Planning an indoor navigation service for a smartphone with Wi-Fi fingerprinting localization
Summary
Research has been done on how to make use of Wi-Fi fingerprints in a smartphone based indoor navigation application, which uses fingerprints as a positioning technique, and which does not uses geometric maps or coordinates as a guideline. Since the fingerprints are directly translated to relative locations, such as “Room 2.200”, the method of positioning has been changed to localization instead of positioning. The focus is on the confirmation of the appropriate location regarding the requested route in topological sense: a user traverses a route constructed by nodes and edges without any explicit coordinates. The framework of the Open Geospatial Consortium’s (OGC) Open Location Services (OpenLS) has been used a guideline to set up a prototype. Parameters cannot be used, since those involve maps and WGS84 coordinates, while the project aims at the non-geometric features.
Signal strengths from a set of Media Access Controls (MAC) which transmits the Wi-Fi signals, have been recorded at 40 locations at the OTB research centre in Delft. In the application a scan will be compared to the recorded signal strengths and if there is a match, the location will be returned. This location is being matched with the requested route and it is possible to tell whether the user is at the correct location or not. As with the nature of multipath, one has to take into account the matching signal has to be within a range of the recorded signal: if the recorded signal was -65 dBm, sufficient search space has to be created around this signal. In this project, a search space of +4 dBm and -4 dBm has been proved sufficient. This means that a live signal of -67 dBm would provide a matching location, since this is within range of the original recorded signal (-69 < RSSI < -61).
The strength of this methodology is that it is easy to maintain a database with the recorded signals, although the recording, or site surveying, is rather time consuming, and the MACs can be easily replaced or removed. Multipath itself is inherent to changes in the time domain, which results in delays upon live tracking. As such, the fingerprint database requires a frequent update. Another strength is that there is no need for additional data processing: since there are no maps needed, there is a fast processing, as a certain degree of map matching does not apply here.
From the results, it was visible that it is highly recommended to store only mac addresses with their signal strengths, which are in the same physical space as the location fingerprint, as it improved the returning results on location predictions. Moreover, upon surveying, it is recommended to use the same class of receiver as the application is to be installed. Scan results obtained from high-end receivers can be considered as incompatible with scan results obtained from a medium-end receiver such as a smartphone. It is most likely the survey results from the latter will match the live results. As a result, a disadvantage is that separate fingerprint databases needs to be recorded for specific devices.
Further improvements are suggested regarding the translation (geocoding) of the fingerprint locations to absolute positions based on coordinates (allowing graphs or maps to be displayed), using extensions of smart navigation (as now, neutral directions have been used) and by distinguishing certain user properties (such as navigating through restricted areas, navigation for disabled users, or time of arrival estimations). In the developed prototype, room has been reserved for extensions such as these. The database for MAC and locations can also be placed on a server, rather than in an inherent database included in the application, for up scaling towards a large coverage area.
Issues will remain regarding the privacy of MAC addresses, since it is not allowed to record MAC addresses along with location information in several countries, as it is possible to trace identifiable persons, which contradicts legalities.