Towards Sub-Maneuver Selection for Automated Driver Identification.

Automated driver identification systems are a vital prerequisite for many application scenarios, e.g., personalized in-vehicle services, and improved safety measures tailored to the individual driving style of the current driver. Existing approaches for driver identification typically consider driving maneuvers, e.g., taking a turn or braking, to determine driver specific-patterns. However, the influence of the temporal sequence of a driving maneuver, e.g., decelerating, turning, and accelerating, is widely unexplored. Moreover, the existing studies on automated driver identification are conducted in artificial environments, e.g., in driving simulators, and use extra sensors, e.g., heart rate monitors. In this paper, we divide the temporal maneuver sequence into so-called sub-maneuvers. We find that choosing the right sub-maneuver can improve the driver identification performance by up to 7 percentage points for braking and turning maneuvers. Furthermore, we use data collected in real-world settings only. We collect realworld data from MOIA, a ride-hailing service operating in Hamburg, Germany. We exclusively extract features from the CAN (Controller Area Network) bus, a de-facto standard for in-vehicle information transports. We believe that these realistic settings can pave the way for deployed automated driver identification systems.