Directional Hidden Markov Model for Indoor Tracking of Mobile Users and Realistic Case Study

Indoors, mobile users tend to exhibit some level of determinism in their movement patterns during a day, for example when arriving to their office, going for coffee, going for lunch break, picking up print outs, etc. In this work we exploit this determinism to improve the accuracy of indoor localization systems. We consider two Hidden Markov Model (HMM) based filtering algorithms that use previous observations to estimate a user's most likely movement trajectory, given a sequence of inaccurate location coordinates. The proposed Directional HMM algorithm is able to learn user habits by discriminating between different movement directions when populating the state transition probability matrix from training data. The proposed algorithm is compared to a Standard HMM algorithm that does not distinguish different movement directions. Evaluation results for a simple test scenario with two oppositely intersecting trajectories demonstrated a significant improvement of location accuracy with the Directional HMM algorithm. Further results for a scenario with realistic simulation based movement trajectories also showed improvements for 60% of the cases, however only if the HMM models are trained with usually unknown true trajectories. When trained with inaccurate location estimations, the HMM based algorithms showed no benefit compared to just using the localization system.