Automatically Delineating Building Footprints

Our research automatically generates and delineates building footprints using satellite imagery. Our model estimates building counts using digital maps and built-up area detection using deep learning models. Using deep learning models we also carry out building footprint extraction.

These outputs help produce geo-spatial datasets that can be applied directly or indirectly to many urban planning and data-driven policy making endeavors. These services include:

  • Population disaggregation
  • Air quality prediction at grid level
  • Damage assessment at a granular level
  • Measuring and optimizing distances between municipal service points and housing points etc. 
  • Generating and updating society maps or for tax valuation purposes, which can save a lot of manual efforts.
Satellite ImageTFNet

Our pre-processing pipeline includes NePAGG models for building footprint extraction, while TFNet is a newly proposed design of deep learning model that can extract building footprints more accurately at densely connected areas.