Drivers Awareness Evaluation using Physiological Measurement in a Driving Simulator

Increasing the road safety requires monitoring drivers' behaviour and evaluating their awareness. Low awareness related crashes have significantly increased in recent years due to the augmentation of social media and driver assistance systems. Accordingly, an advanced system is required to monitor the driver's behaviour and generate warning alarms if driver's performance degradation is detected. This study aims at evaluating the vehicle and driver's data to determine the performance of drivers the onset of degradation. Physiological signals such as perinasal and palm electrodermal activities, heart rate and breathing rate are measured during the simulated driving. Measurements are coming from healthy subjects (male/female and elderly/young). The lane deviation of the vehicle is treated as the response variable whether driver is impacted by stressor or not. Measured physiological signals are then processed and applied for developing machine learning tool for driver's performance evaluation. A mix of linear and non-linear classification algorithms is used for this purpose. Prediction results indicate that the random forest algorithm outperforms other methods by achieving an area under the curve of 0.92%. Its performance remains quite stable and consistent in multiple simulations. Also, it is shown that perinasal perspiration is the most informative feature.