摘要
无人机飞行控制依赖于卫星导航定位技术,然而在GNSS信号衰减或受干扰的环境下,该技术难以保障飞行任务的安全执行。针对这一技术瓶颈,提出异源影像匹配原理的视觉导航定位方案,为无人机提供了不依赖卫星信号的自主导航解决方案。首先,研究了一种融合多尺度深度学习特征的影像匹配算法,通过深度残差神经网络结构自主训练学习型特征,得到异源影像之间丰富、准确的同名特征点对;然后,构建了PnP导航定位模型,根据无人机影像与参考影像图之间的特征匹配结果预估了影像位姿参数;最后,结合单应矩阵进一步约束异源影像间的投影关系,输出精准的无人机导航定位信息。通过多场景下的飞行实验,结果表明:本文方法能实现异源影像间精确配准,导航定位精度优于3 m,满足无人机视觉导航定位要求。
The flight control of UAVs relies on satellite navigation and positioning technology,but in environments where GNSS signals are attenuated or interfered with,this technology is difficult to ensure the safe execution of flight missions.Aiming at this technological bottleneck,a visual navigation and positioning scheme based on the principle of heterogeneous image matching is proposed,providing unmanned aerial vehicles with an autonomous navigation solution that does not rely on satellite signals.Firstly,an image matching algorithm based on multi-scale deep learning features is studied.Learning features are trained independently using a deep residual neural network structure to obtain rich and accurate corresponding feature pairs between heterogeneous images.Then,PnP navigation and positioning model is constructed,and the image pose parameters are estimated based on the feature matching results between the UAV images and the reference images.Finally,combining the homography matrix to further constrain the projection relationship between heterogeneous images and output accurate UAV navigation and positioning information.Through UAV flight experiments in different scenarios,it is concluded that the method in this paper can achieve accurate registration between heterogeneous images,with navigation and positioning accuracy better than 3 m,meeting the requirements of unmanned aerial vehicle visual navigation and positioning.
作者
马小淞
MA Xiaosong(Shenyang Geotechnical Investigation&Surveying Research Institute Co.,Ltd.,Shenyang 110000,China;Industry University Research Fusion Innovation Base of Space Air Ground Integrated Natural Resources Spatiotemporal Information Service,Shenyang 110000,China)
出处
《测绘与空间地理信息》
2025年第10期43-46,50,共5页
Geomatics & Spatial Information Technology
关键词
无人机
视觉导航
异源影像匹配
深度学习
单应矩阵
UAV
visual navigation
heterogeneous image matching
deep learning
homography matrix
