Omerbegovic, Said (Technische Universität Darmstadt, Germany)
Firl, Jonas (Adam Opel AG, Rüsselsheim, Germany)
Advanced driver assistance systems require accurate knowledge on the current and future states of all surrounding vehicles. Current approaches use the information on actual vehicle dynamics and a motion model, leading to short-time pre-dictions. In this work, we developed a framework to determine long-term predictions of the future vehicle trajectory for extra-urban traffic scenarios in dependence on the present situation. Scene analysis is used to yield the probability of occurrence on a set of maneuvers. Trajectory prediction is accomplished by an approach of Case-based reasoning. Based on the current observation, similar trajectories are retrieved from a database, resulting in long-term predictions for a given maneuver. The information of scene analysis is combined with these trajectory predictions to yield the final situation-adaptive long-term vehicle trajectory prediction. The performance of our method is verified with a comparison to conventional approaches in terms of error analysis over the prediction horizon.