I. Introduction
After three decades of intense research, vision based driver assistance systems have entered our passenger cars and trucks and are a significant pace maker for the progress towards fully automated and cooperative traffic [1]. Like no other sensor, vision sensors—due to the rich information included in images—potentially cover all relevant information that is required for driving. This includes but is by far not limited to lane geometry, drivable road segments, traffic signs, traffic lights, object position and velocity as well as object class. Recent approaches to scene understanding mark the way towards a potential future evolution.