Benchmark NSW: Public Life Sensor Kit

Project Website

Current
Research
Role:Lead Researcher
Conference presentation

2025 CUPUM (London, UK)
2025 ACSA (Boston, US)
2025 ACSP (Minneapolis, US)
Awards
2025 Australian Good Design Awards Gold Winner


Paper

Williams, S., & Kang, M. (2025). Striking a Pose: DIY Computer Vision Sensor Kit to Measure Public Life Using Pose Estimation Enhanced Action Recognition Model. Smart Cities, 8(6), 183.


WHAT

This project introduces the Public Life Sensor Kit (PLSK) — an open-source, DIY, and privacy-preserving computer vision system that uses pose estimation–enhanced models to automatically detect public-space behaviors such as sitting, standing, staying, and socializing. It bridges design research and AI technology, allowing cities to capture the social dynamics of public space in real time without compromising privacy.
WHY

Traditional urban observation methods are labor-intensive and limited in scale, while most commercial computer vision (CV) sensors only count vehicles and pedestrians without capturing social or behavioral nuance. By making advanced sensing methods accessible and ethical, the PLSK aims to empower planners, researchers, and communities to evaluate design interventions with real behavioral evidence.
HOW

Integrates a GoPro with an NVIDIA Jetson Orin Nano edge device to process video locally — without storing footage. It employs a pose-enhanced action recognition model that identifies fine-grained activities in real time and stores outputs as anonymized GeoJSON data.
This modular setup supports on-site calibration, multi-sensor scalability, and open-source adaptation for various urban research applications.
WHERE / WHEN

July–August 2024: Deployment at the University of New South Wales (UNSW) as part of a public-space improvement project with movable benches.
August–October 2025: Extended deployment at The Goods Line Park in Sydney, further testing scalability and environmental adaptability across larger public settings.

RESULT

The pose-enhanced model achieved 97.8% mAP@0.5 accuracy, outperforming both the baseline and commercial sensors. Behavioral data revealed major shifts after the intervention
  • +360% increase in people staying
  • +1,400% increase in sitting
  • New instances of socializing (≈9 people/day)







Minwook KangLinkedIn
CV
minwook@mit.edu
mintheworld.official@gmail.com



 


I do DJing for fun. This one’s mostly organic house, 
perfect for sunrise runs.