Research On Feature Matching Algorithm Of UAV Low-altitude Remote Sensing Image

UAV low-altitude remote sensing technology has the advantages of simple structure,low cost,low risk,flexibility,and strong real-time performance.It is widely used in various industries.Image matching is the key to applying high-resolution UAV images to large-scale terrain mapping and fine three-dimensional modeling.It is the purpose of image matching to quickly and effectively obtain the right amount of point pairs with the same name between sequence image pairs.Traditional SIFT and SURF algorithms detect a large number of feature points in high-resolution images with uneven spatial distribution,low matching efficiency,and low registration accuracy.Directly applying them to UAV low-altitude remote sensing image matching has poor results.This paper improved the process of feature extraction,feature description,and feature matching of two classic algorithms,proposed optimized SIFT-OCT algorithm and U U-SURF algorithm,selected remote sensing images of different ground features for experiments.The effectiveness of the improved algorithm applied to feature matching of UAV low-altitude remote sensing images was verified.The main research contents and results include:(1)The low-altitude remote sensing images of UAVs with different types of features have inconsistent image features and different adaptability to feature matching algorithms.In this paper,combining the features of the image features,we deeply analyzed the matching performance of the four commonly used image matching algorithms(SIFT,SURF,ORB,and AKAZE)on real UAV image pairs of different features,and designed the simulation data of the real image under the changes of scale,rotation,brightness and noise,and evaluated the stability of the algorithm.(2)Aiming at the problem of the large number of feature points detected by the classic matching algorithm in high-resolution images and low matching efficiency,the optimized SIFT-OCT algorithm was proposed.The first set of scale spaces were actively abandoned for feature detection and the image segmentation method was adopted to speed up the detection process.In the feature matching stage,the similarity coefficient was proposed for the secondary screening of matching points,and the RANSAC algorithm was used to calculate the perspective transformation model parameters for precise matching.Four groups of images of different object types in the same UAV sequence image are selected for comparative verification experiments.The results show that the optimized SIFT-OCT algorithm greatly reduced the number of feature extractions,improved image matching efficiency.It is a method suitable for UAV low-altitude remote sensing image matching.(3)In the process of UAV low-altitude remote sensing image matching,there are too many feature detection points and uneven spatial distribution,poor matching point pair spatial distribution quality,and low registration accuracy.A uniform distribution U-SURF algorithm was proposed.The quad-tree index strategy was introduced to ensure the uniform distribution of the sub-block feature detection points while retaining a certain number of feature points.The calculation of the main direction of the feature points was ignored.In the feature matching stage,the low-threshold two-way matching was used for rough matching,and a uniform block strategy was proposed to improve the spatial distribution quality of matching point pairs.A stable basic matrix was used to eliminate mismatched points by epipolar constraint.The homography matrix was calculated as the transformation matrix between images for registration.Five groups of UAV sequence images of different ground object types were selected for comparison experiments,which verified the ability of U U-SURF algorithm to be applied to high-resolution UAV image registration.The matching accuracy rate was increased by 0.4%?8.3%,while the accuracy has been improved by 0.013?0.114.