摘要
针对单靠高精度定位建图成本高、更新慢、不适用于大范围,而只基于即时定位与地图构建(SLAM)建图累积误差高,且传统的聚类方法在数据偏移等输入不好时容易导致聚类错误的问题,该文采用基于众包的方法生产高精度地图,包括车端SLAM和云端地图学习两个模块,致力于构建多源异构成图的关键能力。在传统DBSCAN聚类算法的基础上,优化核心种子聚类方式,分析不同类型的道路要素的数据特性,灵活地使用不同的判别要素代替传统的距离判定,并在长安汽车实际采集的众包数据集上进行了实验。与传统的方法相比,该文方法更加适用于较全面的道路要素聚类,不受道路类型的限制,也更加符合实际的工程需求。
high-precision positioning alone is costly and slow to update,so it does not apply to a large range,while simultaneous localization and mapping(SLAM)alone will face high cumulative errors and slow updates.This paper uses a crowd-sourced method to produce high-precision maps,including two modules of SLAM and cloud map learning,and is committed to building key capabilities of multi-source heterogeneous maps The traditional clustering method is prone to cluster errors when the input such as data offset is not good.Therefore,based on the traditional DBSCAN clustering algorithm,this paper optimizes the core seed clustering mode,analyzes the data characteristics of various road elements,flexibly uses different discriminating elements to replace the traditional distance determination,and conducts extensive experiments on the crowdsourced data set collected by Changan Automobile.Compared with the traditional methods,our method is more suitable for the comprehensive clustering of road elements,is not limited by road types,and is more in line with the actual engineering needs.
作者
石作琴
李健
刘昌锋
王建文
金晔
SHI Zuoqin;LI Jian;LIU Changfeng;WANG Jianwen;JIN Ye(State Key Laboratory of Intelligent Vehicle Safety Technology,Chongqing Changan Zhitu Technology Co.,Ltd.,Chongqing 400023,China)
出处
《测绘科学》
CSCD
北大核心
2024年第4期196-207,共12页
Science of Surveying and Mapping
基金
智能汽车安全技术全国重点实验室开放课题项目(cstc2021jscx-dxwtBX0023)。
