Research And Implementation Of UAV Remote Sensing Image Mosaic Technology

UAV remote sensing platform is widely used in remote sensing data acquisition because of its wide use,simple operation and strong maneuverability.Because the surface coverage area of the UAV image is small and the overlap is large,the single image can not show the whole measurement area,so the image splicing has become the basic work of the UAV remote sensing image processing.In the process of image matching,point matching is used to get the same name pair.In this paper,combined with the features of UAV image,the two algorithms of feature extraction and matching search in point feature matching are studied and discussed,and a more efficient and robust UAV remote sensing image splicing method is proposed.The main work of this paper is as follows:First,in the aspect of feature point detection and description,four typical feature extraction algorithms with local invariance are analyzed with the scale invariance of the image as the starting point.In order to solve the problem of low positioning accuracy and no scale invariance of the feature points of ORB feature extraction 'algorithm,the scale invariance is made by establishing the multi-scale space.And introducing sub-pixel interpolation technology to improve its positioning accuracy to improve the ORB feature extraction algorithm.With the aid of Matlab and opencv platform programming,the improved algorithm and typical algorithm are realized.The improved algorithm is compared with the feature extraction effect of SIFT algorithm,SURF algorithm,ORB algorithm and LDB algorithm.The improved feature extraction algorithm not only preserves the fast superiority of the original algorithm,but also improves the matching precision greatly.Degree.Secondly,in feature matching,considering that the improved ORB extraction algorithm uses binary descriptors in the feature description phase,a close neighbor algorithm based on Hamming distance is used to make a rough matching of the feature points.The matching results of the feature points extracted from the existing algorithm and the improved algorithm are compared and analyzed from the matching speed,efficiency and the correct rate,and the feasibility of the rough matching strategy and the scale invariance of the improved algorithm are verified.In order to further improve the matching accuracy,RANSAC algorithm is introduced in the precise matching stage to purify the matching points,and effectively eliminate the wrong matching points.Finally,based on the improved extraction algorithm and matching search strategy,a stitching method is proposed to achieve two image splicing and multi image splicing.The stitching image is compared with the traditional stitching method by subjective evaluation,structural similarity and edge difference spectroscopy.The method has good visual effect and proves the reliability of the method.