【作者】 乌日娜；

【导师】 白云；

【作者基本信息】 内蒙古工业大学 ， 交通运输工程， 2021， 硕士

【摘要】 公路交通是国民交通重要的通行方式,但随着使用年限的增加,以及各种内外作用的影响,路面性能会逐渐下降直至产生病害影响交通通行。路面性能劣化的表现通常为开裂、变形、松散和车辙。在这些破损中,开裂通常发生在路面性能劣化的早期,而开裂产生的裂缝也是路面最常见的破损形式。因此,准确及时地检测路面裂缝并分析其特征和影响是道路养护工作的重中之重。一直以来,对于路面破损的检测识别均采用检测人员目视加路面检测车巡检相结合的方法,这些方法通常操作繁琐、检测效率低下、费时费力。因此本文采用操作灵活、效率高、成本低的无人机作为新的检测工具来检测路面破损。本文研究采用大疆御MAVIC 2 PRO无人机搭载云台相机哈苏HASSELBLAD进行公路路面图像采集,依据行业标准和实地考察采集了具有横向裂缝、纵向裂缝、斜向裂缝和块状裂缝共10000张路面破损图像制作成新的数据集。并基于深度学习的方法,提出了一种结合路面破损自动检测定位、裂缝分类识别分割提取以及破损参数自动计算的综合检测模型。模型首先运用路面破损检测与识别模型对破损裂缝进行检测定位,分类后获得各条裂缝的类别置信度;然后将裂缝信息传入下一级的路面破损分割与提取网络,对选定的路面破损裂缝进行精确地分割提取。最后根据分割的结果计算破损图像中各条裂缝的长度、宽度和面积,同时利用无人机采集的图像定位信息来对破损图像的具体位置进行显示储存。模型整体通过优化内置参数和改进现有的网络结构,提高了路面破损的分类精确度和分割效果。与传统的单一模型相比,本文提出的综合模型方法不仅提供了路面破损的类别信息,还能提供准确的定位和几何参数信息,为道路自动检测和养护工作提供了新方案。同时无人机数据采集的复杂性也使得检测模型具有一定的泛化能力,对于将来其他路面和桥面等破损检测具有参考价值。更多还原

【Abstract】 Highway traffic is an important way of national traffic,but with the increase of service life,and under the influence of various internal and external effects,the performance of the road surface will gradually decline until the occurrence of diseases affecting traffic.The deterioration of pavement performance is usually manifested by cracking,deformation,looseness and rutting.Among these damages,cracking usually occurs in the early stage of pavement performance deterioration,and the cracks generated by cracking are the most common form of pavement damage.Therefore,accurately and timely detection of road cracks and analysis of their characteristics and influence is the top priority of road maintenance work.The detection and identification of road damage has always adopted the method of combining visual inspection personnel and road inspection vehicle inspection,these methods are usually cumbersome,inefficient and time-consuming.Therefore,this paper uses the UAV with flexible operation,high efficiency and low cost as a new detection tool to detect road damage.In this paper,DJI MAVIC 2 PRO unmanned aerial vehicle is used to carry HASSELBLAD camera for road surface image acquisition,according to the industry standard and field investigation,a total of 10,000 pavement damage images with transverse cracks,longitudinal cracks,diagonal cracks and map crackings were collected and made into a new data set.Based on the deep learning method,a comprehensive detection model combining the automatic detection,location,crack classification,recognition,segmentation and extraction of pavement damage and the automatic calculation of damage parameters was proposed.Firstly,the pavement damage detection and identification model is used to detect and locate the damaged cracks,the class confidence of each cracks was obtained after classification.Then the crack information is transmitted to the pavement damage segmentation and extraction network of the next level,and the selected pavement damage cracks are segmented and extracted accurately.Finally,the length,width and area of each crack in the damaged image are calculated according to the segmentation results,and the image positioning information collected by the UAV is used to display and store the specific location of the damaged image.By optimizing the built-in parameters and improving the existing network structure,the model improves the classification accuracy and segmentation effect of pavement damage.Compared with the traditional single model,the integrated model proposed in this paper can not only provide the classification information of pavement damage,but also provide accurate location and geometric parameter information,which provides a new scheme for the automatic detection and maintenance of roads.At the same time,the complexity of UAV data acquisition also makes the detection model have a certain generalization ability,which is of reference value for other damage detection of road surface and bridge deck in the future.更多还原