Lützeler and Dickmanns 1998, Abstract: The MarVEye camera arrangement allows a new kind of road recognition differentiating between the near and the far range also in model building. The wide simultaneous field of view resulting from the oblique (skewed) view of two wide-angle cameras has many advantages for tight maneuvering; in addition, it allows to separate ego-state recognition relative to the road nearby from road curvature further away.
   Experience with bifocal vision in public traffic has shown that both a wide simultaneous field of view (beyond 100° if possible) and a large look-ahead range (to several hundred meters, if possible) are favorable if performance levels close to those of human drivers are looked for. The corresponding arrangement of a set of multi-focal conventional cameras, dubbed MarVEye, allows to adapt road recognition to special regimes and to introduce redundancy as a new feature, both desirable properties for practical systems. Medium and far range road curvature recognition is achieved by actively controlled tele-cameras and area-based image processing geared to photometric properties of the road surface.

Introduction: A reduced 2-D visual servoing solution has been derived by [7]. Behringer perfected the recursive road recognition approach with the following additional features: Recognition of several lanes, kind of lane markings (solid or broken), lane width adaptation with a linear rate of change constraint, and vehicle pitch eigen-motion with a damped oscillation and a separately modeled drift component for load changes. All these models moved with vehicle center of gravity; Behringer also studied geographically fixed coordinates with piecewise constant clothoids. The points of discontinuity have to be recognized together with the parameters of the road from an adapted concept of generalized likelihood [8].
   Since mid-90s sufficient computing power has become available for area based real-time image processing with several cameras simultaneously; this allows to differentiate image processing according to the type of task at hand....