Driver Identification Using Driving Behavior, Habits and Driver Characteristics 2021-01-0185

In this paper, a driver identification scheme is studied using general driver inputs such as accelerating, braking and steering behavior, in addition to the settings related to driver’s physical characteristics, such as driver’s seat position. Several drivers are selected with various ages, genders and driving skills to participate in the study. Their driving data is collected using the same test vehicle, and while driving on the same routes. This helps eliminate the inherent vehicle to vehicle variations and the impact of the route differences and enables the identification algorithm to focus on the driving behavior. The driving routes are broken down into shorter segments where the driving features are calculated and populated in these segments. To reduce the identification bias towards certain rare events in the ride, the features are reset at the beginning of each trip segment. This additionally helps to ensure that there is no spill of feature values across the segments. Trip segmentation also facilitates splitting the dataset into training and testing sets in a way that route diversity is present in both sets. Several classification techniques are studied to learn the driving characteristics of the participants in the experiments. It is observed that ensemble classifiers generally outperform single classifiers as measured by the metric of area under the curve (AUC) of the receiver-operating characteristics (ROC) curve. It is also observed that classification performance can vary noticeably from driver to driver, making it more difficult to distinguish some drivers with similar driving habits. The identification algorithm is also shown driving data of unseen (held-out) drivers and its performance is studied. Finally, it is observed that inclusion of driver’s seat and mirror position that include some of the driver’s physical attributes improves the identification performance significantly.