摘要
吸烟行为检测对于保障火力发电企业生产区域的设备运行状况、作业人员的生命和财产安全具有重要意义。目前吸烟行为检测主要基于香烟目标的检测,但是在火力发电厂生产区域中大多都是大场景、低清晰度的摄像头,监控视频中的香烟属于极小目标,检测困难度大、检测准确率低。为了解决上述问题,提出了一种基于脸部状态分析的吸烟行为分析方法。首先检测出监控画面中出现的人体脸部,并对人体脸部的状态进行分析,通过人体脸部被手部遮挡的情况判断是否属于类吸烟姿态,同时根据脸、人体和手部的相对位置判断提高检测的准确率;然后,通过时序处理模块对人体脸部一段时间内的状态进行分析,当某一段时间内出现类吸烟姿态的频率大于设定的阈值,就可以认定出现了吸烟行为并发出警报。实验结果表明,该方法在工业场景中的查准率、查全率、平均检测时间均优于目前已有的吸烟行为检测算法。
Smoking behavior detection is of great significance for ensuring the equipment operation conditions,operators’life and property safety in the production area of thermal power generation enterprises.At present,smoking behavior detection is mainly based on the detection of cigarette targets.However,in the production areas of thermal power plants,most of the camaras are of low-definition and installed for large scenes.The cigarettes in the monitoring videos are very small targets,with difficulty in detection and low accuracy of detection.In order to solve the above problems,this paper proposes a smoking behavior analysis method based on facial state analysis.First,the human face appearing in the monitoring video is detected,and the status of the human face is analyzed.By determining whether a person's face covered by his hands is a smoking-like posture,the accuracy of detection is improved by judging the relative positions of face,human body and hand.Then,time series processing module is used to analyze the state of human face in a period of time.When the frequency of smoking-like gestures within a certain period of time is greater than a set threshold,smoking behavior can be identified and an alarm is issued.Experimental results show that the precision rate,recall rate,and average detection time of this method in industrial scenes are better than the existing smoking behavior detection algorithms.
作者
汤文炜
王彪
张宇
于俊清
何云峰
TANG Wenwei;WANG Biao;ZHANG Yu;YU Junqing;HE Yunfeng(CHN Energy Changyuan Wuhan Qingshan Co-Generation Co.,Ltd.,Wuhan Hubei 430082,China;School of Computer Science and Technology,Huazhong University of Science and Technology,Wuhan Hubei 430074,China)
出处
《湖北电力》
2023年第5期154-162,共9页
Hubei Electric Power
基金
国家能源集团长源电子股份有限公司科技项目(项目编号:2021-4)。
关键词
工业场景
吸烟检测
目标检测
时序处理模块
行为分析
industrial scene
smoking detection
target detection
time series processing module
behavior analysis
作者简介
汤文炜(1972),男,湖北孝感人,硕士,高级工程师,主要研究方向为电力企业信息化;通讯作者:何云峰(1977),男,江西景德镇人,博士,副教授,主要研究方向为视频分析和处理。
