测绘学报 ›› 2017, Vol. 46 ›› Issue (2): 188-197.doi: 10.11947/j.AGCS.2017.20160293

• 摄影测量学与遥感 • 上一篇    下一篇

一种快速有效的大数据区域网平差方法

郑茂腾1, 张永军2, 朱俊峰3, 熊小东3, 周顺平1   

  1. 1. 中国地质大学(武汉)国家地理信息系统工程技术研究中心, 湖北 武汉 430074;
    2. 武汉大学遥感信息工程学院, 湖北 武汉 430079;
    3. 北京中测智绘有限责任公司, 北京 100830
  • 收稿日期:2016-07-26 修回日期:2016-11-09 出版日期:2017-02-20 发布日期:2017-03-07
  • 作者简介:郑茂腾(1987-),男,博士,讲师,研究方向为航空航天摄影测量,计算机视觉。E-mail:tengve@163.com
  • 基金资助:
    国家自然科学基金(41601502;41571434);中国博士后科学基金面上项目(2015M572224);中央高校基本科研业务费专项资金(CUG160838)

A Fast and Effective Block Adjustment Method with Big Data

ZHENG Maoteng1, ZHANG Yongjun2, ZHU Junfeng3, XIONG Xiaodong3, ZHOU Shunping1   

  1. 1. National Engineering Research Center for Geographic Information System, China University of Geosciences, Wuhan 430074, China;
    2. School of Remote Sensing and Information Engineering, Wuhan University, Wuhan 430079, China;
    3. Smart Mapping Technology Inc., Beijing 100830, China
  • Received:2016-07-26 Revised:2016-11-09 Online:2017-02-20 Published:2017-03-07
  • Supported by:
    The National Natural Science Foundation of China(Nos. 41601502,41571434),The China Postdoctoral Science Foundation(No. 2015M572224),The Fundamental Research Funds for the Central Universities(No. CUG160838)

摘要: 针对摄影测量影像来源多样化、复杂化、大数据化等趋势,传统区域网平差算法在应对当前复杂多变的数据来源,矩阵排列毫无规律的法方程结构以及大数据量带来的高内存需求和低计算效率等问题上,遇到了前所未有的挑战,为了解决上述难题,本文引入了预条件共轭梯度法以及不精确牛顿解法求解区域网平差过程中的法方程,同时使用一种块状法方程系数矩阵压缩存储格式,构建了全新的区域网平差技术流程。本文方法避免了直接对法方程系数矩阵的求逆,压缩了法方程系数矩阵所需的内存空间,使得本文算法比传统算法所需计算机内存空间大幅减少,平差计算速度明显提升,同时保证了计算精度与传统方法相当。初步试验证明,本文方法对4500张影像、近900万像点数据的平差计算在普通电脑上仅需要约15 min,且计算精度达到子像素级。

关键词: 区域网平差, 预条件共轭梯度, 不精确牛顿解, 稀疏矩阵压缩, 大数据

Abstract: To deal with multi-source, complex and massive data in photogrammetry, and solve the high memory requirement and low computation efficiency of irregular normal equation caused by the randomly aligned and large scale datasets, we introduce the preconditioned conjugate gradient combined with inexact Newton method to solve the normal equation which do not have strip characteristics due to the randomly aligned images. We also use an effective sparse matrix compression format to compress the big normal matrix, a brand new workflow of bundle adjustment is developed. Our method can avoid the direct inversion of the big normal matrix, the memory requirement of the normal matrix is also decreased by the proposed sparse matrix compression format. Combining all these techniques, the proposed method can not only decrease the memory requirement of normal matrix, but also largely improve the efficiency of bundle adjustment while maintaining the same accuracy as the conventional method. Preliminary experiment results show that the bundle adjustment of a dataset with about 4500 images and 9 million image points can be done in only 15 minutes while achieving sub-pixel accuracy.

Key words: block adjustment, preconditioned conjugate gradient, inexact Newton method, sparse matrix compressing, big data

中图分类号: