航空学报 > 2014, Vol. 35 Issue (5): 1336-1346   doi: 10.7527/S1000-6893.2013.0400

一种混合的扩展目标跟踪方法

李波睿1,2, 慕春棣1, 白天明2, 柳志娟3   

  1. 1. 清华大学 自动化系, 北京 100084;
    2. 海军装备研究院, 北京 100036;
    3. 上海宇航系统工程研究所, 上海 201109
  • 收稿日期:2013-07-23 修回日期:2013-09-21 出版日期:2014-05-25 发布日期:2013-09-24
  • 通讯作者: 慕春棣,Tel.:010-62794322 E-mail:muchd@tsinghua.edu.cn E-mail:muchd@tsinghua.edu.cn
  • 作者简介:李波睿男,博士研究生,助理工程师。主要研究方向:扩展目标跟踪、视觉导航。Tel:010-62795448E-mail:lbr07@mails.tsinghua.edu.cn;慕春棣女,教授,博士生导师。主要研究方向:信息融合、飞行器导航制导与控制。Tel:010-62794322E-mail:muchd@tsinghua.edu.cn;白天明男,硕士,工程师。主要研究方向:信息传输与处理。Tel:010-66974140E-mail:btm_zn@126.com;柳志娟女,博士,工程师。主要研究方向:多模型算法、飞机健康管理。Tel:021-24188292E-mail:liuzhijuan6512@163.com
  • 基金资助:

    航空科学基金(20128058006)

A Hybrid Approach for Extended Object Tracking

LI Borui1,2, MU Chundi1, BAI Tianming2, LIU Zhijuan3   

  1. 1. Department of Automation, Tsinghua University, Beijing 100084, China;
    2. Navy Academy of Armament, Beijing 100036, China;
    3. Aerospace System Engineering Shanghai, Shanghai 201109, China
  • Received:2013-07-23 Revised:2013-09-21 Online:2014-05-25 Published:2013-09-24
  • Supported by:

    Aeronautical Science Foundation of China (20128058006)

摘要:

与传统的目标跟踪不同,扩展目标跟踪(EOT)不忽略目标的轮廓特征,同时对目标的质心运动学状态和轮廓特征进行估计。基于随机矩阵的扩展目标跟踪方法用随机正定矩阵来描述目标的轮廓特征,并且建立了适合扩展目标跟踪的量测模型。为了改善目标机动时的跟踪性能,根据椭圆(体)与正定矩阵的关系,提出基于椭圆(体)拟合的扩展目标跟踪方法。进一步地,为了综合上述两类方法的优点,提出一种混合的扩展目标跟踪方法,能够根据目标机动与否在两类方法中进行选择。仿真结果表明,该混合方法的轮廓特征估计误差低于前述两类方法,质心运动学状态的估计性能也更好。

关键词: 目标跟踪, 扩展目标跟踪, 随机矩阵, 椭圆拟合, 矩阵分解, Givens旋转, 信息融合

Abstract:

Different from the traditional object tracking technology, extended object tracking (EOT) doesn't ignore the target's physical extension. Instead, EOT simultaneously estimates both the centroid's kinematical state and the physical extension of the target. A random matrix based EOT approach characterizes the physical extension with a random symmetrical positive definite matrix, i.e. the ellipse/ellipsoid, and establishes a measurement model which is suitable for EOT. In order to improve the tracking performance when the target maneuvers, an ellipse/ellipsoid fitting based EOT approach is proposed based on the relationship between the ellipse/ellipsoid and the symmetrical positive definite matrix. Furthermore, a hybrid approach for EOT is presented to combine the advantages of the abovementioned two EOT approaches. Simulation results show that the hybrid approach can appropriately decide whether the target is maneuvering and choose a better approach. The physical extension estimation error of the hybrid approach is lower than the other approaches, and the estimation performance of the centroid's kinematical state is also better.

Key words: target tracking, extended object tracking, random matrix, ellipse fitting, matrix decomposition, Givens rotation, information fusion

中图分类号: