Road recognition from multifocal vision

The system for visual autonomous vehicle guidance which has been developed at UniBwM, has been installed into VaMoRs-P, a Mercedes 500 SEL sedan passenger car. To meet the requirements implicated by high cruising speed (up to 130 km/h) and low camera position, some changes in the design of the road detection and tracking algorithms had to be implemented. For robust visual road tracking and for reliable estimation vehicle position and heading parameters as well as of the road course, the maximal look-ahead range must be about 100 m. The 4D approach for estimation of the vehicle ego-state and the road course parameters by Kalman-filter techniques has been extended to a road model which is segmented into sections. These segments are modelled to be locally fixed, which leads to a better separation of the vehicle´s ego-motion and the road course dynamics. Rearward viewing has been added to the already existing bifocal forward vision. In this paper, the basic concepts of the road recognition module are introduced and discussed.