| Summary: | This research investigates the use of objective EEG analysis to determine multiple levels of sleepiness in drivers. In the literature, current methods propose a binary (awake or sleep) or ternary (awake, drowsy or sleep) classification of sleepiness. Having few classification of sleepiness increases the risk of the driver reaching dangerous levels of sleepiness before a safety system can prevent it. Also, these methods are based on subjective analysis of physiological variables, which leads to lack of reproducibility and loss of data, when a lack of consensus is reached amongst the EEG experts. Therefore, the doctoral challenge was to determine whether multiple levels of sleepiness could be defined with high accuracy, using an objective analysis of EEG, a reliable indicator of sleepiness. The study identified awake, post-awake, pre-sleep and sleep as the multiple levels of sleepiness through the objective analysis of EEG. The research used Neural Networks, a type of Machine Learning algorithm, to determine the accuracy of the proposed multiple levels of sleepiness. The Neural Networks were trained using driving and physiological behaviour. The EEG data and the driving and physiological variables were obtained through a series of experiments aimed to induce sleepiness, conducted in the driving simulator at the University of Leeds. As the Neural Network obtained high accuracy when differentiating between awake and sleep and between post-awake and pre-sleep, it led to the conclusion that the proposed objective classification based on objective EEG analysis was suitable. However, this study did not reach the highest levels of accuracy when the 4 levels of sleepiness are combined, nevertheless the solutions proposed by the researcher to be carried in future work can contribute towards increasing the accuracy of the proposed method.
|
|---|
