Model-based recognition of intersections and lane structures

In the course of tracking moving vehicles in image sequences recorded by a stationary camera at complex inner-city road intersections, it has turned out to be advantageous to automatically recognize the lane structure of the recorded intersection. Similarly, in the context of vision-based automatic driving it is advantageous to rely on as much knowledge about the actual road and lane structure as can possibly be obtained. Based on the authors´ previous research, they assume that knowledge about the type of intersection ahead of the vehicle is made available by access to an automatic navigation system which is based on a digital map of the road network. In preparation of an appropriate selection and instantiation of a generic lane structure model, the authors use model-based machine vision in order to estimate the location of incoming as well as outgoing lanes at an intersection. The authors use image sequences recorded from a moving vehicle in order to detect and track intersections using a Kalman filter. Results obtained from real world data are presented