Trajectory Prediction of Turning Vehicles based on Intersection Geometry and Observed Velocities

Urban intersections are known to be a hotspot for traffic accidents, and understanding of the dynamic traffic situations at intersections can help to prevent injuries. When an ego-vehicle crosses an intersection, predicting the trajectory of oncoming other vehicles is a requirement for Advanced Driver Assistant Systems. In this paper, we propose a method of trajectory prediction of turning vehicles at urban intersections. Trajectory prediction of turn maneuver vehicles is more difficult than straight-maneuver vehicle because a turning vehicle slows down as it approaches the intersections and speeds up as it leaves the intersections. Furthermore, the variation in velocities depends on factors such as an intersection angle, a corner radius. Our method generates a novel desired velocity model for trajectory prediction that takes into account intersection geometry and observed velocities of other vehicles. Specifically, we assume that the velocity becomes minimum at around the crosswalk, and calculate the velocity model by fitting past sequential velocities and estimated minimum velocity to the cubic function. Our method has the advantage of being able to predict the trajectory at any intersection and from any position. The prediction performance of our method in the real traffic scenarios is better than one of other methods.