2007/09/24

ADASS XVII: Making the Sky Searchable: Automated Astrometry (D. Hogg)

David Hogg is professor at (NYU)

Team:
  • Hogg
  • Dustin Lang (Toronto)
  • Miele (Toronto, Google)
  • Rowens (Toronto, Google)
Non-text queries: "here is a picture... what is this a picture of?"
Astrometry.net solves this problem for astronomy

Problem is easy: sky
Problem is hard: sky is big, astronomical images are small, bandpasses and sensitivities of images do not match those of the astrometric catalogs, we know little about the images

Demo
  • Astronomy picture of the day solved
For some images, we should be able to automatically metatag our data!
Astrometry.net does not want to return wrong results due to artifacts...

What can Astrometry.net solve?
  • ...
How does it work?
  • Use quads of stars to identify hypothesis
  • Indexing: USNO-B, attempt not to over resuse individual stars
  • Indices attempt to have contiguous, overlapping coverage
  • Quads:
    • choose 4 stars, "hash" is a 2D position on the sky, from 2 starts only, then solve for others
    • 4D KD-Tree
  • Fast-lookup
    • Typical index has 10^9 quads
    • build a balanced 4D KDTree in quad space
    • Tree is stored on disk and memory
  • Verify
    • Take each quad, and match other sources, which ones are unlikely by chance?
    • It's a match if it's 1/10^9 change to happen
    • This assumes there is no clustering in the catalogs (uniform - poisson)
Plans
  • Alpha currently available to testers (like HST, us)
  • Beta in the winter
  • All code is open source (vanilla c)
  • Implement precision tools
    • camera calibration
    • deepn and correct standards catalogs
    • overlapping images
    • FITSWCS implementation issues
Evil plans
  • Aggragate and make interoparable all amateur data
  • same for plate archives and historical data
  • coordinate activities of ameteur
    • time resolution all-sky
    • interesting objects
  • build sky maps and new catalogs
  • photometric calibration and mosaicing

Lang et al, Science, submitted

Q/A

Q. (Hanish) Proper motion and bad geometric distrorsions
A. System is parallel: we have indexes for blue stars, red stars, and at different times.
There are different sky, all brute force now. Images with large sheer fail. To explore sheer space is extremely hard and expensive to index objects in this way.

Q. Web interface?
A. Sure, take a look at astrometry.net. Once can script against the service. SSH-tunnel access will be available to users in an interaction-free way.

Q. Update with new catalogs?
A. HST images fail... SDSS, 2MASS, GALEX, etc.... Build heterogeneous catalog of the sky

Q. Comparison with other methods, using catalogs
A. Our solutions are single point comparison with USNO-B. More catalogs.

Q. Who will pay for VO and why?
A. Industrial partnerships with Google or other industrial partners. VO needs this.