Mind the gap: revealing sidewalk networks at scale

Abstract In the face of mounting pressures—including climate change—public authorities and researchers are actively exploring strategies to advance sustainable and active transportation modes. One critical challenge is the transition from car-dominated mobility to walkable urban environments, which promises to reduce congestion, lower pollution, and enhance public health. However, effective planning for this transition is hampered by the lack of reliable, city-wide data on pedestrian infrastructure—a data gap that impedes efforts to improve walkability and accessibility. To address this shortfall, we introduce a novel, deep learning-based scalable pipeline for extracting continuous, geocoded sidewalk networks from high-resolution aerial imagery. Our approach integrates advanced semantic segmentation with geometric and graph-based post-processing to convert raw imagery into structured and interconnected pedestrian networks. This work provides urban planners and policymakers with the robust data needed for informed decision-making and paves the way for more sustainable, human-centred urban mobility strategies.