Results on visual road recognition for road vehicle guidance

UBM has developed a vision-based control system which is capable of detecting obstacles, tracking the road, and automatically performing manoeuvres like lane keeping, lane changing, and collision avoidance. This system has been implemented in a van (VaMoRs). Since 1994, the system has been modified to be implemented in a passenger car (VaMP, a Mercedes S class vehicle). Bifocal vision provides a long look-ahead range for normal cruising-speed driving at 130 km/h. The cycle time of 40 ms is short enough for performing real-time reflex-like manoeuvres. For road recognition, the state of the vehicle and the state of the road are recognized by a recursive estimation procedure. Test rides have been performed on freeways, which demonstrated the high robustness of the system. A long distance test ride has been performed from Munich to Odense (Denmark) and back in November 1995, covering more than 1600 km of autonomous ride. Data of the estimation results (curvature on Autobahn and country road, lane width, yaw, angle, and lateral offset) are presented.