Astrometry robot
Exhibition
How the Astrometry.net robot works
Let's begin with what astrometry is.
Astro means 'relating to the stars' and metry means to do with measurement, so astrometry is the branch of astronomy concerned with measuring the positions, motions, and brightness of celestial objects.
Astrometry by hand
When astronomers analyse a new image, one of their first tasks is often to 'solve the astrometry'. This means they work out where that picture belongs in a standard reference frame, such as a chart of the whole sky.
Astrometry 'by hand' can involve lots of complicated calculations but astromomers need to do it so they can compare images and observations. For digital photos, solving the astrometry means they map particular pixels to positions on the sky.
Astrometry by robot
The Astrometry.net robot makes astrometry much easier! It finds photos of the night sky on Flickr and automatically annotates them with astronomical information using notes and machine tags.
The robot is built on top of Astrometry.net's blind astrometry solver - called 'blind' because all it needs is a photo to work out where the telescope or camera was pointing and which stars the image shows.
Stargazers do a similar thing when they look up at the night sky and orientate themselves using constellations. But the robot can also recognise tiny images that cover one ten-millionth of the sky, containing no stars visible to the unaided human eye.
How does it work?
The robot starts with a large catalogue of star positions. At the moment it uses the USNO-B1.0 catalogue, which was created by the US Navy Observatory. This contains the positions and brightnesses of about one billion stars.
Next, it uses this catalogue to find a large number of 'skymarks' in each photo. Each skymark is composed of four stars, and it describes their relative positions. Think of a skymark as a 'landmark' on the sky – a recogisable object that tells you where you are.
As the robot works through more photos it's building up a bigger and better log of skymarks. The idea is that each image it tries to solve will contain at least one, and eventually many, skymarks.
This big collection of skymarks is called an index, and it solves astrometry by working in much the same way as a book index. A book index lists words and then where that word can be found. The robot's index list skymarks and where in the sky that skymark is.
What happens when you submit a photo
First, the robot runs some image processing steps to find stars. Next, it starts looking for sets of four stars in your photo. For each set of four stars, it checks for a match in the skymarks index.
Often, one skymark can be mistaken for another. So when the robot finds a skymark that seems to match, it does some cross-checking. If the skymark really is a match, it asks, 'where else would I expect to see stars in this image?' If many of the predicted stars really are there, then the match must be correct and the image is solved.
Here's an example...
This is the starting image.
Next it's a star search. The red circles are around stars the robot has found in the photo. The green circles are the positions of stars in the index. The green lines join together four stars that make up a skymark. Lots of red and green circles overlap, so this must be a correct match.
Now that the photo has been correctly identified, the robot can label and tag objects in it. That's it!
The Astrometry.net team
The Royal Observatory, Greenwich is very grateful to the Astrometry.net team who built the robot for the Astronomy Photographer of the Year Flickr group. In particular we'd like to thank:
- Christopher Stumm
- Sam Roweis, University of Toronto
- David W. Hogg, New York University
- Dustin Lang, University of Toronto
- Keir Mierle, University of Toronto
- Jon Barron
