Scholarlink

With the rapid development of Intelligent Transportation System (ITS), real-time monitoring of driver distraction behavior has become essential for improving road safety. This paper proposes a monitoring method for driver distraction behavior based on YOLOv5 combined with Efficient Vision Transformer (ViT) and Effective Channel Attention (ECA). Efficient ViT replaces the YOLOv5 backbone network to effectively reduce the model complexity and computation while maintaining the performance, which is suitable for real-time application scenarios. Adding a layer of ECA mechanism module to the neck network can enhance its attention to the local area, and better capture these drivers’ cell phone playing, water drinking, and smoking actions, thus improving the accuracy and stability of distraction detection. The ablation experiments show that the algorithm in this paper is significantly lower in model complexity than YOLOv5, and the detection accuracies of playing cell phone, drinking water, and smoking are improved by 4.4%, 4.7%, and 5.3%. The average precision mean (mAP) increases by 1.6% to 0.988. This technique can provide powerful technical support for intelligent transportation systems, reduce the risk of traffic accidents, and protect the safety of drivers and passengers.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on Driver Distraction Behavior Monitoring System Based on Transformer

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yujuan Yao,
  • Peng Liu,
  • Zhongrun Lv,
  • Minying Zhou,
  • Wenqiang Huang,
  • Zemeng Zhang

摘要

With the rapid development of Intelligent Transportation System (ITS), real-time monitoring of driver distraction behavior has become essential for improving road safety. This paper proposes a monitoring method for driver distraction behavior based on YOLOv5 combined with Efficient Vision Transformer (ViT) and Effective Channel Attention (ECA). Efficient ViT replaces the YOLOv5 backbone network to effectively reduce the model complexity and computation while maintaining the performance, which is suitable for real-time application scenarios. Adding a layer of ECA mechanism module to the neck network can enhance its attention to the local area, and better capture these drivers’ cell phone playing, water drinking, and smoking actions, thus improving the accuracy and stability of distraction detection. The ablation experiments show that the algorithm in this paper is significantly lower in model complexity than YOLOv5, and the detection accuracies of playing cell phone, drinking water, and smoking are improved by 4.4%, 4.7%, and 5.3%. The average precision mean (mAP) increases by 1.6% to 0.988. This technique can provide powerful technical support for intelligent transportation systems, reduce the risk of traffic accidents, and protect the safety of drivers and passengers.