基于事件字典的行人异常事件检测

摘要

近年来极端行为和暴恐行为严重威胁着公共安全,行人异常事件检测已成为研究的热点问题.为了克服传统方法中对异常事件定义的模糊性、对行人异常行为特征的描述不够准确的缺点,提出一种基于事件字典的行人异常事件检测方法.本文方法的创新之处是构造了一种行人特征描述子,能够有效描述行人的身体各部分的变化规律,利用行人特征描述子提取样本的特征进行聚类分析,构建事件字典,预测事件的类别.在标准数据集BEHAVE Interactions Test Case Scenarios、UMN、UCSD上与LDA(Latent Dirichlet Allocation,潜在狄利克雷分布)和双稀疏字典的对比实验表明,该方法对于异常行为的检测准确有较高的检测准确率.%With the development of modem society,many accidents have taken place in densely populated public places,and more individual extreme behaviors have brought great harm to people's life.Crowd anomaly detection has become an important research subject.In this paper,a new method for detecting abnormal events is proposed.The innovation is the proposed pedestrian feature descriptor,which can effectively describe the variation of different parts of the body.Firstly extract the characteristics of the sample cluster analysis with the proposed feature descriptor,construct the event dictionary,then predict the type of event.The proposed method performs better than LDA (Latent Dirichlet Allocation) and two sparse dictionaries on BEHAVE Interactions Test Case Scenarios,UMN,and UCSD datasets.