基于隐含主题模型的异常行为分析  

Abnormal behavior analysis based on latent topic model

作　　者：赵龙[1] 郭立[1] 谢锦生[1] 刘皓[1] 陆海先[1] 

机构地区：[1]中国科学技术大学电子科学与技术系,合肥230026

出　　处：《中国科学院研究生院学报》2013年第3期387-393,共7页Journal of the Graduate School of the Chinese Academy of Sciences

基　　金：国家自然科学基金(61071173)资助

摘　　要：针对目前多数异常行为分析方法没有考虑场景,提出一种基于隐含主题模型的异常行为分析方法.提取场景的颜色和纹理特征,利用K-means对特征聚类,形成视觉单词,利用pLSA模型将视觉单词分为若干语义主题区域,生成场景描述.组合轨迹特征与场景语描述,生成组合特征向量,再利用CRF对组合特征向量建模,通过训练估计模型参数,利用模型推断,分析异常行为.实验表明,本文方法对特定场景的异常行为可以较为准确地分析.Considering that most of abnormal behavior analysis methods do not consider the scene,we propose a method of abnormal behavior analysis based on latent topic model.Features of the scene are extracted and clustered to visual vocabulary by K-means.The visual vocabulary is divided into semantic topics to describe the scene by pLSA model.The descriptions for the scene and trajectory are combined to form feature vector which is modeled by CRF.Parameters of the CRF model are estimated by training,and abnormal behavior is analyzed by inference.The experiments show that abnormal behavior in a particular scene is accurately analyzed by this method.

关 键 词：隐含主题模型 异常行为分析 PLSA CRF 全局行为 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]