电子学报 ›› 2017, Vol. 45 ›› Issue (3): 513-519.DOI: 10.3969/j.issn.0372-2112.2017.03.001

• 学术论文 •    下一篇

基于模糊空时线索的多目标在线跟踪算法

李俊, 谢维信, 李良群, 刘钧彬   

  1. 深圳大学ATR国防科技重点实验室, 广东深圳 518060
  • 收稿日期:2015-10-13 修回日期:2016-05-19 出版日期:2017-03-25
    • 通讯作者:
    • 李良群
    • 作者简介:
    • 李俊 男,1984年生于湖南益阳.现为深圳大学信息工程学院博士研究生.主要研究方向为计算机视觉与机器学习.E-mail:lijun10@email.szu.edu.cn;谢维信 男,1941年生于广东.教授、博导,1965年毕业于西安军事电讯工程学院,现为深圳大学信号与信息处理学科学术带头人,国家级有突出贡献中青年专家,主要研究方向为智能信息处理、模糊信息处理、图像处理和模式识别.E-mail:wxxie@szu.edu.cn;刘钧彬 男,1982年生于江西赣州.现为深圳大学信息工程学院博士研究生.主要研究方向为视觉跟踪,信号处理.E-mail:junbinliu@163.com
    • 基金资助:
    • 国家自然科学基金 (No.61301074,No.61271107); 广东省自然科学基金 (No.S2012010009417); 广东省科技厅产学研协同创新成果转化项目 (No.509111098127); 深圳市科技计划项目 (No.JCYJ20140418095735618); 国防预研基金项目 (No.91400C800501140C80340)

Online Multiple Target Tracking Algorithm Based on Fuzzy Spatio-Temporal Cues

LI Jun, XIE Wei-xin, LI Liang-qun, LIU Jun-bin   

  1. ATR Key Lab of National Defense Technology, Shenzhen University, Shenzhen, Guangdong 518060, China
  • Received:2015-10-13 Revised:2016-05-19 Online:2017-03-25 Published:2017-03-25
    • Supported by:
    • National Natural Science Foundation of China (No.61301074, No.61271107); National Natural Science Foundation of Guangdong Province,  China (No.S2012010009417); Industry-university-research Collaborative Innovation Achievement Transformation Project of Guangdong Science and Technology Department (No.509111098127); Science and Technology Program of Shenzhen (No.JCYJ20140418095735618); Supported by National Defense Pre-research Foundation of China (No.91400C800501140C80340)

摘要:

多目标在线跟踪是视频监控中的关键问题之一.针对日益增长的智能化视频监控的需求,提出了一种基于模糊空时线索的多目标在线跟踪算法.在该算法中,引入模糊空时多属性特征定义距离函数,利用模糊C均值聚类优化得到交叉隶属度矩阵,实现目标与观测间的数据关联.为了减少错误的轨迹起始,利用空时线索定义了遮挡度函数,判别出新目标并起始相应的目标轨迹.实验结果表明,本文算法能够准确地估计出目标的运动轨迹.本文算法可应用于视频监控、安防以及自动驾驶等领域.

关键词: 视频监控, 在线跟踪, 模糊C均值, 空时线索, 模糊隶属度

Abstract:

Online multi-target tracking is one of the key problems in video surveillance.According to the increasing need of smart monitoring,an online multiple target tracking algorithm based on fuzzy spatio-temporal cues was proposed.The fuzzy spatio-temporal multiple features were introduced to define the distance function,and the fuzzy c-means algorithm was adopted to derive the cross fuzzy membership degree matrix which was used to deal with the data association between the targets and the observations.To reduce the wrong initializations of the targets,an occlusion measurement about the levels of occlusion was defined according to the spatio-temporal cues.The new targets were discriminated from the false alarms by the occlusion measurement,and their corresponding tracks were initialized.Experimental results show that the proposed algorithm can accurately estimate the trajectories of multiple targets.The proposed algorithm can be applied in video surveillance,security,autonomous driving,etc.

Key words: video surveillance, online tracking, fuzzy c-means, spatio-temporal cues, fuzzy membership degree

中图分类号: