Vision based intersection navigation

Much progress has been made toward understanding the autonomous on-road navigation problem using vision based methods. A next step in this evolution is the intelligent detection and traversal of road junctions and intersections. The techniques presented in this paper are based on a data driven, active philosophy of vision based intersection navigation. Traversal is accomplished by imaging relevant parts of the intersection using a combination of active camera control techniques and a virtual active vision tool called a virtual camera. By monitoring the response of the underlying lane keeping system to the created images, intersections and road junctions can be detected and traversed.