WiFi received signal strength (RSS) fingerprinting is a promising method for indoor localization but it faces two significant challenges: a laborious and time-consuming off-line survey process for radio map fingerprints formation, and variability in the WiFi coverage over time. Recently, to address these two challenges, researchers have begun to consider the concept of crowd-sourcing and automatic floor map and radio map construction. In this thesis, we propose an Adaptive Multi-floor Room-level Localization System (AMRLS) which focuses on using massive crowd-sourced WiFi RSS data to identify multi-floor transitions and to recognize different rooms that exist in the coverage areas. The system will automatically determine the existence and locations of the rooms on the floor maps, and es...[ Read more ]

WiFi received signal strength (RSS) fingerprinting is a promising method for indoor localization but it faces two significant challenges: a laborious and time-consuming off-line survey process for radio map fingerprints formation, and variability in the WiFi coverage over time. Recently, to address these two challenges, researchers have begun to consider the concept of crowd-sourcing and automatic floor map and radio map construction. In this thesis, we propose an Adaptive Multi-floor Room-level Localization System (AMRLS) which focuses on using massive crowd-sourced WiFi RSS data to identify multi-floor transitions and to recognize different rooms that exist in the coverage areas. The system will automatically determine the existence and locations of the rooms on the floor maps, and establish the radio signatures inside the rooms. Similarly, the system will automatically identify transitions between floors, cluster crowd-sourced data into individual floors, and construct floor maps for individual floors using the clustered data. To boot-trap the system, all it takes in the off-line stage is for a surveyor to walk randomly through the coverage area on one Base Floor to collect one reference RSS traces. In the on-line stage, unlabeled crowd-sourced user data is gathered to identify different floors and extract room-level information so that the map can be continually refined and updated. Our results show that rooms can be effectively recognized by their RSS fingerprints, and rooms can be localized on the floor map by analyzing RSS traces as users enter and leave a room. Our results also show that the algorithm can be extended to multi-floor crowd-sourced data. The floor maps for different floors along with RSS fingerprints of rooms can also be adaptively updated using crowd-sourced user data.