Research On High-resolution Aerial Image Feature Matching Technology

Image matching is one of the basic issues in photogrammetry.In recent years,with the development of science and technology,Image matching technology has been widely used in various fields.Compared to traditional photogrammetry methods,multivision aerial photography captures the texture of buildings from multiple perspectives,providing a large amount of redundant information for image matching.However,because the oblique image has a larger dip angle,the difficulty of matching is increased.This article focuses on high resolution aerial images,using a variety of feature matching,gray-scale matching,and multi-view image intensive matching algorithms to study the image matching problem of regions with different feature features.The main work and innovation are as follows:(1)As the theoretical basis for studying image matching,explores a various of commonly used,robust feature-based image matching algorithms SIFT,SURF,ORB,and AKAZE in recent years.(2)Analyse and compare the matching of AKAZE-SIFT Algorithm,AKAZE Algorithm and SIFT Algorithm,the results show that the combination of AKAZE algorithm and SIFT descriptor is better than AKAZE algorithm and SIFT algorithm in normal image matching,the matching rate on oblique images is extremely low.Combination of AKAZE algorithm and SIFT descriptor doesn't have rotation invariance,making algorithm improvements,correspondence between the SIFT descriptor and the spatial hierarchy of the feature points extracted by the AKAZE algorithm in the nonlinear scale space.Through experimental verification,the rotation invariance of the AKAZE-SIFT algorithm is achieved.(3)There is a large difference in resolution between vertical and aerial oblique images,the problem of mutual obstruction between features on aerial oblique images is particularly serious.In order to obtain high-density,high-precision dense point clouds,first,using a semi-global matching algorithm to obtain a three-dimensional model pixel-by-pixel disparity map;secondly,using stereo matching post-processing method to refine matchingresults;finally,on this basis,a multi-match primitive multi image matching method(MPM)is used to automatically generate densely-imaged area point clouds.(4)Using ordinary images as experimental data,we observed the stability of feature points and matching effect of SIFT,SURF,ORB and AKAZE algorithms on ordinary images.Experimental data show that the AKAZE detection algorithm has uniform feature points and high stability for feature points extracted from different feature images.The improved AKAZE-SIFT algorithm has a relatively high matching rate on images with different features.Using aeronautical oblique image as experimental data to perform matching study,experimental results show that improved AKAZE-SIFT is better than other algorithms in matching airborne oblique image data with different features.