Mental workload (MW) refers to the level of mental effort required to accomplish multiple concurrent tasks. Due to its correlation with the risk of traffic accidents, the assessment of MW is crucial for Advanced Driver-Assistance Systems. In this study, participants were given two cognitive tests (the Digit Span Test, DST, and the Ray Auditory Verbal Learning Test, RAVLT) while driving in a simulated environment. The tests were chosen to elicit different levels of MW from the drivers based on the task's difficulty. Electroencephalography (EEG), functional near-infrared spectroscopy (fNIRS), and heart rate variability (HRV) were used to assess features indicative of the psychophysiological states of the participants in order to feed a Support Vector Machine (SVM) classifier. Multimodal EEG + fNIRS + HRV feature-based classifiers performed better with respect to SVM fed by unimodal features (DST: 69.4% accuracy, RAVLT: 85.5% accuracy). These results showed the possibility of assessing the MW levels of drivers with high accuracy through multimodal physiological acquisitions, which could be easily implemented in ADAS applications, driving competitions, and automotive cockpit neuroergonomics.
Drivers' mental workload assessment based on machine learning applied to multimodal physiological data
Perpetuini D.Primo
;Cardone D.Secondo
;Tiberio A.;Merla A.Ultimo
2023-01-01
Abstract
Mental workload (MW) refers to the level of mental effort required to accomplish multiple concurrent tasks. Due to its correlation with the risk of traffic accidents, the assessment of MW is crucial for Advanced Driver-Assistance Systems. In this study, participants were given two cognitive tests (the Digit Span Test, DST, and the Ray Auditory Verbal Learning Test, RAVLT) while driving in a simulated environment. The tests were chosen to elicit different levels of MW from the drivers based on the task's difficulty. Electroencephalography (EEG), functional near-infrared spectroscopy (fNIRS), and heart rate variability (HRV) were used to assess features indicative of the psychophysiological states of the participants in order to feed a Support Vector Machine (SVM) classifier. Multimodal EEG + fNIRS + HRV feature-based classifiers performed better with respect to SVM fed by unimodal features (DST: 69.4% accuracy, RAVLT: 85.5% accuracy). These results showed the possibility of assessing the MW levels of drivers with high accuracy through multimodal physiological acquisitions, which could be easily implemented in ADAS applications, driving competitions, and automotive cockpit neuroergonomics.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
