Imagine you could write text entirely made up of satellite imagery. Each letter would be a real world feature from a bird’s eye view. A house in the shape of an “A”, a lake in the shape of a “B”, a parking lot in the shape of a “C” and so on. This is the idea behind the nascent kickstarter funded project Aerial Bold. Its inventors Benedikt Groß and Joey Lee are right now collecting font shapes in satellite imagery for Aerial Bold and you can join the letter hunt from space.

For their planetary search for letterforms Benedikt and Joey with their collaborators at the Institute for Artificial Intelligence at HS Weingarten are developing a machine learning algorithm. The algorithm requires training data - examples of dozens of real world shapes in imagery that represent letters, numbers, and characters like “A’s”, “Z’s”, “6’s”, “❤︎’s”.

To help finding training data, they developed the Letter Finder App. Built on a number of Mapbox libraries and APIs, the web app allows you to systematically scan through aerial imagery and to catalog all the letter shaped features you can find. They use Mapbox.js to pull in Mapbox Satellite imagery to show hi-res “study areas”. These study areas were defined using a simple model that ranks grid cells in the Global Rural-Urban Mapping Project from high to low based on whether or not a grid cell intersects with a major city.

Training letters identified with the Letter Finder App.

The letter hunt is focusing on urban areas. The Aerial Bold team identified urban areas using data from the Global Rural-Urban Mapping Project.

Next, they use Turf.js to do on-the-fly calculations such as defining the size and shape of the minimap area in the app’s sidebar (turf.centroid, turf.envelope, turf.bboxPolygon) and tallying how many letters exist in each grid of the study areas (turf.within). As you go from task to task, the Letter Finder App calls the API to directly pass in each GeoJSON study area to return the location name so that you can see where in the world you might find one letter or more.

Letterforms in numbers: the green bars represent letters with a sufficient number of features (30 or more). Aerial Bold needs more letters in the yellow section!

The Letter Finder App shows the 10 latest found letterforms and their contributors, calling the Mapbox Static API to pull the image of the letterform from the mapbox tiles and to annotate them with the letter’s bounding box. Along with a scoreboard tracking all of the top contributions, there’s also a letter counter, showing the current totals of all letters found.

Looks like they could use some “R’s”, “N’s”, and “K’s” - join the fun and start your own letter hunt from space!

Help identify letterforms in satellite imagery with the Letter Finder App.