【作者】 李楠；

【导师】 赵祥模；

【作者基本信息】 长安大学 ， 交通信息工程及控制， 2017， 硕士

【摘要】 裂缝类病害是道路破损的主要形式,也是道路检测的重要内容。现有的裂缝检测算法多存在通用性不高或效率低下的问题,一种算法在设计时往往只针对特定的路面状况,难以适应复杂多变的路面环境,因此仍需进行改进和优化。针对以上现状,本文在分析现有文献的基础上,对路面裂缝区域初定位和路面裂缝高精度检测算法进行深入研究和软件实现,主要工作内容如下:(1)设计了一种基于深度学习的路面裂缝区域初定位算法首先基于亮度高程模型,开发了一种路面图像自动匀光改进算法,消除路面图像亮度分布不均匀的现象。然后对路面图像进行子块分割,设计了基于Lenet-5卷积神经网络的深度学习模型,利用人工标记好的裂缝子块图像和非裂缝子块图像对模型进行训练,从而实现路面图像内裂缝子块的初定位。(2)设计了一种基于多尺度脊边缘的线状裂缝检测算法通过对路面裂缝的脊边缘特性分析,得到基于各点一阶导数和二阶导数的脊边缘检测原理。然后基于裂缝的多方向和多尺度特性,设计了双向、多尺度的融合检测算法,并通过数学形态学和连通域算法去除伪裂缝。(3)设计了一种基于Hessain矩阵的网状裂缝检测算法基于对Hessian矩阵特征值的分析,根据其对各种二维几何结构表现出的不同特性构造多尺度线状滤波器,通过滤波得到路面图像中各点的线状响应强度。然后根据线状响应强度对图像进行二值化并去除伪裂缝。(4)开发了一种路面破损程度评估算法首先提出一种基于最小生成树的裂缝连接算法,增强裂缝分割结果的连续性。然后开发了一套裂缝几何参数检测算法,对连接后的裂缝二值图像进行参数检测。最后根据检测参数对路面进行破损程度评估。本文设计了从图像预处理到破损程度评估的一系列路面裂缝自动检测中的重要算法。算法充分利用了裂缝的多尺度特性和线状特征,对光线不均和噪声干扰有较强的适应性。实验表明:本文提出的算法在检测精度、抗噪性、通用性方面均优于文中用于对比的几种经典检测算法,具有一定工程应用和推广价值。更多还原

【Abstract】 Crack disease is not only the main form of road diseases,but also an important part of road detection.Due to the complexity of the pavement environment,the existing crack detection algorithms are usually inefficient or not universal,and it is often difficult to achieve the expected result.Therefore,it is urgent to improve and optimize the existing detection algorithms.In view of the above situation,based on the analysis of the existing literature,this paper makes an in-depth study and software realization of the initial location of pavement crack area and the high precision detection algorithm of pavement crack.The main contents are as follows:(1)An initial location algorithm for pavement cracks based on deep learning is designed.Firstly,based on the brightness elevation model,an improved dodging algorithm is developed to eliminate the uneven distribution of pavement image brightness.Then,the pavement images are diveded into sub-blocks,and the improved Lenet-5 convolution neural network model is trained by marked non-crack sub-blocks and crack sub-blocks,so as to realize the initial positioning of the crack sub-blocks in the pavement image.(2)A linear crack detection algorithm based on multi-scale ridge edge is designed.Through the analysis on the ridge edge characteristics of pavement cracks,the principle of ridge edge detection based on the first derivative and second derivative of each point is obtained.Then,based on the multi-directional and multi-scale characteristics of the crack,a bidirectional and multi-scale fusion detection algorithm is designed,and the fake cracks are removed by mathematical morphology and connected domain algorithm.(3)A meshy crack detection algorithm based on Hessian matrix is designed.Based on the analysis of the Hessian matrix eigenvalue,the multi-scale linear filter is constructed according to the different characteristics that the eigenvalue shows on different twodimensional geometric structure.The linear response intensity of each point in the pavement image is obtained by filtering,and then the image is binarized according to the response intensity and the fake cracks are removed.(4)A pavement distress evaluation algorithm is developed.Firstly,a crack connection algorithm based on the minimum spanning tree is proposed to enhance the continuity of crack segmentation results.Then a set of crack geometric parameter detection algorithm is developed to detect the binary image.Finally,the pavement distress is evaluated by the detected parameters.This paper designs a series of important algorithms for pavement crack detection from image preprocessing to damage assessment.The algorithms make full use of the multi-scale and linear feature of cracks,and have strong adaptability to uneven brightness and noise.Through a large number of experiments,the proposed algorithms are superior to the classical detection algorithms for comparison of detection accuracy,noise immunity and versatility,and has certain engineering and promoting value.更多还原