基于改进SIFT算法的无人机遥感影像快速拼接  

Quick stitching of unmanned aerial vehicle remote sensing images based on an improved SIFT algorithm

作　　者：王超[1,2] 雷添杰 张保山 徐瑞瑞 陈东攀 WANG Chao;LEI Tianjie;ZHAGN Baoshan;XU Ruirui;CHEN Dongpan(Institute of Geographical Sciences of Henan Academy of Sciences,Zhengzhou 450052,China;Collaborative Innovation Center of Geographic technology of Wisdom Central Plains,Zhengzhou 450052,China;Institute of Environment and Sustainable Development in Agricultural,Chinese Academy of Agricultural Sciences,Beijing 100081,China;Water Resources Bureau in Minquan County of Henan Province,Shangqiu 476800,Henan,China;Beijing University of Technology,School of Artificial Intelligence and Automation,Beijing 100124,China)

机构地区：[1]河南省科学院地理研究所,郑州450052 [2]智慧中原地理信息技术协同创新中心,郑州450052 [3]中国农业科学院农业环境与可持续发展研究所,北京100081 [4]河南省民权县水利局,河南商丘476800 [5]北京工业大学人工智能与自动化学院,北京100124

出　　处：《华中师范大学学报（自然科学版）》2023年第2期302-309,共8页Journal of Central China Normal University：Natural Sciences

基　　金：河南省重点研发与推广专项项目(212102310424);河南省科学院重大科研聚焦项目(210101007)。

摘　　要：近几十年来,无人机遥感在地球观测领域发展十分迅速,然而,无人机影像快速拼接是阻碍其应用的难题.针对无人机遥感影像的特点与SIFT(scale invariant feature transform)拼接算法的缺点,该文提出了一种基于随机抽样一致性算法RANSAC(random sample consensus)和最小二乘匹配改进的SIFT影像拼接算法.首先采用随机采样法RANSAC进行粗略匹配数据的提纯,剔除伪特征点对,以减少特征点数量,降低时耗;然后再以最小二乘匹配进行更加精确的匹配,最终实现了无人机影像的自动拼接.实验结果表明:基于RANSAC和最小二乘匹配改进的SIFT拼接算法的平均正确匹配率为92.8%,拼接精度由1个像元提高到0.1个像元,同时拼接运算效率也得到了较大的提升.经改进的SIFT拼接算法在海量特征数据库中可以进行快速、准确的匹配、甚至可以达到实时的要求,具有更强的鲁棒性,可以满足低空遥感影像的相对定向高度自动化的需要,应用前景广阔.Unmanned aerial vehicle(UAV)remote sensing has been developing rapidly in the field of earth observation in recent decades.However,quick stitching of UAV images has become a problem of blocks for the wide applications.In this paper,an algorithm based on both random sample consensus(RANSAC)and least-squares match method was proposed to improve the image registration performance of SIFT algorithm.On the one hand,RANSAC was able to remove inaccurate feature point pairs that SIFT detected.On the other hand,given all rough feature matches based on SIFT features,least-squares match was used to carry out precise smatching.The experiment results show that our proposal was able to effectively estimate matching error with an average correct matching rate of 92.8%.Moreover,stitching accuracy was improved from 1.0 pixel to 0.1 pixel,and the stitching efficiency was also elevated.The improved SIFT perform fast and accurate matching in massive feature database,even in real time,and has stronger robustness,which would meet the demand for highly automated relative orientation of low-altitude remote sensing images with broad application prospects.

关 键 词：无人机 遥感影像 SIFT RANSAC 最小二乘匹配 快速拼接 

分 类 号：P407.8[天文地球—大气科学及气象学]