A marking-based, Flexible approach to intersection detection

This paper describes a flexible, marking-based approach to intersection detection. Edge contours are extracted from an image acquired with an onboard camera. The contours are mapped to the flat earth model so as to remove the effect of the perspective mapping. Using properties like spatial context, orientation and length, the individual markings are grouped to contiguous straight lines and finally to lanes. The result is a description of an intersection’s geometry without any assumptions about the number, width and orientation of lanes.