Driver Drowsiness Detection Using Facial Temporal Features and Sequence-Based Deep Learning Models

Driver drowsiness is a major factor contributing to road traffic accidents, motivating the development of reliable and non-intrusive monitoring systems. This paper presents a visionbased driver drowsiness detection framework that combines facial behavioral features with temporal deep learning models. Facial landmarks are extracted using MediaPipe, and fatiguerelated features are organized into short temporal sequences to capture driver behavior dynamics. Several architectures, including LSTM, BiLSTM, GRU, CNN–LSTM, and Transformer models, are evaluated on two public benchmark datasets, UTARLDD and NTHU-DDD, using a subject-independent protocol. Experimental results show that the Transformer-based model achieves the best performance, reaching accuracies of 96% and 92.8% on UTA-RLDD and NTHU-DDD, respectively. These findings demonstrate the effectiveness of combining handcrafted facial features with temporal learning for robust driver drowsiness detection.