Infrastructure-Based Vehicle Maneuver Estimation with Intersection-Specific Models

Advanced Driver Assistance Systems have the potential of making traffic safer, more efficient and convenient. At urban intersections the conflicting intentions of different traffic participants frequently lead to accidents. In order to avoid dangerous situations in such a scenario an ADAS requires an accurate estimation of future driving maneuvers executed by other vehicles. In this work we present a framework that estimates the driving maneuver of vehicles approaching at intersections where the vehicle state can only be observed externally. The high variance in driver behavior that arises from different intersection layouts is met with intersection-specific models that are computed locally. The performance of the estimator and the potential of intersection-specific models are validated with data recorded from real intersections.