Currently, the OpenStreetMap dataset for the Calgary, Alberta, Canada region is lacking most of the city's building footprints. The City of Calgary possesses a dataset of the building footprints, however it opts to sell this data rather than distribute it for free via its open data catalogue. The City of Edmonton does distribute its respective building footprint dataset freely and thus the footprints are available via OpenStreetMap. Utilizing high resolution satellite imagery of the City of Edmonton and the building footprint dataset from OpenStreetMap, it should be feasible to train a neural network classifier to generate building footprint polygons from provided satellite imagery of the Calgary region.
A similar challenge Kaggle competition exists: Dstl's Satellite Imagery competition
Reviews of solution:
- Dstl Satellite Imagery Competition, 1st Place Winner's Interview: Kyle Lee
- Paper by 3rd place team
- Article from 4th place team
- Code from 68th place team https://vooban.com/en/tips-articles-geek-stuff/satellite-image-segmentation-workflow-with-u-net/
- A blog post describing how someone acheived what would have been 2nd place: Satellite Image Segmentation: a Workflow with U-Net
Data Preparation for Satellite Machine Learning
The tool downloads OpenStreetMap QA Tile information and satellite imagery tiles and saves them as an
.npzfile for use in Machine Learning training.
Data pipeline for machine learning with OpenStreetMap
A pipeline to simplify building a set of training data for aerial-imagery- and OpenStreetMap- based machine learning. The idea is to use OSM QA Tiles to generate "ground truth" images where each color represents some category derived from OSM features. Being map tiles, it's then pretty easy to match these up with the desired input imagery.
Question: Does label-maker make this repo redundant? Label-maker states that it "builds on the concepts of skynet-data."
From Mapbox TOS:
You may use Studio or third-party software to trace Mapbox maps solely comprised of satellite imagery ("Mapbox Satellite Imagery") and produce derivative vector datasets (i) for non-commercial purposes and (ii) for OpenStreetMap. If you are on at least a Premium or Commercial plan, you may also use Studio to trace Mapbox Satellite Imagery for commercial purposes, but may only trace up to 100 points, lines, or polygons per year. Except as permitted by this paragraph, you may not trace Mapbox maps.
This appears to cover the Digital Globe imagery: