华南理工大学学报(自然科学版) ›› 2008, Vol. 36 ›› Issue (10): 57-60,66.

• 交通运输工程 • 上一篇    下一篇

基于强化学习的视频车辆跟踪

卞建勇徐建闽裴海龙2   

  1. 1. 华南理工大学 土木与交通学院, 广东 广州 510640; 2. 华南理工大学 自动化科学与工程学院, 广东 广州 510640
  • 收稿日期:2007-10-08 修回日期:2007-11-12 出版日期:2008-10-25 发布日期:2008-10-25
  • 通信作者: 卞建勇(1980-),男,博士生,主要从事非线性控制、智能控制、图像处理与模式识别以及智能交通研究. E-mail:bjyong977@126.com
  • 作者简介:卞建勇(1980-),男,博士生,主要从事非线性控制、智能控制、图像处理与模式识别以及智能交通研究.
  • 基金资助:

    国家“863”高技术计划项目(2006AA11Z211);广州市科技计划项目(B04B2070710)

Video Vehicle Tracking Based on Reinforcement Learning

Bian Jian-yong1  Xu Jian-min1  Pei Hai-long2   

  1. 1. School of Civil Engineering and Transportation, South China University of Technology, Guangzhou 510640, Guangdong, China; 2. School of Automation Science and Engineering, South China University of Technology, Guangzhou 510640, Guangdong, China
  • Received:2007-10-08 Revised:2007-11-12 Online:2008-10-25 Published:2008-10-25
  • Contact: 卞建勇(1980-),男,博士生,主要从事非线性控制、智能控制、图像处理与模式识别以及智能交通研究. E-mail:bjyong977@126.com
  • About author:卞建勇(1980-),男,博士生,主要从事非线性控制、智能控制、图像处理与模式识别以及智能交通研究.
  • Supported by:

    国家“863”高技术计划项目(2006AA11Z211);广州市科技计划项目(B04B2070710)

摘要: 基于视频的车辆跟踪在交通监控领域有着重要的实用价值.为了有效地跟踪视频车辆,文中首先提出了一种结合虚拟检测线的统计背景提取方法,然后运用背景差法提取运动车辆信息,再在运动车辆区域运用SUSAN(Smallest Univalue Segment Assimilating Nucleus)算法提取车辆角点特征,在此基础上运用强化学习进行车辆跟踪,充分发挥了强化学习搜索效率高的特性.实验结果表明:文中方法跟踪情况稳定,跟踪准确率比较高,可以获得很好的跟踪效果.

关键词: 交通监控, 强化学习, 车辆跟踪, 背景提取, SUSAN算法

Abstract:

As the video vehicle tracking is of great importance to the traffic monitoring, a statistical background extraction method combined with the virtual detection line is p video vehicles. Then, the background difference method is employed to extract the information of moving vehicles, and the SUSAN algorithm is adopted to extract the comer feature in the moving vehicle region. Moreover, the reinforcement learning theory with high searching efficiency is applied to the video vehicle tracking. Experimental results show that the proposed method helps to obtain satisfying tracking results due to its good stability and high tracking accuracy.

Key words: traffic monitoring, reinforcement learning, vehicle tracking, background extraction, SUSAN algorithm