摘要
大型安防环境威胁行为识别主要依靠红外热成像技术检测环境场景,对环境干扰因素极为敏感,使得识别结果F_(1)值较低,因此,提出基于物联网(IoT)与机器视觉的大型安防环境多层次威胁行为识别方法。依托于IoT技术搭建大型安防环境监控框架,感知大量现场环境图像。结合机器视觉理论进行图像处理,通过灰度化处理、背景模型建立和前景像素检测,找到大型安防环境图像包含的人员目标。利用多层次卷积神经网络,建立多层特征差异图,将其输入由众多支持向量机组成的多层次识别结构,得到多层次威胁行为识别结果。实验结果表明,所提方法识别结果的F_(1)值为0.92,实现了安防环境中威胁行为的准确检测。
In the process of identifying large-scale security environmental threat behaviors,infrared thermal imaging technology is mainly used to detect environmental scenes,which is extremely sensitive to environmental interference factors,and results in low Fi value of the recognition result.Therefore,a multi-level threat behavior recognition method for large-scale security envi-ronments based on Internet of Things(loT)and machine vision is proposed.This paper builds a large-scale security environ-ment monitoring framework based on IoT technology to perceive a large number of on-site environmental images.This paper combines the machine vision theory to process images.Through grayscale processing,background model establishment and foreground pixel detection,personnel targets contained in large-scale security environment images are identified.The multi-lev-el feature difference map is established by the multi-level convolutional neural network,and input into the multi-level recognition structure composed of numerous support vector groups to obtain the multi-level threat behavior recognition results.The experi-mental results show that the Fr value of the recognition results of the proposed method is 0.92,achieving accurate detection of threat behavior in the security environment.
作者
张晖
李曈昊
冯飞
黄双佑
肖波
ZHANG Hui;LI Tonghao;FENG Fei;HUANG Shuangyou;XIAO Bo(Guoneng Changyuan Hanchuan Power Generation Co.,Ltd.,Hanchuan 431614,China)
出处
《微型电脑应用》
2025年第5期134-137,142,共5页
Microcomputer Applications
基金
国家能源集团科技项目(GJNY-22-81)。
关键词
物联网
机器视觉
特征差异图
威胁行为
动作识别
支持向量机
Internet of Things
machine vision
feature difference map
threatening behavior
action recognition
support vec-tor machine
