Related Experiment Video
Updated: Aug 22, 2025
06:58
An Application for Pairing with Wearable Devices to Monitor Personal Health Status
Published on: February 3, 2022
2.9K
Measuring and modelling perceptions of the built environment for epidemiological research using crowd-sourcing and
Andrew Larkin1, Ajay Krishna2, Lizhong Chen2
1College of Public Health and Human Sciences, Oregon State University, Corvallis, OR, USA.
Journal of Exposure Science & Environmental Epidemiology
|November 12, 2022
Summary
New methods quantify built environment perceptions like nature quality and safety using street-view images and deep learning. This enables large-scale health studies, linking urban environments to well-being.
Area of Science:
- Environmental Health
- Computer Science
- Urban Planning
Background:
- Built environment perceptions (nature quality, beauty, relaxation, safety) are crucial for health but difficult to quantify objectively in large populations.
- Few studies have objectively measured these perceptions due to quantification challenges.
Purpose of the Study:
- To develop and apply crowd-sourced and deep learning methods for measuring built environment perceptions from street-view images.
- To enable the use of these perception measures in epidemiologic studies, specifically within the Washington State Twin Registry (WSTR).
Main Methods:
- Utilized Amazon Mechanical Turk for crowd-sourced image comparisons to quantify perceptions.
- Employed transfer learning with deep learning models, leveraging PlacePulse 2.0 data and refining with WSTR-specific perception data.
- Optimized street-view image sampling for WSTR participant addresses to improve data quality.
Main Results:
- Developed deep learning models with high explanatory power for nature quality (77.6%), beauty (68.1%), relaxation (72.0%), and safety (64.7%).
- Transfer learning improved model performance by an average of 3.8%.
- Perception measures showed weak to moderate correlations with traditional built environment metrics (e.g., NDVI, deprivation, walkability).
Conclusions:
- Successfully developed and validated objective measures for built environment perceptions using crowd-sourcing and deep learning.
- These novel exposure measures are optimized for specific health studies, like the WSTR.
- Future research will investigate the association between these perception measures and mental health outcomes in the WSTR.
