视频监控中针对拥挤人群的人体分割与跟踪  

Human Segmentation and Tracking for Crowded People in Video Surveillance

作　　者：温向兵[1] 满君丰[1] 李倩倩[2] 李长云[1] 

机构地区：[1]湖南工业大学计算机与通信学院,湖南株洲412008 [2]湖南化工职业技术学院信息工程系,湖南株洲412004

出　　处：《小型微型计算机系统》2012年第4期891-895,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61170102)资助;湖南省自然科学基金项目(11JJ4050;11JJ3070)资助;湖南省科技厅计划项目(2011GK3145)资助;湖南省教育厅优秀青年项目(11B039)资助

摘　　要：视频监控中,拥挤人群的相互遮挡给人体分割和跟踪带来很大困难.为了解决该问题,提出人体模型和人体边缘曲线相结合的人体分割方法.针对分割可能造成人体特征值存在较大的缺损、畸变问题,采用具有较高鲁棒性的BP(Back Propaga-tion)神经网络作为跟踪模型.为了提高BP网络的自主学习能力,采用分层Dirichlet过程来判断是否有新类别的人体特征数据产生,进而为BP网络的学习提供决策.通过仿真实验证实:本文提出的遮挡处理方法能够有效解决人体部分遮挡问题,与其他方法相比,具有简单且实时性好的优点;此外,分层Dirichlet过程与BP网络的结合提高了跟踪系统的自主学习能力.Mutual occlusion of crowded people makes human segmentation and tracking more difficult in video surveillance. Thus, a human segmentation method combing human model with body edge curve is presented. Because segmentation may result in serious defect and distortion, robust BP neural network model is used as tracking mode. For improving autonomous learning ability of BP net- work,hierarchical Dirichlet process is adopted to decide whether new types of human characteristics data is generated, which provides decision for BP network learning. The simulation experiment confirms that the method presented in this paper can effectively solve the problem of partial human occlusion. Meanwhile, this method has unique advantage of simplicity and real-time over others. Further- more, the combination of hierarchical Dirichlet process and BP network improves autonomous learning ability of tracking system.

关 键 词：拥挤人群 人体模型 分层Dirichlet过程 BP神经网络 前景边缘曲线 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]