地球信息科学学报 ›› 2017, Vol. 19 ›› Issue (4): 528-539.doi: 10.3724/SP.J.1047.2017.00528

• 遥感科学与应用技术 • 上一篇    下一篇

激光点云与光学影像配准:现状与趋势

张靖(), 江万寿*()   

  1. 武汉大学 测绘遥感信息工程国家重点实验室,武汉 430079
  • 收稿日期:2016-05-30 修回日期:2016-12-20 出版日期:2017-04-20 发布日期:2017-04-20
  • 通讯作者: 江万寿 E-mail:jing.zhang@whu.edu.cn;jws@whu.edu.cn
  • 作者简介:

    作者简介:张 靖(1982-),男,博士,讲师,研究方向为点云数据处理。E-mail:jing.zhang@whu.edu.cn

  • 基金资助:
    国家自然科学基金青年科学基金项目(41301367);高等学校博士点专项科研基金(20130141120066);地理空间信息工程国家测绘地理信息局重点实验室开放研究基金项目(201309)

Registration between Laser Scanning Point Cloud and Optical Images: Status and Trends

ZHANG Jing(), JIANG Wanshou*()   

  1. State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China
  • Received:2016-05-30 Revised:2016-12-20 Online:2017-04-20 Published:2017-04-20
  • Contact: JIANG Wanshou E-mail:jing.zhang@whu.edu.cn;jws@whu.edu.cn

摘要:

激光点云与光学影像是2种重要的遥感数据源,二者的融合能够实现优势互补,具有应用价值。点云与影像的配准是实现二者集成应用的基础,虽然经历了多年的发展仍存在许多问题有待解决。本文首先通过建立点云与影像配准问题的数学范式,将整个配准问题划分为观测值提取、配准模型选择和参数优化3部分,深入分析各部分所面临的难点与挑战;然后对现有的点云与影像配准方法进行回顾与总结,对比分析各类方法的优缺点及适用范围;最后展望了今后的发展方向进行了展望,为后续的研究提供参考。

关键词: 光学影像, 激光探测与测距, 点云, 配准, 综述

Abstract:

LiDAR point cloud and optical imagery are different types of remote sensing data source. They have some unique merits, respectively, that are complementary to each other. Integrating these two dataset has significant value in many applications. However, as the existence of various error sources, point cloud and optical imagery are usually misaligned. For the purpose of further integrated processing, the registeration of point cloud and imagery is a preliminary step which will align them into a unified geo-reference frame. Although after decades of research, this registration problem is far from solved. This paper gave a detailed survey of registration between point cloud and optical images. To obtain thorough understanding of this problem, a general mathematical paradigm for the registration was established firstly. By analyzing the mathematical paradigm, we indicated three main difficulties in this registration problem, and then definitely divided the whole workflow of registration into three key parts which are named: (1) the acquisition of corresponding observations, (2) the selection of transformation models; (3) the optimization of unknowns. Afterwards, we reviewed a series of representative registration methods from the above three aspects. In the acquisition of corresponding observations, the existing methods were classified into area-based method, feature-based method and multiple-view geometry based method. In the stage of transformation models selection, frequently-used models were classified into sensor-based models and empirical models. In the unknowns’ optimization part, two principal optimization methods termed local optimization and global optimization were introduced and the general usages of these optimization in registration were described. Furthermore, we summarized the mentioned registration methods and gave a detailed comparison and analysis including the advantages / shortcomings and the applicable scope. At last, the trends of registration development were forecasted.

Key words: optical image, LiDAR, point cloud, registration, literature review