基于计算机视觉的结构应变无靶标鲁棒监测  

Unmarked robust monitoring of structural strain based on computer vision

作　　者：朱前坤[1,2] 王军营 杜永峰[1,2] 张琼[1,2] ZHU Qiankun;WANG Junying;DU Yongfeng;ZHANG Qiong(Institute of Earthquake Protection and Disaster Mitigation,Lanzhou University of Technology,Lanzhou 730050,China;Western Center of Disaster Mitigation in Civil Engineering of Ministry of Education,Lanzhou University of Technology,Lanzhou 730050,China)

机构地区：[1]兰州理工大学防震减灾研究所,甘肃兰州730050 [2]兰州理工大学西部土木工程防灾减灾教育部工程研究中心,甘肃兰州730050

出　　处：《建筑结构学报》2023年第10期211-221,共11页Journal of Building Structures

基　　金：国家自然科学基金项目(52168041,51868046)。

摘　　要：传统视觉监测方法对结构的现场监测精度往往受控于人造靶标和光照条件等因素,为克服现场光照条件对视觉测量精度的影响,结合基于相位的稠密光流算法和支持向量回归(support vector regression,SVR)算法对现场结构应变的无靶标鲁棒监测。建立基于应变传感器原理的应变转换方法实现结构表面连续应变场的计算,并通过模拟测试试验和现场测试试验验证所提方法可行性。模拟测试试验结果表明:在测量精度相当的情况下,所提方法相较于传统DIC算法的运算速度提升50%,且能够得到更为清晰完整展现细节的应变云图;现场测试试验中,所提方法表现出较好的环境抗干扰能力,与传统测试方法对比的应变测量误差可控制在2.0%以内。相较于传统视觉监测方法,所提方法在保证精度要求的前提下提高了运算速度及鲁棒性,且无需人造靶标,可适用于特定大型工程结构表面应变的现场监测。The accuracy of traditional visual monitoring methods for the monitoring of field structures is often controlled by factors such as artificial targets and lighting conditions.In order to overcome the influence of illumination conditions on the accuracy of visual measurement,a method combining the phase-based dense optical flow algorithm and the support vector regression(SVR)algorithm was proposed to achieve robust monitoring of structural strain in the field.The method consists of image pre-processing based on 2D Gabor filter,phase-based dense optical flow sub-pixel displacement field matching calculation and displacement field smoothing based on SVR algorithm.After that,a strain conversion method based on the principle of strain sensors was used to realize the calculation of continuous strain fields on the surface of the structure,and the feasibility of the proposed method was verified by simulation test experiments and field test experiments.In the simulation test,the proposed method has a computation speed 50%faster than the conventional DIC algorithm with comparable measurement accuracy,and a clearer and more complete strain cloud map can be obtained.In the field test,the proposed method shows better environmental immunity,and the strain measurement error can be controlled within 2.0%compared with the conventional test method.Compared with the traditional visual monitoring method,the proposed method improves the computational speed and robustness while ensuring the accuracy requirements,and it is applicable to the field monitoring of strains on the surface of specific large engineering structures without the need for artificial targets.

关 键 词：结构表面应变 计算机视觉 相位 支持向量回归 准静态分析 应变云图 

分 类 号：TU312.1[建筑科学—结构工程]