Tuesday, July 10, 2018

How Accurate are Bing's Building Footprints?

Recently Microsoft released the data of 124,885,597 computer generated building footprints in the United States. The building footprints were generated by training computer vision algorithms to recognize building geometries on aerial imagery of the USA. Microsoft has made the data free to download under an Open Data Commons Open Database License.

Obviously Microsoft's USBuildingsFootprints is a great resource for U.S. map makers. However before using the data you might want to explore if it is fit for purpose. Computer vision artificial intelligence is an emerging science and is not free from error. You might therefore want to explore the accuracy of the data before using it.

Dan Cookson's New York Buildings allows you to explore the accuracy of Microsoft's building footprint data in New York. It does this be overlaying the data on top of Here aerial imagery. Using the map you can compare the Microsoft building footprints to the actual buildings shown in the aerial imagery. You can also make direct comparisons between Microsoft's building footprints and New York City's open buildings data and building footprints from Open Street Map.

In New York OSM and the local authority data appears to be more accurate than Microsoft's building footprint data. The Microsoft building footprints often seem to conflate a number of buildings in a block into one or more larger buildings. It looks to me like Microsoft's data is reasonably accurate in determining the level of building cover in an area but less accurate in defining every individual building footprint (at least in New York). Of course Microsoft's data might still be fit for purpose. It just depends on how accurate you need your building data to be.

Also See

US Buildings - a low resolution interactive map of all of Microsoft's 124,885,597 computer generated building footprints.
Post a Comment