Non-Invasive Automatic Drowsiness Detection using Independently Recurrent Neural Network

Drowsiness is a critical factor contributing to various accidents, particularly in traffic and occupational settings. To mitigate this issue, we propose an algorithm for automatic drowsiness detection. The algorithm begins with face detection, followed by the identification of facial landmarks and the computation of the Eye Aspect Ratio (EAR) to detect blinks. Blink features are then extracted and preprocessed, and the sequences are classified to determine drowsiness. For the classification of these temporal blink sequences, we utilize the Independently Recurrent Neural Network (IndRNN), an advanced Deep Learning technique derived from Recurrent Neural Networks, designed specifically for temporal data processing. This approach leverages the IndRNN's ability to effectively handle long-term dependencies in sequential data. The performance of the algorithm was evaluated using video data from the University of Texas at Arlington Real-Life Drowsiness Dataset (UTA-RLDD). The results demonstrate that the IndRNN achieved a classification accuracy of 91.88% for temporal blink data.