基于贝叶斯模型的三维变形测量系统  

Three-dimensional deformation measurement system based on Bayesian model

作　　者：许雷[1] 潘常良 赵帅 王效贵[1] 梁利华[1] 许杨剑[1] XU Lei, PAN Changliang, ZHAO Shuai, WANG Xiaogui, LIANG Lihua, XU Yangjian(College of Mechanical Engineering, Zhejiang University of Technology, Hangzhou 310000, Chin)

机构地区：[1]浙江工业大学机械工程学院

出　　处：《激光杂志》2018年第5期26-31,共6页Laser Journal

基　　金：国家自然科学基金项目(No.51375448、No.51375447)

摘　　要：为了快速高精度地获取物体表面形貌以及变形，提出了一种基于贝叶斯模型的三维变形测量系统。基于贝叶斯概率模型计算初始视差，利用牛顿拉普森迭代法对初始时差进行迭代优化，实现了高精度的立体匹配。时序匹配中，采用高效的反向组合高斯牛顿法对位移初值进行非线性优化，实现了数字图像相关快速、高精度匹配；针对优化过程中初值难求的问题，采用可靠性导向方法，为每个点的非线性优化提供了可靠的初值估计，同时也避免了误差的传递。实验结果表明，静态物体三维重建的测量精度可以达到0．208％，应变测量结果与商业软件VIC-3D测量结果相当吻合。与传统测量方法相比，本文系统可以更好的满足实验中三维形貌、应变测量的需求。In order to obtain surface morphology and deformation of objects with high precision, a three - dimensional deformation measurement system based on Bayesian model was developed. In the aspect of stereo matching, the initial disparity was calculated based on the Bayesian probability model. Then, the Newton -Raphson method was used to optimize the initial guess, which realized fast and high precision stereo matching. In time matching, inverse compositional Gauss -Newton （IC -GN） method was used for non -linear optimization to realize fast and high precision matching of the digital image. To solve the problem in calculating initial guess of displacement, Reliability Guided DIC method was used to provide a reliable initial value for non - linear optimization. This method also avoids the propagation of error. Experimental results show that the measurement accuracy of 3D reconstruction of static objects is about 0. 208%, and the strain result is very close between our method and commercial software VIC -3D. Compared with the traditional methods, our system is more suitable for three -dimensional reconstruction and deformation measurement in experiment.

关 键 词：图像处理 视觉测量 数字图像相关 三维重建 应变测量 

分 类 号：TN247[电子电信—物理电子学]