一种优化的泥石流无人机影像匹配算法  

An optimized UAV image matching algorithm in debris flow areas

作　　者：宗慧琳 袁希平[1,3] 甘淑 赵振峰[1] 张晓伦 ZONG Huilin;YUAN Xiping;GAN Shu;ZHAO Zhenfeng;ZHANG Xiaolun(Faculty of Land Resource Engineering,Kunming University of Science and Technology,Kunming 650093,China;Plateau Mountain Spatial Information Survey Technique Application Engineering Research Center at Yunnan Provinces Universities,Kunming 650093,China;College of Geosciences and Engineering,West Yunnan University of Applied Sciences,Dali,Yunnan 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]滇西应用技术大学地球科学与工程学院,云南大理671006

出　　处：《测绘科学》2022年第11期104-112,共9页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(41861054)

摘　　要：针对泥石流区无人机影像在采用AKAZE算法匹配时效果好但计算效率较低的问题,该文提出一种AKAZE特征与BEBLID描述符相结合并融合网格运动统计策略的特征匹配算法。先使用AKAZE算法构建非线性尺度空间来检测局部稳定特征,构建BEBLID描述符对检测到的特征点进行二值描述,接着用暴力匹配法完成初步匹配,再用网格运动统计方法进行第一次粗差剔除,最后加入核线约束,进一步提纯内点,得到可靠匹配结果。将文中算法与经典SIFT算法、ORB算法、AKAZE算法、AKAZE-LATCH算法进行实验对比分析。结果表明:所提方法匹配正确率与AKAZE算法基本接近,匹配速度却提升了40%左右,与ORB算法的速度接近,正确匹配点对数量相较几种对比方法都明显提高,且匹配点空间分布分散均匀,具有明显的优势和实用性。This paper proposes a feature matching algorithm that combines AKAZE features with BEBLID descriptors and incorporates a grid-based motion statistics(GMS)strategy in order to address the issue that the AKAZE algorithm has a good effect but low computational efficiency when matching UAV images in debris flow areas.Firstly,utilizing the AKAZE algorithm to detect local stability characteristics by constructing nonlinear scale space and the BEBLID descriptor to describe the detected feature points first,and then the grid-based motion statistics(GMS)method is used to correct gross errors.Finally,the epipolar constraint is added to further purify the interior point,and reliable matching results are obtained.Results of comparing the proposed algorithm to the traditional SIFT algorithm,the ORB algorithm,the AKAZE algorithm,and the Akaze-Latch algorithm reveal that:The matching accuracy of the proposed method is essentially similar to that of the AKAZE algorithm,but the matching speed is increased by about 40%,which is similar to that of the ORB algorithm.In comparison to other comparison methods,the suggested algorithm considerably increases the number of correct matching point pairs,and the matching points are spread equally in space,which has apparent benefits and is doable.

关 键 词：无人机影像 特征提取 BEBLID描述符 网格运动统计 影像匹配 

分 类 号：P237[天文地球—摄影测量与遥感]