作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 图形图像处理 • 上一篇    下一篇

基于HCA与KAZE的铁路路基GPR图像配准算法

杜翠 1,2,张千里 1,2,刘杰 1,2   

  1. (1.中国铁道科学研究院 铁道建筑研究所,北京 100081; 2.高速铁路轨道技术国家重点实验室,北京 100081)
  • 收稿日期:2016-12-06 出版日期:2018-03-15 发布日期:2018-03-15
  • 作者简介:杜翠(1988—),女,助理研究员、博士后,主研方向为探地雷达技术;张千里,研究员;刘杰,副研究员。
  • 基金资助:
    中国铁路总公司科研开发计划项目(2016G005-B);中国铁道科学研究院基金(2015YJ036)。

Railway Ballast GPR Image Registration Algorithm Based on HCA and KAZE

DU Cui 1,2,ZHANG Qianli 1,2,LIU Jie 1,2   

  1. (1.Railway Engineering Research Institute,China Academy of Railway Sciences,Beijing 100081,China; 2.State Key Laboratory for Track Technology of High-speed Railway,Beijing 100081,China)
  • Received:2016-12-06 Online:2018-03-15 Published:2018-03-15

摘要: 针对铁路路基探地雷达(GPR)检测需要较高精度与时效性的要求,提出一种基于直方图曲率分析(HCA)与KAZE特征的图像配准算法。通过HCA进行阈值分割,提取图像高能量区域,以节约无效区域的配准时间。运用KAZE算法提取图像中的特征点,并根据快速近似最近邻搜索算法进行粗匹配。使用随机抽样一致性算法过滤匹配点对,优化特征匹配过程。实验结果表明,该算法对于存在病害差异、增益差异、地物差异的铁路路基多时相GPR图像均取得较好的配准效果,且配准精度比KAZE算法、ORB算法、SIFT算法明显提高,配准效率比KAZE算法提升8%以上。

关键词: 铁路路基, 探地雷达, 图像配准, 直方图曲率分析, KAZE算法

Abstract: Aiming at the problem that railway ballast Ground Penetrating Radar(GPR) detection need high timeliness and accuracy,an image registration algorithm based on Histogram Curvature Analysis(HCA) and KAZE features is proposed.It extracts high energy region of image using HCA threshold segmentation,in order to save time for registration useless region.KAZE features in GPR images are extracted and features matching is optimized through Fast Library for Approximate Nearest Neighbors(FLANN) algorithm and Random Sample Consensus(RANSAC) algorithm.Experimental results show that the proposed algorithm can get good registration results for GPR images including disease,grayscale and structure difference.And it gains higher accuracy than original KAZE algorithm,ORB algorithm and SIFT algorithm and more than 8% higher efficiency than original KAZE algorithm.

Key words: railway ballast, Ground Penetrating Radar(GPR), image registration, Histogram Curvature Analysis(HCA), KAZE algorithm

中图分类号: