【作者】 李刚；

【导师】 肖新平； 曾祥金；

【作者基本信息】 武汉理工大学 ， 固体力学， 2010， 博士

【摘要】 公路路面的裂缝检测是公路运营后维护与保养的主要问题之一。随着全球经济的持续发展,世界上大多数国家都掀起了一股大力发展基础设施的热潮,其中,公路建设作为交通行业的重头戏而受到普遍重视。由于公路建成的里程越来越长,传统的主要靠人工进行路面检测的方式也越来越满足不了日益增长的检测工作量,也不能适应新时期的检测要求,因此研制和开发全自动、高科技的公路路面检测设备与算法成为了发展交通道路行业的一项重要任务。本文在查阅大量相关文献的基础上,对目前的公路路面图像的裂缝检测算法进行了综述。针对目前的路面图像裂缝检测算法主要局限在基本的图像处理技术、数学形态学、神经网络、小波分析等经典理论上的特点,本文尝试把灰色系统理论应用到对路面图像的预处理与分割中来,为裂缝检测的后续处理打下一定的基础。从路面裂缝产生的力学原因入手,本文在分析路面裂缝图像的灰色特点的基础上,阐明了应用灰色系统理论来解决路面图像裂缝检测问题的合理性和优越性,总结了路面检测设备的发展状况,介绍了灰色系统理论的基本概念,通过对路面图像的去噪、滤波、增强以及边缘检测的基本原理的分析,阐述了10种结合灰色关联分析、灰熵理论、灰色预测模型的路面图像的去噪、滤波、增强、边缘检测的新算法。针对传统的邓氏关联度在应用图像数据计算时,公式中可能出现分母为零、滤波效果不理想等情况,本文结合路面图像数据的特征,提出了一种灰色图像关联度模型,选取图像邻域中的部分数据进行加权平均,并充分利用当前滤波窗口中在本次遍历中新得到的图像像素灰度值。当路面图像中的噪声密度增加时,图像邻域窗口中的噪声点在进行加权滤波计算时也增加到了一个不可忽略的地步。本文提出,在对路面图像进行去噪前,先根据邻域窗口中像素灰度值的灰关联序进行噪声判别,再对当前像素为噪声点的邻域窗口实施非噪声点选取的灰关联去噪或扩大窗口中值滤波。在含有噪声的路面图像中,本文利用邻域中各像素值与邻域中值的灰熵值作为邻域中各像素值的权系数,求邻域各像素的加权平均值作为中心点的新灰度值以此来实现对噪声的滤除。通过对图像邻域窗口中像素的灰度值进行排序,将图像的像素分为三类进行分别处理：正常点不变,大噪声块要扩大窗口进行中值滤波,仅对邻域中含有少量噪声的噪声点实施基于仿射变换的灰色预测滤波。通过计算邻域窗口中中心像素的灰度值与邻域各像素值的灰关联熵,来度量邻域窗口中的局部边缘程度,然后搜索阈值来对路面图像的边缘进行分割。通过计算图像邻域窗口中的16种纹理方向的像素灰熵值,以及邻域中灰熵值的最大值与最小值之差,来找出图像中局部纹理起伏的边缘特征,从而设定阂值实现路面图像的边缘检测。对图像邻域窗口中的四个主要的纹理走向的像素组进行添加辅助点的GM(1,1,C)建模,并将四个模型的拟合值的残差和的最大值与最小值之差作为当前中心点的边缘程度的测度,设定阈值,提取路面图像的边缘。在图像邻域窗口中应用灰色图像关联度选择出邻域中部分与中心像素不同属性的像素并增大它们的均值与中心像素的对比度,以此提高图像增强的效果。通过利用图像邻域窗口中的灰熵值构造图像模糊局部对比度增强的增强指数,来实现对模糊对比度函数的自适应增强。从当前中心点周围的八个方向指向邻域窗口中心选取中心点与邻域均值之差作为原始数据点,并按紧邻均值生成的方式添加辅助点,建立终点固定的离散灰色预测模型,将模型的拟合值作为对局部对比度函数的增强尺度,来自适应地调节图像的局部对比度,提高路面图像的增强效果。本课题来源为：教育部高等学校博士点基金项目：基于广义累加灰生成的极限承载力建模与预测研究(项目编号：200804970005)、国家自然科学基金项目：基于矩阵分析的灰序列生成预测建模及应用研究(项目编号：70471019)、武汉市科技攻关计划项目：基于演化优化技术的Web图像语义模糊分类研究(项目编号：201010621218).更多还原

【Abstract】 Road surface crack detection has been one of the main issues of highway care and maintenance after the highway being put into operation. With the sustainable development of global economy, most countries have set off a wave of enthusiasm about developing the infrastructure, in which road-building, as a highlight of the transport sector, has attracted universal attention. As the mileage of completed highway is becoming longer and longer, traditional manual way of road pavement testing is increasingly unable to meet the growing testing workload, neither can it adapt to the testing requirements of the new era. Therefore, researching and developing the automated, high-tech road surface testing equipment and algorithm has become an important task for the development of road traffic industry.In this paper, via referring to a large number of corresponding literatures, we summarise the current crack detection algorithms of the haighway pavement image. Viewing the fact that the current road surface image crack detection algorithms are mainly limited to the traditional theories of the basic image processing techniques, mathematical morphology, neural networks, and wavelet analysis theory, this paper attempts to apply the grey system theory to the road surface image preprocessing and segmentation with an effort to lay a foundation for the follow-up treatment of crack detection.Beginning with the analysis of the mechanical causes of road pavement crack, the paper has clarified the rationality and superiority of applying the grey system theory to solving the problem of pavement image crack detection based on the analysis of the grey characteristics of the pavement crack image, has sumed up the development of road testing equipment and has introduced the basic concept of grey system theory. By analyzing the basic principles of the pavement crack image denoising, filtering, enhancement, and edge detection, we propose ten new kinds of algorithms about the pavement image denoising, filtering, enhancement and edge detection combining with the grey relational analysis, grey entropy, grey prediction theory respectively.Viewing the fact that when applying the Deng’s grey relational degree to the image data computation, the denominator in the formula may appear zero, the filtering effect is not very satisfactory, and taking into account the characteristics of the road image data, this paper propose a new grey relational degree model, named grey image relational degree, which selects the part of the data in the neighborhood of image to carry out weighted average operation, and makes full use of the new pixel gray value in the neighborhood window resulting from the current traversal.When the noise density of the road surface image increases, the noise points in the image neighborhood window turn to an extent that can not be ignored when computing the central weighted value. This paper propose that, before pavement image denoising, the noise and non-noise points should be distinguished according to grey relational order of pixel gray value in the neighborhood window. And then for the pixel that is noise point as the center of the neighborhood window, the non-noise pixels around the central pixel in the neighborhood are chosen to carry out the grey relational noise reduction operator, or expand the window to carry out the median filter operator. In a noise-containing pavement image, this paper uses the entropy between the central pixel and each other pixel of the beighborhood as the weighted exponent to compute the weighted average value of all pixels in the neighborhood as the new value of central pixel in the neighborhood to achieve the noise filtered. By sorting the pixel gray value of the image neighborhood window, we divide the pixels of the image into three groups to deal with them respectively:those remain the normal unchanged, those expand the window for median filtering for large noise block, and those only make the noise with a neighborhood containing a small amount of noise operated by grey prediction filtering based on a affine transformation.Through computing the grey relational entropy between the central pixel of the neighborhood window and each adjacent pixel of the neighboerhood, we measured the local edge degree of the neighborhood window, and then searched the threshold value to nake the edge of the pavement image that has been segmented from the background. Aftering calculating the pixel grey entropy values of sixteen kinds of texrure in the image neighborhood, we computed the difference between the maximum and minimum of the gray entropy in order to find out the edge feature of local texture variances, and then set a threshold to make the edge of pavement detected. We made GM(1,1,C) models for the pixel groups with four main texture toward in the image neighborhood window by adding auxiliary point and by supposing the difference of the maximum and minimum of rediduals summation of fitting values of four models be the extent of measure of the current central point,then we set thresholds and extracted the edge of the pavement image.In the image neighborhood window, we selected the part of pixels which are with property different from the central pixel by using the gray image relational degree, and increased the contrast between their average value and the central pixel value in order to improve the effect of image enhancement. Using the grey entropy value of the neighborhood of the image, we constructed the enhancement factor of the image fuzzy local contrast enhancement in order to make the fuzzy local contrast function adaptively enhanced. From eight directions of the current central pixel, pointing to the centre point of the surrounding neighborhood window, we chose the absolute value of difference between the centre poit and the average value of all the pixels in the neighborhood as the original data points, generated auxiliary data points according to the way of using the mean value of pixels adjacent, established a discrete grey prediction model with a fixed end point, made the fitted values of the model as the scale of local contrast enhancement to adaptively adjust the image local contrast, and improved the effect of the pavement image enhancement.This dissertation was from Specialized Research Fund for the Doctoral Program of Higher Education of China:Study on Modeling and Prediction of the Ultimate Bearing Capacity Based on the Generalized Accumulated Generating Operation (NO.200804970005), the National Natural Science Fund Project:Study on Modeling and Prediction of the Short-term Traffic Flow in the Road Network based on Grey Generation Space Modeling (NO.70971103), and the Wuhan Science and Technology Research Projects:Fuzzy Classification of Web Image Semantics based on Evolutionary Optimization(NO.201010621218).更多还原