Visual navigation of an autonomous vehicle along a line

This paper describes an approach that enables an autonomous vehicle to drive along a line on the ground using machine vision. Our approach has two parts: the image processing that detects the line and the steering control responsible for following the line. The procedure for visual navigation has three steps. First, one-dimensional, horizontal differentiation of a live video image detects a line in the field of view. A series of points represents this line discretely. Next, the procedure finds the cubic curve fitting these points. Finally, a steering control algorithm, called the extended target point following algorithm, uses the cubic curve to visually navigate the autonomous vehicle along the line. We implemented this approach on a vehicle with onboard computers and conducted several experiments in which the vehicle drove autonomously along a line at speeds up to 0.15m/sec. The success of these experiments demonstrate the feasibility of the algorithm.