景象匹配无人机视觉定位  

Drone-based Scene Matching Visual Geo-localization

作　　者：袁媛 孙柏[1,2] 刘赶超 YUAN Yuan;SUN Bo;LIU Gan-Chao(School of Computer Science,Northwestern Polytechnical University,Xi'an 710072;School of Artificial Intelligence,Optics and Electronics(iOPEN),Northwestern Polytechnical University,Xi'an 710072)

机构地区：[1]西北工业大学计算机学院,西安710072 [2]西北工业大学光电与人工智能研究院,西安710072

出　　处：《自动化学报》2025年第2期287-311,共25页Acta Automatica Sinica

基　　金：国家自然科学基金(62273282,62201471)资助。

摘　　要：无人机因其高度的灵活性,在临地安防、灾后救援、地质勘测、农业植保等领域发挥着重要作用,因此受到越来越多的关注.定位导航作为无人机中的关键技术,对于无人机能否顺利执行任务至关重要.当前主要的定位导航算法包括全球导航卫星系统、惯性定位以及景象匹配定位导航等.首先,景象匹配定位导航方法利用计算机视觉技术,对无人机飞行时采集的航空影像进行数字化特征编码;随后,通过构建相似性度量与检索模型,将航空影像特征与预先获取的遥感地图库特征进行相似性度量,从而完成景象匹配;最后,根据无人机航空影像与遥感卫星地图的匹配结果,获取相应的地理位置信息,并将其更新为无人机的定位结果.景象匹配定位导航方法摆脱了定位系统对定位信号的依赖,实现了无人机飞行定位的自主化.鉴于此,以景象匹配算法中的特征提取方式为线索,分别针对基于模板匹配、基于手工特征以及基于度量学习的景象匹配,梳理其发展过程,并总结景象匹配定位导航方法中的关键问题.最后,针对景象匹配算法的发展现状,总结无人机景象匹配定位方法中亟待解决的问题.Drones play an important role in vicinagearth security,post-disaster rescue,geological survey,agricultural plant protection,and other fields due to their high flexibility,and they receive increasing attention.As a key technology in drones,positioning and navigation are crucial for whether the drone can successfully perform tasks.Currently,the main positioning and navigation algorithms include the global navigation satellite system,inertial positioning,and scene matching positioning and navigation.Among them,the scene matching positioning and navigation method uses computer vision technology to encode the digital features of aerial images collected during the flight of drones.Then,by constructing a similarity measurement and retrieval model,it measures the similarity between the aerial image features and the pre-obtained remote sensing map library features to complete the scene matching.Finally,based on the matching results of drone aerial images and remote sensing satellite maps,it obtains the corresponding geographic position information and updates it as the positioning result of the drone.The scene matching positioning and navigation method eliminates the dependence of the positioning system on positioning signals and realizes the autonomy of drone flight positioning.This paper follows the feature extraction methods in the scene matching algorithm and outlines the development process of scene matching based on template matching,manual feature-based,and metric learning-based approaches while summarizing the key problems in the positioning and navigation methods of scene matching.Finally,this paper summarizes the urgent problems that need to be solved in drone scene matching localization methods based on the current development status of scene matching algorithms.

关 键 词：临地安防 无人机 视觉定位 景象匹配 度量学习 多视角变化 

分 类 号：V279[航空宇航科学与技术—飞行器设计] TP391.41[自动化与计算机技术—计算机应用技术]