基于图像处理的路面裂缝自动检测技术研究

发布时间：2018-02-27 03:38

本文关键词： 裂缝检测图像预处理图像分割特征提取支持向量机　出处：《长安大学》2014年硕士论文　论文类型：学位论文

【摘要】：随着经济的发展，我国公路交通事业高速发展，因此，对公路的养护工作也提出了更高的要求。公路在建成后受到气候、地质条件、通行量、载荷量等因素的影响，导致公路产生不同程度的裂缝，因此，相关部门需要对公路进行定期的检测和养护。国内在对路面裂缝检测时大部分还是使用传统的人工检测方法，但这种传统的方法效率低、误差大，而且对交通有较大的影响，检测人员的人身安全也不能完全保证。因此，对路面自动检测技术的研究迫在眉睫，以便于节省费用，，延长使用年限，提高公路的服务水平。本文主要研究裂缝图像的处理技术，分为图像预处理、图像分割、特征提取三部分。图像预处理部分，本文采用最近邻插值法将图像缩小，变为原图像的1/4；采用四种不同类型的结构元素依次对路面裂缝图像进行中值滤波，平滑去噪；采用基于图像背景提取的灰度校正算法校正图像光照不均。图像分割部分，本文采用Ostu阈值分割算法对路面裂缝图像进行分割，并进行适当改善；采用连通域白色像素点阈值去噪算法去除二值图像的噪声；将数学形态学和白色像素点阈值去噪算法相结合，利用多种形态学算法交替处理，提取出裂缝，最后运用迭代细化方法对裂缝进行了细化。特征提取部分，本文根据不同类型裂缝的特征选取了裂缝像素面积、水平投影、垂直投影、矩形度作为裂缝的特征值，利用裂缝像素面积能够准确判断图像中有无裂缝。本文用裂缝的四个属性作为支持向量机分类器的特征向量，采用高斯径向基核函数RBF，运用“一对多”的多分类算法，对95幅测试样本进行识别，总识别率为85.26%。最后简要分析了造成误判的原因。在相同样本条件下，对比了BP神经网络和支持向量机的分类效果，结果表明，支持向量机的分类精确度要优于BP神经网络。为了满足道路养护对路面裂缝数据参数的需求，分别计算了横向和纵向裂缝的长度以及块状和网状裂缝的最小外接矩形面积，进一步计算了路面状况指数PCI，得出路面破损的程度等级，从而确定养护策略。
[Abstract]:With the development of economy, the highway traffic in our country is developing rapidly. Therefore, the maintenance of highway is also required. The highway is affected by the factors of climate, geological conditions, traffic volume, load and so on. As a result of varying degrees of cracks in the highway, the relevant departments need to carry out regular inspection and maintenance of the highway. Most of the traditional manual methods are still used in the detection of pavement cracks in China, but this traditional method is inefficient. The error is large, and it has a great influence on traffic, and the personal safety of the examiner can not be completely guaranteed. Therefore, it is urgent to study the automatic detection technology of road surface in order to save the cost and prolong the service life. Improve the service level of the highway. This paper mainly studies the processing technology of crack image, which is divided into three parts: image preprocessing, image segmentation and feature extraction. Changing to 1 / 4 of the original image; adopting four different structural elements to filter the pavement crack image in turn to smooth the noise; using the gray correction algorithm based on the image background extraction to correct the uneven illumination of the image. In this paper, the Ostu threshold segmentation algorithm is used to segment the pavement crack image, and the white pixel threshold de-noising algorithm in connected domain is used to remove the noise of the binary image. Combining mathematical morphology with white pixel threshold denoising algorithm, the cracks are extracted by alternate processing of various morphological algorithms, and the cracks are refined by iterative thinning method. In this paper, the crack pixel area, horizontal projection, vertical projection and rectangular degree are selected as the characteristic values of cracks according to the characteristics of different types of cracks, and the crack pixel area can be used to accurately judge whether there are cracks in the image. In this paper, four attributes of cracks are used as feature vectors of SVM classifier, Gao Si radial basis function RBFand "one-to-many" multi-classification algorithm are used to identify 95 test samples. The total recognition rate is 85.26. Finally, the causes of misjudgment are briefly analyzed. Under the same sample condition, the classification results of BP neural network and support vector machine are compared. The results show that the classification accuracy of support vector machine is better than that of BP neural network. In order to meet the requirement of pavement crack data parameters for road maintenance, the length of transverse and longitudinal cracks and the minimum external rectangular area of block and mesh cracks are calculated, respectively. Furthermore, the pavement condition index PCI is calculated, and the degree of pavement damage is obtained, and the maintenance strategy is determined.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;U418

【参考文献】