Real Time Drowsiness Detection System based on ResNet-50

The staggeringly high frequency of traffic/road accidents is well-known. Drowsy driving, which is the risky combination of driving with sleepiness or weariness, is one of the leading causes of such accidents i.e., roughly 100,000 crashes, or about 9.5 percent of all crashes, according to the National Safety Council (NSC). A driver's fatigue can have multiple causes such as lack of sleep, long journey, restlessness, alcohol consumption and mental pressure. Early study in this domain includes detecting the same using Machine learning and multilayer perceptron. This paper proposes a model with increased accuracy, where the cameras are used to capture the facial landmarks. To make it easier to predict drowsiness, those images are made to go through a Convolutional Neural Network (CNN). To proceed, they are collected and thoroughly analyzed for the driver's head and facial expressions in drowsy conditions. Then, using the ResNet-50 transfer learning neural network model, datasets are built and a system to detect and classify drowsiness is developed. Finally, the experiment's conclusion validates the proposed model's efficacy. The proposed model shows an accuracy of 0.9860.