Evaluation of a Multi-Sensor Approach for Drowsiness Monitoring

For the past few years, a significant interest has been paid to crew monitoring systems to enhance flight safety. One of the key aspects is assessing the level of drowsiness of crew members. To address this, some airlines adopt a prescriptive approach through the use of biomathematical models, while others consider testing solutions based on analyzing the crew's neurophysiological signals. In this paper, our contribution is threefold. First, we introduce a novel drowsiness monitoring system based on cardiac and ocular analyses. It is autonomous, objective and provides a non-invasive and continuous monitoring. Second, we present a protocol alongside experiments designed to test this system. Third, we conduct an in-depth analysis of the performances of the aforementioned algorithm compared to mono-sensor approaches on the collected test dataset. The results demonstrate that, on the test dataset, the multi-sensor model offers promising performance and correlation with reference drowsiness metrics, with 75% accuracy and 0.88 Area Under Curve in a three classes classification and a correlation coefficient of 0.57, outperforming mono-sensor models on the same dataset.