无人机热红外影像快速拼接的改进SFM算法研究  

Research on Improved SFM Algorithm for Rapid Stitching of UAV Thermal Infrared Images

作　　者：胡露太 李英冰[1] 张涛[1] 张沛 HU Lutai;LI Yingbing;ZHANG Tao;ZHANG Pei(School of Geodesy and Geomatics,Wuhan University,Wuhan 430079,China)

机构地区：[1]武汉大学测绘学院,湖北武汉430079

出　　处：《测绘与空间地理信息》2024年第11期33-36,40,共5页Geomatics & Spatial Information Technology

基　　金：国家重点研发计划项目——山区和边远灾区应急供水与净水一体化装备(2020YFC1512401)资助。

摘　　要：无人机热红外成像技术在场景监控、物体识别等领域应用广泛,但是该影像具有分辨率低、倾角大、画幅小等缺点,因此无人机热红外影像拼接比较困难。本文提出了一种改进SFM(运动恢复结构)的无人机热红外影像的快速拼接算法(FastSFM)。FastSFM利用无人机热红外影像序列具有的时空顺序性,采用串联式影像匹配代替整体影像两两匹配,再根据稀疏矩阵的特性对Levenberg-Marquardt光束法平差方法进行优化,利用线性公式代替矩阵运算,从而提高拼接效率。通过实地采集影像数据进行了SFM和FastSFM拼接对比实验,20张热红外影像拼接时间从521 s提高到39 s,50张红外影像拼接时间从3 102 s提高到58 s,且拼接精度也有一定提高。结果表明,FastSFM能够实现无人机热红外影像的快速拼接。UAV thermal infrared imaging technology is widely used in scene monitoring,object recognition and other fields,but the image has shortcomings such as low resolution,large inclination,and small frame size,so it is difficult to stitch UAV thermal infrared images.In this paper,a fast stitching algorithm(Fast-SFM) for UAV thermal infrared images with improved SFM(Structure From Motion) is proposed.We take the advantages of the spatiotemporal sequence of UAV thermal infrared image sequences,uses serial image matching instead of pairwise matching of overall images,then optimizes the Levenberg-Marquardt bundle adjustment method according to the characteristics of sparse matrices,uses linear formulas instead of matrices operation,and thereby improves the stitching efficiency.We conducted a comparison experiment of SFM and Fast-SFM stitching by collecting image data in the field.The stitching time of 20 thermal infrared images was increased from 521 s to 39 s,and the stitching time of 50 infrared images was increased from 3 102 s to 58 s.The results show that Fast-SFM can realize rapid stitching of UAV thermal infrared images.

关 键 词：无人机热红外影像 影像拼接 运动恢复结构 影像匹配 光束法平差 

分 类 号：P231[天文地球—摄影测量与遥感]