A research group from Czechoslovakia has been doing some nice work with ground-based guidance of the Parrot AR.Drone using image processing. From their video description: "This video demonstrates the ability of the AR drone to takeoff and land on a moving vehicle. The ARDrone bottom camera is used to recognize a simple pattern on the moving robot and the drone either howers above or lands on the pattern.
"This video shows a fully autonomous unmanned aerial vehicle following a short pre-learned path. The quadcopter is first manually guided around by a teleoperator. During the teleoperated flight, image processing algorithms (SURF,SIFT) recognize salient objects from the image of the on-board camera.
Positions of these objects are estimated and put in a threedimensional map.
In the autonomous mode, the quadcopter loads a relevant map and matches the mapped objects to the currently visible ones. In this way, the quadcopter is able to estimate it position with a precision higher than conventional GPS."
(via I Heart Robotics)
Comments
Im working on adapting my steropsis (currently outputs a 3D point cloud of objects) program to use this.