Developing vehicle-based advanced warning system for driver drowsiness based on a hybrid algorithm

Publication Type:
Thesis
Issue Date:
2010
Full metadata record
Files in This Item:
Filename Description Size
Thumbnail01Front.pdf13.22 MB
Adobe PDF
Thumbnail02Whole.pdf157.58 MB
Adobe PDF
Fatigue is a major public health issue causing substantial emotional and financial burden on society. Driver fatigue is identified in nearly 20-30% of road fatalities, and can cost around AUD 3 billion per year. Providing drivers with early warning systems for fatigue could minimise fatigue-related road accidents. A car driving simulator study was conducted and physiological data such as electroencephalography (EEG), eye activity, movement sensor data, video and questionnaire information were obtained for the purposes of developing a drowsiness detection algorithm. The study was conducted at the Monash University Accident Research Centre (MUARC) where sixty non-professional drivers aged between 20- 60 years were recruited. The study was conducted in the afternoon and the driving sessions lasted up to 3 hours of monotonous day and night driving scenarios with realistic scenery. The preliminary analysis identified sections of data where clear episodes of drowsiness were evident. The analysis revealed that it was possible to detect drowsiness from a combination of physiological signals consisting of EEG, car seat movements and eye activity. Once the association between episodes of drowsiness and various signals were established, statistical analysis was performed on the entire data set. Two types of EEG processing were e1nployed at this stage based on EEG alpha power and alpha burst analysis. A significant association was established between the probability of drowsiness and EEG alpha activity, with alpha burst duration resulting in a better association. Drowsiness detection algorithms based on these two methods were then developed. The association established between drowsiness and the seat movement signals was far less than that between drowsiness and the alpha signals. The seat movement signals were then combined with both methods of alpha analysis. Adding seat movement signal to either of the two EEG methods resulted in improved associations with drowsiness with alpha burst association still being superior. The algorithm based on the combinations of alpha burst and seat movements formed the basis for the new hybrid algorithm. Subjective measures of drowsiness, lifestyle and behaviour were also examined in this research and validated against video ratings of fatigue. It was shown that increased anxiety, anger and an unhealthy diet were associated with an increased probability of drowsiness. The findings of this research can serve as a foundation for designing future vehicle-based fatigue countermeasure devices as well as highlight potential difficulties and limitations. Such driver fatigue studies will also benefit from further investigations of driver lifestyle and behavioural factors.
Please use this identifier to cite or link to this item: