华南理工大学学报(自然科学版) ›› 2023, Vol. 51 ›› Issue (1): 134-144.doi: 10.12141/j.issn.1000-565X.210778

所属专题: 2023年交通运输工程

• 交通运输工程 • 上一篇    

基于语义分割模型的三维沥青路面车辙异常分析方法

王艾迪郎洪1 丁朔陆键洪小刚2 温添1   

  1. 1.同济大学 道路与交通工程教育部重点实验室, 上海 201804
    2.山西高速公路工程检测有限公司, 山西 太原 030008
  • 收稿日期:2021-12-06 出版日期:2023-01-25 发布日期:2023-01-02
  • 通信作者: 郎洪(1994-),男,博士后,主要从事智能交通系统、交通安全等的研究。 E-mail:honglang@tongji.edu.cn
  • 作者简介:王艾迪(1997-),女,博士生,主要从事智能交通系统、交通安全研究。E-mail:aidiwang@tongji.edu.cn
  • 基金资助:
    国家自然科学基金资助项目(71871165)

Rutting Abnormality Analysis Method for 3D Asphalt Pavement Surfaces Based on Semantic Segmentation Model

WANG AidiLANG Hong1 DING Shuo1 LU Jian1 HONG Xiaogang2 WEN Tian1   

  1. 1.Key Laboratory of Road and Traffic Engineering of the Ministry of Education, Tongji University, Shanghai 201804, China
    2.Shanxi Expressway Engineering Inspection Co. , Ltd. , Taiyuan 030008, Shanxi, China
  • Received:2021-12-06 Online:2023-01-25 Published:2023-01-02
  • Contact: 郎洪(1994-),男,博士后,主要从事智能交通系统、交通安全等的研究。 E-mail:honglang@tongji.edu.cn
  • About author:王艾迪(1997-),女,博士生,主要从事智能交通系统、交通安全研究。E-mail:aidiwang@tongji.edu.cn
  • Supported by:
    the National Natural Science Foundation of China(71871165)

摘要:

车辙深度是路面状况评价和道路养护的一个重要指标。现有车辙深度测量方法未考虑复杂路况下坑槽、松散、裂缝和桥梁接缝等对车辙计算的影响,其有效性和适用性受到了限制,测量结果的真实性也有待进一步验证。有鉴于此,文中提出了一种基于语义分割模型的沥青路面车辙异常检测与校正方法。采用三维线激光技术采集道路横断面高程数据,利用包络线算法提取车辙深度。对于最大车辙深度超过10 mm的横断面,搭建基于深度学习的语义分割框架,提出改进的DeepLabV3+网络对病害类型进行自动辨识和像素定位,并结合最大车辙深度高程点,设计基于拉格朗日插值的校正规则来对异常车辙进行校正。研究结果表明,改进的DeepLabV3+模型能较为准确地识别和定位造成车辙检测异常的路面病害,其对5种路面特征和病害的综合检测准确率达81.63%,在均交并比(MIoU)和大部分交并比(IoU)上的表现均优于U-Net、PSPNet、DeepLabV3+模型。现场验证结果表明,文中方法不仅能够自动分析车辙异常的原因,还能通过对异常车辙的校正排除其他病害的影响,从而较大程度地还原实际车辙深度水平。文中研究成果可为路面预防性养护提供科学数据支撑。

关键词: 三维线激光技术, 车辙异常检测, 语义分割模型, 车辙深度, 异常值校正

Abstract:

Rutting depth is an important indicator for the evaluation of pavement conditions and road maintenance. The existing rutting depth measurement method fails to consider the influence of potholes, raveling, cracks and bridge joints under complex road conditions, and its effectiveness and applicability are limited, and the authenticity of the measurement results remains to be further verified. In view of this, an abnormality detection and correction method based on semantic segmentation model is proposed. The road cross section elevation data is collected by three-dimensional line laser technology, and the rutting depth is extracted by the envelope algorithm. For cross-section of the maximum rut to more than 10 mm, this paper builds a semantic semantic division framework based on deep learning, proposes an improved DeepLabV3+ network to automatically identify and pixel positioning of the disease type, and designs a correction rule based on Lagrangian interpolation to correct the abnormal rut by combining the maximum rut depth elevation points. The results show that the improved DeepLabV3+ model can more accurately identify and locate pavement distress causing rutting abnormality, and its comprehensive detection accuracy of five pavement characteristics and distress reaches 81.63%. Its performance in Mean Intersection over Union (MIoU) and Intersection over Union (IoU) is better than that of U-Net, PSPNet, and DeepLabV3+ models. The field validation results show that the method in the paper can not only automatically analyze the causes of rutting abnormality, but also exclude the influence of other distress by correcting the abnormal rutting, so as to restore the actual rutting depth level to a greater extent. The research results in this paper can provide scientific data support for pavement preventive maintenance.

Key words: 3D line laser technology, rutting abnormality detection, semantic segmentation model, rutting depth, abnormal value correction

中图分类号: