测绘学报 ›› 2021, Vol. 50 ›› Issue (5): 641-651.doi: 10.11947/j.AGCS.2021.20200506

• 摄影测量学与遥感 • 上一篇    下一篇

多分支网络联合的倾斜影像仿射不变特征匹配

张传辉1, 姚国标1, 张力2, 艾海滨2, 满孝成1, 黄鹏飞1   

  1. 1. 山东建筑大学测绘地理信息学院, 山东 济南 250100;
    2. 中国测绘科学研究院, 北京 100830
  • 收稿日期:2020-10-15 修回日期:2021-02-10 发布日期:2021-06-03
  • 通讯作者: 姚国标 E-mail:yao7837005@163.com
  • 作者简介:张传辉(1997-),男,硕士生,研究方向为深度学习影像匹配。E-mail:dplzch2020@163.com
  • 基金资助:
    国家自然科学基金(41601489);山东省自然科学基金(ZR2015DQ007;ZR2020MD025);山东省高等学校青创人才引育计划(0031802)

Affine invariant feature matching of oblique images based on multi-branch network

ZHANG Chuanhui1, YAO Guobiao1, ZHANG Li2, AI Haibin2, MAN Xiaocheng1, Huang Pengfei1   

  1. 1. College of Surveying and Geo-Informatics, Shandong Jianzhu University, Jinan 250100, China;
    2. Chinese Academy of Surveying and Mapping, Beijing 100830, China
  • Received:2020-10-15 Revised:2021-02-10 Published:2021-06-03
  • Supported by:
    The National Natural Science Foundation of China (No. 41601489);The National Natural Science Foundation of Shandong Province (Nos. ZR2015DQ007;ZR2020MD025);The Talent Introduction Plan for Youth Innovation Team in Universities of Shandong Province (No. 0031802)

摘要: 倾斜立体影像间存在较大仿射变形甚至透视畸变,导致现有的宽基线立体影像匹配算法容易失效或仅获得稀少匹配。基于此,本文提出一种多分支卷积网络联合的倾斜立体影像仿射不变特征匹配算法。算法首先使用Hessian算子提取初始特征邻域,接着引入三分支网络(TN)模型,并通过深度学习策略获得仿射不变特征邻域。为提高非同名相似特征的匹配效果,提出多边约束损失函数并训练多分支描述符网络(MDN)模型,继而得到具有较高区分度的描述符。然后采用最/次近邻距离比率(NNDR)测度获得同名特征,并通过随机采样一致性(RANSAC)算法剔除可能的误匹配点。最后,多组低空无人机倾斜立体影像试验结果验证了算法的有效性。

关键词: 倾斜立体影像, 卷积神经网络, 仿射不变特征, 深度学习, 影像匹配

Abstract: The available wide-baseline image matching algorithms have been prone to failure or only producing few matches, due to the complex affine deformation and perspective distortion. On this basis, we proposed a novel affine invariant feature matching algorithm for oblique stereo images based on multivariate network. In our method, we applied the Hessian algorithm to extract initial feature regions, then we constructed triplet network (TN) model, and obtained affine invariant feature regions through deep learning. To improve the matching performance of similar features, we proposed multilateral constraint loss function to train multi-branch descriptor network (MDN) model, and then generated deep learning descriptors with higher discrimination for image features. Afterwards, the conjugate features were produced by the matching metric of nearest/next distance ratio (NNDR), and eliminated possible mismatch points through random sampling consistency (RANSAC) algorithm. Finally, experiments on oblique stereo images acquired by unmanned aerial vehicle verified the effectiveness of the proposed approach.

Key words: oblique stereo images, convolutional neural network, affine invariant feature, deep learning, image matching

中图分类号: