The Journal of Science and Engineering Applications (JSEA)

Abstract

Abstract

Driver fatigue is one of the most serious factors of road accidents on roads all over the world, and real-time monitoring of driving fatigue is an imperative intervention needed to ensure traffic safety. This study describes a real-time driver's fatigue detecting system based on an embedded CNN that processes data from an accelerometer and gyroscope to perform high-accuracy classification of behaviors associated with driver fatigue. It gives a robust solution for real-time analysis with very low computational overhead by taking advantage of the capability of CNNs in extracting spatial features from raw sensor data. In this regard, the model architecture is optimized for embedded systems and hence can work compatibly in resource-constrained environments. Extensive experiments on publicly available datasets prove that the proposed system performs much better in terms of accuracy and reliability compared to traditional machine learning models. These results highlight the potential of embedded deep learning frameworks in scalable solutions to real-world problems that may help avoid the risks of driving in fatigue. The proposed model was compared with advanced architectures, including LSTM, Transformer, and Hybrid CNN-LSTM models. While the proposed CNN achieved superior efficiency (93.8% accuracy and 50ms inference time), Transformer models demonstrated better long-term dependency capture but required higher computational resources. This comparison highlights the trade-offs between efficiency and accuracy in real-time applications.