基于人体骨骼信息的列车司机行为智能识别  

Intelligent Recognition of Train Driver Behavior Based on Human Skeleton Information

作　　者：王志毅[1] 刘昕武 WANG Zhiyi;LIU Xinwu(Shuohuang Railway Development Co.,Ltd,Suning,Hebei 062350,China;Zhuzhou CRRC Times Electric Co.,Ltd,Zhuzhou,Hunan 412001,China)

机构地区：[1]朔黄铁路发展有限责任公司,河北肃宁062350 [2]株洲中车时代电气股份有限公司,湖南株洲412001

出　　处：《机车电传动》2020年第4期90-93,共4页Electric Drive for Locomotives

摘　　要：文章提出了一种基于骨架识别的人体姿态捕捉预测算法,用于识别列车驾驶员违章行为。该方法是将司机作业监控视频拆分成一组连续的图片组合,然后通过构建深度学习模型网络预测每一帧图片中的人体姿态,并使用逻辑回归模型对司机行为进行分类,最终,通过将识别出的司机行为与LKJ数据相关联,裁定司机行为是否违章。该方法比目前机务段靠人工抽检的方式更加高效合理,但其识别效果受限于车载摄像头清晰度。A human poseture capture and prediction alcognition was proposed, which was used to identify violated driver. Firstly, driver behavior or action captured by camera was split into a set of successive frames with human poseture. Then each poseture in a frame was recognized by deep learning network and a type of behaviors composed by a set of poseture was classified via logistic regression model. Finally, the association between detected driver behaviors and locomotive LKJ data was applied in the recognition of violated driver. Compared with the daily random-check approach by inspectors in locomotive depots, this way was more reasonable and efficient. However, the accuracy of it was limited by the resolution of surveillance cameras in the driving cab of locomotive.

关 键 词：人体骨骼 深度学习 姿态识别 LKJ 列车 

分 类 号：U268.4[机械工程—车辆工程]