基于高速视觉的结构运动测量算法研究与应用

发布时间：2018-06-27 12:53

本文选题：视觉测量 + 高速摄像　；参考：《中国科学技术大学》2017年博士论文

【摘要】：测量技术作为现代工业技术的基础技术,其测量的精度和效率是衡量国家科学技术发展水平的重要因素,工业制造技术的进步与加工工艺的提高对测量方法的性能提出了更高的要求。随着数字成像技术与图像处理芯片的发展,图像传感器在成像精度和速度上均得到较大提升,以机器视觉和深度学习为代表的分析算法的进步使得图像信息的提取更加多元化。基于视觉的测量技术作为一种新兴的测量手段,以其非接触、高精度、全场测量、无负载效应和自动化程度高等诸多优点,受到越来越多国内外研究人员和学者的关注。视觉测量技术已经在机械零件的形状与尺寸检测、目标识别与追踪、结构运动分析和生理信号采集等领域得到了广泛应用。随着高速成像技术的发展,视觉测量中相机的采样帧率不断提升,使其能够对被测物体瞬间发生的变化以及更高频率的运动信息进行记录与分析。本文以高速视觉测量系统作为研究对象,对高速视觉测量系统的硬件结构设计和信息提取算法进行了深入研究,并且针对结构不同的运动情况提出一系列行之有效的测量方案。本文首先阐述了视觉测量的研究意义与发展现状,对高速视觉测量系统的硬件构架进行了详细介绍,分析了不同硬件的选型标准和工作特点。通过合理搭配硬件结构,视觉测量系统可以完成不同尺度和不同距离下的目标运动参数测量与分析。文章同时对视觉测量系统中相机的几何成像原理进行了推导与分析,明确了在使用标靶的简单相机标定中影响测量精度的主要误差来源。常规的图像处理算法往往具有人工参数调整、自动化程度低、鲁棒性差等不足,文中针对结构视觉测量中物体的平动、旋转和微小振动等问题提出了一系列简单有效的信息提取算法。首先通过将简单梯度优化和曲面拟合优化与多种整像素精度的图像匹配算法相结合,提升了原始振动提取算法的测量精度和运算效率。使用优化后算法的视觉测量系统与传统的光栅尺位移传感器相比测量数据的误差小于1%,算法帧间位移提取耗时降低至1ms以下,可以满足实际拍摄帧率在1000fps左右的实时测量需求。其次,文章针对物体旋转参数测量问题给出了径向和轴向两种情况下的解决方案,将Lucas-Kanade与ECC图像对齐算法及其快速反向组合算法应用于物体的旋转参数测量中,提出了一种新的非线性扭曲函数描述圆柱结构径向旋转测量中目标图像与模板图像的扭曲变化关系,优化了参数在运算时的迭代方式,提升了算法的运行效率。最后,文章将视觉测量系统用于提取目标物体表面微小的亮度变化和振动信息中,通过对影像放大算法的剖析发现,放大视频中的微小变化量的前提在于有效提取该信息,针对影像放大算法执行效率低的问题,文章提出一种基于SVD奇异值分解的快速微振动提取算法,能够在去除冗余信息的同时有效提取图像序列中像素信息的微小变化。所提出的相关测量算法被集成在搭建的高速视觉测量系统中,在实际工程测量中取得了良好的效果。文中分别使用该装置对叉车方向盘系统和铁路桥声屏障结构的振动,故障轴承驱动时的旋转参数以及声音激励产生的微小振动进行了测量与分析,实验结果表明所开发的高速视觉测量系统在实际测量时工作稳定可靠,具有重要的应用价值。
[Abstract]:As the basic technology of modern industrial technology, measurement technology is an important factor in measuring the level of national science and technology development. The progress of industrial manufacturing technology and the improvement of processing technology put forward higher requirements for the performance of measurement methods. With the development of digital image technology and image processing chip, image transmission is the key factor. The sensor has greatly improved the accuracy and speed of imaging. The progress of the analysis algorithm represented by machine vision and depth learning makes the extraction of image information more diversified. The vision based measurement technology is a new measuring method, with its non-contact, high precision, full field measurement, no load effect and high automation. Many advantages have attracted more and more attention from researchers and scholars at home and abroad. Visual measurement technology has been widely used in the fields of shape and size detection of mechanical parts, target recognition and tracking, structural motion analysis and physiological signal acquisition. With the development of high-speed imaging technology, the sampling frame of the camera in visual measurement In this paper, the high speed vision measurement system is taken as the research object, and the hardware structure design and information extraction algorithm of the high-speed visual measurement system are studied in this paper. In this paper, a series of effective measurement schemes are put forward. Firstly, the research significance and development status of visual measurement are expounded. The hardware framework of the high speed vision measurement system is introduced in detail, and the selection standards and work characteristics of different hardware are analyzed. The visual measurement system can complete the different scales through the rational matching of hardware structure. The principle of the geometric imaging of the camera in the visual measurement system is derived and analyzed, and the main source of error which affects the measurement accuracy in the calibration of the simple camera using the target target is clarified. The conventional image processing algorithms often have artificial parameters adjustment and automation. In this paper, a series of simple and effective information extraction algorithms are proposed for the translation, rotation and micro vibration of objects in structural vision measurement. First, the original vibration extraction is improved by combining simple gradient optimization and surface fitting optimization with a variety of pixel precision image matching algorithms. The measurement accuracy and efficiency of the algorithm. Compared with the traditional grating scale displacement sensor, the error of the visual measurement system using the optimized algorithm is less than 1%, and the time time of the inter frame displacement extraction is reduced to less than 1ms, which can meet the real time measurement requirement of the actual shooting frame rate of about 1000fps. Secondly, the article is aimed at the rotation of the object. The solution of two kinds of radial and axial conditions is given in the parameter measurement problem. The Lucas-Kanade and ECC image alignment algorithm and its fast reverse combination algorithm are applied to the measurement of the rotation parameters of the object. A new nonlinear distortion function is proposed to describe the distortion of the target image and the template image in the radial rotation measurement of the cylindrical structure. The iteration method of the parameter in operation is optimized and the efficiency of the algorithm is improved. Finally, the visual measurement system is used to extract the small brightness change and vibration information of the object surface. By analyzing the image magnification algorithm, it is found that the premise of enlarging the small change in the visual frequency is to extract it effectively. Information, in view of the low efficiency of image amplification algorithm, a fast micro vibration extraction algorithm based on SVD singular value decomposition (SVD) is proposed, which can effectively extract the small change of pixel information in the image sequence while removing redundant information. The proposed correlation measurement algorithm is integrated in the built high speed vision measurement system. Good results have been achieved in practical engineering measurement. The device is used to measure and analyze the vibration of the sound barrier structure of the steering wheel system and the railway bridge, the rotation parameters and the small vibration caused by the sound excitation. The experimental results show that the developed high-speed visual measurement system is real. The measurement work is stable and reliable, and has important application value.
【学位授予单位】：中国科学技术大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】