For the last few years, a great deal of interest has been paid to crew monitoring systems in order to address potential safety problems during a flight. They aim at detecting any degraded physiological and/or cognitive state of an aircraft pilot or crew, such as visual tunneling, also called inattentional blindness. Indeed, they might have a negative impact on the performance to pursue the mission with adequate flight safety levels. One of the usual approaches consists in using sensors to collect physiological signals which are then analyzed. Two main families exist to process the signals. The first one combines feature extraction and machine learning whereas the second is based on deep-learning approaches which may require a large amount of labeled data. In this work, we focused on the first family. In this case, various features can be deduced from the data by different approaches: spectrum analysis, a priori modeling and nonlinear dynamical system analysis techniques including the estimation of the self-affinity of the signals. In this paper, our purpose was to uncover whether the self-affinity of the pilot gaze direction can be related to his cognitive state. To this end, an experiment was carried out on thirteen subjects in a pilot activity representative environment based on a modified version of the software MATB-II. The scenarios were designed to elicit different levels of mental workload eventually associated to attentional tunneling. A database to train the machine learning step was first created by recording the gaze directions of the subjects with an eye-tracker. The self-affinities of these signals were extracted with the Detrended Fluctuation Analysis method. They constituted the inputs of the classifier. Then, other signals were analyzed and classified. Preliminary results showed promising abilities to detect visual tunneling episodes for different levels of mental workload.
展开▼
