Research On Close-range High-accuracy Vision Navigation Technology For UAV Aerial Refueling

Aerial refueling is an important technology to extend the flight time of Unmanned Aerial Vehicle(UAV),which can enable the ability of UAV to carry out missions all-weather and no-rest.Thus,it has great advantages in long-term non-stop aerial surveillance,reconnaissance and strike.The relative position perception in close docking is the key to successful aerial refueling.Visual navigation has been applied to short-range relative navigation because of its ability to simulate human eye perception.At present,there are still some problems such as inadequate positioning accuracy,reliability and real-time performance for aerial refueling visual navigation of UAV.As for this,this paper studies the key technologies of visual navigation system,such as cone drogue detection and tracking,marker-assisted visual navigation,visual pose calculation and so on,aiming to promote the development of visual navigation technology for aerial refueling of UAV,and enhance the comprehensive performance of UAV aerial refueling vision navigation system.Firstly,the characteristics of color-shape feature localization method and machine learning method for drogue target detection are analyzed.In view of the contradiction between these two methods in real-time performance and high reliability,a fast drogue detection and localization method based on cascaded multi-dimensional image features is proposed.The color feature,shape feature,image statistical features and edge features are sequentially detected and extracted by cascaded structure,thus the high real-time performance and recognition reliability are guaranteed.Then,the characteristics of the inverse compositional image alignment(ICIA)method and discriminant correlation filtering(DCF)method for fast tracking of drogue are analyzed.Aiming at these problems,such as sensitivity problem under scale change,occlusion and deformation,and drifting problem with time,a multi-patches fusion based ICIA drogue tracking method and a drogue feature points positioning method based on median tracking initialized polar transformation are proposed,which can ensure high real-time performance and improve the positioning accuracy of drogue target tracking.After that,aiming at the problem of point positioning problem in marker-assisted visual navigation,this paper proposes a visual calibration method for spatial point location under multi-view scene determination,which realizes the accurate measurement of 3D coordinates of spatial points.In addition,aiming at the problem of feature point alignment in visual navigation with general geometric distribution,a general feature point correspondence algorithm based on singular value decomposition(SVD)of coordinate matrix is proposed,which avoids the exponential search of feature point correspondence in traditional methods or the need for manual design of the complicated matching strategy when searching for feature point correspondence.Afterwards,aiming at the problem of monocular single-circle vision navigation,this paper deduces a simpler formula for calculating monocular vision pose with single circle feature from the similarity between the coefficient matrix of the world system and the coefficient matrix of the image system.Aiming at the problem of vision pose estimation under arbitrary multi-circle and multi-point combined features,a general point and circle feature fusion method for vision pose estimation is proposed.By fusing point features and circular features in the same frame,the optimal estimation of vision pose parameters under arbitrary multi-circle and arbitrary multi-point combination can be obtained.At last,three different types of test platforms,such as labeled image set,indoor high-precision position assessment platform and outdoor low-altitude flight simulation,are established.The proposed visual navigation method in this paper are tested by experimental data from different aspects.The results show that the proposed methods can improve the precision,realtime performance and reliability of vision navigation in UAV aerial refueling,and these methods provide important technical reference for vision navigation applied in UAV aerial refueling.