欢迎访问中国科学院大学学报,今天是

中国科学院大学学报 ›› 2013, Vol. 30 ›› Issue (2): 244-250.DOI: 10.7523/j.issn.1002-1175.2013.02.016

• 信息与电子科学 • 上一篇    下一篇

基于MUGG的轨迹建模与异常检测

桂树1,2, 郭立1, 陆海先1, 谢锦生1   

  1. 1. 中国科学技术大学信息科学技术学院, 合肥 230022;
    2. 电子工程学院, 合肥 230037
  • 收稿日期:2012-01-19 修回日期:2012-03-08 发布日期:2013-03-15
  • 通讯作者: 桂树
  • 基金资助:

    国家自然科学基金(61071173)资助

MUGG-based modeling of trajectories and anomaly detection

GUI Shu1,2, GUO Li1, LU Hai-Xian1, XIE Jin-Sheng1   

  1. 1. College of Information Science and Technology, University of Science and Technology of China, Hefei 230022, China;
    2. Electronic Engineering Institute, Hefei 230037, China
  • Received:2012-01-19 Revised:2012-03-08 Published:2013-03-15

摘要:

构建视频场景中目标轨迹分布的概率模型——混合单边广义高斯模型,通过计算目标轨迹的信息量分析目标轨迹是否异常.该方法不依赖场景的先验知识,模型建立过程无监督,且模型能实时更新以适应时变环境.实验表明,该方法的有效性和鲁棒性,具有一定的应用价值.

关键词: 异常检测, 混合单边广义高斯模型, 轨迹学习, 轨迹距离

Abstract:

A probabilistic model named MUGG (mixture of unilateral generalized Gaussians) is designed for modeling the distribution of trajectories in visual scene. Information of trajectory is calculated to determine whether the trajectory is abnormal. This method is unsupervised and independent of prior knowledge.It is fit for time-varying environment with the real-time updated model. Its availability and robustness shown by experiments proves the application value.

Key words: anomaly detection, MUGG, trajectory learning, trajectory distance

中图分类号: