Rosebrugh Bldg, Toronto, ON M5S 3G9
Room: RS 211
In Ontario, 20-25% of all car accidents are related to drowsy or fatigued driving, making it one of the top five causes of accidents on Ontario’s roads. Existing systems for drowsy driving detection are typically expensive, dependent on various parameters such as vehicle type, lighting conditions, or road geometry, or are often inconvenient for the driver. Electroencephalogram (EEG) is commonly used to detect drowsy driving. However, most studies based on EEG employ segments that are 30s long, which is too long to detect drowsiness, especially in the context of drowsy driving detection. Therefore, a reliable and high-resolution drowsiness detection system is yet to emerge. This study proposes an efficient and high-resolution EEG-based algorithm to detect drowsiness. At sleep onset, alpha (8-13 Hz) and beta (13-30 Hz) band powers of EEG decrease and delta (1-4 Hz) increases. In this study, we employ EEG data from an overnight polysomnography study to extract relative power features of the three bands- alpha, beta, and delta from 3s long signal segments. Subsequently, these features are used to develop a sigmoid function based wake probability model, which determines the probability of wakefulness of each of the 3s long EEG segments. The sigmoid wake probability model is validated using various metrics and arousal and deep sleep segments. Upon its successful validation in a driving study, the proposed model will lead to the development of a high resolution, convenient, and accurate drowsiness detection system based on wearable EEG headbands.