ABSTRACT
ABSTRACT: Outdoor positioning still merely relies on satellite-based signals where problems quickly arise when the user enters “urban canyons”. The availability of satellite signals can dramatically decrease, e.g., due to various shadowing effects. Alternative concepts for localization in urban areas are either not accurate enough or are economically not affordable. The original contribution of the presented work is to propose visual positioning to complement standard sensors. The method is based on image matching and Kalman filtering to render positioning more robust. In the visual learning stage, a mobile platform with calibrated cameras must navigate through the urban area to capture a sequence of geo-referenced images. In the operation stage, images are classified with respect to their geo-information, i.e., the associated coordinates within a global world model. The recognition of a current location is performed using a Kalman filter based estimator for voting and smoothing of the image based position classification.
