Smartphone-based indoor localisation with multimodal sensing

Location is essential information to a growing number of context-driven applications. Obtaining this information outdoors is relatively easy using the GPS, however indoor localisation is slightly more challenging. By combining different sensing modalities available on smartphones we are discovering new methods to overcome this challenge. In this presentation I will describe the underlying considerations driving the design of HiMLoc (a hybrid indoor localisation system) and the push towards addressing its limitations. 

HiMLoc combines two established localisation methods, Pedestrian Dead Reckoning with a user activity recognition component and WiFi Fingerprinting, integrated through a particle filter. Limiting factors in activity recognition and fingerprints matching pushed us to explore deep learning methods for more robust solutions.