Deep learning analysis of street panorama images to evaluate the streetscape walkability of neighborhoods for subsidized families in Seoul, Korea

Abstract

Many studies have examined the relationship between the location of subsidized housing and a variety of socioeconomic opportunities in the surrounding neighborhoods. However, there has been little research on whether public housing promotes the subsidized population’s access to walkable neighborhoods. This study fills the gap in prior literature by examining spatial patterns of public housing as classified by program attributes and the neighborhood- and eye-level environmental characteristics of the surrounding neighborhoods. In particular, this research estimated visual walkability at pedestrian eye-level using semantic segmentation techniques built on a deep learning network and Google Street View datasets. Based on these estimations, we employed binary logistic regression models to determine whether neighborhoods with public housing ensure favorable walkable environments for subsidized families in Seoul, Korea. Our findings showed that walkability differed between long-term and short-term housing. Long-term public housing was primarily located in areas with low 4-or-more leg intersection density and street pavement but a high density of crosswalks. Conversely, short-term public housing tended to be located in neighborhoods with poor greenery and openness, and having lower street intersection and crosswalk densities. These results may provide insight for housing developers and planners regarding the uneven spatial distribution of public housing and help better retrofit neighborhood walkability for subsidized families.