【作者】 赵晶晶；

【导师】 白瑞林；

【作者基本信息】 江南大学 ， 检测技术与自动化装置， 2011， 硕士

【摘要】 伴随着现代工业科技的快速发展,工业对测量精度要求越来越高,视觉测量技术在工业生产中有重要地位。它具有速度快、精度高、非接触、自动化程度高等优势,该技术近年来在非接触测量领域发挥着巨大作用。视觉测量技术不仅包含机器视觉的一般内容,如视觉感知、图像处理、图像分析和模式识别等,也包含着测量领域的特殊性,如空间几何尺寸的精确检测、定位和识别等。摄像机标定技术源于20世纪中叶,经过几十年发展,该技术已经很成熟,从传统的标定技术发展到自标定技术,在制造业、电子信息、印刷行业、钢铁冶金、国防等行业应用。随着自动化水平的快速发展,摄像机标定技术在国内市场有较大发展潜力,本文研究重点是摄像机标定技术在机器视觉测量中应用。本课题来源于产学研合作项目“嵌入式机器视觉控制器的研究与开发”(合同号:07398),基于嵌入式机器视觉平台进行研究,主要用于工业视觉测量领域。结合工程思想,本课题提出一种基于单幅图像二维平面标定方法。该方法首先进行光路调整,主要分为两个步骤:光强调整,能有效防止X靶标边缘信息丢失,而导致X靶标亚像素角点检测的不准确;光轴与物面垂直度调整,有效防止摄像机CCD平面与被测物面之间存在夹角引起测量误差。本方法采用了改进的Harris算子提取像素级角点,有效降低算法的运算量;再利用空间矩方法进行X靶标亚像素角点的提取,精度比较高,亚像素精度能达到0.1像素。该标定方法运算量小,精度高,稳定性好,已经在嵌入式机器视觉工业现场进行在线测量,测量精度能够达到0.05mm。基于单幅图像的标定方法存在不足之处——要求测量平面与标定平面必须在同一个平面,本课题提出基于多个特征点的标定方法。该方法不需要辅助设施——X靶标,虽然需要从不同角度连续拍摄多幅图片,但操作较为方便,成本低,能够解决不同平面以及摄像机与被测量平面之间存在一定夹角情况下的标定,在工业测量行业使用更为广泛。后续将针对三维测量进一步研究,利用结构光作为辅助设施,进行工业生产的三维测量与三维重建,具有较大发展空间。更多还原

【Abstract】 With the rapid development of modern industrial technology, vision measurement technology plays an increasingly important role in industry. It has lost of advantages, such as the fast speed, high precision, non-contact, high automation. In recent years, it plays an increasing role in the field of non-contact measurement. Vision measurement includes not only the general content in the machine vision, such as visual perception, image processing, image analysis and pattern recognition and so on, but also has the particularity of measurement, such as the precise detection of space geometry, location and identification.Camera calibration originated from the mid-20th century, and after decades of development, it has been more and more mature, developed from traditional calibration technique to self-calibration technique. Calibration can been applied in the manufacturing, electronic information, the printing industry, iron and steel metallurgy, national defense and so on. With the rapid development of automation, calibration has great development potential. The paper focused on calibration in machine vision applications.This topic came from the research cooperation project "Research and development of embedded machine vision controller" (contract number: 07398). With engineering thinking, this paper presented a new calibration method based on a single image. First, the calibration needed the optical path adjustment. It was divided into two steps: light intensity adjustment system can effectively prevent the edge of the target information losing, which can lead to sub-pixel corner detection inaccurate. Verticality adjusting system between optical axis and object surface can effectively prevent the angle between the CCD plane and the object surface, which lead to measurement error. Corners detections of this calibration used a modified Harris operator extracted pixel corners. It effectively reduced the algorithm computation. Then it used the space moment method to sub-pixel corner extraction. It had high accuracy and the accuracy of the sub-pixel corner can reach 0.1 pixels. Calibration technique had high precision, good stability, low computation. It has been applied in the embedded machine vision industry. Measurement accuracy can reach 0.05mm.The calibration based a single image had a defect that the measuring plane must be in the same flat with the calibration plane. So the topic proposed a new calibration based on a number of feature points. The plan did not require auxiliary facilities-X target. Although it needed to capture multiple images, it had easy operation and lower costs. In future, I need to further research three-dimensional measurements. We will use structured light as auxiliary facilities to three-dimensional measurement and three-dimensional reconstruction.更多还原