Using Convolutional Neural Networks to Derive Neighborhood Built Environments from Google Street View Images and...

Area of Science:

• Environmental Health
• Urban Planning
• Computer Vision

Background:

• Assessing neighborhood built environment characteristics at scale is challenging, limiting large-scale health research.
• Lack of data hinders understanding neighborhood features as structural determinants of health nationally.

Purpose of the Study:

• To use Google Street View images for characterizing built environments across the US.
• To examine the influence of these built environments on chronic diseases and health behaviors.

Main Methods:

• Processed 164 million Google Street View images nationwide.
• Employed Convolutional Neural Networks (deep learning) to extract built environment features.
• Validated model accuracy at 82%+ for neighborhood characteristics.

Main Results:

• Lower urban development (e.g., single-lane roads) linked to chronic conditions and worse mental health.
• Walkability/urbanicity indicators (crosswalks, sidewalks, cars) associated with better health (less depression, obesity, hypertension, high cholesterol).
• Street signs, streetlights linked to fewer chronic conditions; chain-link fences to poorer mental health.

Conclusions:

• Built environments supporting social interaction and physical activity promote positive health outcomes.
• Computer vision models accurately identify neighborhood features from street imagery.
• This approach enhances the feasibility, scale, and efficiency of neighborhood health studies.