<![CDATA[Drowsiness detection systems play a crucial role in safe driving, considering the high rate of traffic accidents caused mainly by drowsiness. Several drowsiness detection systems built using the eye aspect ratio (EAR), percentage of eyelid closure (PERCLOS), and convolutional neural network (CNN) methods still have limitations in terms of accuracy and response time. This study aimed to overcome these problems by applying the You Only Look Once version 9 (YOLOv9). This method has advantages in terms of speed and accuracy because it can detect objects in real-time in one processing stage. The dataset was collected independently from several sources in a real environment inside the vehicle with various lighting and viewing angles; then, labeling, preprocessing, and modeling processes were conducted. The model performance was evaluated based on precision, recall, F1 score, and mean average precision (mAP) metrics. The best model was optimized using several optimization techniques to determine the most optimal results. The results indicate that the YOLOv9 model trained using Nesterov-accelerated adaptive moment estimation (Nadam) optimization has a better image processing speed than other models. This model yielded a precision, recall, F1 score, mAP@50, mAP@50, mAP@50-95, and processing speed of 99.4%, 99.6%, 99.5%, 99.5%, 85.5%, and 52.08 FPS, respectively. The developed model can detect drivers’ drowsiness signs, such as closed eyes, yawning, abnormal head positions, and unnatural hand movements, in real time. However, this model still has limitations in detecting drivers wearing sunglasses, so further development is needed to improve its performance in these conditions.]]>
Copyrights © 2025
