Counting African Homes

A home doesn't always look like one.

· data,machine learning

Imagine the following scenario: there’s a horrible natural disaster in Africa. You’d like to get an idea of the number of households affected. How would you go about counting them ? 

One possible approach would be to use satellite imagery. But as it turns out, satellite data of Africa can be very difficult to interpret. 

Here’s a few quick reasons why:

  1. Some African houses are very small
  2. Tents, a common form of housing in some areas, are sometimes only a few pixels in size on satellite imagery and therefore very difficult to correctly identify
  3. Natural features can easily be confused with homes.

Some homes in Africa simply don’t look the way we think they would, and the richness of the natural surroundings can make their identification a very hard task indeed.

Lucky for us, Google has just made these kinds of tasks a lot easier with the release of the Google Open Building Database!

The Google Open Building Database

This new database, recently released by Google, aims to count as many African homes as possible. It contains an impressive 516M building detections, across an area of 19.4M km2 (64% of the African continent). 

For each building in their dataset, the Google team includes the polygon describing its footprint on the ground, a confidence score indicating how sure they are that the identified object is a building, and a code corresponding to the center of the building.

As mentioned earlier, being able to count homes can really matter. 

There are many great potential applications for the Google Open Building Database, such as population mapping, humanitarian response, environmental science, or even vaccination planning. Can you come up with any other applications?

 To learn more about this database, check out it's page on Google research.


Tous Les Articles

Vous y êtes presque...

Nous venons de vous envoyer un e-mail. Veuillez cliquer sur le lien contenu dans l'e-mail pour confirmer votre abonnement !