GNSS-denied UAV Visual Navigation Research

With the rapid development of GNSS jamming technology,GNSS,the global satellite navigation system,is facing increasingly severe challenges.In the complex electromagnetic environment GNSS rejection conditions,due to the limitations of UAV communication and perception capabilities,it cannot accurately sense the surrounding environment and thus cannot work properly,and the limited autonomous navigation capability seriously hinders the application of UAVs.Therefore,exploring the autonomous navigation of UAVs under GNSS denial conditions is one of the hot directions for future research.With the rapid development of computer vision,vision sensors are becoming more flexible and cheaper,and vision-based approaches show great advantages in the field of UAV navigation.In this thesis,we address the needs of autonomous navigation and positioning of rotary-wing UAVs under GNSS denial conditions,study image matching algorithms,introduce visual pose solving models,propose a visual navigation and positioning method based on fast matching of UAV images and high-resolution satellite images,and propose a Kalman filter fusion positioning solution for multi-source information such as visual data,inertial guidance data and altimeter data to realize the autonomous navigation and positioning of UAVs under GNSS denial conditions.The proposed Kalman filter fusion solution can achieve safe and reliable autonomous navigation and positioning of UAVs under GNSS denial conditions.Although domestic and foreign researchers have made a series of research results in the research of UAV navigation and positioning based on visual scene matching,there are still problems such as poor positioning accuracy and long positioning time,especially with the development of high-resolution satellite images,the matching process between UAV images and digital image maps is time-consuming,and the existing matching algorithms cannot meet the demand of UAV visual navigation and positioning.In this paper,through in-depth analysis and summary of the shortcomings and challenges of existing methods,the following four aspects of research work are carried out.(1)In-depth study of image matching algorithms,comparison of the advantages and disadvantages of various feature point alignment algorithms,and screening of the best combination of feature point detectors and descriptors based on the needs of UAV visual navigation.(2)A new method of fast image matching for UAV vision navigation is proposed.Firstly,we conduct a priori pre-processing research and feature sparsification research on digital image maps,screen high-quality feature information,establish a priori feature database,and provide technical support for fast image alignment of UAV visual navigation;Then,a new fast matching method of "Downsampling + Prior Extraction + SURF" fast image matching method for visual navigation(DPES)is proposed to achieve fast alignment of UAV images with digital image maps,which has slightly lower accuracy but obvious speed advantage compared with the standard SURF algorithm,and the alignment processing speed can be more than 60 Hz for digital image maps of 4000 pixels.Finally,a new fast matching algorithm of Sub-pixel based on Sparse features and Two-time matching is proposed to achieve fast and high-precision alignment between UAV images and digital image maps,with comparable accuracy and obvious speed advantage compared with the standard SURF algorithm,and the alignment processing speed can reach more than 20 Hz for digital image maps of 4000×4000 pixel level.(3)Research on visual pose solving method based on image matching.Based on the assumption of flat scene to give the height information of digital image map feature points,while taking into account the feature point distribution characteristics,a visual Pn P pose estimation method based on random sampling consistency algorithm(RANSAC)is proposed,and the visual pose solution accuracy meets the needs of UAV visual navigation in this study.(4)In view of the different update frequencies of information such as visual inertial guidance,variable observation equations are used to enable the fusion of multi-source information of different frequencies,which improves the accuracy and reliability of navigation and positioning information,and the rationality and feasibility of the multi-source information fusion scheme proposed in this chapter are verified through simulation experiments.