基于目标边界视觉追踪的结构多水准变形状态识别  

Multi-level structural deformation estimation by vision-based object boundary tracking

作　　者：任晨辉 单伽锃 刘钰汶 吴昊[1] 周德源[1] Ren Chenhui;Shan Jiazeng;Liu Yuwen;Wu Hao;Zhou Deyuan(Tongji University,Shanghai 200092,China;Shanghai Engineering Research Center for Resilient Cities and Intelligent Disaster Mitigation,Shanghai 200092,China)

机构地区：[1]同济大学,上海200092 [2]上海韧性城市与智能防灾工程技术研究中心,上海200092

出　　处：《土木工程学报》2023年第10期11-19,共9页China Civil Engineering Journal

基　　金：国家自然科学基金(51878483);上海市青年科技启明星计划(20QC1400700)。

摘　　要：工程结构不同水准变形状态的监测与识别对于评估其结构性能具有重要意义。传统的结构健康监测方法大多具有接触式测量、离散布置、局部观测的特点,无法充分考虑工程结构的连续体物理特征,获得结构多水准连续型变形信息。基于视觉追踪的测量技术因其非接触、低成本、易布置等优点逐渐成为结构健康监测的新兴方法之一,此外,近年来计算机视觉领域也提出了多种目标检测模型可以鲁棒识别显著视觉对象。该文结合结构变形机理、视觉测量技术与显著目标检测算法,从图像中提取并追踪结构目标区域的连续边界,实现了结构多水准变形状态识别,并通过足尺钢筋混凝土梁落锤冲击试验对提出的方法进行了验证。研究结果表明,提出的方法可以稳定追踪结构边缘在毫秒级别的变化状态,准确识别包括竖向位移、构件转角和截面曲率在内的结构多水准变形状态,为进一步评估结构性能提供了重要数据与信息。The monitoring and identification of multi-level deformation states of engineering structures is of great significance to evaluate their structural performance.Most of the traditional structural health monitoring methods have the characteristics of contact measurement,discrete layout and local observation,but they cannot fully consider the physical characteristics of the continuum of engineering structures and obtain the multi-level continuous deformation information of structures.The measurement technology based on visual tracking has gradually become one of the new methods for structural health monitoring because of its advantages of non-contact measurement,low cost and easy layout.In addition,in recent years,a variety of target detection models in the field of computer vision have been proposed to robustly identify outstanding visual objects.In this work,combined with the structural deformation mechanism,visual measurement technology and salient object detection algorithm,the continuous boundary of the structural target region is extracted and tracked from the image,and the multi-level deformation state recognition of the structure is realized.The proposed method is verified by the full-scale drop hammer impact test on reinforced concrete beams.The results show that the proposed method can stably track the change state of the structural edge at the millisecond level,and accurately identify the multi-level deformation state of the structure,including vertical displacement,member corner and section curvature.It provides important data and information for further evaluation of structural performance.

关 键 词：结构监测 变形识别 显著性目标检测 梁曲率 

分 类 号：TU375.1[建筑科学—结构工程]