Jacob Kruse, Yuhao Kang, Yu-Ning Liu, Fan Zhang, and Song Gao. “Places for play: Understanding human perception of playability in cities using street view images and deep learning.” Computers, Environment and Urban Systems 90 (2021): 101693.
Abstract: Play benefits childhood development and well-being, and is a key factor in sustainable city design. Though previous studies have examined the effects of various urban features on how much children play and where they play, such studies rely on quantitative measurements of play such as the precise location of play and the duration of play time, while people’s subjective feelings regarding the playability of their environment are overlooked. In this study, we capture people’s perception of place playability by employing Amazon Mechanical Turk (MTurk) to classify street view images. A deep learning model trained on the labelled data is then used to evaluate neighborhood playability for three U.S. cities: Boston, Seattle, and San Francisco. Finally, multivariate and geographically weighted regression models are used to explore how various urban features are associated with playability. We find that higher traffic speeds and crime rates are negatively associated with playability, while higher scores for perception of beauty are positively associated with playability. Interestingly, a place that is perceived as lively may not be playable. Our research provides helpful insights for urban planning focused on sustainable city growth and development, as well as for research focused on creating nourishing environments for child development.
- Our deep learning model was able to produce playability scores whose distribution closely matched that of the training data.
- Using images labeled by our deep learning model, we produced a map of playability scores for Boston, Seattle, and San Francisco.
- Downtown areas in the three cities studied had high lively scores but low playability scores.