测绘学报 ›› 2016, Vol. 45 ›› Issue (2): 205-212.doi: 10.11947/j.AGCS.2016.20140610

• 摄影测量学与遥感 • 上一篇    下一篇

均值漂移与卡尔曼滤波相结合的遥感影像道路中心线追踪算法

曹帆之, 朱述龙, 朱宝山, 李润生, 孟伟灿   

  1. 信息工程大学地理空间信息学院, 河南 郑州 450000
  • 收稿日期:2014-12-05 修回日期:2015-08-20 出版日期:2016-02-20 发布日期:2016-02-29
  • 作者简介:曹帆之(1992-),男,硕士生,研究方向为遥感图像处理。
  • 基金资助:
    国家自然科学基金(41401462)

Tracking Road Centerlines from Remotely Sensed Imagery Using Mean Shift and Kalman Filtering

CAO Fanzhi, ZHU Shulong, ZHU Baoshan, LI Runsheng, MENG Weican   

  1. Institute of Geosgatial Information, Information Engineering University, Zhengzhou 450000, China
  • Received:2014-12-05 Revised:2015-08-20 Online:2016-02-20 Published:2016-02-29
  • Supported by:
    The National Natural Science Foundation of China (No. 41401462)

摘要: 基于模板匹配的道路追踪方法是道路提取中较实用的一类方法,但传统模板匹配方法主要以相关系数作为相似性测度,对车辆、树荫等遮挡敏感,不适用于高分辨率遥感影像道路提取。针对这一问题,本文采用一种稳健的相似性测度,设计了一种基于均值漂移的道路中心点匹配算法,克服了传统模板匹配对遮挡敏感的缺点;然后运用卡尔曼滤波,实现高分辨率遥感影像道路中心线追踪。试验表明,该方法能够准确提取高分辨率遥感影像道路中心线,对车辆、树荫等遮挡具有稳健性。

关键词: 高分辨率遥感影像, 道路提取, 道路中心线追踪, 模板匹配, 均值漂移, 卡尔曼滤波

Abstract: Road tracking based on template matching is one class of practical methods of road extraction. However, the conventional methods of template matching mainly utilize correlation coefficient as the similarity measure. As a result, these algorithms are sensitive to occlusions caused by vehicles and trees and are unsuitable for road extraction from high-resolution remotely sensed imagery. To address this problem, this paper designs a road center matching algorithm based on mean shift utilizing a robust similarity measure, which overcomes the sensitivity of correlation coefficient matching to occlusions; then Kalman filter is utilized to track road centerlines from high-resolution remotely sensed imagery. Experimental results demonstrate that the proposed method can extract road centerlines from high-resolution remotely sensed imagery accurately and is robust to occlusions caused by vehicles and trees.

Key words: high-resolution remotely sensed imagery, road extraction, road centerline tracking, template matching, mean shift, Kalman filter

中图分类号: