【作者】 刘琰；

【导师】 李言；

【作者基本信息】 西安理工大学 ， 机械电子工程， 2021， 硕士

【摘要】 制造自动化技术的飞速发展,对制件自动检测的需求日益增大。在制造过程中,制件的检测长期依赖坐标测量机及人工检具进行检测。对注塑件的微小缺陷,很大程度上还依赖人工目检,强度大,效率低。机器视觉作为一种非接触式测量系统,由于视场大,柔性程度高,适应范围广等特点在工业自动化检测领域受到越来越多关注。本文针对电子连接器注塑件的工业检测需求,研究基于机器视觉的微小尺寸检测问题,具有重要的理性意义和工程实践价值。建立了基于工业相机采集的注塑件尺寸检测系统和图像处理流程,通过高像素工业相机获得注塑件的灰度图像,利用远心镜头及合理的背板光源等方式,解决光学畸变,获得清晰图像。本文提出了基于三维高斯函数的改进BM3D算法,该算法利用三维高斯函数代替硬阈值,根据图像灰度值变化自适应计算阈值大小,与传统BM3D相比,解决了硬阈值大小不会随着图像灰度值变化而自适应变化的问题,对高斯噪声的去除效果更好,同时最大程度地保留边缘像素点。采用峰值信噪比的方法对不同条件下采集的图像进行去噪效果检验,结果表明,改进BM3D算法均优于传统算法。本文提出了基于最大类间方差法(OTSU)的改进Canny边缘检测算子,解决了传统Canny算子中的梯度阈值根据人工经验自行设置,无法精确得到合适的梯度阈值的问题。通过样张对比,改进Canny算子对边缘区域的识别更准确,产生更少的虚假边缘;针对电子连接器微小特征测量需求,本文提出了基于三灰度区间模型的改进Zernike矩亚像素边缘检测算法,利用三灰度区间模型模拟实际边缘灰度值渐变特征,通过样张对比,改进的Zernike矩方法对于亚像素级别边缘定位更加准确,微小特征边缘连接更加完整。建立了尺寸检测标定系统,通过标准棋盘格标定板计算标定系数,对多边形孔和圆形孔采用最小二乘法拟合,根据方程计算得到像素坐标和微小特征边缘像素尺寸,利用标定系数将像素尺寸转化为实际物理尺寸。通过与手动测量尺寸对比,本文建立的尺寸检测系统检测误差为±0.17mm,满足工业检测的要求。更多还原

【Abstract】 With the rapidly development of manufacturing automation technology,the demand for automatic inspection of parts is increasing.In the manufacturing process,the inspection of parts has long relied on coordinate measuring machines and manual inspection tools for inspection.For small defects of injection molded parts,manual visual inspection is largely dependent on the strength and low efficiency.As a non-contact measurement system,machine vision has attracted more and more attention in the field of industrial automation inspection due to its large field of view,high flexibility,and wide adaptability.In this paper,according to the industrial inspection requirements of electronic connector injection parts,the research of micro size detection based on machine vision has important rational significance and engineering practice value.The size detection system and image processing flow of injection molding parts based on industrial camera are established.The gray image of injection molding parts is obtained by high pixel industrial camera.The optical distortion is solved and the clear image is obtained by using telecentric lens and reasonable back light source.This paper proposes an improved BM3D algorithm based on a three-dimensional Gaussian function.The algorithm uses a three-dimensional Gaussian function to replace the hard threshold,and adaptively calculates the threshold according to the change in the image gray value.Compared with the traditional BM3D,it solves the problem that the hard threshold does not vary with the image.The problem of adaptive change due to gray value changes has a better effect on removing Gaussian noise,while retaining edge pixels to the greatest extent.The peak signal-to-noise ratio method is used to test the denoising effect of images collected under different conditions.The results show that the improved BM3D algorithm is better than the traditional algorithm.This paper proposes an improved Canny edge detection operator based on the Maximum.Between-Class Variance Method(OTSU),which solves the problem that the gradient threshold in the traditional Canny operator is set according to manual experience,and the appropriate gradient threshold cannot be accurately obtained.Experiments show that the improved Canny operator recognizes the edge area more accurately and produces fewer false edges;for the measurement requirements of the small features of electronic connectors,this paper proposes an improved Zernike moment sub-pixel edge detection algorithm based on the three-gray interval model.The three-gray interval model is used to simulate the actual edge gray value gradient feature.The experimental results show that the improved Zernike moment method is more accurate for sub-pixel edge location and the edge connection of small features is more complete.Dimension detection and calibration system is established.Calibration coefficients are calculated by standard checkerboard grid calibration plate.Polygon holes and circular holes are fitted by least square method.Pixel coordinates and small feature edge pixel sizes are calculated according to the equation.Pixel sizes are converted into actual physical sizes by calibration coefficients.By comparing with manual dimension measurement,the dimension detection system established in this paper has a detection error of mm,which meets the requirements of industrial testing.更多还原