摘要
异常行为检测在视频分析中非常重要.本文提出了一种基于主题模型和三支决策的方法来检测视频中的异常行为.首先利用改进的主题模型来产生低维特征,这些低维特征能够表示视频中的原子行为;然后采用三支决策的方式,对于确定属于异常和正常的行为做出立即决策,而对于疑似异常的行为则进行进一步的分析以减少误分率.并且通过实验证明该方法能够产生有意义的低维特征,并且具有更好的分类性能.
Abnormal behavior detection is important in video analysis.In this paper,we propose a method based on topic model and three-way decision rough sets to detect abnormal behaviors in video.Firstly,we use the improved topic model to generate low-dimensional features,which can represent the atomic activities in the video.Then,three-way decision rough set method is used.It is available to make an immediate decision for determined normal and abnormal behavior.For those suspected abnormal behaviors,further analysis is needed in order to reduce the misclassification rate.Experiments prove that our approach can generate meaningful low-dimensional features and has better classification performance.
出处
《南京大学学报（自然科学版）》
CAS
CSCD
北大核心
2013年第4期475-482,共8页
Journal of Nanjing University（Natural Science）
基金
国家自然科学基金(61170180
61035003)
江苏省自然科学基金(BK2011005)
关键词
视频行为
异常检测
主题模型
三支决策
video behavior
anomaly detection
topic model
three-way decision
