Research On The Application Of Train Driver Sleep-Deprived Driving Detection Algorithm

The safety of people and property can be gravely jeopardized by accidents,despite railway transportation’s essential part in economic growth.Accurately recognizing the fatigue of a driver and issuing timely warnings has become a challenge for the transportation industry,due to fatigue driving being a major factor in traffic accidents.When train drivers feel tired,they exhibit different fatigue states and behavioral characteristics.Therefore,this article designs a fatigue driving detection system by analyzing the driver’s facial features.The system uses the YOLOV5 algorithm and facial keypoint positioning to determine the fatigue state.To improve driver focus,the system also includes distracted driving detection based on abnormal behavior.When the system detects driver fatigue or abnormal behavior,it issues a warning to ensure driver focus and prevent safety accidents.The main research content is as follows:(1)CA(Coordinate Attention)attention mechanism module is incorporated into the Backbone network’s C3 module to tackle the issue of the original YOLOv5 network structure’s low attention and detection rate in the facial area.This module divides the input features into multiple channels,weights each channel,and finally concatenates the results of different channels to improve the model’s attention to the facial area and thus improve detection accuracy and speed.(2)To ensure rapid and effective multi-scale feature fusion for facial recognition,the YOLOv5 algorithm’s network structure has been augmented with the Bi FPN(weighted bidirectional feature pyramid network)structure,supplanting the PANet structure.Feature network layers are thought of as two-way paths,with the layers of varying depths being joined together to maintain the essential characteristics of small scales,which is advantageous for enhancing the detection precision of small targets.Experiments have demonstrated that,in comparison to the original algorithm,the m AP is augmented by 1%,the recall rate by 0.9%,and the accuracy by 1.2%-all of which are improvements.(3)A Dlib68 library-based facial key point detection model is employed to acquire the coordinates of the driver’s facial key points.The number of blinks and yawns is determined by comparing the MAR,EAR,and threshold values,and the PERCLOS algorithm is used to determine whether the driver is fatigued by combining the number of blinks and the frequency of yawns.At the same time,an improved YOLOv5 detection model is used to detect video stream information and check for abnormal distracted driving behavior by the driver.(4)In order to facilitate the use of this method for fatigue and distracted driving detection,a UI interface for fatigue driving detection system is designed.Through the UI system interface,the driver’s fatigue and distracted abnormal behavior can be monitored in real time and timely warnings can be given.In a timely fashion,the driver can take pertinent steps to avert traffic mishaps.