摘要
针对运动网格统计(GMS)等特征匹配算法误匹配集中出现,且在视点变化较大的情况下,由于匹配点过少而导致匹配失败的问题,提出了一种使用改进GMS算法结合RANSAC计算单应矩阵的特征匹配算法。首先,对原始GMS算法网格进行分配权重,设置可变阈值以获取足够匹配点,通过对改进GMS算法得到的匹配点集使用RANSAC拟合单应矩阵,并对初始暴力匹配集进行细筛选生成最终的匹配点集。在公开的特征匹配数据集上的实验表明,匹配准确率和召回率分别提升了22.69%,18.58%,算法更适用于视点变化较大的场景。
Aiming at the problem that feature matching algorithms,such as moving grid statistics(GMS),have concentrated mismatching,and the matching fails due to too few matching points when the viewpoint changes greatly,a feature matching algorithm using improved GMS algorithm combined with RANSAC to calculate homography matrix is proposed.Firstly,the original GMS algorithm grid is assigned weights,and the variable threshold is improved to obtain enough matching points.The matching point set obtained by the improved GMS algorithm uses RANSAC to fit the homography matrix,and the Brute Force Matching(BFM)set is finely selected to generate the final matching point set.Experiments on public feature matching datasets show that the matching precision and recall are improved by 22.69% and 18.58% respectively,and the algorithm is more suitable for scenes with large viewpoint changing.
作者
朱世宇
陈志华
ZHU Shiyu;CHEN Zhihua(National Key Laboratory of Transient Physics Nanjing University of Science and Technology,Nanjing 210000,China)
出处
《电光与控制》
CSCD
北大核心
2023年第7期51-56,共6页
Electronics Optics & Control
