Computer vision for promoting healthy living

Meta Design Lab, Lee Kwan Yew Center for Innovative Cities, Changi General Hospital

The project used primary-source data such as public space observation (both human and using AI), interviews and life-logging on seniors to investigate the relationship between the built environment and the levels of physical activity of residents, with a specific focus on promoting walking among the elderly. From this data and other related projects conducted by Dr. Belinda Yuen’s team, we aimed to build a guide of best practices to promote elderly wellness in the community though urban configuration and design, and to transform these into urban design interventions.

I was project coordinator for the computer vision aspect of the project, and liaised with multiple stakeholders such as Changi General Hospital and the Tampines Town Council. I also performed site visits and camera installation at multiple locations. Much of my focus was in creating proofs-of-concept using computer vision techniques for crowd estimation activity identification, gait analysis and physical health assessment from body posture.

2021 - Present
Research
Programming
Bianchi Dy, Arivazhagan Karunakaran, Sam Joyce
Aerial view of the study site radius Proposed mount points for cameras in study sites

The study site was in Tampines. Blue dots indicate where we would place cameras for observation.

Preview of DecisionVis Web Tool, showing steps in order: 
    Upload data, filter data, set chart settings, then view charts Preview of DecisionVis Web Tool, showing steps in order: 
    Upload data, filter data, set chart settings, then view charts

Structure of the study and our approach to evaluating videos for gait and posture

Preview of DecisionVis Web Tool, showing steps in order: 
    Upload data, filter data, set chart settings, then view charts Preview of DecisionVis Web Tool, showing steps in order: 
    Upload data, filter data, set chart settings, then view charts Preview of DecisionVis Web Tool, showing steps in order: 
    Upload data, filter data, set chart settings, then view charts

We used computer vision to capture different aspects of activities, e.g. objects interacted with, posture, gait. We decided to drop the use of demographic elements due to privacy concerns. I focused on using AlphaPose and other similar algorithms to track gait and posture for health analysis.