Leveraging machine-vision for activity recognition utilising indoor localisation to support aging-in-place

Student thesis: Doctoral Thesis


The world is facing an unprecedented challenge where the oldest segment of society has now become the fasting growing segment of society. This is placing a large burden on existing healthcare systems who are struggling to deal with the increase in the elderly. Thus, the concept of Ambient Assisted Living to facilitate aging-in- place has come to the forefront as a potential solution to ease the burden on healthcare systems. A novel solution to this challenge using a single, wearable egocentric camera is presented. This allows a unique first-person viewpoint of the environment to be established which, through the use of fiducial markers, allows the occupant’s location and current activity to be established. A study is presented assessing the technical feasibility for accurate indoor localisation to be established through the use of fiducial markers placed on key objects throughout the environment. This resulted in an effective technique to determine the current location of an occupant within an indoor environment. The tool developed within this study was then used throughout the subsequent studies as a core component of this research.

A subsequent study then sought to determine if it was possible to determine if an occupant/object interaction was a genuine interaction or a result of a cluttered environment or via navigation of the environment. The Intelligent System for Detecting Inhabitant-object Interactions (ISDII) tool was developed to determine if an interaction was genuine through the use of distance estimation to the object of interest. This study also provided a comparison between the tool developed in the previous study vs. an off the shelf algorithm. This study resulted in the improved performance by reducing the number of False Positives that were detected within the video stream improving precision.

A final study was carried out to not only determine the location of the occupant but to estimate their current activity. Due to the use of a wearable camera a lot of noise was introduced into the data via motion blur which resulted in missing or incorrect marker detection. Dempster-Safer theory was implemented to deal with uncertainty that was present in the data to determine the belief that an activity was being carried out. This study demonstrated the ability to reliably detect the correct activity with an 84% success rate when tested on unreliable data.

The incorporation of these findings into the wider body of knowledge may aid in the development of future systems with the goal of solving the challenge of aging-in-place.
Date of AwardOct 2023
Original languageEnglish
SponsorsDepartment for the Economy
SupervisorChristopher Nugent (Supervisor), Haiying Wang (Supervisor) & Mark Donnelly (Supervisor)


  • Wearable technology
  • Assisted living
  • AmI

Cite this