부산대학교 도서관

In this work, we present an approach to estimate route intentions of traffic participants, which can additionally be utilized to infer information about occluded areas from the behavior of visible road users. The approach predicts a traffic scene from a past state to the current time frame while incorporating all combinations of possible road user intentions and uses these predictions in combination with the real observed behavior to update the intention estimation in a Hidden Markov Model (HMM). After showing the suitability of the approach to efficiently estimate the intention of the road users, its ability to infer information about traffic participants in occluded areas is demonstrated. The available context information about visible road users is used to reason about the presence of other, hidden road users. Therefore, the observed scene development is compared to possible hypothetical scene developments induced by the presence or absence of the hidden road users. Our approach considers information about vehicles, trucks, pedestrians and bicyclists, as well as additional regulatory elements like traffic signals or road regulations. We evaluate the approach in multiple simulated and real world traffic scenarios.