23.11.23 - TILE2NET: A New Perspective on Urban Sidewalk Mapping

In the predominantly car-centric urban landscapes of North America, the availability of pedestrian infrastructure is limited, making it sometimes challenging to reach destinations on foot. As public administrations seek to reduce their carbon footprint and decrease dependence on automobiles, having an accurate mapping of sidewalks becomes crucial. Unfortunately, most U.S. cities lack such comprehensive mapping.

Enter #TILE2NET, which was developed by Massachusetts Institute of Technology researchers to fill this gap. Unveiled earlier this year, this open-source tool is aimed at obtaining these data, using nothing but aerial imagery!

The implementation works in consecutive steps: starting with an orthorectified aerial image, semantic segmentation is used to identify sidewalks, crosswalks, and footpaths, predicting a label for each pixel. The resulting raster is then converted into a polygon shapefile, which is further simplified into a graph network.

The results are promising, as the methodology was able to recognise 90% or more of the sidewalks and crosswalks in the studied area.

Although #TILE2NET was trained on images from U.S. cities, where the ground truth was already available, the implementation is meant to be adaptable to other city types, making it a possible asset in urban planning.

This tool showcases how machine learning techniques can facilitate policy change by unveiling previously unavailable information about our public space.

Find out more about this story on:
https://lnkd.in/g8sfiNDZ

or explore the implementation:
https://lnkd.in/emqN-u2Q

Image credit:
https://lnkd.in/efWQgTwg

#urbanplanning #machinelearning #publicspace #environment