Research And Implementation Of UAV Visual Navigation Key Technology

UAV(Unmanned Aerial Vehicle)visual navigation is one of the most important autonomous navigation technologies.Computer vision technology processes image information from camera,combineing with data from UAV's other sensors to calculate the UAV current location and other information so that UAV can complete given mission.Because of low cost of vision navigation and not relying on GPS,vision navigation is widely used in all walks of life.In this paper,the low altitude UAV vision navigation system in small area is considered and designed,and the whole system is divided into three parts: hardware,navigation software and auxiliary software.The hardware consists of four-rotor UAV,airborne computer,gimbal,camera and digital radio.Navigation software is composed of visual positioning and visual tracking.Auxiliary software is composed of ground station,flight management software and data transmission protocol.During visual positioning,the ORB feature is used.Extract the ORB feature points of the two images,and the descriptors of the feature points are generated.Firstly,calculate two images' s descriptors violent matching using Hamming distance.After that,cross matching algorithm and the PROSAC algorithm are used to filter the rough matching,and the similarity between the two images is measured according to the number of matching point.Finally,the clustering algorithm based on density is used to classify the matching feature points,and the two classes with the largest area are selected to represent two images' s overlap area.A good offset calculation result is obtained.The accuracy is improved by 10% compared to the results before density clustering.This paper uses KLT algorithm for visual tracking.To trace the images' s large amount of motion,the KLT algorithm is combined with the Gaussian pyramid to calculate the offset from the top of the pyramid and then using offset as the initial value to calculate the next layer image offset.Based on the pyramid model,the algorithm has achieved good results in continuous frame image test.When the system is implemented,take the base gallery scheme into use to reduce the computational time of the visual navigation.The feature points and descriptors of the reference maps are extracted and stored in a text file in advance,which is 90% time less than that of the direct extraction in images matching.Visual navigation using visual tracking and visual matching.Every ten seconds visual matching is used to calculate the coordinates of UAV coordinates,correcting visual tracking cumulative error.