摘要
协同即时定位与地图构建(SLAM)建立在多无人机的联合感知能力之上,通过局部地图的交互融合构建一个增量式全局环境地图,以提高多无人机任务协同的准确性、实时性和鲁棒性.针对多无人系统协同定位与构图中数据高效共享与利用的难题,面向快速、准确、大范围多机协同SLAM需求,本文提出了一种基于集中式架构的多无人机局部地图数据高效共享和融合的SLAM方法——Data Sharing Oriented Multi-UAV Collaborative SLAM(DSM-SLAM).该方法创新性地提出了:(1)基于共享局部地图的两步重定位机制,实现了多机间全局地图的高效闭环检测;(2)基于层次聚类的地图融合选择机制,提高了地图融合的快速性和准确性,最终,在增强单无人机跟踪定位鲁棒性的同时,实现了高效地闭环检测,提升了多无人机视觉SLAM系统的数据协同处理能力.本文不仅基于ROS实现了DSM-SLAM的原型系统,而且采用公开数据集KITTI对该方法展开了一系列实验验证,结果表明,基于上述两种机制设计的DSM-SLAM方法,不仅能够快速高效地恢复单无人机跟踪定位,而且可以自适应选择数据融合的顺序,有效增强了全局地图构建的完整性和准确性.
Built on the joint perception of multi-UAVs,collaborative Simultaneous Localization And Mapping(SLAM)can construct an incremental global environment map by means of the interaction and fusion of local maps,which can improve the accuracy,real-time and robustness of multi-UAV task coordination.Aiming at the problem of efficient data sharing and utilization,and the demand of fast,accurate and large scale area in cooperative SLAM of multi-UAVs,this paper proposes a novel centralized architecture based multi-UAV collaborative SLAM method—DSM-SLAM(Data Sharing Oriented Multi-UAV Collaborative SLAM),to achieve the efficient share and fusion.DSM-SLAM innovatively proposes:(1)a two-step relocation mechanism based on local map sharing,which achieves efficiently loop closure detection for the global map among multi-UAVs;(2)a map-fusion selection mechanism based on hierarchical clustering,which improves the speed and accuracy of map fusion.The proposed method can not only enhance the robustness performance of single UAV tracking and localization,but also can promote the collaborative processing capability of the multi-UAV visual SLAM system.We implement our DSM-SLAM system based on the Robot Operating System(ROS)and perform the experimental evaluation based on the public KITTI dataset.The experimental results demonstrate that the proposed DSM-SLAM can achieve relocalization of single UAV quickly and efficiently,and enhance the integrity and accuracy of global map construction by the adaptive selection of data fusion.
作者
史殿习
杨卓越
金松昌
张拥军
苏向东
李睿豪
SHI Dian-Xi;YANG Zhuo-Yue;JIN Song-Chang;ZHANG Yong-Jun;SU Xiang-Dong;LI Rui-Hao(National Innovation Institute of Defense Technology,Academy of Military Sciences,Beijing 100166;School of Computer Science,National University of Defense Technology,Changsha 410073;Tianjin Artificial Intelligence Innovation Center,Tianjin 300457;College of Computer Science,Inner Mongolia University,Hohhot 010021)
出处
《计算机学报》
EI
CAS
CSCD
北大核心
2021年第5期983-998,共16页
Chinese Journal of Computers
基金
国家重点研发计划(2017YFB1001901)资助.
作者简介
史殿习,男,1966年生,博士,研究员,博士生导师,中国计算机学会(CCF)会员(14099M),主要研究领域为人工智能、分布式计算、云计算及大数据处理等.E-mail:dxshi@nudt.edu.cn;杨卓越,女,1993年生,硕士,主要研究方向为人工智能、分布式计算及云计算等.E-mail:yangzhuoyue18@163.com;通信作者:金松昌,男,1987年生,博士,助理研究员,主要研究方向为人工智能、数据挖掘、社区发现等.E-mail:jsc04@tsinghua.org.cn;张拥军,男,1972年,博士,研究员,主要研究领域为高性能计算、分布式应用系统等.E-mail:zhang-vic@263.com;苏向东,男,1984年生,博士,讲师,主要研究方向为人工智能、自然语言处理等.E-mail:cssxd@imu.edu.cn.李睿豪,男,1991年生,博士,助理研究员,主要研究方向为自主机器人、视觉SLAM、人工智能等.
