基于无人机的路桥病害检测方法研究

发布时间：2018-01-13 02:18

本文关键词：基于无人机的路桥病害检测方法研究　出处：《中国科学院大学(中国科学院遥感与数字地球研究所)》2017年硕士论文　论文类型：学位论文

【摘要】：路桥病害的实时检测对于路桥的安全具有重要意义,也是公路巡检和维护工作的主要依据之一。路桥病害的数量和严重程度直接反映了路桥健康状况,同时,路桥病害的改变预示着路桥表层或者深层的病变,进而影响路桥的使用寿命和安全系数。线状病害、网状病害和坑洞作为路桥早期病害,是路桥表面的常见病害,其相关参数(如线状裂缝的形状、面积和位置等)是评估路桥健康状况的重要参数。传统路桥检测是采用路桥检测车加巡检员目视相结合的方法,具有检测效率低下、费时费力的缺点,而无人机作为新型的检测平台,具有采集影像角度灵活、成本低、效率高的特点。随着低空摄影测量和图像处理技术的发展,路桥病害的检测方法产生了显著变化,特别是无人机的发展对整个路桥病害检测体系产生了深刻影响。因此,研究基于无人机的路桥病害检测方法及其精度评估,对于路桥健康状况的评估和对应的巡检方案的安排都具有重要意义。本研究基于路桥养护公司提供的路桥病害数据和大疆S900无人机搭载的SONY ILCE-7R相机拍摄得到的影像数据,首先采用无人机影像处理降低无人机影像模糊和影像高噪声,然后应用手持相机拍摄得到的路桥病害影像训练线状裂缝、网状裂缝和坑洞这三类早期和基础性的路桥病害的多部件形变模型,然后应用路桥部件的多部件形变模型在无人机影像上,通过检测得到路桥病害区域,再基于路桥病害区域影像应用边缘检测方法提取出路桥病害信息,最后对基于无人机影像的路桥病害检测系统进行了介绍和实力验证。通过分析无人机影像处理、路桥病害模型、模型检测和路桥信息提取方法和路桥病害检测系统,本研究得出了以下结论:(1)提出了一套创新性的基于无人机的路桥病害检测技术。本研究针对无人机影像特性比较了多种影像处理算法以提高影像质量;验证了描述路桥病害的形态、结构、纹理特性的多部件形变模型,及其模拟路桥病害的可行性,并建立了路桥病害模型库;检测出路桥病害影像区域并采用边缘检测算法来提取出路桥病害信息,显著提高了路桥病害检测和信息提取的效果;最后阐述了路桥病害检测系统功能、算法和实施验证。(2)通过研究无人机影像特性和影像预处理方法,有效改善了无人机影像质量。在多参数条件下比较了去模糊概率模型和标准各向异性扩散这两种算法,证明了标准各向异性扩散函数在保存影像边缘特征的前提下,在消除无人机影像模糊方面表现更好。(3)通过研究路桥病害影像特性和特征表达方法,创新性地提出将模拟了路桥病害影像形态、结构、纹理特性的多部件形变模型应用到路桥病害检测中,并比较了病害模型的模拟泛化性能。研究结果表明,线状裂缝模型模拟泛化能力良好,模型的平均精度AP=0.64891;网状裂缝模型模拟泛化能力较差,平均精度AP=0.5658;坑洞模型模型的泛化能力最强,平均精度AP=0.68254。在尽量全面地提取出路桥病害的要求下,如果要求路桥病害检准率较高,当召回率(recall)值在0.6时,精度(precision)值均在0.2~0.4之间。(4)不同病害模型应用在无人机影像上,在响应度阈值的设定和检测效果上存在差异,边缘分割效果也具有显著差别。研究结果表明,线状裂缝响应度为-0.95时查全率最高,网状裂缝查全率最高出现在响应度为-0.99处,坑洞响应度为-0.9时查全率最高。对检出区域应用局部自适应阈值分割和最小连通域去除法提取路桥病害的边缘、形状等参数信息,与经典的Canny算法、Sobel算法相比,本实验提出的思路可以更准确地提取出路桥病害信息。(5)通过编程实现基于无人机影像的路桥病害检测系统,在系统设计、功能实现和实例验证方面进行了阐述。实现结果表明,本文算法有效可靠、系统功能良好,具有良好的应用前景。
[Abstract]:The real-time detection of Luqiao disease has important significance for the safety of Luqiao, is one of the main basis for the highway inspection and maintenance work in Luqiao. The number and severity of disease directly reflects the health status of Luqiao, at the same time, Luqiao disease change indicates that Luqiao surface or deep lesions, thereby affecting Luqiao's service life and the safety coefficient of linear. Disease, disease and potholes as reticular early disease is a common disease in Luqiao, Luqiao on the surface of the relevant parameters (such as linear crack shape, size and location) is an important parameter to evaluate the health status of Luqiao. The traditional detection method is adopted to detect the Luqiao Luqiao car with a combination of visual inspection, has low detection efficiency and the shortcomings of time-consuming, and as a new platform for UAV detection, image acquisition with flexible angle, low cost and high efficiency. With the characteristics of low altitude photography The development of measurement and image processing technology, detection method of Luqiao disease has changed significantly, especially the development of UAV has a profound impact on the whole of Luqiao disease detection system. Therefore, the research of Luqiao disease detection method and accuracy assessment based on UAV, is of great significance to assess the inspection program in Luqiao health and the corresponding arrangement. This study based on the image data of UAV with Luqiao disease data of Luqiao and Xinjiang maintenance company provides S900 SONY ILCE-7R camera was first used, UAV image processing to reduce the UAV blurred images and images with high noise, linear crack image training in Luqiao disease and then apply the hand-held camera to get. Multi component deformation model of mesh cracks and potholes of these three types of early and basic diseases in Luqiao, Luqiao and many parts of parts of the application form In the model of UAV images, obtained by detecting the disease region of Luqiao, Luqiao area image detection method and application of edge extraction based on the disease of bridge disease information, at the end of the Luqiao disease detection system based on UAV images are introduced and the strength verification. Through the analysis of the UAV image processing, Luqiao model and Luqiao model of disease, detection the information extraction method and Luqiao disease detection system, this study draws the following conclusions: (1) put forward a set of innovative Luqiao disease detection technology based on UAV. Based on the characteristics of UAV image comparison of a variety of image processing algorithms to improve image quality; verify the description of Luqiao disease morphology, structure, multi component deformation model of texture characteristic, and the feasibility of simulation of Luqiao disease, and established the Luqiao disease model library; detection of bridge disease image and mining area Using edge detection algorithm to extract the bridge disease information, improve the Luqiao disease detection and information extraction; finally elaborated Luqiao disease detection system function, algorithm and implementation of the verification. (2) through the research on the characteristic of a UAV image and image preprocessing method, effectively improve the UAV image quality in multi parameter. Under the condition of comparison to diffusion of the two algorithms of fuzzy probability model and standard anisotropy, proved that the standard anisotropic diffusion function in the premise of preserving image edge features, in the elimination of UAV image fuzzy performance better. (3) the expression method through the study of Luqiao disease image features and characteristics, put forward the simulation Luqiao disease image morphology, structure, texture and multi component deformation model is applied to Luqiao disease detection, and compared the simulation model of disease generalization performance. The results table The linear crack model generalization ability is good, the average accuracy of AP=0.64891 model; mesh crack model generalization ability is poor, the average accuracy of AP=0.5658; the generalization ability of the model pits the strongest, the average accuracy of AP=0.68254. in the extraction of the bridge as a comprehensive disease request, if Luqiao disease rate is higher when the precision, recall rate (recall) value of 0.6, accuracy (precision) value was between 0.2~0.4. (4) different disease model is applied in UAV images, there are differences in the response threshold and detection result, edge segmentation also has significant difference. The results show that the linear response to -0.95 recall crack the highest recall the highest reticular crack at -0.99 in response to -0.9 response, potholes. The highest recall on detection of local adaptive threshold segmentation and regional application of minimum connected Go to the Luqiao disease domain division to extract edge, shape and other parameters, and the classical Canny algorithm, Sobel algorithm, the proposed method can accurately extract the bridge disease information. (5) through the programming of Luqiao disease detection system based on UAV images, in the system design, function realization and verification are described. The result shows that this algorithm is effective and reliable, the system function is good, has a good application prospect.

【学位授予单位】：中国科学院大学(中国科学院遥感与数字地球研究所)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U446

【参考文献】