Extraction of Risk Knowledge from Time To Collision Variation in Roundabouts

Abstract

Roundabouts are intersections which require understanding the intentions of other road users to be crossed safely. In this paper, we investigate on the nature and variation of driving risk in roundabouts, to allow connected vehicles to quickly assess a personalized and real-time level of risk associated with crossing a roundabout. First, Time To Collision (TTC) information is extracted from real roundabout vehicle tracks. Then, a supervised machine learning model to assess the probability for a given vehicle to exit the roundabout at the next available exit is trained. Finally, a risk metric is defined based on TTC thresholds and risk probability, which is found to show a strong correlation with the coefficient of variation of TTC values over a roundabout. Once integrated in knowledge-centric frameworks such as Vehicular Knowledge Networking, the obtained risk knowledge has a potential to support driver assistance systems in roundabouts.