Vision-based Refinement of GPS Location and Compass Orientation

Abstract

Most commercially available Global Positioning Systems (GPS) do not provide location information that is sufficiently accurate to be used for practical location based services (LBS). Especially when there are high building structures nearby, GPS location measurements are known to be erroneous. In this paper, we present a computer vision based method for refining user’s two dimensional location and one-dimensional orientation starting from inaccurate GPS and digital compass measurements. Our method utilizes corner positions of buildings in the digital map and the building vertical edges in the captured images. Once calculated, refined user’s position and orientation can be used as an initial value for sensor-based tracking in accordance with user’s panning motion.