Mapping the walk: A scalable computer vision approach for generating sidewalk network datasets from aerial imagery
Computers, Environment and Urban Systems, 2023
Skip to content
tile2net
Mapping the Walk: A Scalable Computer Vision Approach for Generating Sidewalk Network Datasets from Aerial Imagery
Automatic generation of sidewalk networks from aerial imagery
Computers, Environment and Urban Systems 2023
Tile2Net is an end-to-end tool for automated mapping of pedestrian infrastructure from aerial imagery. We trained a semantic segmentation model to detect roads, sidewalk, crosswalk, and footpath from orthorectified imagery. The results are then converted to geo-referenced polygons and finally a topologically interconnected centerline network is generated. This work is as an important step towards a robust and open-source framework that enables comprehensive digitization of pedestrian infrastructure, which we argue to be a key missing link to more accurate and reliable pedestrian modeling and analyses. By offering low-cost solutions to create planimetric dataset describing pedestrian environment, we enable cities with a tight budget to create datasets describing pedestrian environment which otherwise would not be possible at a comparable cost and time.
For a detailed overview of tile2net, see its GitHub repository.
Publications
How to cite
@article{hosseini2023tile2net,
title = {Mapping the walk: A scalable computer vision approach for generating sidewalk network datasets from aerial imagery},
volume = {101},
doi = {10.1016/j.compenvurbsys.2023.101950},
journal = {Computers, Environment and Urban Systems},
publisher = {Elsevier BV},
author = {Hosseini, Maryam and Sevtsuk, Andres and Miranda, Fabio and Cesar, Roberto M. and Silva, Claudio T.},
year = {2023},
month = apr,
pages = {101950}
}