Detecting Driver Cognition Alertness State From Visual Activities in Normal and Emergency Scenarios

Current driving behavior studies have limits in obtaining driver's state information, and recent studies involving driver state focus on driver distraction or inattention. But it is more common that drivers operate in an intermediate subconscious state, where the drivers are neither cognitively fully focused nor distracted. There was little research to address this topic. In this study, the driver cognition alertness state information, which indicates driver subconscious alertness, is detected from eye and iris activities by non-contact computer vision methods. And a novel analysis is conducted and reveals the strong correlation between driver performance and driver cognition alertness state from experiment results. In detail, the driver cognition alertness state is quantified by the proposed metric-iris movement index, which is calculated from iris-eye relative displacements. The developed computer vision method produces high precision results in detecting faces, eyes, and irises in the experiment based on multiple deep learning networks applied in cascade, and enables displacement tracking of eye and iris targets. A filtering method is proposed and removes artifacts due to eye blinks in displacement measurements. The driving performance is estimated by a proposed performance evaluation method in a series of traffic scenarios designed with normal and emergent conditions.