A novel machine learning framework for drowsiness detection using an electrostatic wearable sensor and hyperdimensional computing

This work proposes a wearable system for the specific application of drowsiness detection in activities of daily living. The system is based on a new family of electrostatic sensors (Qvar) supported by smart algorithms. It records the electroencephalogram (EEG) from a single-channel and extracts a number of features from the spectral bands of the EEG signal, in addition to entropies. An innovative approach based on the emerging brain-inspired Hyperdimensional Computing paradigm is used and compared with two conventional Machine Learning algorithms. The algorithms are all trained on a public dataset and tested on our data recorded by the Qvar system. As a result, the new algorithm outperforms consistently the conventional ones, while being computationally more efficient and highly parallel. Thanks to the single channel implementation and the low-power consumption, the Quar system candidates for ubiquitous real-time EEG monitoring, ensuring minimal invasiveness and long-time autonomy, and paves the way to facilitate extensive screening processes, and expand the range of applications for EEG technology.