一种基于改进ORB的无人机影像特征匹配方法  

A method of UAV image feature matching based on improved ORB

作　　者：唐迪 甘淑[1,2] 袁希平 TANG Di;GAN Shu;YUAN Xiping(Faculty of Land and Resources Engineering,Kunming University of Science and Technology,Kunming 650093,China;Plateau Mountain Spatial Information Survey Technique Application Engineering Research Center at Yunnan Provinces Universityies,Kunming 650093,China;West Yunnan University of Applied Sciences,Dali 671006,China)

机构地区：[1]昆明理工大学国土资源工程学院,昆明650093 [2]云南省高校高原山地空间信息测绘技术应用工程研究中心,昆明650093 [3]滇西应用技术大学,云南大理671006

出　　处：《测绘工程》2024年第6期33-40,共8页Engineering of Surveying and Mapping

基　　金：国家自然科学基金资助项目(62266026)。

摘　　要：针对高原山区区域的无人机影像受到地形起伏,纹理信息不突出的影响,导致ORB算法的匹配正确率偏低且不稳定的问题,文中提出适应高原山区区域的无人机影像特征匹配的改进算法。该方法集成了ORB算法、小波影像金字塔、BRISK关键点描述符、GMS算法和随机抽样一致算法(RANSAC)模型约束改进的方法。首先将ORB算法中的高斯金字塔替换为小波影像金字塔并进行特征点检测;然后,结合BRISK描述符生成适合影像特征的二进制描述符;其次使用暴力匹配算法和GMS算法,对特征点进行第一次约束优化提取;最后利用RANSAC算法对匹配点对进行第二次约束优化提取。试验结果表明,文中算法在高原山区区域无人机影像特征匹配的实验中,匹配正确率和匹配耗时均有着显著的优势。相较于传统ORB算法,文中算法的匹配正确率和匹配耗时有较大提升。The UAV image in the plateau mountainous area is affected by terrain fluctuation and the texture information is not prominent,the matching accuracy of ORB algorithm is low and unstable.This paper presents an improved algorithm for UAV image feature matching in the mountainous region of the plateau.This method integrates the ORB algorithm,wavelet image pyramid,BRISK key point descriptor,GMS algorithm and random sampling consensus(RANSAC)model constraint improvement methods.The specific process of the method is as follows:firstly,Gaussian pyramid in ORB algorithm is replaced by wavelet image pyramid and feature point detection is carried out.Then we combine BRISK to generate binary descriptors suitable for image features.Secondly,Brute-Force Matching algorithm and GMS algorithm are used to extract the feature points for the first time.Finally,the RANSAC algorithm is used to extract the second constraint optimization of the matching points.The experimental results show that the algorithm in this paper has significant advantages in matching accuracy and running time in the experiment of UAV image feature matching in the plateau mountainous area.Compared with the traditional ORB algorithm,the matching accuracy and running time of the proposed algorithm have been greatly improved.

关 键 词：无人机影像 特征提取 小波图像金字塔 二进制鲁棒不变特征点描述算法 基于网络的运动统计算法 

分 类 号：P237[天文地球—摄影测量与遥感]