测绘学报 ›› 2014, Vol. 43 ›› Issue (12): 1266-1273.doi: 10.13485/j.cnki.11-2089.2014.0172

• 学术论文 • 上一篇    下一篇

光流特征聚类的车载全景序列影像匹配方法

张正鹏1,2, 江万寿2, 张靖2   

  1. 1. 辽宁工程技术大学 测绘与地理科学学院, 辽宁 阜新 123000;
    2. 武汉大学 测绘遥感信息工程国家重点实验室, 湖北 武汉 430079
  • 收稿日期:2013-04-12 修回日期:2014-07-09 出版日期:2014-12-20 发布日期:2014-12-23
  • 通讯作者: 江万寿 E-mail:jws@lmars.whu.edu.cn
  • 作者简介:张正鹏(1981-),男,讲师,博士生,研究方向为车载移动测量技术.zhangzhengpeng2004@126.com
  • 基金资助:

    国家973计划(2012CB719904);国家自然科学基金(41101452);高等学校博士学科点专项科研基金(20122121120003);测绘地理信息公益性行业科研专项(201412007)

An Image Match Method Based on Optical Flow Feature Clustering for Vehicle-borne Panoramic Image Sequence

ZHANG Zhengpeng1,2, JIANG Wanshou2, ZHANG Jing2   

  1. 1. School of Geomatics, Liaoning Technical University, Fuxin 123000, China;
    2. State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China
  • Received:2013-04-12 Revised:2014-07-09 Online:2014-12-20 Published:2014-12-23

摘要:

提出一种光流特征聚类的车载全景序列影像匹配方法.采用非参数化的均值漂移特征聚类思想,以SIFT多尺度特征匹配点的位置量和光流矢量,构建了影像特征空间的空域和值域;利用特征空间中对应的显著图像光流特征为聚类条件,实现了全景序列影像的匹配;最后以全景极线几何约束为条件进行粗差的剔除.通过相同、不同内点率以及不同数据的试验对比分析,本文方法在匹配正确点数和正确率方面要优于经典的Ransac法和金字塔Lucas-Kanade光流法,尤其在场景复杂造成的低内点率情况下,算法表现较为稳定,并可较好地剔除由重复纹理、运动物体、尺度变化等产生的匹配点粗差.

关键词: 车载全景影像, 光流聚类, 均值漂移, SIFT特征

Abstract:

An image match method based on optical flow feature clustering is presented for vehicle-borne panoramic image sequence. The spatial domain and range domain of image feature space are built by the coordinates of SIFT multi-scale feature matching point and optical flow vector, then the panoramic image match is finished by Mean Shift attached to the optical flow clustering constraint condition in image feature space. Finally, panoramic geometric constraint of Ransac method is used for gross error detection. Several panoramic images are selected and used for experiment. The experiments of analysis and comparison were carried out in the conditions of the same inlier ratio, different inlier ratio and different data. The results show that the proposed method in the number and accuracy of correct matching points are superior to classic Ransac method and Pyramid Lucas-Kanade method, especially in the complex scene in low inlier ratio cases, the algorithm performance is relatively stable, and have better constraint effect for the gross error usually caused by repeat texture, moving objects and scale change.

Key words: vehicle-borne panoramic image, optical flow clustering, mean shift, SIFT feature

中图分类号: