《电视技术》第46卷第12期(总第565期)51PARTS&DESIGN器件与设计文献引用格式:李干涛,朱岸平,李火星.乘务员值乘状态监测技术研究[J].电视技术,2022,46(12):51-55.LIGT,ZHUAP,LIHX.Researchonthemonitoringtechnologyoflocomotivecrewon-dutystate[J].VideoEngineering,2022,46(12):51-55.中图分类号:TP311.1文献标识码:ADOI:10.16280/j.videoe.2022.12.012乘务员值乘状态监测技术研究李干涛1,朱岸平1,李火星2(1.中国铁路南昌局集团有限公司机务部,江西南昌330000;2.郑州畅想高科股份有限公司,河南郑州450000)摘要:利用多维度视频融合感知技术对机车乘务员值乘状态进行实时分析,通过乘务员前方图像采集设备和后上方图像采集设备实时获取视频,先对视频图像进行动态自适应校正,再利用mobilenet-yolov3轻量级网络检测视频中乘务员的人脸多维特征信息、玩手机以及驾驶室车门开闭等特征,采用Hu矩不变量特征辅助分析乘务员的盹睡状态,从而及时提醒司机规范驾驶行为,实时预警,对保障行车安全具有重大意义。关键词:多维度视频融合;轻量级网络;Hu矩不变量;3D-ResnetResearchontheMonitoringTechnologyofLocomotiveCrewOn-DutyStateLIGantao1,ZHUAnping1,LIHuoxing2(1.LocomotiveDepartmentofChinaRailwayNanchangBureauGroupCo.,Ltd.,Nanchang330000,China;2.ZhengzhouThinkFreelyHi.Tech.Co.,Ltd.,Zhengzhou450000,China)Abstract:Themulti-dimensionalvideofusionperceptiontechnologyisusedtoconductreal-timeanalysisonthedutystatusoflocomotivecrewmembers.Thevideoisobtainedinrealtimethroughtheimageacquisitionequipmentinfrontofthecrewmembersandtheimageacquisitionequipmentattheupperback.First,thevideoimageisdynamicallyselfadaptivelycorrected.Then,themobileyolov3lightweightnetworkisusedtodetectthemulti-dimensionalfeatureinformationofthecrewmembers’facesinthevideo,mobilephones,cabdooropeningandclosingandotherfeatures,Humomentinvariantsareusedtoassisttheanalysisofthesleepingstateofthestewards,soastotimelyreminddriverstoregulatedrivingbehaviorandgivereal-timewarning,whichisofgreatsignificancetoensuredrivingsafety.Keywords:multi-dimensionalvideofusion;lightweightnetwork;Humomentinvariants;3D-Resnet0引言长期以来,机车乘务员因工作场所处于相对封闭的司机室...