Vehicle control by computer vision

A method is described for guiding an autonomous vehicle along roadways based on visual signals. The vehicle follows a white guideline on a flat road sensed by a computer-controlled display (CCD) camera. The camera, mounted behind the windscreen of the vehicle, looks forward and down, detecting an area between 5 and 25 m ahead. An image processor (based on Transputers T800) scans the image for the borderline of the road. The line coordinates detected by the image processor are transmitted via transputer links to another transputer network. The implemented algorithm uses geometric coordinate-transformation and a dynamical model (Kalman filter) to identify the road curvature ahead and the relative position of the vehicle on the road. The reference angle for the electrical steering servo is computed from the lateral deviation, the yaw-angle deviation, the road-curvature, and the velocity of the vehicle. Altogether, it takes about 70 ms per image for recognition and control. Results for an experimental system are reported