| 路面破损图像自动处理技术研究进展 |
| |
| 引用本文： | 徐志刚,车艳丽,李金龙,赵祥模,潘勇,王忠仁,韦娜,宋宏勋.路面破损图像自动处理技术研究进展[J].交通运输工程学报,2019,19(1):172-190. |
| |
| 作者姓名： | 徐志刚 车艳丽 李金龙 赵祥模 潘勇 王忠仁 韦娜 宋宏勋 |
| |
| 作者单位： | 长安大学 信息工程学院,陕西 西安,710064;石家庄华燕交通科技有限公司 西安分公司,陕西 西安,710000;长安大学 信息工程学院, 陕西 西安 710064;中咨数据有限公司, 北京 100083;美国加州交通厅 养护处, 加利福尼亚萨克拉门托 95835 |
| |
| 基金项目： | 国家重点基础研究发展计划项目2018YFB010510401陕西自然科学基础研究计划项目2013JQ8017交通部基础应用项目2015319812060中央高校基本科研业务费专项资金项目310824163202中央高校基本科研业务费专项资金项目300102248403 |
| |
| 摘 要： | 总结了路面破损图像自动处理技术的重要研究成果, 分析了该领域关键技术的研究进展, 包括路面破损检测系统、图像处理算法和识别算法评估; 比较了不同路面破损检测系统与目标自动识别算法的检测精度和适用性, 给出了路面破损图像自动处理技术未来可能的主要研究方向。研究结果表明: 在路面破损检测系统方面, 从早期基于摄影技术的图像采集到目前的3D激光扫描技术, 路面图像采集技术更加便捷和高效, 但破损图像自动分析和目标自动识别算法仍然存在挑战; 在路面破损图像处理算法方面, 传统的路面裂缝目标分割算法已由过去的基于单一特征(灰度、边缘形状等) 的检测方法演化到多特征融合检测方法和图优化检测方法, 还出现了一些精细化的裂缝目标连接与恢复算法, 大幅提高了裂缝检测精度, 但需要的计算资源和人工先验知识库也随之不断增大; 在路面裂缝处理算法评估和比较方面, 主要利用人工分割来评价自动识别结果, 目前迫切需要建立一个面向全球开放的大型路面破损图像数据库, 以客观、有效地评估现有各种路面破损图像处理算法; 基于2D图像特征分析的路面破损图像自动识别算法很难在识别精确性、算法通用性和实时性方面同时取得最佳效果; 近年来, 大量学者开始尝试借助深度学习神经网络自动识别路面破损, 但该技术仍处于活跃的演进过程中; 在提高路面破损自动识别精度和效率方面, 3D激光扫描技术和基于人工智能的深度学习技术的发展将对未来路面破损图像自动识别技术的最终突破产生重大推进作用。
|
| 关 键 词： | 路面养护 路面破损 图像处理 破损识别 裂缝检测 破损分类 |
| 收稿时间： | 2018-08-22 |
Research progress on automatic image processing technology for pavement distress |
| |
| XU Zhi-gang,CHE Yan-li,LI Jin-long,ZHAO Xiang-mo,PAN Yong,WANG Zhong-ren,WEI Na,SONG Hong-xun.Research progress on automatic image processing technology for pavement distress[J].Journal of Traffic and Transportation Engineering,2019,19(1):172-190. |
| |
| Authors: | XU Zhi-gang CHE Yan-li LI Jin-long ZHAO Xiang-mo PAN Yong WANG Zhong-ren WEI Na SONG Hong-xun |
| |
| Affiliation: | 1.School of Information Engineering, Chang'an University, Xi'an 710064, Shaanxi, China2.Shijiazhuang Huayan Transportation Technology Co., Ltd., Xi'an Branch, Xi'an 710000, Shaanxi, China3.Zhongzi Data Co., Ltd., Beijing 100083, China4.Division of Maintenance, California Department of Transportation, Sacramento 95835, California, USA |
| |
| Abstract: | The important research achievements on the automatic image processing technology for pavement distress were summarized. The research progress of key technologies in this field was analyzed, including the pavement distress detection system, image processing algorithm and evaluation of recognition algorithm. The detection accuracy and applicability were compared for the different pavement distress detection systems and target automatic recognition algorithms. The possible future research directions of automatic pavement distress image processing technology were presented. Research result shows that in the aspect of pavement distress detection system, from early image acquisition based on the photography technology to the current 3D laser scanning technology, the pavement image acquisition technology becomes more and more convenient and effective. However, there still exist some challenges in the automatic analysis on distress images and automatic recognition algorithm on targets. In the aspect of pavement distress image processing algorithm, the traditional algorithms of segmenting pavement distress targets evolve from the methods using single feature (such as grayscale and edge shape) to multi-feature fusion-based methods and graph optimization-based detection methods. Furthermore, there emerges some dedicated algorithms for recovering or connecting cracks, greatly improving the detection accuracy of crack recognition. Nonetheless, as the complexity of these algorithms grows up, the required computational resources and the size of prior knowledge base both sharply increase. In the aspect of evaluation and comparison of crack processing algorithms, manual segmentation is mainly used to evaluate automatic recognition results. At present, it is urgent to establish a large-scale pavement distress image database opening to the world, so as to objectively and effectively evaluate various existing image processing algorithms for pavement distress. Automatic image processing algorithms for pavement distress based on 2D image features analysis is difficult to achieve the best results with detection accuracy, algorithm versatility and real-time performance simultaneously. In recent years, a large number of scholars begin to use the deep learning neural network to automatically recognize pavement distress, but the technology is still in an active evolution process. In the aspect of improving the accuracy and efficiency of automatic recognition for pavement distress, the 3D laser scanning technology and the deep learning technology based on artificial intelligence will greatly promote the final breakthrough on automatic image recognition technology for pavement distress in the future. |
| |
| Keywords: | |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
| 点击此处可从《交通运输工程学报》浏览原始摘要信息 |
|
点击此处可从《交通运输工程学报》下载全文 |
|
