Scholarlink

Recently, the utilization of machine vision techniques for detecting abnormal driving behaviors has gained significant importance in ensuring the safety of vehicle occupants. This detection can significantly enhance driving safety and mitigate the occurrence of accidents. Nevertheless, previous research in this domain has predominantly focused on facial expression recognition, neglecting a comprehensive analysis of overall posture. This paper introduces three novel models: YOLO-v8-CBAM, YOLO-v8-CBAM-SimAM, and Gold-YOLO-v8, which integrate attention mechanisms and the Gold-YOLO framework into the YOLO-v8 network. The objective of these models is to enhance the accuracy and timelines of abnormal driving behavior detection. Common abnormal driving behaviors were categorized as closed eyes, yawning, head turning, and talking on the phone. A comparative analysis was conducted utilizing YOLO-v8n, YOLO-v8s, Y-OLO-v8n-CBAM, YOLO-v8n-CBAM-SimAM, and Gold-YOLO-v8, yielding accuracy rates of 93.496%, 93.514%, 93.675%, 93.285%, and 94.387%, respectively. The experimental results demonstrate that the proposed models accurately and promptly identify abnormal driving behaviors.

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

Abnormal Driving Detection Algorithm Based on Improved YOLO-v8 with Self-attention and GOLD-YOLO Mechanism

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Pengfei Li,
  • Xinrui Shao,
  • Ziwei Yin,
  • Chen Tao,
  • Zeping Zhu,
  • Jiangchen Li,
  • Zhichao Xu

摘要

Recently, the utilization of machine vision techniques for detecting abnormal driving behaviors has gained significant importance in ensuring the safety of vehicle occupants. This detection can significantly enhance driving safety and mitigate the occurrence of accidents. Nevertheless, previous research in this domain has predominantly focused on facial expression recognition, neglecting a comprehensive analysis of overall posture. This paper introduces three novel models: YOLO-v8-CBAM, YOLO-v8-CBAM-SimAM, and Gold-YOLO-v8, which integrate attention mechanisms and the Gold-YOLO framework into the YOLO-v8 network. The objective of these models is to enhance the accuracy and timelines of abnormal driving behavior detection. Common abnormal driving behaviors were categorized as closed eyes, yawning, head turning, and talking on the phone. A comparative analysis was conducted utilizing YOLO-v8n, YOLO-v8s, Y-OLO-v8n-CBAM, YOLO-v8n-CBAM-SimAM, and Gold-YOLO-v8, yielding accuracy rates of 93.496%, 93.514%, 93.675%, 93.285%, and 94.387%, respectively. The experimental results demonstrate that the proposed models accurately and promptly identify abnormal driving behaviors.