Research On Key Technologies Of UAV Navigation And Positioning System Based On Vision

UAV(Unmanned Aerial Vehicle)visual navigation technology is one of the current research hotpots in the field of computer vision.Its purpose is to use the visual information of aerial images to achieve continuous,stable and accurate positioning of UAV.With the continuous upgrading of hardware platforms,UAV are widely used in many fields such as national defense,military,disaster detection,and geological exploration.Most UAV rely on the equipped Global Navigation Satellite System(GNSS)for navigation and positioning.However,radios that acquire the system are prone to signal loss or deliberate attacks.UAV are expensive,and positioning systems play a vital role in their stability and safety during flight.Vision-based navigation and positioning technology does not rely on satellite positioning systems,and is of great significance for unmanned aerial vehicles to achieve autonomous flight tasks when GNSS signals are lost.This article analyzes and designs the UAV visual positioning system in outdoor environment,conducts in-depth research on visual information extraction and positioning estimation,and proposes a series of improvement measures.The main research contents and contributions of this article are as follows:(1)This thesis studies a visual information processing algorithm based on ORB features.The algorithm includes image feature extraction,matching filtering,and descriptor generation.This thesis proposes improvement measures for the image matching and filtering stage.First,the ORB features of the captured image are extracted,and then the improved RANSAC algorithm is used to filter the coarse matches between adjacent frames,while eliminating outlier feature points.Compared with the traditional matching method,the matching accuracy is improved,and the retained feature points can more accurately represent the visual information of the image.(2)This thesis designs a visual localization(BVL)process,which includes two steps of data collection and localization estimation.Data collection is the process of storing image feature descriptors and positioning information in the database,while localized estimation is the process of finding the visual database that best matches the current scene observed by the UAV.These two steps can be performed simultaneously and continuously.The image coordinates with the highest similarity obtained will be(3)used as reference positioning information in the positioning estimation process.(3)This thesis studies an improved particle filter localization algorithm.Aiming at the problem that the standard particle filter positioning algorithm is likely to affect the positioning estimation when processing particle sampling,an adaptive sampling method is proposed so that it can set the number of particle samples according to the motion state.At the same time,the similarity between the current image and the reference image is converted into a discrete probability distribution to improve the weight distribution of the particles.(4)Obtain satellite image video frames based on the ROS system,build a satellite image visual database,and program the positioning algorithm.The experimental results show that the technical solution proposed in this thesis can meet the needs of visual navigation and positioning.