In contrast to previous multisensory approaches, we choose a vision-centered approach. The detection of the heading changes by vision algorithm is generally much better than that reported for the built-in digital compass in a Smartphone. On the other hand, vision-based position aiding by storing key frames together with the corresponding position on the map is usually the last strategy used to limit the drift inherent in others sensors. So, we first explore the pedestrian localization possibilities by using only vision, even if it means adding other sensors in future.