R2DS: A novel hierarchical framework for driver fatigue detection in mountain freeway

Area of Science:

• Computer Science
• Engineering
• Transportation Safety

Background:

• Driver fatigue is a significant safety concern on mountain freeways.
• Existing fatigue detection systems often lack a balance of accuracy, speed, and robustness.
• Practical applications require comprehensive solutions addressing these key performance metrics.

Purpose of the Study:

• To propose a novel, real-time, and robust fatigue detection system for drivers.
• To enhance driving safety, particularly in challenging environments like mountain freeways.
• To develop a system that integrates accuracy, speed, and robustness in fatigue detection.

Main Methods:

• A three-layered framework: facial feature extraction, eye region extraction, and fatigue detection.
• Utilized a deep cascaded convolutional neural network for face and eye key point detection.
• Employed face tracking, validation sub-modules, and a finite state machine for efficient and stable operation.
• Applied ellipse fitting for pupil shape analysis and PERCLOS for fatigue determination.

Main Results:

• Achieved a comprehensive fatigue detection accuracy of 95.87%.
• Demonstrated a high processing speed with an average rate of 32.29 ms/f for 640x480 pixel images.
• The system effectively extracts facial features, eye regions, and determines driver alertness.

Conclusions:

• The proposed Real-time and Robust Detection System offers a significant advancement in driver fatigue monitoring.
• The framework's integration of speed, accuracy, and robustness addresses limitations of previous systems.
• This research provides a foundation for next-generation driver fatigue detection systems, improving safety on mountain freeways.