AIoT-based drowsiness level detection system using a hybrid F-LSNN network

Drowsiness while driving is considered one of the leading causes of traffic accidents worldwide. This behavior not only results in property damage but also leads to permanent injuries or even threatens the lives of drivers and other road users. Furthermore, drowsy driving can occur silently, often going unnoticed by drivers until they are startled awake by external stimuli or warning sounds. To address this issue, various studies have focused on developing safe drowsiness detection and warning systems by collecting data on drowsiness states and signs, combined with artificial intelligence algorithms, sensors, and IoT devices. Aiming to build a real-time system with minimal latency and improve accuracy over previous research, we have developed a fuzzy deep neural network named F-LSNN, which is a hybrid network based on neuro-fuzzy and long short-term memory for extracting the hidden features of drowsiness from the images, ensuring that only relevant features are used for classification. This approach relies on drowsiness signs and scales to determine drowsiness severity. In addition, we integrate an emotion detection model to enhance accuracy and reduce latency in real-time performance. The paper also introduces an AIoT system integrated with the proposed model, enabling the automatic handling of sensor and software failures. Moreover, the study evaluates model performance by comparing it with advanced deep learning architectures, including Vision Transformer, Fuzzy Deep Neural Network, and Long-Short Term Memory (LSTM). Experimental results demonstrate that the F-LSNN method can achieve an accuracy of up to 99% while reducing prediction time by half compared to previous studies. The proposed method highlights the potential of the system to contribute to the development of safe driving systems and intelligent transportation systems in the future.