摘要
无人机图像纹理丰富、特征显著,在机器视觉三维重建及机器人导航中应用广泛,但其视角变化大,且易倾斜。传统的尺度不变特征变换(SIFT)算法和Affine SIFT(ASIFT)算法等图像特征匹配算法误差较大,难以满足应用要求。针对该问题,提出了一种基于ASIFT的改进算法。首先用ASIFT算法模拟图形畸变,然后利用SIFT算法中的k d树算法对最邻近特征点进行快速搜索匹配,最后加入随机抽样一致算法,得到匹配对的参数模型,同时对不符合模型的误差匹配对进行剔除。实验结果表明,该算法可以优化匹配效果,提高匹配速度。
The texture and feature in UAV images are rich,which are widely used in machine vision,3D reconstruction,and robot navigation,but their perspectives vary greatly and are easy to tilt.This paper addresses the problem of matching UAV images.Solutions to this problem are often extracted and registered by SIFT and Affine-SIFT methods,which result in large matching errors.To address this issue,an improved algorithm based on ASIFT is proposed.Specifically,the ASIFT algorithm is used to simulate the distortion of the image,and then utilizes the k d tree algorithm of SIFT algorithm to quickly search and match the nearest neighbor feature points.Finally,RANSAC algorithm is joined to obtain the parameter model of the matching pair,and the error matching pair for the non-conformity model is eliminated.The experiment results clearly demonstrate that the proposed method can effectively optimize the matching effect and improve the matching speed.
作者
孙东阁
陈辉
SUN Dongge;CHEN Hui(School of Automation Engineering,Shanghai University of Electric Power,Shanghai 200090,China)
出处
《上海电力大学学报》
CAS
2020年第3期275-279,共5页
Journal of Shanghai University of Electric Power
基金
国家自然科学基金(51705304)
上海市自然科学基金(16ZR1413400)。