“道路指纹”关键技术及其在智能车路系统中的应用  

Road Fingerprint:Key Technologies and Applications for Intelligent Vehicles and Infrastructure Systems(IVIS)

作　　者：胡钊政[1] 陶倩文 黄刚[1] 王相龙[1] HU Zhaozheng;TAO Qianwen;HUANG Gang;WANG Xianglong(Intelligent Transportation Systems Research Center,Wuhan University of Technology,Wuhan 430063,China)

机构地区：[1]武汉理工大学智能交通系统研究中心,武汉430063

出　　处：《交通信息与安全》2020年第5期39-49,共11页Journal of Transport Information and Safety

基　　金：国家重点研发计划子项目(2018YFB1600801);湖北省技术创新项目重大专项(2016AAA007);武汉市科技局项目(2020010601012165、2020010602012003)资助。

摘　　要：针对智能网联汽车与车路协同系统中的高精度定位核心技术问题,提出了“道路指纹”的概念与表征模型,并在“道路指纹”的基础上提出了面向智能车路系统的高精度定位方法。“道路指纹”是通过车载传感器数据提取的高稳定性与高辨识度的道路场景特征信息。在“道路指纹”表征模型中,分别从表征的唯一性、计算的快速性、特征的稳定性以及表征的精准性等4个方面完成建模工作。其中,针对表征唯一性需求,提出基于多视角(包括俯视、前视、侧视等)与多传感器的表征方法;针对计算快速性要求,提出了全局特征与语义特征的表征方法;还提出基于深度卷积神经网络(D-CNN)的深度学习特征提取方法,大幅度提高特征表征的鲁棒性;最后,通过提取路面的局部特征,实现特征的精准性(亚像素精度)表征。通过对上述特征进行层次化组织,完成“道路指纹”的表征建模。通过对道路上各个节点进行“道路指纹”计算与建模,并同步获取节点的传感器位姿、场景结构信息,完成道路指纹库构建工作。在定位过程中,首先通过车载传感器获取的数据实时完成“道路指纹”计算,然后通过匹配道路指纹库,完成车辆的高精度位置计算。在开发的“道路指纹”技术基础上,分别从视觉道路指纹定位、LiDAR道路指纹定位以及道路资产管理等3个应用案例给出了该技术的应用前景。所提出的“道路指纹”技术,为解决智能车路系统中的高精度定位问题,特别是卫星信号盲区下的高精度定位问题,提供了一种新的解决思路。To address the core problem of high-precision localization for intelligent vehicles and infrastructure systems(IVIS),this paper first proposes the concept of road fingerprint and its representation models and proposes an accurate localization method for IVIS based on the proposed road fingerprint.The term road fingerprint is stable and distinctive features extracted from the captured data of the road.In this paper,the road fingerprint is represented from four guidelines:①uniqueness;②fastness;③robustness;④accuracy.Multi-view and multi-sensor based methods are proposed to enforce the uniqueness of the road scenario.Furthermore,holistic and semantic features are used for fast computation of road fingerprint.Moreover,the Convolutional Neural Network(CNN)is utilized to enhance the robustness of feature representation.Finally,local features are used to represent pixel-level or point-level features that can describe detailed features of road scenes.Through the hierarchical organization of the above features,road fingerprint is thus accomplished.By computing road fingerprint at every position in the roadways,together with the geometric information and the poses of the sensors,a database of road fingerprint is established.In the step of localization,a real-time road fingerprint calculation is firstly completed through the data obtained from the vehicle sensor.Then thecomputed road fingerprint from current sensed data is matched within the road fingerprint.The position is thus inferred from the best match.The proposed road fingerprint technology is further demonstrated with three typical applications,which illustrates that the proposed method has good potential for IVIS.The proposed road fingerprint suggests a novel solution to the classic high-accuracy localization problem for IVIS,especially for the satellite signal blind areas.

关 键 词：智能车路系统 高精度定位 道路指纹 特征提取 

分 类 号：U495[交通运输工程—交通运输规划与管理]