Research On Scene Matching Algorithms For UAV Visual Positioning

In unmanned aerial vehicle(UAV)visual positioning,the effect of scene matching directly determines the precision of visual positioning.Generally,the high resolution remote sensing images are used as reference images,the UAV aerial images are used as real-time images,and the UAV position is determined by scene matching.However,because the imaging principle of the reference image and the real time image is diverse and shooting conditions are vastly different,so there is noticeable discrepancy between the image gray level,making it difficult to match the images.The line feature can use the joint structure information between images to solve the matching problem effectively,so we study the visual positioning algorithm based on line feature matching deeply.First,the basic process of visual localization is described,and the importance of scene matching algorithm is stated.According to the basic order of feature matching,the line feature extraction algorithms are studied first.Through comparison experiment,it is found that the performance of line feature extraction algorithm based on curve fitting is most stable.And the line feature matching algorithm based on RANSAC is explored,the shortcomings of the algorithm are pointed out,and the optimization method by grouping line feature is proposed to speed up finding corresponding line features.In the end aiming at the problem that size difference is quite large between the reference image and the real time image,a method of scale unification is proposed.Using the from-coarse-to-fine matching strategy,a matching method,suitable for UAV positioning with high precision and good real-time performance,is obtained.The main contents and innovation are as follows:1.The basic principle of UAV visual positioning is expounded.The basic components of UAV and visual positioning system are presented,the basic process of visual positioning is introduced,illustrating the important position of scene matching technology.The formula of calculating image transform parameters using homologous features is deduced.And last the images used in UAV visual positioning are analyzed,showing the difficulties in the UAV visual positioning.2.Various line feature extraction algorithms,line feature matching algorithm based on RANSAC and the method using homologous lines to determine transformation parameters are studied.The specific implementation steps of different line feature extraction algorithms are studied,some properties of extraction algorithm under the influence of the noise are compared with synthetic images,it is found that the line feature detection algorithm based on curve fitting is best.Next,the method to find homologous lines by RANSAC is studied,shortages of the algorithm are pointed out,and the method based on features grouping is illustrated to speed up the feature matching.3.The scene matching in the UAV visual location is completed.The method able to uniform image scale with the scale stability of line feature,the line feature extraction algorithm based on the curve fitting and the improved RANSAC feature matching algorithm are used to roughly match images of the smallest images.Then the corresponding relationship between the line features in larger image of the adjacent layers is determined in turn,more accurate transformation parameters can be calculated.When a matching image achieve the size of the original image,the least square method is used to improve the matching result.