Lessons learned in generating ground truth for indoor positioning systems based on wi-fi fingerprinting

Wi-Fi fingerprinting, a well-established indoor positioning technique, relies on a simple rule: the position attached to a fingerprint can be estimated from known positions of similar fingerprints. One crucial step of fingerprinting is to generate the radio map (reference dataset of fingerprints with well-known positions) required for estimating the position at the localization step. Generating this radio map is time-consuming and it might be prone to errors in large environments, where people have to manually survey an environment with several buildings and similar rooms and corridors. This chapter shows the lessons learned after developing and deploying real working indoor positioning systems in large scenarios under two very different contexts: in-home monitoring and pedestrian navigation. Our experience shows that advanced strategies and tools are required to minimize the user’s fatigue when they are generating the radio map and reduce the wrongly labeled measurements that might poison the radio map.