【作者】 周虎；

【导师】 杨建国；

【作者基本信息】 东华大学 ， 机械制造及其自动化， 2011， 博士

【摘要】 精密测量技术是工业发展的基础和先决条件,测量的精度和效率在一定程度上决定了制造业乃至科学技术发展的水平。基于计算机视觉的影像测量系统以现代光学为基础,融光电子学、计算机图像学、信息处理、自动控制等科学技术为一体的现代测量技术,正成为一种提高检测效率和保证产品质量的关键技术,具有广阔的应用前景。本文结合上海市科技攻关项目(09DZ1121600)《支持产品数字化制造的高精度影像测量系统研究与开发》,对机器视觉测量关键技术进行了系统、深入的研究,开发了基于机器视觉的影像测量仪软硬件系统。论文的主要内容和成果如下:(1)在数字图像采集方面,设计了包括底光源和多角度直射环型上光源的四通道光源照明结构,开发了照明控制子系统。为了获取最优待测零件的图像质量,设计并开发了基于遗传算法的智能照明控制系统,对不同待测零件自适应地寻找最优照明控制参数,并根据待测零件的不同类别建立专家系统,以加速控制参数的寻优过程。(2)在超视场零件的图像获取方面,开发了基于互信息测度的模板匹配图像拼接算法,对相似测度通过基于二元三次样条插值函数的数值拟合,获得了亚像素级的配准精度。对基于模板匹配的拼接算法进行了一系列优化:通过采用黄金分割法快速定位配准点;通过多空间分辨率、多尺度的分层搜索方法以提高搜索效率;采用基于高精度精密光栅反馈的工作台闭环运动控制系统以直接把匹配位置锁定在极小的区域。(3)为了提高拼接算法的鲁棒性和实时性,提出了一种基于边缘特征和行程编码索引的最优特征模板提取策略。最优模板包含了尽可能多的图像特征,模板中参与运算的只是具有边缘位置特征和灰度变化显著的像素点,通过行程编码方法构建索引模版以指定特征像素集合,从而显著提高了配准精度和匹配速度。(4)提出了一种针对混合噪声的自适应滤波算法,通过判断窗口中心像素的噪声类别来选择滤波方法,通过信噪比改善因子验证了算法的效果。(5)在边缘检测和定位方面,提出了一种基于多尺度的自适应LoG边缘检测改进算法,通过计算不同区域灰度共生矩阵的惯性矩来自适应地选择滤波尺度和零交叉阈值,以最大限度地滤除噪声并保持边缘。为了提高边缘点的亚像素定位精度,对传统的基于高斯插值的亚像素边缘点定位算法进行了改进:在边缘法向通过加权拉格朗日函数插值进行灰度重采样,沿具有更高灰度差分的梯度方向进行高斯插值以获得高精度亚像素边缘点。(6)研制了高精度机器视觉测量系统的软硬件装置,通过改进的二次标定法对系统进行标定,并采用标准尺寸的量块和量规以及精密机械零件进行了对比测量实验,测量结果达到了微米级精度。更多还原

【Abstract】 The technology of precision measurement is the base and the precondition for the industrial development.The developing level of manufacturing as well as the scientific technology depends largely on the accuracy and efficiency of measurement.Image measurement based on machine vision is a modern techonolgy which integrates the discipline of optoelectronics,computer graphics,information processing and automation control.It is becoming a key and promising technology for increasing the inspection efficiency and essuring product quality.Rely on Shanghai key scientific and technological projects "The Research and development of high precision image measurement support Digitized Manufacture"(ID:09DZ 1121600),this paper made a systematic and deep investigation on the key technology of image measurement.An image measurement instrument based on machine vision was developed and many innovation fruits were achieved.The main research contents and innovations of the paper were as follows:(1)In the aspect of digital image acquisition,this paper designed an illumination architecture which includes a back light and three direct ring front light lamp houses.An intelligent illumination control system based on GA was put forward in order to acquire the best quality image of the object.The optimal illumination control parameters aimed at different objects to be measured could be found with the help of intelligent illumination control system.The fuzzy expert system based on the category of objects to measure was built to accelerate the searching process of optimal control parameters.(2)In order to acquire the panorama image of the workpiece which exceeds the field of view(FOV)of the camera,this paper put forth the template matching algorithm based on the mutual information measure.Sub-pixel matching accuracy was obtained using polynomial spline interpolation of cubic polynomial with two unknown according to the similarity compare.The template matching algorithm was optimized through three methods.The first point,we put forward the use of golden cut method to speed up the localization of the matching point.Second,we build up the structure of image pyramid which has different spatial resolution and different scale images in different layers using mean filter,then perform matching in different layers to improve the search efficiency.Third,the closed loop worktable motion control system with feedback of precise grating was utilized to minimize the search area.(3)The optimal template pick-up strategy using run length encoding index was invented and it made a breakthrough in the robustness and speed of template matching algorithm.The optimal template contains image characteristics as much as possible while only the pixels which have edge character and distinct gray value changes participate in the matching operation.Both the precision and speed of template matching algorithm were improved remarkably.(4)For image denoising,to filter the hybrid noise of workpiece image,an adaptive filtering method for hybrid noise is proposed.This method firstly uses specific rule to judge the noise type,then select corresponding distinguished filter to eliminate noise according to noise type.Finally the effectiveness of the filter was tested and verfied by analyzing the improvement factor of signal to noise ratio(SNR).(5)In the aspect of edge detection and localization,we put forward a multi-scale self adaptive LoG algorithm which improved the traditional operator.The filter scale and zero-crossing threshold can be selected flexibly according to the inertia moment of the grey symbiosis matrix in different image areas.Thus the noise can be filtered and the edge can be maintained to the maximum extent.In order to acquire sub-pixel precise edge,this paper improved the traditional Gaussian interpolation based edge localization algorithm.Perform gray value resampling in the edge normal using weighting Lagrange interpolating function,then execute Gaussian interpolation to acquire sub-pixel edge points with higher accuracy.(6)Final,the equipment of image measurement based on machine vision was developed and calibration was performed using improved calibration method.We use standard gauge block and other precise machine parts to test and analyzing the precision of the measuring system.Experiments show the system is stable and could reach the accuracy of micrometer.更多还原