MAPPING THE WALK: A SCALABLE COMPUTER VISION APPROACH FOR GENERATING SIDEWALK NETWORK DATASETS FROM AERIAL IMAGERY

A SCALABLE COMPUTER VISION APPROACH FOR GENERATING SIDEWALK NETWORK DATASETS FROM AERIAL IMAGERY

Abstract

After a century of car-oriented urban growth (Walker & Johnson, 2016), cities around the world are implementing policies and plans that aim to make their neighborhoods and streets more walkable and transit oriented. Renewed attention to walkability is driven simultaneously by the impending climate crisis, public health concerns, and a strive for economic competitiveness. With more than a third of all CO2 emissions attributable to the transport sector (EPA, 2021), it has become clear that climate goals will not be reached unless urban populations start driving less and relying more on walking and public transportation (Cervero, 1998; Speck, 2013). From a health perspective, more walkable cities have been found to have lower obesity and inactivity-related conditions, respiratory diseases, and lower overall public health expenditures (Frank & Engelke, 2001; Grasser et al., 2013; Zapata-Diomedi et al., 2019). Economically, walkable and transit-served city environments have also become an important draw for a competitive workforce (Moretti, 2012; Glaeser, 2010) and now command some of the highest-priced real estates in American cities (Leinberger & Lynch, 2014).

Despite the growing, multi-pronged importance of pedestrian-oriented city design, the necessary geospatial data for pedestrian infrastructure mapping and modeling remains far behind vehicular infrastructure data. Digital mapping of vehicular road networks expanded rapidly in the 1990s, led by Federal legislation (President Clinton 1994), municipal governments’ investments, as well as private companies such as Navteq and TomTom that operationalized roadway mapping in cities across the world. Assembly and wide-scale dissemination of such data has been instrumental to numerous technologies that use road network data as a key input: mapping and routing applications (e.g., Google Maps, TransitApp), transportation service technologies (e.g. Uber, Amazon Prime), urban transportation models and policies (e.g., metropolitan and urban Travel Demand Models, congestion charging systems in various of cities), as well as mobility data specification standards (e.g., Google’s General Transit Feed Specification, and the City of Los Angeles’ Mobility Data Specification).