一种适用于道路裂缝识别的自动化道路提取方法  

An Automated Road Extraction Method Suitable for Road Crack Recognition

作　　者：许志鸿 梁艳平[1,2] 修祺 牛一杰 XU Zhihong;LIANG Yanping;XIU Qi;NIU Yijie(School of Traffic and Transportation,Beijing JiaoTong University,Beijing 100044,China;Key Laboratory of Transport Industry of Big Data Application Technologies for Comprehensive Transport,Beijing JiaoTong University,Beijing 100044,China)

机构地区：[1]北京交通大学交通运输学院,北京100044 [2]北京交通大学综合交通运输大数据应用技术交通运输行业重点实验室,北京100044

出　　处：《综合运输》2025年第2期130-136,共7页China Transportation Review

基　　金：自然科学横向课题(T10A800050)。

摘　　要：为了提高道路裂缝的识别精度,建立了一种基于分水岭算法与GrabCut算法的自动化道路提取框架,可以满足复杂环境下的结构化与非结构化道路的识别与提取。该框架首先确定图像背景、前景区域并利用分水岭算法初步处理图像,其次采用Canny算子与概率霍夫变换作为中间衔接部分,自动搜寻并确定含道路区域的矩形框,最后利用GrabCut算法将含道路区域的矩形框与交互式矩形框相对应,以此来确定道路区域,进而达到自动化道路提取的目的。自动化道路提取对存在反光、有部分遮挡等复杂环境下的结构化路面、非结构化路面提取效果均较为理想,且在晴、阴天气下均有较好的表现。后续采用YOLOv8模型对原始图像和经过自动化道路提取处理后的图像分别进行道路裂缝识别训练,以两组数据集的各个损失值和m AP50值进行评价,处理后图片相较于原始图像的损失值均降低了26%以上,mAP50指标提升了38.2%,说明自动化道路提取能有效增强道路裂缝识别的精度。To enhance the precision of road crack identification,an automated road extraction framework based on the watershed algorithm and GrabCut algorithm is established.The framework can identify and extract structured and unstructured roads under complex environments.Initially,the framework determines the background and foreground regions of the image and utilizes the watershed algorithm for preliminary image processing.Subsequently,the Canny operator and probabilistic Hough transform serve as intermediate connectors to automatically search for and define rectangular regions containing road features.Finally,the GrabCut algorithm associates these rectangular regions with interactive ones to delineate the road areas,thereby achieving the goal of automated road extraction.The automated road extraction exhibits promising results for both structured and unstructured road surfaces in complex environments with reflections and partial obstructions,performing well under various weather conditions.Subsequent to this,the YOLOv8 model is employed to train road crack detection on both the original images and those processed through automated road extraction.Evaluation is conducted using various loss values and mAP50 values of the two datasets.Results show that the processed images exhibit a reduction in loss values of over 26%compared to the original images,with an improvement in the mAP50 metric by 38.2%,indicating that automated road extraction effectively enhances the accuracy of road crack identification.

关 键 词：自动化道路提取 分水岭算法 GrabCut算法 概率霍夫变换 YOLOv8 

分 类 号：U418.6[交通运输工程—道路与铁道工程]