摘要
针对监控视频异常活动检测算法检测准确率与鲁棒性较低的问题,提出了一种基于词袋模型与无向图建模的视频异常活动检测算法.(1)将输入视频划分为大小相等的视频片段,提取每个视频片段的时空兴趣点;(2)生成一个局部活动的无向图集,图的顶点表示时空兴趣点,边表示兴趣点之间的关系;(3)分别对局部异常活动和全局异常活动进行分类处理,识别出异常活动.基于公共数据集UMN的仿真实验结果表明,本算法对视频监控中异常活动具有较好的检测准确率.
In order to improve the detection accuracy and robustness of abnormal behavior of surveillance video,a abnormal behavior detection algorithm of surveillance video based on bag of word and undirected graphs has been proposed.Firstly,the incoming video is split into video clips of equal size and space-time interest points in each video clip are extracted.Secondly,a set of undirected graphs of local activities is generated,the vertices of the graphs are space-time interest points and the edge represents the relationship between the vertices.And lastly,the local abnormal behaviors and the global abnormal behaviors are classified respectively,so that the abnormal behaviors are recognized.Simulation experimental results based on the public UMN dataset show that the proposed algorithm realizes a better detection accuracy to abnormal behavior of surveillance video.
作者
潘志安
PAN Zhi-an(Information Technology College,Hubei Polytechnic Institute,Xiaogan Hubei 432000,Chin)
出处
《西南师范大学学报（自然科学版）》
CAS
北大核心
2018年第7期60-66,共7页
Journal of Southwest China Normal University(Natural Science Edition)
基金
国家自然科学基金项目(61370092)
关键词
图核
二分类支持向量机
视频监控
异常活动检测
公共安全
graph kernel
binary support vector machine
surveillance video
abnormal behavior detection
public safety
