【作者】 刘源泂；

【导师】 孔建益；

【作者基本信息】 武汉科技大学 ， 机械设计及理论， 2011， 博士

【摘要】 本文针对机器视觉钢板表面缺陷检测技术中的成像优化和深度信息特征提取问题,以如何提高成像质量为切入点,深入研究了钢板生产工艺、表面光学性质及缺陷最优化成像的内在规律和相关理论,基于立体视觉的方法研究实现了钢板表面深度信息提取等关键技术,为提升机器视觉钢板表面检测系统性能提供了理论依据,拓宽了钢板表面自动化检测的思路和技术手段。本文主要研究内容和成果如下:1)建立了一个准确描述钢板表面光散射性质的机器视觉表面光照模型。基于钢板生产工艺和表面光学特征,结合BRDF光照模型理论,对不同钢板表面光散射特征进行测试和实验,得到光源条件、光线入射角、表面粗糙度与光散射分布的关系。结果表明钢板表面散射光受入射角和表面粗糙度影响较大,镜向反射峰值明显,并呈现指数函数分布的规律,因此建立了基于粗糙度因子和微平面高斯分布的表面半经验散射模型,通过非线性拟合优化确定了模型中各参数。2)以大量钢板表面典型缺陷样本的分析研究为基础,针对缺陷成像光路型式最优化问题,设计实验方案并开发专用平台进行成像实验,量化了成像光路中各参数对图像的影响;提出基于图像多特征的综合像质评价体系,得到各缺陷最佳成像方案,建立了钢板表面缺陷最优成像流程模型,对各子方案组合优化得出整个系统成像方案,作为钢板表面成像的理论研究平台,为提高整个检测系统性能提供依据。3)针对工业应用中如何利用立体视觉提取钢板表面三维深度信息,研究了摄像机非线性标定方法。建立基于LENZ径向畸变模型的非线性针孔成像几何模型,分析了面阵和线阵摄像机的模型标定内外参数,利用HALCON的矩阵网格状圆靶平面标定板及机器视觉的函数库平台,基于两步法思想设计算法并实现了单台摄像机的非线性标定,以双面阵摄像机构架实现立体视觉系统的标定,通过误差和精度分析验证了算法准确性和灵活性,可有效应用于工业机器视觉系统中。4)针对立体视觉中另一个核心技术即立体匹配算法进行研究。基于灰度相关的区域匹配方法能生成稠密视差图,实现了基于归一化互相关相似度量函数(NCC)的区域灰度相关的匹配算法,结合钢板表面深度提取的要求,分析优化算法参数提高算法精度;并与均值图像金字塔相结合,通过控制匹配窗口尺寸、最小特征值、视差范围阈值及相似度阈值等多个参数,提出一种灵活高效的分层区域匹配的优化NCC算法。5)结合相关研究成果,针对实验系统进行成像参数分析与成像系统设计,以及设备选型和安装调试,模拟工业现场环境设计开发了基于机器视觉的钢板表面检测实验系统。该实验系统由小型带钢传输实验台、机器视觉成像系统及图像采集与处理平台三部分组成,是重要的理论和实验研究平台。更多还原

【Abstract】 In this dissertation, aiming at the problems of imaging optimization and the depth-feature information extraction in the technology of steel plate surface defects detection based on machine vision, with improving the image quality as the starting point, the interaction effect laws and related theories have been researched which of the steel plate production process, the surface optical properties and to optimum imaging of surface defects. And the key techniques of steel plate surface depth extraction were achieved based on the methods of stereo vision to enhance the performance of the steel surface inspection system based on machine vision and provide a theoretical basis, and widened the ideas or techniques of automatic steel plate’s surface inspection system. The main works and achievements of the thesis are as follows:1) Building an accurate surface illumination model for machine vision to describe the light scattering distribution of steel plate surface. The model is based on steel production process and surface optical properties and combined with (Bidirectional Reflectance Distribution Function) BRDF lighting model theory. Through the experiments light scattering characteristics of different steel plate surface, the rules of relationship between the light incidence angle, surface roughness and laws of light scattering under a particular light-source conditions were found. The results showed that there was an apparent specular reflection peak on steel surface, and surface light scattering was influenced greatly by light incidence angle and surface roughness, of the law of exponential distribution functions. Thus the improved semi-empirical light scattering mathematical model which based on roughness factor and surface Gaussian distribution of micro-plane components has been formed, and through non-linear model fitting and optimization to determine the parameters.2) On the basis of analysis and study on many samples of steel plate typical surface defects, aimed at the optimization problem of imaging optical pattern for steel plate surface defects, we designed the imaging experiment scheme and development the special experiment equipment to research that quantifies the effects of optical imaging parameters to the surface defects images. Further comprehensive evaluation system of image quality based on the images’multi-features was proposed to obtain the optimal imaging program for every defect. Furthermore, through the optimization imaging process model of steel plate surface defects each sub-imaging schemes were analyzed, and the portfolio optimization imaging system design scheme was obtained. As the theories research platform of steel surface imaging for defects, the model provided a basis to improve the performance of the detection system.3) In view of achieving three-dimensional depth information extraction using Stereo Vision on the industrial applications, firstly researched the camera nonlinear calibration technology. Building the nonlinear camera pinhole models for area scan CCD camera and line scan CCD camera based on the LENS distortion model, determined the exterior parameters and interior parameters of camera calibration. Using HALCON planar calibration board which was circular targets of two-way array and functions library improved the calibration method for single camera based on principle of two-step and achieved it. And the camera calibration method of binocular stereo vision system which was composed by the structure of two area scan cameras was studied, as an example achieved the calibration process. The accuracy of these parameters and stability of the algorithm were validated through experiments and accuracy analysis. The calibration method is flexible and good portability to be effectively used in industrial machine vision systems.4) For the other key technologies of stereo vision, we researched the stereo matching algorithm. The regional matching methods based on gray correlation can generate the dense disparity map. Therefore to improve the algorithm accuracy, the regional matching algorithm based on gray correlation based on normalized cross-correlation (NCC) as the similarity measure function was realized to perform the stereo matching for examples of steel plate, and combined with the requirements of steel surface depth extraction, algorithm parameters were analyzed to be optimized. Further a region-based hierarchical matching algorithm based on the mean image pyramid was proposed combining the NCC method. The algorithm accuracy and computational efficiency have been controlled efficiently through comprehensive setup of the parameters such as matching window size, the smallest eigenvalue, and the disparity search space, similarity measure threshold and others.5) On the basis of conclusions in this research, according to the detection indicators of experimental system, we analysis the imaging parameters and design the imaging system, equipment selection, installation and commissioning. A machine vision-based experimental system of the steel plate surface inspection was built with the establishment of the industrial conditions. The system consists of three parts: small strip transmission equipment, machine vision imaging system, platform of image acquisition and processing. The detection test system is an important basis of research on theoretical and experiment.更多还原