Real time image segmentation and drowsiness detection using deep learning

Driver drowsiness is a paramount of road accidents worldwide. Traditional approaches to detect drowsiness rely on measuring the driver’s vital signs, but these methods have limitations and do not consider the driver’s visual keeping an eye on the road. Recently, machine learning approaches have been proposed to detect driver drowsiness using visual cues such as eye closure, head pose, and facial expression. In this study, we provide a deep learning method to identify driver drowsiness utilizing fully-connected-neural-network (FCNN) for classification, convolutional-neural-network (CNN) for feature extraction. The proposed approach was evaluated using the publicly available Drowsy Driver Detection dataset and achieved an accuracy of 97.5%. The accuracy and memory for the drowsy state were 97.3% and 97.7%, respectively, while the precision and recall for the alert state were 97.6% and 97.2%, respectively. The suggested strategy has the ability to be implemented in real-time systems to alert drivers and prevent accidents caused by drowsiness. Traditional methods of detecting driver drowsiness have relied on measuring the driver’s vital signs, such as heart rate and eye blink rate. However, these methods have limitations as they require specialized sensors, which can be intrusive and uncomfortable for the driver. Moreover, these methods do not consider the driver’s visual attention to the road, which is critical in preventing accidents caused by drowsiness.