Longitudinal Risky Driving Behavior Recognition Based on Naturalistic Driving Study

摘要

Traffic safety is a primary concern for drivers. Risky driving behavior recognition is essential to reduce the accident risk for drivers by reminding them to improve their driving behavior or driving style. In this research, a field naturalistic driving study has been conducted using a car-following system. The lead car was controlled automatically and performed consistently throughout the experiment, and the volunteer drivers were asked to drive naturalistically to follow the leading one. The operation data, including speed, acceleration, position of pedal, headway, etc., for each driver were recorded together with the leading car's traveling data simultaneously. In order to reveal risky driving behavior, a driving stability estimation method based on car-following model was adopted in practice, and compared with time to collision (TTC) estimation method accordingly. We found that driving stability analysis could monitor the real-time reaction time and sensitivity of a driver, which may provide a risk alarm in a more proactive way, and TTC would link to collision risk immediately. This study can help drivers to understand and improve their driving behavior, avoid collision during driving. In addition, it can also shed light on user-based insurance (UBI) pricing or driver training and management.