A new navigation method based on measurements of image tokens and Kalman filtering is presented. An image token is the central projection of a landmark, a point on the terrain surface. This surface being described by an elevation map, a Kalman filter processes the measurements to update estimates of camera position and orientation, and landmarks. The method has been implemented for off-line simulations of aeroplane navigation. Preliminary tests indicate a performance at least comparable to that of satellite navigation systems. The implemented algorithm also seems to have high tolerance against noise and modeling errors.