基于改进YOLOV7与StrongSORT算法的列车司机手比行为检测

列车司机驾驶行为的规范性直接影响到列车行车安全与状态，但当前对列车司机手比行为的检测仍存在不足。为了对列车司机手比行为进行有效检测，本文利用动车组模拟驾驶系统所采集的司机乘务作业影像，结合采用融合注意力机制的You Only Look Once Version 7 (YOLOV7)神经网络模型与Strong Simple Online and Realtime Tracking (StrongSORT)算法，对动车组司机驾驶过程中的5种手比行为进行检测。实验结果表明：本文算法可以有效提升对列车司机乘务作业时不同类型手比行为的检测效果，其中检测精确率平均提升1.2%，检测召回率平均提升1.9%。本文提出的算法将有助于改进铁路院校、机务部门对列车司机日常训练、工作考评的效果，提高列车行车过程中的安全性。

Gesture Behavior Detection of Train Drivers Based on Improved YOLOV7 and StrongSORT

The standardization of train driver''s driving behavior directly affects the running safety and state of train operation.At present,the detection of train driver''s gesture behavior is insufficient.In order to effectively detect the gesture behavior of train drivers,this article uses the images of drivers'' crew work collected by the EMU simulation driving system,combined with the You Only Look Once Version 7 (YOLOV7) neural network model using the fusion attention mechanism and the Strong Simple Online and Realtime Tracking (StrongSORT) algorithm,five kinds of gesture behaviors of EMU drivers during driving were tested.The experimental results show that the algorithm in this article can effectively improve the detection effect of different types of gesture behaviors when the train driver is working.The detection accuracy rate is increased by 1.2% on average,and the detection recall rate is increased by 1.9% on average.The algorithm proposed in this article will help to improve the effect of daily training and work evaluation of train drivers by railway colleges and locomotive departments,and improve the safety in the process of train operation.