Ultra-Wide Band Radar Empowered Driver Drowsiness Detection with Convolutional Spatial Feature Engineering and...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computer Vision
• Automotive Safety

Background:

• Drowsy driving presents substantial risks, including impaired cognitive function and increased accident potential, leading to severe outcomes like injuries or fatalities.
• Artificial intelligence (AI) offers a promising solution for real-time driver drowsiness detection, aiming to prevent accidents and improve driver performance.
• There is a critical need for accurate and timely drowsiness detection systems to reduce the incidence of fatigue-related road accidents.

Purpose of the Study:

• To develop and evaluate an AI-based system for accurate, real-time detection of driver drowsiness.
• To investigate the efficacy of using ultra-wideband radar data processed through deep learning and machine learning models for drowsiness detection.
• To enhance the performance of drowsiness detection models through data augmentation techniques.

Main Methods:

• Ultra-wideband (UWB) radar data was collected over five-minute intervals and segmented into one-minute image chunks.
• A two-dimensional Convolutional Neural Network (2D-CNN) was employed to extract spatial features from the grayscale radar images.
• Extracted features were used to train and test various machine learning classifiers, including an ensemble model (RF-XGB-SVM) combining Random Forest, XGBoost, and Support Vector Machine.
• Generative Adversarial Networks (GANs) were utilized for data augmentation to improve model accuracy.

Main Results:

• The initial ensemble classifier, RF-XGB-SVM, achieved an accuracy of 96.6% with a k-fold cross-validation score of 97% and a low standard deviation of 0.018.
• Data augmentation using Generative Adversarial Networks (GANs) led to improved accuracy across all tested models.
• The augmented dataset further enhanced the performance of the RF-XGB-SVM model, achieving a final accuracy of 99.58%.

Conclusions:

• The proposed AI approach effectively detects driver drowsiness using processed UWB radar data.
• The ensemble model RF-XGB-SVM, particularly after GAN-based data augmentation, demonstrates high accuracy and robustness for real-time drowsiness detection.
• This research highlights the potential of AI and radar technology in developing advanced driver-assistance systems to significantly improve road safety.