In the context of mobile robot applications for public transportation in cities, there are many challenging issues, one of these being the ability to guarantee a continuous positioning with an acceptable and controlled level of uncertainty under poor conditions of satellites visibility. In the meanwhile, the local environment is more and more accurately described with the very last geographical data bases. The idea that has been studied and implemented at IFSTTAR consists in detecting and eliminating among current GPS raw data the possible Non-Line-Of-Sight satellite measurements, using the knowledge of the 3D environment around the vehicle. The paper describes the principle of the method, its validation and the promising results that have been obtained on the resulting PVT solution. Next step will be the final real-time demonstration of the French National project CityVIP next summer in Paris.