Robust and Efficient Residential Indoor Localisation for Healthcare

  • Michal Kozlowski

Student thesis: Doctoral ThesisDoctor of Philosophy (PhD)


This thesis contains a treatise on residential indoor localisation for pervasive health monitoring. Contained therein, are all of the aspects to consider, when developing an indoor positioning system. This work encompasses the evaluation of various popular sensor modalities, which are currently popular amongst the community. In addition, it considers the possible space for fusion combinations between said sensors, clearly displaying the current preferences of said community with respect to pressing application challenges, which include accuracy, efficiency and robustness. It also presents a novel dataset, which aims to address the lack of high resolution localisation data in the wider literature. The dataset comprises of a number of real residential houses, parametrised and prescribed using popular methods relating to Radio Frequency fingerprinting. Following this dataset, the thesis focuses on various issues pertaining to
training and parametrisation, especially in relatively constrained spaces of residential abodes. It proposes a novel system which alleviates all of the above issues and provides an improvement on the overall data collection and training tasks, through Simultaneous Localisation and Mapping in the service of Radio Frequency fingerprinting. The thesis then addresses the most important community challenges, by suggesting novel algorithms and methodologies aiming to mitigate their effects. In order to improve robustness and accuracy, a study is performed, which fuses Radio Frequency and Accelerometer data. Results demonstrate, that activity information is beneficial in face
of network adversity, in addition to accuracy improvement and information about the well-being of the participant. The following study of adaptive sensor utilisation techniques through Reinforcement Learning, focuses on accuracy and efficiency of Wireless Sensor Networks. This work shows, that when presented with sensors of varying degree of efficiency, such as wearables and cameras, the system is able to perform weak training, over the lifetime of the infrastructure, whilst at the same time making the system energy aware. This helps the system to remain relatively maintenance free and unobtrusive for potential patients. Finally, in order to alleviate the issues concerning robustness and efficiency, this thesis will present an examination of efficient sensor selection methods in
a Wireless Sensor Network environment. It confirms, that there exist a finite number of sensors which provide near-optimal service for indoor localisation. The results also suggest that data from real world measurements is best to benchmark this type of challenges, as opposed to toy examples. The thesis then summarises all of the above work, and provides indicators for possible future research avenues.
Date of Award23 Jan 2020
Original languageEnglish
Awarding Institution
  • University of Bristol
SponsorsEngineering and Physical Sciences Research Council
SupervisorRaul Santos-Rodriguez (Supervisor), Ian J Craddock (Supervisor) & Robert J Piechocki (Supervisor)

Cite this