A Feasibility Study of Using Google Street View and Computer Vision to Track the Evolution of Urban Accessibility

Previous work has explored scalable methods to collect data on the accessibility of the built environment by combining manual labeling, computer vision, and online map imagery. In this poster paper, we explore how to extend these methods to track the evolution of urban accessibility over time. Using Google Street View's "time machine" feature, we introduce a three-stage classification framework: (i) manually labeling accessibility problems in one time period; (ii) classifying the labeled image patch into one of five accessibility categories; (iii) localizing the patch in all previous snapshots. Our preliminary results analyzing 1633 Street View images across 376 locations demonstrate feasibility.