Detection of Alcohol Inebriation from Eye Movements

Abstract

Today, the most convenient way of estimating an individual's blood-alcohol concentration requires a breathalyzer device and intense user cooperation, which severely limits the scope of potential applications. We develop and study a machine-learning model that detects alcohol inebriation based on a person's eye gaze and eye closure. We investigate the relative contribution of individual features derived from eye gaze and eye closure to the model. In order to train and experimentally evaluate the model, we collect— and share—a new data set with participants in baseline and alcohol-intoxicated states. We find that the model can in fact detect the consumption of a moderate amount of alcohol; the accuracy grows significantly with increasing blood alcohol concentration. The most relevant features turn out to relate to the velocity and acceleration profiles during fixations and saccades. From our proof-of-concept study, we can conclude that contactless inebriation detection based on eye gaze is in fact possible, albeit data need to be collected on an industrial scale to reach practical applicability. Potential applications of contactless inebriation detection include the detection of impaired drivers or operators of other hazardous machinery as well as health-monitoring applications.