An Intelligent Multi-Sourced Sensing System to Study Driver’s Visual Behaviors

Abstract

Understanding how changes in visual and attentional behaviors impact driving as we age is still a subject studied by the research community. However, little attention has been paid to using sensing and AI techniques to conduct such studies. We present a multi-sourced intelligent sensing system that infers the visual point of attention (VPoA) associated with five vehicle’s cockpit zones with an accuracy of 98%. The VPoA is inferred from the pitch, yaw, and roll angles of head movements captured with inertial sensors and a facial recognition application. The system also includes a tablet-based application that automatically collects data from the driving context, e.g., speed and location. It also enables an annotator to add observed drivers’ actions, e.g., interactions with a passenger. We conducted a naturalistic study with 15 younger adults and 15 older adults to demonstrate the system’s efficacy to identify visual behavior patterns similar to those identified in previous studies that have used traditional data collection methods. A new finding is that the younger group looks more frequently at the lap than the elderly group, independently if a passenger was present. The Lap was the VPoA associated with using the cellphone.