Research On Feature Matching Algorithm In UAV Visual Navigation

In the case of GPS failure of UAV(Unmanned Aerial Vehicle),visual navigation is an important auxiliary navigation method.Visual navigation is widely used because the strong autonomy,high precision and high precision and high efficiency to complete the auxiliary navigation task.Among them,the UAV vision-assisted navigation and positioning technology based on feature matching has great research significance because of its simple equipment and high positioning accuracy,and can be combined with inertial systems to form a fully autonomous high-performance navigation and positioning system.Because the image feature matching algorithm has the problems of instability,high time-consuming,and poor robustness,this paper conducts in-depth research on traditional feature matching algorithms,and proposes some improved algorithms,which are used in the navigation and positioning of UAVs.The main improvements are as follows:First of all,considering that feature points should be used to calculate the position of the UAV in the navigation and positioning of the UAV based on feature matching,too concentrated feature points will cause position solution errors,this paper proposes a grid-based uniform ORB feature detection algorithm.The algorithm uses sub-pictures that divide the picture into grid units,and then performs feature point detection and extraction,so that the grid-based uniform ORB feature detection algorithm can reduce the phenomenon of concentration when extracted feature points.Experiments have shown that compared with the ORB algorithm,the feature point repetition rate of the proposed algorithm has increased by 2%,the number of correct matches has increased by 8.6%,and the accuracy of UAV positioning has increased by 21.9%.Secondly,this paper proposes a GMS consistency screening algorithm that incorporates local structure preservation.This algorithm takes into account the limitation that the accuracy of GMS depends on the number of feature points,and proposes a combination of feature vector screening in local neighborhoods and GMS statistical screening to eliminate mismatches.Experiments show that compared with GMS algorithm,the GMS consistency screening algorithm that incorporates local structure preservation can bring an average 4.3% improvement in the feature matching accuracy and in this paper can bring an average 18.7% improvement in the accuracy of the UAV positioning.At last,considering the UAV visual navigation positioning algorithm,high-precision and a large number of correct matching points can increase the accuracy of navigation and positioning.In this paper,a GMS consistency screening algorithm based on weighted grid of feature number is proposed,considering the importance of undifferentiated grid in GMS,some correct matching errors are eliminated.The paper proposes that the mesh weight is weighted by the feature quantity,and the comprehensive score of the mesh weight and feature quantity is used to screen.Experimental results show that compared with GMS algorithm,the GMS consistency screening algorithm based on weighted grid of feature number can improvement by 6.6% on the matching accuracy and the estimation accuracy of UAV flight height is improved by 23%.The precision of UAV visual navigation has increased.