A dense multi-pooling convolutional network for driving fatigue detection.

Area of Science:

• Computer Science
• Artificial Intelligence
• Transportation Safety

Background:

• Driver fatigue is a significant factor in traffic accidents, especially for large vehicle operators.
• Current vision-based fatigue detection systems struggle with accuracy in complex scenarios like wearing glasses or the presence of passengers.

Purpose of the Study:

• To develop an accurate and timely vision-based driver fatigue detection method for complex driving conditions.
• To improve traffic safety by addressing the limitations of existing fatigue detection technologies.

Main Methods:

• A novel approach combining Driver's State Detection (DSD), a Dense Multi-Pooling Convolutional Network (DMP-Net), and Driving Fatigue Detection (DFD).
• Real-time facial landmark detection using MTCNN and a face detection bounding box filtering algorithm.
• Fatigue assessment based on frequency of mouth opening (FM) and percentage of eyelid closure over pupil (PERCLOS).

Main Results:

• The proposed DMP-Net achieves high accuracy with fewer parameters, outperforming existing methods in accuracy and speed.
• The integrated system demonstrated exceptional accuracy: 99.25% on CEW, 99.24% on ZJU, and 99.12% on a self-collected dataset.
• The method provides real-time fatigue detection with high accuracy, surpassing current state-of-the-art approaches.

Conclusions:

• The novel network and facial action analysis method effectively detects driver fatigue in real-time.
• This approach offers a robust solution for complex driving scenarios, enhancing driver safety.
• The system's high accuracy and efficiency make it a promising tool for preventing fatigue-related accidents.