Convolutional neural network-based driver drowsiness detection system

Driver drowsiness and fatigue form the two main causes for road accidents around the globe, affecting more people who fall in the age group 18-45. In this paper, a proposed Driver Safety System (DSS) is aimed at detecting real-time drivers' signs of fatigue. The system captures video at all times of the face of the driver and develops each frame into grayscale images using the HAAR CASCADE algorithm, which is a very reliable object detection tool. MobileNet takes these images to deep recognition and tracking of closed eyes per frame. Inside the decision block, a counter logs the period over which the eyes close that puts a flag on the drowsy driver, raises an alert, and then restarts the counter. For higher accuracy, the DSS integrates other Convolutional Neural Networks (CNN) models that are implemented, such as ResNet50, MobileNetV2, and VGG16. Using pre-trained layers, it enhances the system to more accurately classify distracted behavior. Optimized as a High Precision Low Power (HPLP) prototype, it functions at its best under consistent lighting with a homogeneous background, so there are no reflections or interference from the background. Testing has shown that this approach, based on CNN, significantly outperforms existing techniques with better accuracy. The DSS would process video in real-time and identify fatigue much in advance, thus proving its meaningful contribution to safer driving through timely alert mechanisms.