摘要
为了提升自动驾驶车辆的感知效率和准确率,解决协同感知算法中对协同条件的限制和多源数据融合等问题,本文引入基于激光雷达的即时定位与建图(Simultaneous Localization and Mapping,SLAM)算法,提出面向自动驾驶的多车协同SLAM框架.首先,车辆运行单车SLAM,构建本地约束并共享地图和位姿数据.同时车辆接收并处理其他车的数据,若其他车辆与本车已建立坐标系转换关系则直接完成数据融合,否则基于重叠区域相似点云配准解算多车坐标系转换关系.采用图的连通分支和生成森林理论跟踪数据融合情况并构建多车回环约束,基于通用图优化(General Graph Optimization,G2O)理论对全局地图优化.真实场景与KITTI数据集的实验结果表明,本文的框架无需构建包含所有车辆相对位姿的全局坐标系或满足多车相遇等约束条件,即可实现多车协同SLAM,并在SLAM的效率和准确率等指标上具有优势.
In order to improve the perception efficiency and accuracy of autonomous driving vehicles,and to solve the limitations which correspond to the conditions and the multi-source data fusion,this paper advocates a cooperative SLAM(Simultaneous Localization and Mapping)framework for autonomous driving.Firstly,each vehicle runs single SLAM algorithm,builds the local constraints and shares the map and pose.Meanwhile,each vehicle receives and handles data from others.If the coordinate relationship has been established,the coming data is directly fused.Otherwise,the rela⁃tionship is computed based on the point clouds registration in overlapping areas.The connection components and spawn for⁃est are adopted to track and build the multi-vehicle loop constraint.General Graph Optimization algorithm(G2O)is used to optimize the global map.Experiments based on real world and KITTI dataset show that our framework outperforms relevant SLAM systems,relaxing the conditions of providing relative poses in initial stage and vehicles encountering.
作者
李博洋
刘思健
崔明月
赵治豪
黄凯
LI Bo-yang;LIU Si-jian;CUI Ming-yue;ZHAO Zhi-hao;HUANG Kai(School of Computer Science and Engineering,Sun Yat-sen University,Guangzhou,Guangdong 510006,China;Shenzhen Institute,Sun Yat-sen University,Shenzhen,Guangdong 518057,China)
出处
《电子学报》
EI
CAS
CSCD
北大核心
2021年第11期2241-2250,共10页
Acta Electronica Sinica
基金
广州市科技计划(No.202007050004)
深圳市基础研究(No.JCYJ20180507182508857)
国家重点研发计划(No.2018YFB1802400)。
关键词
协同SLAM
车车协同通信
数据融合
回环检测
激光雷达
自动驾驶
cooperative slam
vehicle-to-vehicle(V2V)
data fusion
loop detection
light detection and ranging
au⁃tonomous driving