Intention prediction of car drivers at inner city junctions

In this work an approach to reliably predict a driver's intention at a T-intersection is presented. Knowing the intention of oncoming drivers is especially helpful at unregulated intersections. We specifically focus on a scenario where the turning pattern generates a situation in which no driver has the right of way. Our approach utilizes only the trajectories of the vehicles that are involved. These can be observed directly and we therefore do not rely on any explicit communication between vehicles. From the trajectories several features are defined that serve as input for the classification algorithms. We use Random Forests and Conditional Random Fields in this work. To gain behavioral data of human drivers in unregulated situations, a study in a driving simulator was conducted.