北京航空航天大学学报 ›› 2015, Vol. 41 ›› Issue (12): 2240-2249.doi: 10.13700/j.bh.1001-5965.2014.0827

• 论文 • 上一篇    下一篇

基于边缘信息的灰度目标跟踪算法

郑海超1, 毛峡1, 梁晓庚1,2   

  1. 1. 北京航空航天大学 电子信息工程学院, 北京 100191;
    2. 中国空空导弹研究院, 洛阳 471000
  • 收稿日期:2014-12-30 修回日期:2015-04-03 出版日期:2015-12-20 发布日期:2016-01-04
  • 通讯作者: 毛峡(1952-),女,浙江义乌人,教授,moukyou@buaa.edu.cn,主要研究方向为计算机视觉、模式识别和人机交互. E-mail:moukyou@buaa.edu.cn
  • 作者简介:郑海超(1986-),男,河北承德人,博士研究生,zhenghaichao_buaa@163.com
  • 基金资助:
    高等学校博士学科点专项科研基金(20121102130001);中央高校基本科研业务费专项资金(YWF-14-DZXY-015);航天支撑技术基金

Gray target tracking algorithm based on edge information

ZHENG Haichao1, MAO Xia1, LIANG Xiaogeng1,2   

  1. 1. School of Electronic and Information Engineering, Beijing University of Aeronautics and Astronautics, Beijing 100191, China;
    2. China Airborne Missile Academy, Luoyang 471000, China
  • Received:2014-12-30 Revised:2015-04-03 Online:2015-12-20 Published:2016-01-04

摘要: 提出了一种基于边缘信息的跟踪算法,其可以实现对剧烈变化的灰度目标的精确跟踪.首先,利用基于双同心圆窗口算子的非线性边缘检测算法得到高质量的边缘信息;其次,为了解决单一边缘特征空间不能充分表征目标的难题,提出了一种通过组合边缘图像构建特征空间的方法,以便为目标建模提供充分信息;再次,在构建的特征空间中使用核估计方法对目标进行建模;在目标定位阶段,利用Kalman滤波器对目标进行预估后,再由Mean Shift算法在预估位置邻近区域实现目标定位;最后,提出了一种基于形态学的动态模型更新策略,使得算法不仅可以获得精确的目标区域,还可以实现对目标尺寸和形状变化的自适应.实验结果表明,本算法不仅可以有效跟踪剧烈变化的灰度目标,而且跟踪窗口可以实现对目标尺寸和形状的自适应.

关键词: 计算机视觉, 目标跟踪, 灰度目标, 跟踪算法, 边缘检测, Mean Shift

Abstract: To precisely track the gray targets undergoing drastic changes in the image sequence, a new tracking algorithm based on edge information was proposed. Firstly, obtained by the two-concentric-circular-window operator, a nonlinear edge detection algorithm was proposed to get high quality edge information. Secondly, a novel method to construct feature space by synthesizing edge images was proposed in order to solve the problem that single edge feature space was not able to characterize the target thoroughly. The proposed method provided enough information to construct target model. Then, an approach to construct the target model with the kernel-based estimation method was proposed in constructed feature space. In target localization stage, the target position was preliminarily predicted by Kalman filter, and then the Mean Shift algorithm is utilized to locate the target in the region around the predicted position. Finally, a new dynamic model update strategy based on morphological operations was proposed. It can offer the proposed algorithm the ability to obtain precise target region and automatically adjust to the changing target size and target shape. Experimental results demonstrate that the proposed algorithm can perform well in image sequences where the targets undergo drastic changes. Meanwhile, the proposed algorithm can obtain the precise target region, and the track window can automatically adjust to the changing target size and target shape.

Key words: computer vision, target tracking, gray target, tracking algorithm, edge detection, Mean Shift

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发