大区域无人机图像序列全局配准方法  

Global registration method of image sequence taken by UAV in large region

作　　者：周行[1] 周东翔[1] 彭科举[1] 郭瑞斌[1] 樊玮虹[1] 

机构地区：[1]国防科学技术大学电子科学与工程学院,湖南长沙410073

出　　处：《现代电子技术》2015年第21期14-19,共6页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(61375032);国防科学技术大学预研资助项目(ZDYYJCYJ20140601)

摘　　要：由于无人机搭载的摄相机视角有限,在进行大区域航拍作业的后期数据处理中不可避免地要进行大量无人机图像的拼接。这里提出一种基于图像分布估计的多图像全局配准技术,首先通过运动估计和运动分解对图像序列的链式分布进行估计,然后通过精确SURF特征匹配对图像分布进行优化。得到的分布估计图的每一条边表征了一个图像对的匹配关系,根据这些匹配关系进行捆绑调整,求解每幅图像到最终拼接平面的最优变换矩阵,从而实现大区域多图像的全局配准。实验所得的拼接结果表明该技术配准效率高、效果良好。For the limited visual angle of the camera carried by UAV,a large number of mosaicing of images taken by UAV is inevitable in the later processing of the large region aerial photo mission. A multi-image global registration technology based on image distribution estimation is proposed. The chain distribution of the image sequence is estimated by using motion estimation and motion decomposition,and then the image distribution is optimized by precise SURF feature matching. Each edge of the obtained distribution estimation images characters the matching relation of an image pair. The binding adjustment is conducted according to the matching relation to solve the optimal transformation matrix from each image to the final mosaic plane,and achieve the multi-image global registration in the large region. The mosaic results from the experiment show that the method has high registration efficiency,and better registration result.

关 键 词：分布估计 无人机 全局配准 捆绑调整 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP391.4[自动化与计算机技术—计算机科学与技术]