Physical environment features that predict outdoor active play can be measured using Google Street View images

Area of Science:

• Environmental Science
• Urban Planning
• Computer Science

Background:

• Children's outdoor active play is crucial for development.
• Play behavior is influenced by physical and social environmental features.
• Traditional data sources struggle to capture these nuanced environmental features.

Purpose of the Study:

• To assess the feasibility of using machine learning with Google Street View (GSV) images.
• To measure various physical and social environmental features relevant to play.
• To evaluate the generalizability of these models across different urban contexts.

Main Methods:

• Developed machine learning models to detect natural features, traffic (pedestrian, vehicle, bicycle), traffic signals, and sidewalks using GSV imagery.
• Trained models in one city and validated their performance in a second, distinct city.
• Compared model performance for time-invariant versus time-variant environmental features.

Main Results:

• Models demonstrated good performance for time-invariant features like sidewalks and natural elements.
• Performance significantly decreased for time-variant features, such as traffic, when models were applied outside their training context.
• Generalizability challenges were observed when testing models in a new urban environment.

Conclusions:

• GSV images offer a potential automated data source for environmental feature measurement.
• Machine learning applied to GSV data can support the development of predictive models for play behavior.
• Further research is needed to improve model robustness for dynamic environmental features and cross-city applications.