测绘学报 ›› 2018, Vol. 47 ›› Issue (6): 790-798.doi: 10.11947/j.AGCS.2018.20170630

• 高精度高效率数字摄影测量 • 上一篇    下一篇

基于特征尺度分布与对极几何约束的高清影像快速密集匹配方法

赵红蕊, 陆胜寒   

  1. 清华大学3S中心土木系地球空间信息研究所, 北京 100084
  • 收稿日期:2017-12-09 修回日期:2018-03-19 出版日期:2018-06-20 发布日期:2018-06-21
  • 作者简介:赵红蕊(1969-),女,博士,教授,研究方向为遥感与摄影测量。E-mail:zhr@mail.tsinghua.edu.cn
  • 基金资助:
    国家自然科学基金(41571414)

Dense High-definition Image Matching Strategy Based on Scale Distribution of Feature and Geometric Constraint

ZHAO Hongrui, LU Shenghan   

  1. Institute of Geomatics, Department of Civil Engineering, 3S Centre, Tsinghua University, Beijing 100084, China
  • Received:2017-12-09 Revised:2018-03-19 Online:2018-06-20 Published:2018-06-21
  • Supported by:
    The National Natural Science Foundation of China (No.41571414)

摘要: 本文重点阐述了基于机器视觉的智能摄影测量的效率基础问题之三:高清影像快速智能匹配处理。图像特征匹配是影响数字摄影测量坐标空间计算效率的基础数据处理过程。为了解决高分辨率数据匹配校验计算成本更高及相似特征干扰等影像产品生成效率负面影响问题,本文通过研究影像尺度不变特征的数学本质,结合多视图相机几何模型,推导并验证了图像特征点的降采样尺度分布规律。根据图像空间几何关系在降采样尺度上的匹配映射关系,缩减图像匹配过程中的计算量并筛选有效待匹配点集,将特征点数量105量级的快速全局特征距离初匹配时长限制在亚秒级。在此基础上结合特征尺度分布信息改进的对极几何约束,改进特征匹配算法,辅助缩小匹配搜索范围,通过特征索引与分区并行处理,实现高清影像同名特征的高精度快速密集匹配,提升特征点基数、匹配特征点对数量与正确率。本文使用intel i7-4720HQ与NVIDIA GTX970M进行试验,基于尺度分布特性的特征匹配方法,以亚秒级的计算时间,获取符合多约束条件的103量级的匹配点对,为数字影像的快速高精度处理提供了一种新思路,在充分满足数字摄影测量的精度的基础上可提高其产品生成效率。

关键词: 尺度不变特征, 尺度比例约束, 对极几何, 密集匹配

Abstract: This paper mainly expounds the basic issue three about the intelligent photogrammetry based on machine vision:the intelligent and fast matching process among high-definition images.Image feature matching,as a fundamental data processing procedure,plays an important role in the computational efficiency of computing the digital photogrammetry coordinate space.There are challenges for image feature match including high computational expense due to high-resolution data and similar feature interference.Concerning these problems,the mathematical nature of the invariant feature of image scale was studied,and the geometric model of multi-view camera was used to derive and verify the scale distribution of image feature points.The information interaction process of the scale component in the feature extraction and the matching process was determined.Through the equal-scale feature matching,the calculation amount in the image matching process was reduced and the effective information was retained,which greatly reduced the time of the initial distance matching progress among 105 points in 1 second.On this basis,combining the feature scale distribution and the geometric constraint,the improved feature matching algorithm was used to reduce the matching search range under the limited time and the computational scale.Fast and dense matching achieves through the feature index and the partition parallel processing.Using intel i7-4720HQ and NVIDIA GTX970M,the experiment shows that the feature matching method based on the scale distribution feature has a great advantage in improving the speed and accuracy of automatic image matching and matches thousands of points in less than one second.It provides a new idea for the fast and high precision processing of digital images,which can not only meet the accuracy of digital photogrammetry but greatly improve the efficiency of production.

Key words: scale invariant feature, scale constraint, epipolar geometry, dense matching

中图分类号: