An Efficient Approach of Drowsiness Detection Based on ResNet50 and Xception Architectures

This paper delves into a critical aspect of road safety by investigating a method to detect drowsiness in images. The accurate utilization of this technology by fatigued drivers holds the potential to prevent numerous fatal accidents, as the proposed models exhibit swift responsiveness in identifying instances of driver drowsiness. Within the realm of eye closure detection, two drowsiness models, namely ResNet50 and Xception, are employed to portray various eye state classifications. Nonetheless, the scarcity of readily available and reliable eye datasets poses a formidable challenge. Leveraging the power of ResNet50, deep facial features are extracted, resulting in an impressive accuracy of 98.06%. Similarly, the utilization of Xception yields a commendable accuracy of 95.69% when evaluated with the MRL dataset. These experimental outcomes firmly establish the higher accuracy and cost-effectiveness of eye closure estimation, thereby underscoring the efficacy of the proposed framework for drowsiness recognition.