Detection of Distracted State Based on Head Posture and Facial Expression

Research and development of automatic driving have been progressing actively. Level 3 autonomous driving requires shifting of driving activity between the system and a driver. Such shifting poses high risks of a severe traffic accident if a driver is in a distracted state. Nevertheless, no distracted state detection method has been established. For our earlier study, research was conducted to extract characteristic driving behavior patterns in distracted states by quantifying eye movement. Results indicated a challenging task for further development: detecting gaze information with high accuracy. For this study, time-series changes of head posture and facial expressions in a driving concentration state and distracted state are quantified using hierarchical growth type recurrent SOM and a U-matrix. We assess the possibility of detecting driving behavior patterns that involve head posture and facial expressions and which characterize a distracted driver state.