| dc.description.abstract | In most real-time scenarios, it is highly essential to evaluate the level of hypovigilance in drivers, pilots, security guards and sportsmen to ensure efficient performance in their work. Driver hypovigilance is one of the major causes for road accidents. Drowsiness and distraction are two major components of hypovigilance. Most real time detection systems use the conventional classifier based approach to distinguish different levels and proposed the measurement index to differentiate lower numbers of hypovigilance levels (such as drowsy versus awake, distraction versus neutral, etc.). Furthermore, existing detection systems are bulky and costly. In this thesis electroencephalogram (EEG) signal analysis in the time-frequency domain with application to driver hypovigilance recognition is developed. Five stimuli are considered for devolving an intelligent hypovigilance detection system. 14 wireless multi channel EEG are placed over the entire scalp through international 10-20 system. The EEG dataset is developed with 50 subjects (43 males and 7 females) and the discrete hypovigilance states (neutral, low distraction, medium distraction, high distraction, awake, drowsy, high drowsy, sleep stage 1) are evoked by using audio and visual stimuli (media player, GPS, mental think, SMS, and driving for 1hour). Distraction, sleepy, and hypovigilance indices were derived from EEG frequency bands; delta (0-4Hz), theta (4-8Hz), alpha (8-12Hz), and beta (14-32Hz) using hybridization of Discrete Wavelet Transform (DWT) and FFT. Two statistical features (Spectral centroid (SC), Power Spectral Density (PSD)) derived from EEG frequency bands using different wavelets (db4, db8, sym8, and coif5) are used to classify the hypovigilance levels using three classifiers namely; Probabilistic Neural Network (PNN), K Nearest Neighbour (KNN) and subtractive fuzzy classifier. As a result of this study, the average of distraction and sleepy index detection rate were 88.75% and 85% respectively, both based on db4. On the other hand, subtractive fuzzy classifier based distraction and drowsiness achieves maximum classification rate of 79.21% based on sym8, and 84.41% based on db4, respectively. The embedded system (TS7800) has been used in this research as real time hypovigilance detection system based on hybridization of discrete wavelet transform and fast Fourier transform. The conventional filter bank method had been investigated and compared with hybrid method. The Results of this research indicated that db4 based hypovigilance detection system gave best average detection rate using index method. While classification method showed that sym8 and db4 gave high accuracy results when they were applied for features extraction of distraction and drowsiness states, respectively. The output of this thesis is the newly distraction, drowsiness, and hypovigilance indices obtained by hybridization of DWT and FFT, in addition to hardware real time detection system based on embedded system. | en_US |
