Scholarlink

<p>Driver fatigue is one of the leading causes of traffic accidents, making its early detection essential for accident prevention. In real-world scenarios, driver movements such as head turning and yawning can partially occlude facial features, leading to facial information loss. To address these issues, we propose a novel Multi-stream Facial Feature Fusion Convolutional Neural Network (MSFˆ3-CNN) architecture. Our approach utilizes the Multi-task Cascaded Convolutional Neural Network (MTCNN) module to extract both global facial information and local eye features. The left and right eye regions are processed by the MobileNetV3-small network, while the full-face image is handled by the MobileNetV3-large network. A Global Feature Extraction Module (GFEM) and a Stream Fusion Module (SFM) are then used to combine the extracted features, which are subsequently classified via a fully connected layer. Experimental results demonstrate the superior performance of our method, achieving a 4.24% improvement in accuracy on Driver Fatigue Dataset (DFD) and a 0.8% gain in accuracy on the CEW dataset, compared to single-stream baselines. This method exhibits strong robustness and lightweight characteristics, providing a reliable solution for fatigue detection and accident prevention.</p>

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

Driver fatigue detection based on facial multi-feature fusion computing

  • 期刊论文

  • 发表日期：

    2025年11月8日

查看全文
  • Shuwan Cui,
  • Hao Li,
  • Xiaozhen Liu,
  • Chengyu Yang,
  • Hui Gao,
  • Jvnhao Yang,
  • Bo Yang

摘要

Driver fatigue is one of the leading causes of traffic accidents, making its early detection essential for accident prevention. In real-world scenarios, driver movements such as head turning and yawning can partially occlude facial features, leading to facial information loss. To address these issues, we propose a novel Multi-stream Facial Feature Fusion Convolutional Neural Network (MSFˆ3-CNN) architecture. Our approach utilizes the Multi-task Cascaded Convolutional Neural Network (MTCNN) module to extract both global facial information and local eye features. The left and right eye regions are processed by the MobileNetV3-small network, while the full-face image is handled by the MobileNetV3-large network. A Global Feature Extraction Module (GFEM) and a Stream Fusion Module (SFM) are then used to combine the extracted features, which are subsequently classified via a fully connected layer. Experimental results demonstrate the superior performance of our method, achieving a 4.24% improvement in accuracy on Driver Fatigue Dataset (DFD) and a 0.8% gain in accuracy on the CEW dataset, compared to single-stream baselines. This method exhibits strong robustness and lightweight characteristics, providing a reliable solution for fatigue detection and accident prevention.