Deep Learning Approaches for Drowsiness Prediction and Detection

Road accidents in India are an escalating issue, predominantly caused by drowsy driving, posing risks to individual safety and the well-being of fellow travellers. While there have been advancements in detecting driver drowsiness, a system that predicts drowsiness holds greater potential for real-world conditions. This study introduces a novel approach to address this critical issue by monitoring facial movements in real time to not only detect but also predict drowsiness in Indian drivers. The approach involves extracting and computing derivative features from facial analysis, which are then utilized to train various deep learning models for precise driver drowsiness prediction. Optimization is carried out using the UTA real-life drowsiness dataset, specifically on Indian subjects. Tested across a diverse dataset of Indian drivers, the method exhibits notable accuracy in drowsiness prediction. The results found that the decision tree model outperforms both multi-linear regression and lasso regression models, achieving the highest R-squared value of 93.65% and the lowest error metrics