lsd 特征点匹配代码_论文推荐 | 陈敏：面向城区宽基线立体像对视角变化的结构自适应特征点匹配...

《测绘学报》

构建与学术的桥梁 拉近与权威的距离

面向城区宽基线立体像对视角变化的结构自适应特征点匹配

陈敏¹, 朱庆¹, 何海清², 严少华¹, 赵怡涛¹

1. 西南交通大学地球科学与环境工程学院, 四川 成都 611756; 2. 东华理工大学测绘工程学院, 江西 南昌 330013

收稿日期：2018-06-08；修回日期：2019-01-23

基金项目：国家自然科学基金(41631174;41971411;41501492);国家重点研发计划课题(2016YFB0502603;2016YFB0501403);中央高校基本科研业务费专项资金(2682017CX084;2682016CY02)

第一作者简介：陈敏(1986-), 男, 博士, 副教授, 研究方向为多源遥感影像处理与分析。E-mail:minchen@home.swjtu.edu.cn

摘要：提出一种基于结构自适应特征的城区宽基线影像特征点匹配方法。首先，对影像提取点特征和直线特征，挖掘点特征与其邻域内直线特征之间的几何关系，构建结构自适应的特征区域和特征描述符，并通过双向匹配策略获得初始匹配结果。然后，基于初匹配结果估计影像基础矩阵，构建核线约束的结构自适应特征匹配算法进行二次匹配。最后，将已匹配特征作为控制基础设计匹配扩展算法，进一步增加匹配点数量。本文方法以特征点邻域几何结构为出发点，构建自适应的特征区域，能够在显著的影像视角变化下，为同名特征点提取影像内容一致的特征区域，进而获得相似的特征描述符。试验结果证明，与传统算法相比，本文方法在城区宽基线影像上能够同时获得更多的正确匹配特征和更高的匹配正确率。

关键词：城区宽基线影像 特征点匹配 结构自适应特征区域 视角变化 匹配扩展

Structureadaptive feature point matching for urban area wide-baseline images with viewpoint variation

CHEN Min¹, ZHU Qing¹, HE Haiqing², YAN Shaohua¹, ZHAO Yitao¹

1. Faculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu 611756, China;2. School of Geomatics, East China University of Technology, Nanchang 330013, China

Foundation support: The National Natural Science Foundation of China (Nos. 41631174;41971411;41501492); The National Key Research and Development Program of China (Nos. 2016YFB0502603; 2016YFB0501403); The Fundamental Research Funds for the Central Universities (Nos. 2682017CX084; 2682016CY02)

First author: CHEN Min(1986—), male, PhD, associate professor, majors in multi-source remote sensing images processing and analysis. E-mail:minchen@home.swjtu.edu.cn.

Abstract: A feature point matching method based on structureadaptive feature is proposed for urban area wide-baseline images. Firstly, interest points and straight lines are detected from images. Structure adaptive feature region and descriptor are constructed by exploring the geometric relationship between the interest point and the straight lines located in the local neighborhood of the point, and initial matching results are obtained by using bi-directional matching strategy. Secondly, the fundamental matrix is estimated from the initial matching results. An epipolar geometry constrained structure adaptive feature matching method is proposed to match those features have not been matched in the initial matching step. Finally, a matching expansion method is proposed based on the previous matching results to improve the matching performance.The proposed matching method can generate similar feature regions and descriptors for corresponding features under significant image viewpoint variation benefiting from the proposed structure adaptive feature region construction method. The experimental results demonstrate that the proposed method provides significant improvements in correct matches number and matching precision compared with other traditional matching methods for urban area wide-baseline images (e.g. unmanned aerial vehicle images and oblique images) with viewpoint change and occlusion.

Key words: urban area wide-baseline image feature point matching structure adaptive feature region viewpoint variation matching expansion

影像匹配是遥感影像处理与应用的关键步骤之一。经过数十年的发展，研究人员提出了许多适用于不同类型影像的匹配方法^[1-5]。现有影像匹配方法大体上可分为两类：基于灰度的匹配方法和基于特征的匹配方法^[6-7]。基于灰度的匹配方法能够获得亚像素级精度，但对影像灰度变化和几何变形比较敏感^[8]。相对而言，基于特征的匹配方法能够较好地克服基于灰度的匹配方法对影像灰度变化和几何变形稳健性不足的问题。随着SIFT(scale invariant feature transform)算法^[9]的成功，基于特征的匹配方法受到了越来越多的