北京航空航天大学学报 ›› 2009, Vol. 35 ›› Issue (12): 1487-1490.

• 论文 • 上一篇    下一篇

基于聚类最近数据关联的多目标跟踪算法

丁文锐1, 李红光2, 李新军3   

  1. 1. 北京航空航天大学 无人驾驶飞行器设计研究所, 北京 100191;
    2. 北京航空航天大学 机械工程及自动化学院, 北京 100191;
    3. 北京航空航天大学 无人驾驶飞行器设计研究所, 北京 100191
  • 收稿日期:2008-12-01 出版日期:2009-12-31 发布日期:2010-09-14
  • 作者简介:丁文锐(1971-),女,辽宁鞍山人,高级工程师,ding@buaa.edu.cn.
  • 基金资助:

    航空基金资助项目(2008ZC51029);武器装备预研资助项目

Multi-target tracking based on cluster min-distance data association

Ding Wenrui1, Li Hongguang2, Li Xinjun3   

  1. 1. Research Institute of Unmanned Aerial Vehicle, Beijing University of Aeronautics and Astronautics, Beijing 100191, China;
    2. School of Mechanical Engineering and Automation, Beijing University of Aeronautics and Astronautics, Beijing 100191, China;
    3. Research Institute of Unmanned Aerial Vehicle, Beijing University of Aeronautics and Astronautics, Beijing 100191, China
  • Received:2008-12-01 Online:2009-12-31 Published:2010-09-14

摘要: 提出了一种基于聚类最近数据关联的多目标跟踪算法.建立了基于目标位置、目标大小和目标灰度的3层匹配距离,将3个距离加权综合得到目标匹配的距离函数.以目标链为基准寻找与其存在最小匹配距离关系的观测目标作为目标链的数据后继,再以此观测目标为基准寻找与其存在最小匹配距离关系的目标链作为该目标的数据前驱.当且仅当匹配对象之间前驱与后继关系同时成立时认为二者匹配成功.将已有目标链分为4类,将当前帧观测目标分为2类.分析各类数据间可能存在的匹配关系,利用上述方法进行匹配运算.对于目标遮挡等情况,基于目标运动轨迹的瞬时直线性和均值滤波原理,将某时间段内的目标质心坐标作为输入数据得到回归直线,预测下一时刻的目标质心位置.目标大小和灰度预测数据由该时间内均值滤波得到.本算法在多人体跟踪实验中取得良好效果.

Abstract: A multi-target tracking algorithm based on cluster min-distance data association was proposed. Three matching distances about target position, target size and target gray value were established, weighted and integrated into final target matching distance function. The algorithm searched for the target list-s follow-up data which had min-matching distance relationship with observed target and the observed target-s precursor data which had min-matching distance relationship with target list. Only when the follow-up relationship and the precursor relationship were both true, the matching relationship was true between them. All the target lists were divided into four clusters. All the observed targets were divided into two clusters. The probable relationships between all clusters were analyzed, with carrying out the above computation. If the target was blocked, a target forecasting algorithm was executed. Based on the theory of instantaneous linear and mean filtering, the position data of target list were input to gian a regression line for forecasting the target-s position of next frame. The target-s size and gray value data were gained by mean filtering for the moment. A good result was presented on multi-human tracking.

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