Research state-of-the-art algorithms in statistical machine learning and computer vision, and develop a computer vision system that can classify images of the sky.
Some years ago, All-Sky-Cameras were deployed at all Thirty Meter Telescope (TMT) candidate sites. The images gathered by these cameras were used to assess the cloud statistics for each site. The effectivness of this process relied on a huge amount of human effort. Your project is to develop a computer vision system that can automate this analysis, and that can be deployed in night-to-night observation planning at sites around the world.
Supervisory panel would include Prof. Tony Travouillon.
Create a computer program that can be deployed (globally) at astronomical observatories, and that is accurate at classifying images of the sky.
- Must be comfortable programming, and running empirical experiments.
- Most be familiar with CV libraries, such as OpenCV.
- We would expect the candidate to be able to run a kernel-PCA pipeline and cluster images. Then run a multinomial logistic regression on PCA features, as a baseline.
- We are happy to supervise students who want to study connectionist architectures (e.g. deep learning).
Warren Skidmore, Matthias Schöck, Eugene Magnier, David Walker, Dan Feldman, Reed Riddle, Sebastian Els, Tony Travouillon, Edison Bustos, Juan Seguel, Joselino Vasquez, Robert Blum, Paul Gillett, Brooke Gregory. Using All Sky Cameras to determine cloud statistics for the Thirty Meter Telescope candidate sites. Ground-based and Airborne Telescopes II. International Society for Optics and Photonics. 2008.
Research experience in computer vision and astronomy.