基于立体视觉的结构大变形全过程非接触动态测量方法

To dynamically measure large deformation of structural models or actual structures in a contactless way, a stereovision-based large deformation measurement method will be proposed here. The bundle adjustment calibration algorithm will be studied to optimize two cameras of stereovision system. The four-parameter mathematical distortion model for stereovision imaging will be established, the least square method will be used to solve the initial value of the distortion model. Moreover, the bundle adjustment calibration algorithm will locally optimize the interior and exterior parameters of two cameras, also globally optimize solid circle panel. And the comparision test of precision calibration will be performed. Subsequently, a marker-based stereovision measurement approach for large deformation of structures will be researched. The combination algorithm of Canny-MCP will be adopted in edge detection of circle marker, whose precision will be reach 0.01 pixel. Kalman filtering will be used to match and track circle marker, then 3D displacement equations of structures will be obtained. In addition, a stereovision measurement method for large deformation of structures based on image correlation will be studied. The threshold of correlation coefficient will be presented, FDT approach will be employed to match correlation image under large deformation, and the shape function in finite elment method will be used to calculate strain filed under large deformation. Afterwards，the stereovision measurment software for large deformation of structures will be compiled, and the stereovision measurment system will be integrated by combining the hardware device. Accordingly, the performance of the integrated system will be achieved by the prcesion test. Furthmore, the dynamic measurment experiments for large deformation of structural models and actual modles will be conducted with the integrated system. Experimental results will verify that the proposed method is feasible, and the integrated stereovision system is reliable. Finally, a non-contact dynamic measurement approach will be provided to detect the whole process of large deformation of structures.

针对结构模型或实际结构的大变形非接触动态测量需求，提出结构大变形立体视觉测量方法。研究立体视觉双摄像机的光束平差优化标定算法，建立4参数立体视觉成像畸变模型，使用最小二乘法求解畸变模型初值，使用光束平差法局部优化左右摄像机内外参数，全局优化实心圆标定板，进行精度标定对比试验。研究基于标定点的结构大变形立体视觉测量算法，提出精度可达0.01像素的Canny-MCP组合算法进行圆形标志点边缘检测，采用卡尔曼滤波算法跟踪标志点，给出结构三维位移公式。研究基于图像相关的结构大变形立体视觉测量算法，提出相关系数阈值，采用分段位移法进行大变形时相关图像匹配，使用有限元形函数计算大变形下应变场。编译结构大变形立体视觉测量软件，集成立体视觉测量系统，进行系统精度试验。完成结构模型及现场结构的大变形动态测量试验，验证本方法的可行性及集成系统的可靠性，为土木工程结构大变形全过程测量提供一种非接触动态测量方法。

针对结构模型或实际结构的大变形非接触动态测量需求，提出结构大变形立体视觉测量方法。研究基于实心圆靶标的立体视觉系统标定算法，标定方法第一步选用线型针孔模型获得相机的初始内外参数，第二步考虑畸变模型，分别对左右摄像机内外参数进行LM优化，获得外参数近似解。第三步以实心圆靶标对角线上两实心圆的距离作为约束条件，对摄像机外参数进行LM优化，获得外参数最优解。编译基于实心圆靶标的立体视觉系统标定软件，完成标定精度对比试验研究。研究基于圆形标识的结构大变形立体视觉测量算法，完成圆形标志点边缘检测，采用卡尔曼滤波算法跟踪标志点，给出结构三维位移公式。研究基于图像相关的结构大变形立体视觉测量算法，引入相关系数阈值来判定相关图像的匹配失效，进而对参考图像进行更新。通过初始时刻立体匹配、分块匹配算法及大变形匹配算法得到试件表面网格所有节点的离散位移值，使用滑动平均方法对离散点位移做平滑处理，进而构造形函数求解位移场，并带入格林应变张量完成应变场求解。在上述算法基础上，编译结构大变形立体视觉测量软件，集成结构大变形立体视觉测量系统，使用三维电控移动平台完成了系统精度试验。完成地震波作用下三层框架模型倒塌立体视觉监测试验及CFRP混凝土试件粘结滑移监测试验，开发宋家庄高铁立交桥挠度变形监测系统，并进行现场挠度监测试验。试验结果验证了立体视觉结构大变形测量方法的有效性和可行性，验证了集成的立体视觉测量系统的可靠性，为土木工程结构大变形全过程测量提供一种非接触动态测量方法。