Implementation and evaluation of particle filtering for indoor positioning

We implement the classic k-nearest neighbours probabilistic indoor positioning algorithm and compare its performance against two more modern variants, namely the Compass method and a particle filtering approach. Our implementation runs on off-the-shelf mobile hardware and uses the Android platform to realistically appraise the effectiveness of all algorithms in a practical setting. The parameters used in the implementation are outlined in detail and an investigation of optimal values is conducted to ensure adequate algorithmic performance. Our results indicate that while an indoor positioning estimate to within a few meters can be achieved there is significant variance in accuracy. The methods examined do not differ significantly in performance and generally follow the same effectiveness trends, which suggests that further research is needed to draw confident conclusions on their efficacy. We make the implementation of the positioning systems as well as the evaluation tools available to the community to encourage further research and aid duplication of results.