Inferring Motion and Location Using WLAN RSSI

We present novel algorithms to infer movement by making use of inherent fluctuations in the received signal strengths from existing WLAN infrastructure. We evaluate the performance of the presented algorithms based on classification metrics such as recall and precision using annotated traces obtained over twelve hours effectively from different types of environment and with different access point densities. We show how common deterministic localisation algorithms such as centroid and weighted centroid can improve when a motion model is included. To our knowledge, motion models are normally used only in probabilistic algorithms and such simple deterministic algorithms have not used a motion model in a principled manner. We evaluate the performance of these algorithms also against traces of RSSI data, with and without adding inferred mobility information.