Non-Invasive Driver Drowsiness Detection System.

Area of Science:

• Automotive Safety
• Biomedical Engineering
• Signal Processing

Background:

• Driver drowsiness significantly impairs cognitive function, increasing the risk of severe road accidents.
• Current driver monitoring systems often rely on intrusive methods, necessitating non-invasive alternatives.
• Real-time drowsiness detection is crucial for mitigating accident rates and their associated consequences.

Purpose of the Study:

• To classify driver drowsiness states using respiration rate detection.
• To evaluate the efficacy of non-contact, impulsive radio ultra-wideband (IR-UWB) radar for this application.
• To establish a benchmark for UWB radar in driver fatigue monitoring.

Main Methods:

• Acquired chest movement data from 40 subjects using a lab-based IR-UWB radar system at a 5m range.
• Extracted respiration per minute from the radar signals to create a labeled dataset.
• Trained and compared various machine learning models, including Support Vector Machine (SVM), Decision Tree, and others.

Main Results:

• The Support Vector Machine model achieved the highest classification accuracy of 87% for distinguishing drowsy from non-drowsy states.
• Respiration rate, derived from IR-UWB radar, proved to be a viable indicator of driver drowsiness.
• The dataset comprised respiration rate, age, and drowsiness labels for model training.

Conclusions:

• Impulsive radio ultra-wideband (IR-UWB) radar can effectively detect driver drowsiness through respiration monitoring.
• This non-contact method offers a promising avenue for developing advanced driver assistance systems.
• Further research can leverage this ground truth for validating UWB radar in real-world driver safety applications.