Trade.Information

Topic Name: A visual GPS for cars or robots

Category: Robotics

Research persons: Jean-Philippe Tardif, Yanis , Kostas Pavlidis Daniilidis

Location: Pennsylvania, United States

Details

Usually accurate, but GPS can be wrong in some circumstances: dense urban environment, tunnels, interference waves, etc.. To resolve this problem, researchers from the University of Pennsylvania have developed a method, using an omnidirectional camera, to determine the path through visual observation. At the 2008 edition of the International Conference on Intelligent Robots and Systems, held in Nice, three researchers from the University of Pennsylvania in the United States (Jean-Philippe Tardif, Yanis and Kostas Pavlidis Daniilidis) have introduced a system to estimate the trajectory of a car with an omnidirectional camera capable of filming a field of 360 degrees. The system used, the Ladybug, manufactured by the Canadian firm Point Gray, is actually composed of 6 small cameras. With such a field, it becomes possible to extract much more information than using a conventional camera. Also interesting for robotics to estimate the trajectory, the system detects points of interest in images and then follows the sequence of images. To obtain more robust results, researchers have developed a new method of analysis that determines the rotation and translation separately. Its principle is based on mapping of 3D points obtained by triangulation with their projections in a new image. With the help of large scale experiments, the authors have shown that this approach can yield better results than existing technology. It is possible to estimate the trajectory of a vehicle as accurately as GPS, or more. Thus, a car fitted with this system, which would be supplemented by an electronic card, could be guided out of its GPS reception. In addition, the method allows a posteriori to reconstruct the 3D structure of the environment . It is also interesting for various applications in robotics.