Fizyolojik sinyaller kullanarak sürücünün stres seviyesinin simülasyon tabanlı analizi

The aim of this thesis study is to detect driver's stress with the help of physiological signals. A racing game experiment is designed to understand stressors on the road. One subject played five different levels of a racing game while wearing ECG and EDA sensors. By using the knowledge gained from the racing game experiment about road stressors, a driving simulation was implemented. Four subjects tried to follow a specific route in the simulation with the ECG and EDA sensors attached. The same four subjects also played a simulation game called City Car Driving wearing always the physiological sensors. Using the physiological signals and machine learning algorithms, stress detection was made for all three simulation experiences: The Racing Game Experiment, Simulation Experiment, and City Car Driving Experiment. Also, a publicly available drivers' physiological signals dataset called DriveDB was used for stress detection. The following accuracy rates of stress detection were obtained: 70.77% for the Racing Game Experiment, 86% for the Simulation Experiment, 85% for the City Car Driving Experiment. The analysis of DriveDB dataset yield an accuracy of 69% for detecting low, medium and high stress levels (3 level estimation) vs 91% for the binary stress level detection (stressed vs non stressed).