摘要
公共安全事件监控视频中的异常目标通常为中小型个体,其监测过程也易受到周围公共场所环境的干扰。提出突发公共安全事件监控视频异常行为监测方法。建立符合人体异常行为特点的增强型Cucker-Smale群体运动模型,描述人体异常行为的目标运动函数。以上述函数为基础,采集监控视频中与突发公共安全事件相关的异常行为数据。利用光流块为单位提取异常行为数据特征,并将异常行为数据特征与K-means聚类算法结合,全面分类异常行为数据特征,实现突发公共安全事件监控视频异常行为的监测。仿真结果表明,研究方法能够准确的识别出监控视频中的异常行为,且AUC值最高可达0.983,说明所提方法具有更好的应用性能。
The abnormal targets in public safety event monitoring videos are usually small and medium-sized individuals,and their monitoring process is also susceptible to interference from the surrounding public environment.In this paper,a method of monitoring abnormal behaviors in surveillance videos of sudden public security incidents was presented.First of all,we built an enhanced Cucker-Smale group motion model in line with the characteristics of human abnormal behavior to describe the target motion function of human abnormal behavior.Based on above functions,we collected the abnormal behavior data related to sudden public security incidents in the surveillance videos.Moreover,we used the optical flow block as a unit,and extracted the abnormal behavior data features.And then,these features were combined with K-means clustering algorithm.Finally,we comprehensively classified the abnormal behavior data features,thus achieving the monitoring on abnormal behavior in surveillance videos of sudden public security incidents.Simulation results show that the proposed method can identify the abnormal behavior in surveillance video accurately.The AUC value is up to 0.983.Therefore,the method has better application performance.
作者
杨传杰
殷洁
汪雁
武文亚
YANG Chuan-jie;YIN Jie;WANG Yan;WU Wen-ya(China Fire and Rescue Institute,Beijing 102202,China;School of Geographical Science,Nantong University,Jiangsu Nantong 226607,China)
出处
《计算机仿真》
2024年第1期243-246,292,共5页
Computer Simulation
基金
南通市社科基金项目(2021BNT022)。
关键词
突发公共安全事件
监控视频
异常行为
光流块
聚类算法
Sudden public security incidents
Surveillance video
Abnormal behavior
Optical flow block
Clustering algorithm
