UbiSpot - A user trained always best positioned engine for mobile phones

Abstract

We implemented a positioning engine for mobile phones that can be trained by the users to recognize places as personal landmarks by their wireless communication fingerprint. Our always-best-positioned approach integrates heterogeneous sensor data, such as Bluetooth (BT) device addresses, WLAN MACs, GSM cell ids and GPS coordinates, if available. As an alternative to measuring the signal strength of wireless access points, our positioning engine measures the relative frequency of their appearance and disappearance over time, which closely correlates to their distance. The user can add new places as symbolic names to a hierarchical location model at any time using their mobile phone. For each place, the wireless sensor fingerprint can be trained by the user to define a landmark. Once landmarks have been trained, the positioning engine continuously matches the current sensor profile against the database of learned fingerprints and chooses the most likely place. In case that no BT or WLAN APs are visible, the hierarchical data model can at least derive a higher-level description of the current region based on GSM or GPS as fallback strategy in the sense of being always best positioned. We evaluated the positioning accuracy in our university's lab environment in terms of hits and misses and investigated the effect of various time window sizes for the frequency measurement of the fingerprint. The symbolic location model can be applied for example to adapt the mobile device to different contexts, e.g. automatically mute the ringtone in meeting rooms, trigger location-dependent rules and events, or disclose the current location to friends. (Abstract)