Data fusion for robust indoor localisation in digital health

This paper offers an approach for the combining of signals from multiple sensors observing everyday activities in a digital health care monitoring context. The IoT environment presents a number of advantages for indoor localisation. The amalgamation of several passive sensors can be used to provide an accurate location. This location often bears unique signatures of activity, especially when considering residential environments. However, it is only the basic human instincts, such as periodicity and routine, that make this possible. The fact that behaviours and actions recur naturally is an important assumption in this paper. The study proposes a method, whereby semantic information about the location is learned from an additional source. This method deals with the question of robust indoor localisation prediction by extracting additional activity information available from a wrist worn acceleration sensor. A number of different fusion models are considered, before choosing and validating the model which provides highest improvement in accuracy and robustness over the baseline example. The performance of the methods is examined on different unique datasets, which closely resemble residential living scenarios.