基于边缘响应优化SIFT算法在无人机影像匹配中的研究  

An improved SIFT algorithm is applied to UAV image matching research

作　　者：魏休耘 甘淑[1,2] 袁希平 高莎[1] WEI Xiuyun;GAN Shu;YUAN Xiping;GAO Sha(Faculty of Land and Resources Engineering,Kunming University of Science and Technology,Kunming 650093,China;Engineering Research Center of Plateau and Mountain Spatial Information Mapping Technology,Universities of Yunnan Province,Kunming 650093,China;Yunnan Key Laboratory of Cloud Data Processing and Application of Mountain Real Scenic Spots,West Yunnan University of Applied Technology,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省高校山地实景点云数据处理及应用重点实验室,云南大理671006

出　　处：《测绘工程》2024年第6期1-10,共10页Engineering of Surveying and Mapping

基　　金：国家自然科学基金资助项目(62266026)。

摘　　要：无人机(UAV)影像匹配过程中常规的尺度不变特征变换(SIFT)算法在处理UAV影像时会受边缘响应影响,导致特征点匹配的准确度与精度降低。为了进一步提高UAV影像匹配的准确度与精度,提出了一种优化的SIFT图像匹配算法。以云南省禄丰市恐龙谷为研究区,选择两组研究区无人机影像以及一组公开无人机影像作为实验数据。首先,基于尺度不变性构建高斯差分金字塔,并进行空间极值检测实现特征点的定位。其次,使用Canny算法对图像进行边缘检测,在边缘检测的基础上对图像进行边缘响应点去除,对去除边缘响应后的特征点使用FLANN匹配器进行特征点匹配。然后,在图像匹配的基础上使用最佳近似比进行匹配点的第一次筛选,随后,利用随机抽样一致性(RANSAC)算法对匹配点进行第二次筛选,确定最佳比值阈值,并计算出RANSAC筛选后的均方根误差(RMSE)与准确率。最后,将优化SIFT算法与SIFT、ORB、SURF算法进行对比分析。结果表明,优化的SIFT算法不仅特征点匹配耗时小于其他3种算法,在匹配精度、准确度上以及物方精度上也优于其他3种算法。优化后的SIFT算法在A、B、C 3组数据匹配的精度分别是93.13%、84.09%、92.73%,RMSE值分别是0.6490、1.1805、0.7726。因此优化的SIFT算法能够降低边缘响应对传统SIFT算法带来的影响,提高了图像匹配的精度和准确度。In the UAV image matching process,the conventional Scale-Invariant Feature Transform(SIFT)algorithm is susceptible to edge response when handling UAV images,leading to decreased accuracy and precision in feature point matching.To enhance the accuracy and precision of UAV image matching,we propose an optimized SIFT image matching algorithm.The study area chosen is Dinosaur Valley in Lufeng City,Yunnan Province,utilizing two sets of UAV images and one set of public UAV images as experimental data.Initially,a Gaussian difference pyramid is constructed based on scale invariance,and spatial extremum detection is conducted to locate feature points.Subsequently,the Canny algorithm is applied to detect the image’s edges,and edge response points are eliminated based on this edge detection.The FLANN matcher is employed to match feature points after the removal of edge responses.Following image matching,the best approximation ratio is used for the initial screening of matching points.Subsequently,the Random Sampling Consistency(RANSAC)algorithm is applied for a second screening,determining the optimal ratio threshold.The Root Mean Square Error(RMSE)and accuracy rate after RANSAC screening are calculated.Finally,the optimized SIFT algorithm is compared with SIFT,ORB,and SURF algorithms.Results indicate that the optimized SIFT algorithm not only requires less time for matching feature points compared to the other three algorithms but also surpasses them in terms of matching accuracy,precision,and object-side accuracy.The accuracy of the optimized SIFT algorithm for data matching in groups A,B and C is 93.13%,84.09%and 92.73%,respectively,with corresponding RMSE values of 0.6490,1.1805 and 0.7726.Thus,the optimized SIFT algorithm effectively mitigates the impact of edge response on the traditional SIFT algorithm,enhancing the precision and accuracy of image matching.

关 键 词：图像匹配 优化SIFT算法 边缘响应 RANSAC算法 

分 类 号：P237[天文地球—摄影测量与遥感]