a structural surface crack detection method based on three-dimensional reconstruction.This method not only improves the accuracy of crack edge rec-ognition but also realizes the integration of crack regions and parameter calculation.The location of cracks can be determined by mapping two-dimensional cracks onto three-dimensional models.The main work in-cludes:planning the photographic path,collecting image data,and constructing an image dataset targeting building cracks according to the facade height and structural characteristics of the building;establishing a three-dimensional model of the exterior facade area with cracks using Structure from Motion SfM reconstruc-tion technology,and obtaining object distance information without the need for ranging equipment based on the distance from the camera center to the crack fitting plane;identifying and segmenting structural surface cracks using cascaded neural networks;calculating crack parameters using the pixel resolution obtained from the real-world 3D model.The experimental results demonstrated that the proposed method could accu-rately identify cracks in building facades while simultaneously calculating crack widths and lengths,ulti-mately achieving crack integration and localization.The relative error of crack width is less than 1.5％,in-dicating good practical value.
关键词
混凝土裂缝检测/深度学习/三维实景模型/像素解析度Key words
concrete crack detection/deep learning/3D realistic model/pixel resolution引用本文复制引用
基金项目
国家重点研发计划(2022YFC3005502)
国家自然科学基金(51979174)
国家自然科学基金联合基金(U2040221)
中央级公益性科研院所基本科研业务费专项(Y321004)
出版年
2024
NETL
NSTL
万方数据