基于机器视觉的零件平面尺寸自动测量

【作者】 马文娟；

【导师】 李歧强；

【作者基本信息】 山东大学 ， 模式识别与智能系统， 2006， 硕士

【摘要】 零件尺寸的检测是机械行业中最常见的工作之一。目前国内的零件尺寸检测以人工检测为主，这既增加了工人的劳动强度，又难以保证零件的加工质量，尤其对于形位误差这种复杂检测，靠人力特别费时，检测精度也不高。近年来，随着机器视觉在工业检测中的应用，运用计算机技术对零件尺寸进行检测，是进行现代化生产的必然趋势。 本文应用机器视觉理论，研究了零件尺寸自动检测系统，包括摄像机、图像采集卡、个人计算机及软件系统，可实现零件部分形位尺寸(直线度、平行度、圆度和同心度)的自动测量。本文的工作主要有三部分：图像采集、图像处理及尺寸测量。图像采集部分主要是为了获得零件的图像信息以进行后续的图像处理，后两部分是本文研究的重点。 图像处理分为图像预处理、边缘提取与特征检测三步，其中边缘提取是整个检测系统的关键，其方法的好坏直接影响系统的速度和准确度。本文通过对传统的边缘检测方法进行比较研究，采用了阈值分割法和数学形态学方法，得到连续单像素边缘，不仅运算量小，定位精度也较高。 本文主要研究直线度、平行度、圆度和同心度的检测方法，因此需要检测的特征为直线和圆。对于直线的检测，本文基于变分辨率图像金字塔的策略，用多级霍夫变换法与线性回归法相结合的方法检测直线特征。该方法在保证精度不变的前提下减少了传统霍夫变换法的存储量和计算量，其适用范围较广，可以检测任意直线，但仍然需要较多计算时间，为此本文又提出了基于随机霍夫变换法的搜索方法。该方法计算量小，比较适用于检测图像边缘的直线特征。对于圆的检测，常用的方法仍然是霍夫变换法及其改进算法。本文在随机霍夫变换法的基础上运用圆的性质对算法做了改进，减少了计算量。应用该方法可以提取复杂图像中的圆。 尺寸测量部分是在检测出形位特征后，根据各形位公差带的定义，分别进行了各形位误差的计算。论文最后通过具体实验验证了本文的方法，并对误差做了具体分析。更多还原

【Abstract】 Dimension detection of parts is one of ordinary jobs of mechanical industry. At present, the detection of parts’ dimensions is done manually, which not only increases the workers’ labouring intension, but also is difficult to guarantee the quality of parts. Especially for the manual detection of form-position error, it takes much time and the precision is not high. In the recent years, with the application of machine vision in industry detection, it is a necessary trend of modern manufacture to apply computer technology to detect parts’ dimensions.Based on the theory of machine vision, this paper presents a system of automatic detection of parts’ form-position dimensions, which includes vidicon, image collecting clip, Personal Computer and software system. It can realize the automatic measurement of some form-position dimensions of parts. This paper is composed of three parts: image collecting, image processing and dimension measuring. Image collecting is to get images of the parts for the following image processing, the last two parts are the keystones of this paper.Image processing has three steps: pretreatment of images, edge extraction and character detection. As a key step, edge extraction affects the processing speed and precision of the whole system. By comparing the traditional methods of edge detection, this paper applies thresholding method for image segmentation and mathematical morphology for edge extraction to process the images and get continuous and single-pixel edges. This method has less calculation and high location precision.The detecting methods of straightness, parallelism, roundness and concentric degree are studied in this paper, so we need to detect beeline and circle. For the detection of beeline, based on the tactic of changed distinguishing rate, this paper applies Multi-level Hough更多还原