Place Learning via Direct WiFi Fingerprint Clustering

Most current mobile devices are able to determine their location, which has become part of the contextual information available to applications. However, in many cases, the exact position of the device in terms of longitude and latitude is not necessary. On the contrary, applications might benefit more from a discrete context variable that indicates the ``place'' in which the device currently is. To realize this, the continuous device's trajectory needs to be clustered into discrete locations. Besides, the device's location is often not measured directly, but rather inferred from other measurements, such as the list of available WiFi access points. Since similar WiFi measurements lead to similar estimates of the position, it appears that the conversion into geographical coordinates is an unnecessary step in the identification of places. In this paper, we describe a density-based clustering approach that allows to learn significant places directly from a set of raw WiFi measurements.