Research On Technology Of Positioning And Attitude Measurement For UAV Vision-Based Navigation

Because of its simple structure,low cost,strong viability and good mobility,UAV is widely used not only in civil fields,such as aerial photography,disaster detection and urban three-dimensional mapping,but also in military aspects like battle field reconnaissance and target strike.With the fast development and popular application of the UAV,its autonomous navigation has attracted wide attention and has become an important research topic for researchers in the areas of UAV application and navigation.At present,its navigation system is mainly based on GNSS and INS.However,since GNSS is susceptible to electromagnetic interference and INS has big cumulative errors,they cannot completely meet the needs of UAV navigation.With the advantages of low cost,little volume,light weight,mutual exchange to circumstance and high attitude accuracy,vision-based navigation can effectively improve the relative position and attitude accuracy and reliability of UAV autonomous navigation.By taking UAV autonomous navigation as the research topic,the author concentrates studies on the absolute position and orientation measurement method with image of single artificial signs and the relative position and orientation measurement method with image sequence of natural characteristics,on the basis of systematic analysis of the fundamental principle and methodology of UAV autonomous navigation.The main contents and innovative points of this paper are listed as follows.(1)The technical processes of absolute position and orientation measurement and relative position and orientation measurement for vision-based navigation are studied,including knowledge of camera projection,the theoretical basis of absolute position and orientation measurement based on the estimation of homographic matrix and relative position and orientation measurement based on the estimation of intrinsic matrix.(2)To the artificial ordinary circular marks,the RANSAC algorithm based on the estimation of homographic matrix is studied and implemented for matching image points and the correspondent 3D points holistically.By establishing the mapping relationship between image coordinates and their coordinates of artificial marks,the absolute position and orientation information of UAV is obtained via homographic matrix.The experiments with rotary-wing UAV show that the absolute position and orientation measurement proposed in this paper is correct and effective.(3)On the basis of analyzing the robustness and computation efficiency of the SIFT algorithm,the SIFT matching algorithm is improved and fast matching is reached for image sequence with natural characteristic points.By setting proper pixel step at the stage of feature point detection in scale space,the number of extreme points to be detected is reduced and the distributive uniformity of feature points is improved.With the determination of image contrast threshold based on the gray scale of detected feature points,the points with low contrast is eliminated effectively.Considering the feature of little change in displacement and rotation between adjacent UAV images,a method of excluding mismatch points is put forward according to the topological relation confinement of characteristic points.The RANSAC algorithm is applied for estimating the fundamental matrix and the relative position and orientation parameters are finally estimated with the intrinsic matrix.(4)The accuracy of the position and orientation measurement for UAV vision-based navigation is analyzed.The paper analyzes the effect of various factors in the solution process on the accuracy of position and orientation parameters,and lays emphasis on the influences of the image position error on the measurement of position and orientation.A single variable error model and the covariance propagation model are established.(5)An experimental system for positioning and attitude measurement of UAV vision-based navigation is designed and built to verify the position and orientation measurement methods proposed in this paper.With the images of artificial signs and images with natural characteristic points captured by rotary-wing UAV and fixed-wing UAV respectively,the paper experiments on a single image and image sequences.The absolute position and orientation measurement method takes the error of re-projection as the standard of accuracy,while the relative position and orientation measurement method achieves a better result by comparing the POS system information.The reliability and accuracy of the two position and orientation measurement methods is verified by the experiments.