Hidden Markov Model based Mobility Learning for Improving Indoor Tracking of Mobile Users

Indoors, a user's movements are typically confined by walls, corridors, and doorways, and further he is typically repeating the same movements such as walking between certain points in the building. Conventional indoor localization systems do usually not take these properties of the user's movements into account. In this paper we propose a Hidden Markov Model (HMM) based tracking algorithm, which takes a user's previous movements into account. In a quantized grid representation of an indoor scenario, past movement information is used to update the HMM transition probabilities. The user's most likely trajectory is then calculated using and extended version of the Viterbi algorithm. The results show significant improvements of the proposed algorithm compared to a simpler moving average smoothing.