Real-Time Driver Drowsiness Detection and Classification on Embedded Systems Using Machine Learning Algorithms

Abstract

Traffic accidents caused by driver drowsiness are a significant concern. An automatic, contactless device capable of early detection and identification of a driver's drowsy state could greatly enhance their safety. This study presents a real-time driver drowsiness detection and classification system implemented on Jetson Nano and Jetson TX2 embedded systems using machine learning algorithms, including Logistic Regression, Naive Bayes, K-Nearest Neighbors (KNN), Decision Tree, Random Forest, and Multi-Layer Perceptron (MLP). Feature extraction is performed on images obtained from video segments within the dataset, followed by a normalization process. The normalized features are classified using machine learning algorithms, and the results are reported. A 10-fold cross-validation model is employed during the experiment, and the grid search hyperparameter optimization (GSHPO) method is used to fine-tune the classifier algorithms' parameters for the proposed system. The MLP classifier outperformed the other classifiers, achieving an accuracy, F1-score, and AUC (the area under the receiver operating characteristic curve (ROC)) of 0.91, 0.91, and 0.90, respectively. The developed system is implemented on Jetson Nano and Jetson TX2 embedded systems, and the frames per second (FPS) results are provided for comparison. The high accuracy of this hardware-based system in detecting drowsy driving, along with its portability for in-vehicle use, is a critical aspect of this work.