Classification of Drivers' Workload Using Physiological Signals in Conditional Automation

The use of automation in cars is increasing. In future vehicles, drivers will no longer be charge the main driving task and may allowed to perform a secondary task. However, they might requested regain control car if hazardous situation occurs (i.e., conditionally automated driving). Performing increase drivers' mental workload consequently decrease takeover performance level exceeds certain threshold. Knowledge about driver's state hence useful for increasing safety vehicles. Measuring continuously essential support driver limit number accidents situations. This goal can achieved using machine learning techniques evaluate classify real-time. To usefulness physiological data as an indicator driving, three signals from 90 subjects were collected during 25 min fixed-base simulator. Half participants performed verbal cognitive induce while other half only had monitor environment car. Three classifiers, sensor fusion levels segmentation compared. Results show that best model was able successfully condition with accuracy 95%. some cases, benefited sensors' fusion. Increasing (e.g., size time window compute indicators) increased windows smaller than 4 min, but decreased larger min. conclusion, study showed high accurately detected conditional based on 4-min recordings respiration skin conductance.