Research On Stitching Technology Of UAV Remote Sensing Images

UAV is widely used in many fields,such as agriculture,reconnaissance intelligence,city planning,aerial photography,disaster rescue,surveying and mapping,because of its advantages such as high flexibility,small size,high timeliness,low cost and low risk.However,UAV remote sensing platforms are limited by camera focus and flight altitude,which results in less coverage of a few single remote sensing images,which is less likely to be satisfied.In order to get a panoramic image with wide field of view,two or more remote sensing images of unmanned aerial vehicles(uavs)that overlap with each other need to be mosaic.In this paper,the UAV Remote Sensing Image Mosaic Technology as the research goal,focusing on the UAV remote sensing image mosaic technology,especially image preprocessing,feature point extraction and matching.The main work is as follows:(1)Image acquisition and preprocessing.The image of flue-cured tobacco planting base in Pingba district of Anshun,Guizhou Province was collected by using the unmanned aerial vehicle(UAV)with Sony 35 mm F1.8 Lens sensor as the remote sensing data acquisition platform.Because the original image is relatively dark and the details may be lost,the Retinex algorithm is used to enhance the image to make the details more prominent and provide support for feature point extraction.(2)Image registration.In this paper,Sift Algorithm is used for UAV remote sensing image registration.Due to the influence of Pixel gradient variation,SIFT description operator relies too much on the gradient direction of neighboring pixels,which leads to the inaccurate principal direction and the inaccurate feature points.Based on the SIFT feature extraction problem,this paper uses least squares to extract feature points to fit the better transformation Matrix,which improves the time by 688 ms and the accuracy by 5.6%.The Sift Algorithm only considers the descriptors formed by the neighboring pixels of the feature points,but not the whole structure of the image,which leads to the mismatching of many feature point pairs.In order to solve the SIFT matching problem,this paper uses the k-d tree BBF Algorithm to match feature point pairs,which improves the matching time by 2160 ms and the accuracy by6.5%.RANSAC algorithm is used to eliminate the false matching points to achieve the goal of purification.Finally,the accurate matching of remote sensing image feature points is realized.(3)Image fusion.The image after registration is spliced to form a panoramic image.This article uses Poisson Blendingfor image mosaics.After Poisson Blending,the overlapping parts of the images are smoothly transferred,and the final panoramic image is good at the visual level and achieves the desired goal.