作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2014, Vol. 40 ›› Issue (12): 182-187,194. doi: 10.3969/j.issn.1000-3428.2014.12.034

• 人工智能及识别技术 • 上一篇    下一篇

基于均值漂移聚类的扩展目标量测集划分算法

刘风梅,葛洪伟,杨金龙,李鹏   

  1. 江南大学物联网工程学院,江苏 无锡 214122
  • 收稿日期:2014-01-06 修回日期:2014-02-26 出版日期:2014-12-15 发布日期:2015-01-16
  • 作者简介:刘风梅(1987-),女,硕士研究生,主研方向:目标跟踪技术;葛洪伟,教授、博士、博士生导师;杨金龙,副教授、博士;李 鹏,硕士研究生。
  • 基金资助:
    国家自然科学基金资助项目(61305017);江苏省自然科学基金资助项目(20130154)。

Extended Target Measurement Set Partition Algorithm Based on Mean Shift Clustering

LIU Fengmei,GE Hongwei,YANG Jinlong,LI Peng   

  1. School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China
  • Received:2014-01-06 Revised:2014-02-26 Online:2014-12-15 Published:2015-01-16

摘要: 在噪声环境下,存在扩展目标数未知且变化的多扩展目标跟踪量测集难以划分、计算代价高的问题。为此,提出一种基于均值漂移聚类的量测集划分算法。通过迭代更新中心点,使其收敛于局部最优,并引入极大似然估计技术估计每个划分子集中的目标数,对于目标数大于1的子集采用模糊C均值聚类算法进行二次划分,使得划分的量测子集与各个扩展目标一一对应。实验结果表明,该算法在多扩展目标量测集划分性能上明显优于传统的距离划分和K-means++划分算法,尤其是在保持跟踪精度的前提下量测集划分数和计算代价明显降低,且能较好地划分紧邻扩展目标的量测集。

关键词: 多扩展目标跟踪, 量测集划分, 均值漂移聚类, 极大似然估计, 距离划分, 紧邻的扩展目标

Abstract: Taking into account the difficulties of measurement set partition of the multiple extended target due to the unknown target number and the disturbance of the clutter.A novel measurement partition algorithm based on the mean shift clustering is proposed.The local optimum is obtained by iterating update the center point.The maximum likelihood estimation technique is introduced to estimate the number of targets for each cell,if the number is larger than one.It splits the cell into small cells by Fuzzy C-Mean(FCM) clustering algorithm until the cell corresponding to target number.Experimental results show that the proposed algorithm improves the performance of multiple Extended Target Tracking(ETT) compared with distance partition and K-means++ partition,especially effectively reduces partition number and computational cost without losing tracking accuracy,and has a good performance for spatially close targets measurement partition.

Key words: multiple Extended Target Tracking(ETT), measurement set partition, mean shift clustering, maximum likelihood estimation, distance partition, spatially close extended target

中图分类号: