Driver Fatigue Detection using CWT-Extracted Features and a Deep Learning Approach (CNNLSTM)

Driver fatigue is a leading reason behind traffic accidents globally, resulting in significant threats to public safety and substantial economic costs. Electroencephalography (EEG) has shown to be a critical tool for identifying driver fatigue as it can record brain activity associated with sleepiness, which gives it an advantage over other physiological modalities. Although raw EEG data can provide insightful information, accurate fatigue identification requires strong feature extraction techniques because of the inherent complexity of the data. This emphasizes how urgent it is to investigate revolutionary deep-learning architectures that can successfully extract discriminative features from raw EEG data. This study provides an innovative framework for driver fatigue identification from EEG that combines continuous wavelet transform (CWT) with convolutional neural networks (CNN) and long-short term memory (LSTM). In order to gain the advantage over the drawbacks of manual feature extraction, this study generates time-frequency spectrum representations of EEG data using the CWT. Following extracting information, each channel’s time-frequency images are concatenated and fed into a CNN-LSTM architecture. This combination models the temporal and spatial characteristics of the EEG data and automatically learns discriminative features for identifying drivers’ normal and fatigued states. A publicly available EEG dataset with recordings from twelve subjects is used to evaluate the proposed CWT-CNN-LSTM architecture. The result shows a prominent classification accuracy of 98.34% for both the average of each subject and combined subjects. These findings show how well the CNN-LSTM framework captures EEG patterns associated with fatigue, which may result in more reliable driver fatigue detection systems and improved traffic safety.