Detection of Intoxicated Drivers Using Online System Identification of Steering Behavior

Impaired driving is known to be among the leading causes of death and injury on roads; however, the existing measures to address this menace appear to be insufficient. This paper presents a novel method to detect intoxicated driving and lays a foundation that can be implemented in future cars to derive personalized models of drivers and to detect not only intoxicated driving but also other reckless driving styles. We employ system identification techniques to develop models for sober and impaired drivers. A total of 200 sets of data from various subject drivers were collected in a high-fidelity driving simulator. The lateral preview error and the steering wheel angle were considered the input and output of a driver, respectively. We will demonstrate that the autoregressive noise integration moving average with exogenous input (ARIMAX) model best fits the data to describe the steering behavior of drivers. The positions of model poles are shown to be a good indicator of intoxicated driving behavior. An aggressive driving style due to impaired driving leads to the migration of dominant poles toward the instability region. The Kalman filter and online identification techniques are used to update the driver model during driving. The poles of this updated model are used for the detection of impaired driving.