公路路面裂缝检测与识别技术研究

发布时间：2018-06-04 11:30

本文选题：图像预处理 + 图像分割　；参考：《大连海事大学》2017年硕士论文

【摘要】：伴随着公路的使用,很多早期的公路出现了破损、弯沉等问题,严重影响了公路的运载能力、行车安全和使用年限。在检测中,通过公路的各种定量指标来估量其预期工作年限与维护工作量,其中路面裂缝的估量是最重要的指标。伴随着计算机技术的高度发展及图像处理技术水平的迅速提升,基于图像处理的路面病害自动检测技术被提出来应用到实际中。在实际路面裂缝检测中,裂缝本身方向性比较复杂,另外路面会有油污、障碍物及阴影等,这样会导致图像噪声比较严重,因此实现有效的检测算法是路面裂缝图像检测中的难点。目前的图像处理算法对于裂缝图像的处理效果还不能完全满足实际需求,尤其对裂缝的检测有待完善。本文对路面裂缝图像预处理、裂缝图像分割和裂缝识别分类进行了研究。(1)路面裂缝图像预处理。根据路面裂缝图像的特征,提出了一种改进的中值滤波算法和一种基于区域对比度的增强算法,对路面裂缝图像进行处理。对比常用的去噪增强算法实验结果,本文算法很好地消除了噪声,增强了裂缝区域,提高了图像的质量。(2)路面裂缝图像分割处理。采用了一种基于子块的图像差分方法,对路面裂缝图像进行相位移动后再极值平滑得到背景差分图像,得到的裂缝边缘清晰,连续性较好。之后根据裂缝的线性特征,采用连通域的几何特征对油污等具有一定纹理特征的噪声进一步去除。(3)裂缝识别分类。裂缝的投影特征和像素个数有区别,根据这个特点对裂缝图像进行特征提取,然后采用BP神经网络对裂缝图像进行分类,设计不同隐含层个数的网络进行训练,选择正确率高的网络进行分类。最后对裂缝骨架化,计算裂缝的参数,进行路面损坏状况指数PCI计算,判断路面损坏程度。
[Abstract]:Along with the use of highway, many of the early highway appeared damage, deflection and other problems, which seriously affected the transport capacity of the highway, driving safety and service life. In the detection, the expected working life and maintenance workload are estimated by various quantitative indexes of highway, among which, the estimation of pavement cracks is the most important index. With the development of computer technology and the rapid improvement of image processing technology, the automatic detection technology of road surface diseases based on image processing has been put forward and applied to practice. In the actual pavement crack detection, the orientation of the crack itself is quite complex, in addition, the road surface will have oil stains, obstacles and shadows, which will lead to serious image noise. Therefore, effective detection algorithm is a difficult problem in pavement crack image detection. The current image processing algorithm for crack image processing effect can not fully meet the actual needs, especially for crack detection needs to be improved. In this paper, the preprocessing of pavement crack image, crack image segmentation and crack recognition classification are studied. According to the characteristics of pavement crack image, an improved median filter algorithm and an enhancement algorithm based on regional contrast are proposed to process the pavement crack image. Compared with the experimental results of the commonly used denoising enhancement algorithm, the proposed algorithm can eliminate noise, enhance the crack area and improve the image quality of pavement crack image segmentation. A subblock based image differential method is used to obtain the background differential image by phase shifting and extreme value smoothing. The crack edge is clear and the continuity is good. Then according to the linear characteristics of cracks, the geometric features of connected regions are used to remove the noises with certain texture features such as oil pollution and so on. The projection feature and pixel number of cracks are different. According to this feature, the feature extraction of crack image is carried out, and then BP neural network is used to classify the crack image, and the network with different number of hidden layers is designed for training. The networks with high accuracy are selected for classification. Finally, the crack skeleton is used to calculate the crack parameters, and the pavement damage index (PCI) is calculated to judge the pavement damage degree.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U418.6;TP391.41

【参考文献】