地球信息科学学报 ›› 2019, Vol. 21 ›› Issue (4): 588-599.doi: 10.12082/dqxxkx.2019.190001

• 论文 • 上一篇    下一篇

基于Dense SIFT特征的无人机影像快速拼接方法

杨佳宾1(), 姜永涛1, 杨幸彬2, 郭广猛1   

  1. 1. 南阳师范学院 河南省自然灾害遥感监测重点实验室,南阳 473061
    2. 北京建筑大学 测绘与城市空间信息学院,北京 100044
  • 收稿日期:2019-01-01 修回日期:2019-03-14 出版日期:2019-04-24 发布日期:2019-04-24
  • 作者简介:

    作者简介:杨佳宾(1995-),男,河南郑州人,本科生,主要从事数字摄影测量与计算机视觉方向研究。 E-mail:yangbin187513@163.com

  • 基金资助:
    国家自然科学基金项目(41604009、41071280);南阳师范学院青年项目(18060)

A Fast Mosaic Algorithm of UAV Images based on Dense SIFT Feature Matching

Jiabin YANG1,*(), Yongtao JIANG1, Xingbin YANG2, Guangmeng GUO1   

  1. 1. Key laboratory of Natural disaster and remote sensing of Henan province, Nanyang Normal University, Nanyang 473061, China
    2. College of Surveying and Urban Spatial Information, Beijing University of Civil Engineering and Architecture, Beijing 100044, China
  • Received:2019-01-01 Revised:2019-03-14 Online:2019-04-24 Published:2019-04-24
  • Contact: Jiabin YANG E-mail:yangbin187513@163.com
  • Supported by:
    National Natural Science Foundation of China, No.41604009, 41071280;Nanyang Normal University Youth Project, No.18060

摘要:

特征匹配是无人机影像拼接过程的关键步骤,针对传统的特征匹配方法在影像拼接过程中获取匹配点少、特征点分布不均匀、匹配耗时长等问题,本文提出一种基于Dense SIFT特征的无人机影像快速拼接算法。首先,利用影像POS信息构建连接矩阵以引导匹配过程;然后在降采样影像上进行影像分块,利用Dense SIFT算子获取初始匹配点,并采用两次NCC方法分别实现降采样影像和原始影像上匹配点的精化;最后,基于共线方程将影像投影至物方面上,完成影像的快速拼接。本文选取2组无人机影像进行拼接实验,将本文算法与SIFT和SURF匹配拼接方法进行对比,结果表明:在影像特征点匹配方面,本文方法获取匹配点数量是SIFT和SURF算法的5倍以上,且匹配点分布更加均匀;在影像拼接结果方面,本文方法不仅能够较快完成影像拼接,而且有效避免了拼接影像中的“重影”现象,保证了较好的拼接质量。

关键词: 影像拼接, Dense SIFT, 连接矩阵, NCC, 共线方程, 光束法平差, 无人机

Abstract:

The UAV (Unmanned Aerial Vehicle) photography is a new remote sensing system emerging in recent years. It plays an important role in the rapid emergency response of natural disasters. However, due to the large amount of UAV image data, the traditional method for image matching and mosaic is low accuracy and time-consuming. Feature matching is one of key steps in UAV image mosaic. Traditional matching algorithms have several problems, including less feature points, feature maldistribution, and time-consuming. To solve these problems, a fast image mosaic algorithm based on Dense SIFT feature matching is proposed. Firstly, the connection matrix is build based on POS (Position and Orientation System) data to conduct the matching process. The UAV images are then down-sampled. Secondly, image segmentation is performed on the down-sampled images. Then the Dense SIFT operator is used in overlap area of down-sampled images to obtain the initial matching points which are eliminated through matching by the RANSAC (Random Sample Consensus) algorithm and refined by the NCC algorithm on the original and down-sampled images, respectively. Finally, processed images are projected to the object coordinate system based on collinear equation which is calculated by the bundle adjustment method. By contrast, the SIFT (Scale-Invariant Feature Transform) and SURF (Speeded Up Robust Feature) algorithms and the Pix4Dmapper Photogrammetry software are used to test the quality and efficiency of the Dense SIFT algorithm. Two groups of UAV images mosaic experiment results indicate: (1) The Dense SIFT algorithm can be used to obtain about five times more evenly distributed matching points than the SIFT and SURF algorithm at the same time; (2) The Dense SIFT algorithm can be used to effectively improve the quality of the images mosaic by removing the phenomenon of ghosting; (3) It takes about half the time of Pix4D mapper software to complete the same image mosaic test using the Dense SIFT algorithm. This indicates that the presented algorithm has a high image mosaic quality and fast processing speed, which can play an important role in the rapid emergency response of natural disasters.

Key words: UAV image mosaic, Dense SIFT feature matching, connection matrix, NCC algorithm, collinear equation, bundle adjustment