Publication

Abstract : There has been an alarming increase in the number of accidents that occur due to drowsiness. Research so as to analyze and rectify this state has become a necessity. Here, a system has been developed which would detect drowsiness with sufficient reliability using electroencephalogram (EEG) signals. The study was conducted on nine males and nine females, whose EEG waveforms are obtained with the help of a wireless neuro-headset, while subjects underwent a virtual driving experiment. Preprocessing of the raw EEG signals is done by Principal Component Analysis (PCA). The resultant signals are then normalized and smoothed. Subsequently, the preprocessed signal is segmented and the time-frequency features are extracted using Continuous Wavelet Transform (CWT). Transfer learning approach is adopted in this paper to discriminate between the 3 states, ‘Drowsy’, ‘Asleep’ and ‘Awake’. An existing pre-trained ResNet50 model, fine-tuned using the scalograms obtained from CWT, is utilized for evaluating the state of driver