A Sensing Architecture Based on Head-Worn Inertial Sensors to Study Drivers’ Visual Patterns ^†

Most studies on driving behaviors use video-cameras and simulators. It involves human observers to code the video data to be later analyzed, which can be a demanding task. We propose a sensing architecture to conduct studies on driving behaviors under naturalistic conditions. It includes smart glasses and a classifier algorithm to infer the vehicle’s cockpit’s spot drawing drivers’ visual attention. Thus, our architecture facilitates annotating the collected datasets with codes corresponding to classes of the cockpit’s spots. We have collected data with the sensing architecture from 15 young drivers to study how glances duration and frequency to cockpit’s spots are correlated with driving speed. Our results suggest that the incidence of drivers’ glances at all spots is less on high-speed roads than in low-speed roads. And that even though participants limited their interaction with the audio system, this is the spot that most eye fixation demanded to interact with.