计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (21): 278-286.DOI: 10.3778/j.issn.1002-8331.2006-0052

• 工程与应用 • 上一篇    下一篇

电力巡线无人机航拍图像匹配算法研究

马耀名,陈艺琳,李万禹   

  1. 辽宁工程技术大学 电气与控制工程学院,辽宁 葫芦岛 125105
  • 出版日期:2021-11-01 发布日期:2021-11-04

Research on Aerial Image Matching Algorithm of Power Patrol UAS

MA Yaoming, CHEN Yilin, LI Wanyu   

  1. Faculty of Electrical and Control Engineering, Liaoning Technical University, Huludao, Liaoning 125105, China
  • Online:2021-11-01 Published:2021-11-04

摘要:

基于局部特征的图像匹配算法是电力巡线无人机航拍图像匹配算法中最为实用的一种方法。针对传统匹配算法构建尺度空间会致使图像边缘信息丢失或者效率较低等问题,提出一种基于高斯曲率尺度空间的航拍图像匹配算法。借助高斯曲率滤波器构建一阶尺度空间,利用FAST算法提取特征点并选择特征采样区域,再以对特征采样区域建立二阶尺度空间并提取二阶尺度空间层内LIOP描述符,随后二阶尺度空间两两层LIOP描述符做差值并二值化处理,累加二值化值得到ASV-LIOP描述符完成匹配。在航拍图像上,使用SIFT、ORB、KAZE、AKAZE、改进KAZE等算法与所提对比实验,实验表明,所提算法正确匹配率平均提高5%左右,匹配效率约降低50%,可应用对稳定性要求较高且实时性较低的场景。

关键词: 电力巡线无人机, 高斯曲率尺度空间, ASV-LIOP描述符

Abstract:

Image matching algorithm based on local features is the most practical method in aerial image matching algorithm for power patrol UAV. An aerial image matching algorithm based on Gaussian curvature scale space is proposed to solve the problem that the traditional matching algorithm can cause the loss of image edge information or low efficiency. The first order scale space is constructed with the help of Gaussian curvature filter, then the feature points are extracted by FAST algorithm and the feature sampling region is selected. The second order scale space is established for the feature sampling region and the descriptors within LIOP second order scale space layer are extracted. Then the second order scale space pairwise LIOP descriptors are differentiated and binarized. Finally, the cumulative binarization is worth matching ASV-LIOP descriptors. On aerial images, algorithms such as SIFT, ORB, KAZE, AKAZE, and improved KAZE are used to compare with the proposed experiments. Experiments show that the accuracy of the proposed algorithm increases by 5% on average and the matching efficiency decreases by 50%.

Key words: power patrol UAV, Gaussian curvature scale space, ASV-LIOP descriptors