Abstract

Due to their diverse driving features, human drivers have different driving skills, expertise, and attitudes. In order to prevent traffic accidents, it is crucial to develop an efficient sleepiness detection algorithm. Driver fatigue has been a major problem jeopardizing road safety. The challenge of identifying abnormal human driving behaviour has been tackled by a number of research projects using computer vision techniques to analyze the driver's frontal face and the dynamics of the vehicle. Nevertheless, complex characteristics of driving behaviour are not captured by the traditional approaches. But since the development of deep learning architectures, a great deal of research has also been done to use neural network algorithms to assess and identify tiredness in drivers. This research presents a novel framework for driver sleepiness recognition based on YoloV5 architectures. In order to extract Region of Interest (ROI), a unique YoloV5 pre- trained architecture is proposed for face extraction. This work presents binary image classification, which are trained and verified using the public dataset and custom dataset, in response to the shortcomings of earlier models. The model's training and validation accuracies were 96.2% and 97.4%, respectively. It achieves 95.5% accuracy. Experiments carried out indicate that our approach has a great deal of potential for real-world use in intelligent transportation systems.