Personalized AI-Based Temporal Fatigue Analysis Using Computer Vision and Adaptive Risk Score Computation for Driver Monitoring

This research proposes a Personalized AI-Based Temporal Fatigue Monitoring Framework using Computer Vision, Artificial Intelligence, Machine Learning, and temporal behavior analysis. The framework extracts fatigue-related facial indicators, including Eye Aspect Ratio (EAR), Mouth Aspect Ratio (MAR), blink rate, PERCLOS, and head pose. These features are analyzed continuously over time and compared with a personalized driver baseline to calculate an Adaptive Fatigue Risk Score. The research considers publicly available driver drowsiness datasets, including NTHU-DDD, YawDD, and DROZY. The proposed methodology includes data preprocessing, facial feature extraction, temporal analysis, personalized behavior learning, fatigue risk scoring, and performance evaluation using Accuracy, Precision, Recall, F1-Score, False Alarm Rate, and Frames Per Second (FPS). The framework aims to improve driver fatigue monitoring by accounting for individual behavior and reducing unnecessary alerts. The performance values presented in this paper are illustrative and require validation using final experimental data. Keywords: Artificial Intelligence, Driver Fatigue Detection, Computer Vision, Temporal Behavior Analysis, Adaptive Fatigue Risk Score, Machine Learning, PERCLOS, OpenCV, MediaPipe.