Object tracking in urban intersections based on active use of a priori knowledge: Active interacting multi model filter

The perception of moving objects in urban environments poses new challenges for sensors and tracking algorithms. Tracking algorithms must estimate the movement of a great number of objects simultaneously while taking into account longitudinal and crossing, as well as turning traffic. In this process, the early recognition and stable tracking of a turning maneuver is a particular challenge. We present an efficient tracking algorithm that faces this task based on laser scanner data. It uses three different dynamic models for each object simultaneously: One for an object driving straight ahead, one for turning left and one for turning right. We use a priori knowledge to recognize the objects' maneuvers early and to decide which of the three dynamic models is currently appropriate. Based on the objects' current positions in the intersection, we determine possible turning maneuvers and use them actively in the filtering process. This method enables an early recognition of the object movement and a stable tracking of longitudinal, crossing, and turning traffic.