【作者】 何涛；

【导师】 史铁林； 钟毓宁；

【作者基本信息】 华中科技大学 ， 机械制造及其自动化， 2011， 博士

【摘要】 机器视觉是一门多领域交叉学科,它从客观对象中提取图像信息,用于检测、测量和控制,有非接触、速度快、稳定性好、精度高的技术优点。近二十年大量的视觉系统装备在高端制造设备和生产线上。图像处理技术是机器视觉的核心,是研究的热门话题,也是先进制造领域的深入及广泛应用的瓶颈问题。问题主要体现在图像处理技术缺乏针对使用对象而开发的专用解决方案。为此,本文开展有针对性的图像测度研究,以解决机械产品图像测量和缺陷识别的在线检测技术难题,有重要的意义和价值。本文主要工作有：(1)分析了机械产品在线视觉检测技术研究和应用现状,确定了研究问题和研究内容,创新提出了基于图像结构维数的描述子新概念和模型。首先,从理论和技术的层面分析了影响机器视觉技术不能完全满足制造在线检测要求的因素,主要是图像描述不足以表示工业产品特征,因此,研究机械产品与其数字图像的一致性映射特征。产品本质是几何形体,其映射特征是具有几何形状和几何结构特点的图像。应用拓扑学基础理论,构建了基于基元空间集合表示数字图像图形特征的理论体系。(2)分析了机械产品的现场图像具有不完全性和非线性特性,提出了基于测度理论重建图像基元的新方法。视觉系统中的光源等视觉系统单元,以及测试对象本身的几何形态不一致,都使得工业产品的基元图像不完整、不一致、差异性大,严重影响了机器视觉技术在测量中的应用。本文结合形态学的重建图像基元的理念,利用分形几何的幂律和分维理论、计盒维数计算技术,提出了基于测度理论的重建图像基元新方法,有效解决了现场图像的不完整性和非线性问题,为机械产品图像的测量问题提供了新方法。(3)针对复杂的机械产品图像,研究了分形几何中的豪斯多夫测度和维数理论,发现形态不规则的相似性图像的测度规律,提出了构建纯元空间的图像模型和方法。图像分割是把图像分成若干个有意义的区域,图像特征的提取也是图像的再分割。图像分割得越合理,图像分析和图像理解就简单且可靠性高,计算速度快,有利于在线检测。结合拓扑学的集合空间分割和重建理论,在综合研究图像轮廓提取方法的基础上,研究了分形几何中的豪斯多夫测度和维数理论,应用基于“周长+面积”的计算方法,有效解决了复杂边界图形的相似性度量问题,为图像特征的进一步分割和构建纯元空间提供了理论依据。(4)研究了图像的几何特征空间转换问题。机械产品图像的几何特征空间的建立是产品尺寸测量和缺陷识别的理论基础,可是,工业产品图像的非线性使得图像的几何特征空间不易于从图像中直接和获取。研究了基于零件生成的几何结构形状,采用信号降维技术,提出了基于几何形状的图像特征转换模型,降低了工业产品图像的结构维数,提高了图像理解能力和计算速度。(5)开发了一个面向制造现场的可重组机器视觉图像检测软件和几个在线视觉检测系统。构建了一个基于测度理论研究成果的、面向制造现场的可重组机器视觉图像处理软件平台,简述了这个图像检测软件在开发检测系统中的应用。本文的研究成果多次成功地应用在制造现场的在线检测系统,这些系统提高了生产效率,有效地控制了生产过程中的产品质量。当然,基于几何形态的图像研究还有待于进一步的深入和完善,本文还提出了今后工作的主要研究内容。更多还原

【Abstract】 It is real-time, in-machine and online measurement method that satisfys requirements on high-speed, precision, and quality. Machine vision (MV) extracts information from images of objectives for detecting, measuring and controlling. In the last two decades, it is wildly applied in high-end manufacturing equipments and production lines, for its advantages of non-contact, high speed, good stability, and high accuracy. Image processing is not only the core of MV and a hot topic in recent researches, but also is a bottleneck of applications in advanced manufacturing field. The bottleneck reflects on the lack of image processing solution for a special object. So, image measure for special solutions to mechanical parts’image is researched in this dissertation.The dissertation mainly includes:1. Current online MV-based inspection methods for the mechanical product are analyzed. New concepts and descriptors based on image structural dimensions are proposed. Firstly, factors that MV did not satisfy online inspection requirements are summarized to a result of poor reasonable image descriptions for mechanical parts’images. Secondly, mechanical parts and their digital images are consistency mapping. Thirdly, a theory model for space set of primitives in an image is proposed to elaborate geometrical features based on Topology.2. Incomplete and nonlinear characteristics of a mechanical part image are analyzed. Nonlinear factors in a MV system affect its applications in measurement. In the dissertation, new methods based on measure theory are used to reconstruct image primitives. They combine both image primitive’s reconstruction concept in morphology, and power law and box dimension computing method in fractal geometry.3. A measurement law of similar objects’images with irregular shape is found from the research of both Hausdorff measurement and dimension theory in fractal geometry, to describe a structure primitive’space. Becacuse of the complexity of mechanical parts’images, the efficiency and rationality of image segmentation is correlated with both image analysis and image understanding. The correlation is showed by extracting image features and regions of interest. Topological set space segmentation and reconstruction theory, as well as Hausdorff measurement theory, are used to solve measurement problems for complex and similar contours, with a method of "Perimeter+Area" to structure primitives’ space.4. A problem to extract geometric feature from images’ feature space directly is difficult because of nonlinearity, which is solved by image feature space conversion with signal dimension reduction technology.5. A reconfigurable software based on machine vision for manufacturing is developed and several online MV inspection systems are designed.The above researches have been successfully applied to finish online inspections for several manufacturing processes The applications help the manufacturers to improve production efficiency and control product quality of the production process. Certainly, more efforts are need for study of image geometrical shape. At last, the dissertation proposes the lines of future work.更多还原