| 摘要: |
| 针对全球卫星导航系统（Global Navigation Satellite System，GNSS）拒止或间歇可用环境下，多智能体协同导航面临的通信受限与误差累积问题，提出了一种基于交叉视图匹配的轻量化分布式协同导航定位方法。该方法结合目标检测算法与深度相机，设计了一种同时考虑语义信息与几何阈值的多视角匹配策略，以获取智能体间的交叉观测信息。所提方法基于卡尔曼滤波，融合了测距数据、视觉惯性里程计信息、交叉观测信息以及间歇可用的GNSS数据，通过智能体间共同观测到的环境特征建立跨视角几何约束。空地协同实验结果表明，在GNSS拒止环境下，相较于仅依赖惯性测量的导航方式，无人机在引入交叉观测信息后，绝对位置均方根误差（Root Mean Square Error，RMSE）由121.56 m降低至25.01m，误差降低约79.4%；相较于仅依赖视觉惯性里程计的导航方式，无人机在引入交叉观测信息后，绝对位置RMSE由11.30 m降低至7.31 m，定位精度提升约35.31%；对于无人车平台，交叉视图协同观测同样能够在视觉-惯性导航基础上进一步降低定位误差。 |
| 关键词: 无人机集群 GNSS拒止 空地协同定位 卡尔曼滤波 交叉视图匹配 多智能体协同 |
| DOI：10.20079/j.issn.1001-893x.251229005 |
|
| 基金项目:装备预先研究共用技术项目（50911030202） |
|
| A Cross-view Constrained Air-Ground Collaborative Navigation and Localization Algorithm |
| AN Yi,QIAO Tao,YANG Hong,YAN Jiaxin,ZHOU Zebo |
| (1.Southwest China Institute of Electronic Technology，Chengdu 610036，China;2.CRRC Qingdao Sifang Co.，Ltd.，Qingdao 266111，China;3.School of Aeronautics and Astronautics，University of Electronic Science and Technology of China，Chengdu 611731，China) |
| Abstract: |
| To address the problem of communication constraints and error accumulation in multi-agent collaborative navigation under Global Navigation Satellite System(GNSS)-denied or intermittently available environments，a lightweight distributed collaborative localization method based on cross-view matching is proposed.The method integrates object detection algorithms with depth cameras to design a multi-view matching strategy that simultaneously considers semantic information and geometric thresholds，thereby obtaining cross-observation information among agents.Built upon the Kalman filter，the proposed approach fuses ranging data，visual-inertial odometry，cross-observation information，and intermittently available GNSS data.Cross-view geometric constraints are established through environmental features commonly observed by multiple agents.Experimental results in an air-ground collaborative scenario demonstrate that in GNSS-denied environments，compared with navigation relying solely on inertial measurements，the absolute position root mean square error(RMSE) of the unmanned aerial vehicle(UAV) is reduced from 121.56 m to 25.01 m after incorporating cross-observation information—a reduction of approximately 79.4%.Compared with navigation relying solely on visual-inertial odometry，the absolute position RMSE of the UAV decreases from 11.30 m to 7.31 m after integrating cross-observation，improving localization accuracy by about 35.31%.For the unmanned ground vehicle platform，cross-view collaborative observation further reduces localization error on the basis of visual-inertial navigation. |
| Key words: UAV swarm GNSS-denied air-ground collaborative localization Kalman filter cross-view matching multi-agent collaboration |
《电讯技术》编辑部 