3-D Triangulation From One Landmark
Larry E. Banta, U. Kumari · 2005
Reliable navigation of mobile robots demands that the machine be able to extract location information from its environment, Numerous researchers have used machine vision to follow road edges, to read specialty designed signposts or to triangulate position from the sighting of multiple landmarks. This paper describes a method for position estimation based on monocular vision and triangulation from self-motion. The method is computationally simple due to a manipulation of the transforriation mathematics, and does not require the placement of specialty designed signs or landmarks in the environment. "'Natural" landmarks such as comers, light fixtures, or stationary equipment can be used for triangulation. If the position of the landmark is known in some global coordinate frame, the method can be used for fixing the robot's absolute position. The technique can also be used for circumnavigation of obstacles or for other types of "relative" navigation such as docking or close quarters maneuvering. Travel on non-flat surfaces is accounted for by compensation for vehicle pitch and roll. The method requires input from pitch and roll clinometers, pan/tilt position sensors, odometric sensors and the calculation of one position parameter from independent sensors such as ultrasonic range sensors, optical range sensors, a fluxgate compass or another reliable source independent of the vision system. Keywords: mobile robot, navigation, landmark, vision, triangulation.