【作者】 黄琳；

【导师】 廖飞宇； 骆勇鹏； 陈培建；

【作者基本信息】 福建农林大学 ， 建筑与土木工程（专业学位）， 2022， 硕士

【摘要】 结构位移是进行结构安全评估的重要参数之一。有效、快速且精准地获得位移数据可提高工程结构评估结果的实时性及可靠性。随着计算机视觉和深度学习的快速发展,图像自动定位问题取得巨大突破,为结构位移的测量提供新的思路。采用目标检测算法与角点检测算法对测量到的图像进行自动识别和定位,结合相机标定来计算结构位移,从而大大减少人工工作量,提高测量精度和效率。为此,本文对基于深度学习与计算机视觉的结构位移测量方法展开深入研究,主要研究内容及成果如下:(1)对图像模式与视频模式下的相机标定算法进行深入研究。对试验所用相机进行标定,得到两种模式下的相机参数,获取待测结构上目标特征物的世界坐标与图像坐标之间的映射关系。(2)提出基于YOLO V4算法的静动态位移测量方法。利用YOLO V4算法对结构表面特征物(如六角螺栓)进行自动检测和定位,在保证检测定位精度的前提下对YOLO V4算法进行改进,从而缩小网络结构、减小训练显存并缩短训练时间。采用三层框架结构静动荷载试验验证所提算法的准确性,通过坐标转化得到结构的静动态位移数据。结果表明两种算法均能较为准确的完成结构位移测量。(3)提出基于亚像素角点检测的位移测量识别算法。针对Shi-Tomasi算法进行优化设计,在特征物初始角点位置展开最小二乘法迭代计算,获取精确的亚像素级角点坐标,进而获取结构的静动态位移。试验结果表明:算法的测量结果与位移计的测量结果吻合良好。更多还原

【Abstract】 Structural displacement is one of the most important parameters for structural safety assessment.Obtaining displacement data effectively,quickly and accurately can improve the real-time ability and reliability of engineering structure evaluation results.With the rapid development of computer vision and deep learning,the problem of automatic image positioning has made a great breakthrough.Computer vision and deep learning provide a new choice for the measurement of structural displacement.Target detection algorithm and corner detection algorithm are used to automatically identify and locate the measured image.Combining with camera calibration to calculate the structural displacement,target detection algorithm and corner detection algorithm can greatly reduce the manual workload and improve the measurement accuracy and efficiency.Therefore,this paper proposes a structural displacement measurement method based on computer vision and deep learning.The main research contents are as follows:(1)Making a profound study of camera calibration algorithm in image mode and video mode.The camera parameters in the two modes are obtained,which can reflect the mapping relationship between the world coordinates and image coordinates of the target feature on the structure to be tested.(2)A static and dynamic displacement measurement method based on YOLO V4 algorithm is proposed.The YOLO V4 algorithm is used to detect and locate the structural surface features(such as hex bolts)automatically.On the premise of ensuring the detection and positioning accuracy,the YOLO V4 algorithm is improved to reduce the network structure,reduce the training memory and shorten the training time.The accuracy of the proposed algorithm is verified by the static and dynamic load test of three-story frame structure.The static and dynamic displacement data of the structure are obtained through coordinate transformation.The results show that both algorithms can accurately complete the structural displacement measurement.(3)A displacement measurement recognition algorithm based on sub-pixel corner detection is proposed.The least square iterative calculation is carried out at the initial corner position of the feature to obtain the accurate sub-pixel corner coordinates.On this basis,the static and dynamic displacement of the structure is obtained.The experimental results show that the measurement results of the algorithm are in good agreement with those of the displacement meter.更多还原