面向高速公路的公里桩检测及误检改进  

Kilometer piles detection and improvement of error detection for expressway

作　　者：吴镇涛 罗文婷 李林 秦勇[3] 陈文婷 温王鹏 胡喜生 WU Zhentao;LUO Wenting;LI Lin;QIN Yong;CHEN Wenting;WEN Wangpeng;HU Xisheng(College of Transportation and Civil Engineering,Fujian Agriculture and Forestry University,Fuzhou 350108,China;College of Transportation Engineering,Nanjing University of Technology,Nanjing 211816,China;State Key Laboratory of Rail Transit Safety and Control,Beijing Jiaotong University,Beijing 100084,China)

机构地区：[1]福建农林大学交通与土木工程学院,福州350108 [2]南京工业大学交通运输工程学院,南京211816 [3]北京交通大学轨道交通安全与控制国家重点实验室,北京100084

出　　处：《交通科技与经济》2023年第1期50-56,共7页Technology & Economy in Areas of Communications

基　　金：国家重点研发计划项目(2021YFB3202901);福建省高校产学合作重大项目(2020H6009)。

摘　　要：公里桩是高速公路的位置信息载体,自动检测公里桩是实现高速公路路面信息定位的前提。针对高速公路公里桩自动检测中存在的问题,通过车载道路巡检设备采集公里桩数据信息,训练公里桩检测模型。结合Hu不变矩法和公里桩独特的图像特征,提出基于二维图像的公里桩自动检测及误检改进方法,为交通标志的日常巡检作业提供新思路。以测试段高速公路为例进行研究,结果表明:YOLOv5s检测模型对公里桩的召回率达到100%,准确率为52.7%,运用Hu不变矩法对检测结果进行误检排除,召回率降低5.6%,准确率提升35.7%,检测速度和精度能够满足实际巡检需求,并可为其他目标物检测定位提供帮助。Kilometer piles are the carrier of location information on expressway, and automatic detection of kilometer pilesare is the premise of realizing expressway pavement information positioning.Aiming at the problems in automatic detection of kilometer piles of expressway, on-board road inspection equipment is used to collect the data information of kilometer piles, and the detection model of kilometer piles is trained.Combined with Hu moment invariant method and the unique image characteristics of kilometer piles, an improved method of kilometer piles automatic detection and error detection based on two-dimensional images is proposed, which provides a new idea for the daily inspection of traffic signs. Taking the expressway in the test section as an example, the results show that the recall rate and accuracy rate of the YOLOv5s detection model for kilometer piles are 100% and 52.7%. After the Hu moment invariant method is used to eliminate the false detection of the detection results, the recall rate is reduced by 5.6% and the accuracy rate is increased by 35.7%. The detection speed and accuracy can meet the needs of actual patrol inspection and also provide help for the detection and positioning of other objects.

关 键 词：道路巡检 公里桩检测 深度学习 二值化处理 轮廓提取 HU不变矩 

分 类 号：U417.9[交通运输工程—道路与铁道工程]