Self-learning navigation maps based upon data-driven models using recorded heterogeneous gps tracks

We present our innovative approach to keep navigation maps up to date by deducing map changes from recorded GPS tracks using adequate models and rules. First, we describe, how models for receiver, mobility and terrain can be generated from adequately preprocessed recorded GPS tracks. These models are used by a server in order to predict plausible extensions of available navigation maps. In order to allow for multimodal track sources (pedestrians, automobilists, bicyclists, horseback riders, etc.), geometrical matches have to be further checked for plausibility. We give examples of such plausibility rules we have developed for this purpose. The main benefits of our development are better maps and better guidance for various classes of possible users, from pedestrian, over cross-country skier, to bus driver, to name just a few.