Constructing personalized situation awareness dataset for hazard perception, comprehension, projection, and action of drivers

Human drivers have dominant control authorities in the Traffic-Driver-Vehicle (TDV) loop. Inferring driver behaviors is beneficial to assess the potential risk and design the humanized decision-making algorithms for Advanced Driver Assistance Systems (ADASs). Within the Intelligent Traffic System (ITS) community, extensive research has been conducted on hazard perception, anomaly detection, trajectory prediction, and intended maneuvers of drivers. Nevertheless, few studies integrated these topics into a multistage cognitive process by reason of the limited application of the theoretical framework. We designed a novel experimental framework based on Situation Awareness (SA) theory, in which perception, comprehension, and projection are believed to be antecedents to decision making. A publicly available dataset, Personalized Situation Awareness of Drivers (PSAD) was proposed. Our dataset consists of 2724 real-world accidental video clips with detailed scenes and driver behavior annotations. Personalized behavior characteristics of six experienced drivers were labeled by individual annotations in a multistage cognitive process and self-reported driving behavior. To the best of our knowledge, PSAD is the first public driver behavior dataset integrating SA framework. It can foster discussions on a better understanding of the cognitive mechanism in critical traffic scenes and provide data reference for personalized collision avoidance strategy of ADASs. The PSAD dataset is available online at https://github.com/Shun-Gan/PSAD-dataset.