测绘学报 ›› 2016, Vol. 45 ›› Issue (10): 1171-1181.doi: 10.11947/j.AGCS.2016.20160068

• 摄影测量学与遥感 • 上一篇    下一篇

利用网络图进行高分辨率航空多视影像密集匹配

闫利, 费亮, 陈长海, 叶志云, 朱睿希   

  1. 武汉大学测绘学院, 湖北 武汉 430079
  • 收稿日期:2016-02-24 修回日期:2016-06-21 出版日期:2016-10-20 发布日期:2016-11-08
  • 通讯作者: 费亮 E-mail:lfei@whu.edu.cn
  • 作者简介:闫利(1966-),男,博士,教授,研究方向为摄影测量与遥感。E-mail:lyan@sgg.whu.edu.cn.
  • 基金资助:

    测绘地理信息公益性行业科研专项(201512008);中央高校基本科研业务费专项资金(2015214020201)

A Multi-view Dense Matching Algorithm of High Resolution Aerial Images Based on Graph Network

YAN Li, FEI Liang, CHEN Changhai, YE Zhiyun, ZHU Ruixi   

  1. School of Geodesy and Geomatics, Wuhan University, Wuhan 430079, China
  • Received:2016-02-24 Revised:2016-06-21 Online:2016-10-20 Published:2016-11-08
  • Supported by:

    The National Public Welfare Foundation for Surveying, Mapping and Geoinformation Industry Research (No.201512008);The Fundamental Research Funds for the Central Universities (No.2015214020201)

摘要:

提出了一种基于网络图的高分辨率航空多视影像密集匹配算法。首先利用影像间的重叠关系和方向确定候选立体像对并构建网络图;在立体像对密集匹配阶段,引入导向中值滤波采用由粗到精的改进半全局匹配(SGM)算法进行双向视差图生成;最后基于所有立体像对构建的网络图完成多视影像间的密集点云生成及融合。试验选取了ISPRS的Vaihingen航空影像和ISPRS/EuroSDR项目的苏黎世倾斜下视影像进行试验,结果表明:本文算法对高分辨率多视影像密集匹配是有效可行的,无论在匹配完整性、效率、精度上都能获取较好的结果,重建的密集点云平均反投影误差的中误差可以达到亚像素级精度,实际精度可以达到1.5倍GSD,并且在建筑物、植被、水体等视差不连续、弱纹理或重复纹理区域也取得了较好的匹配结果。

关键词: 高分辨率, 多视影像, 密集匹配, 半全局匹配, 网络图

Abstract:

A multi-view dense matching algorithm of high resolution aerial images based on graph network was presented. Overlap ratio and direction between adjacent images was used to find the candidate stereo pairs and build the graph network, then a Coarse-to-Fine strategy based on modified semi-global matching algorithm (SGM) was adapted to calculate the disparity map of stereo pairs. Finally, dense point cloud was generated and fused using a multi-triangulation method based on graph network. In the experiment part, the Vaihingen aerial image dataset and the oblique nadir image block of Zürich in ISPRS/EuroSDR project were used to test the algorithm above. Experiment results show that out method is effective for multi-view dense matching of high resolution aerial images in consideration of completeness, efficiency and precision, while the RMS of average reprojection error can reach subpixel level and the actual deviation is better than 1.5 GSD. Due to the introduction of guided median filter, regions of sharp discontinuities, weak textureness or repeat textureness like buildings, vegetation and water body can also be matched well.

Key words: high resolution, multi-view, dense matching, SGM, graph network

中图分类号: