We deal in this paper with an original visual servoing method intended to control a nonholonomic wheeled mobile robot. These works are a part of a complete image-based framework for navigation of a mobile robot, using only a monocular vision system embedded on the robot. The navigation system uses a set of relay images to enable the robot to reach a goal configuration expressed as a desired attitude of visual landmarks in the image plane. This paper particularly focuses on a suitable image-based control law of a nonholonomic robot for navigating in image database, with the assumption that the ground is obstacle free. We show how computing a suitable homographic transformation between the current camera image and a desired one allows to estimate an error vector on the state of the robot relative to a virtual straight line. This line is joined and followed thanks to a control law designed according to a well suited chained system of the state of the robot. Simulations illustrate this point, and experimental results for a two wheels driven robot confirm the validity of our approach.