Events & News
Colloquium
| Category | Lecture |
| Date | Monday, December 14, 2009 |
| Time | 1:30 pm |
| Location | GS 906 |
| Speaker | Jonathan Myers |
| Title | MS Thesis Defense |
| Affiliation | University of Arizona |
Solar System Object Searching in "Deep Stacks"
Forthcoming surveys of the solar system will come from next-generation
telescopes such as PanSTARRS and LSST. For most cases, current multiple-hypothesis tracking (MHT) methods are a computationally-efficient and reliable method for tracking and identifying solar system given source-extracted observed points from images of the sky.
However, the variety of science goals for these instruments will result in certain observing cadences which generate data intractable for current MHT approaches. In particular, we focus on "deep stacks" of images in which a single area of the sky is imaged many times in rapid succession. These "deep stack" collections are a poor match for existing multiple hypothesis tracking methods, resulting in exponential growth in number of hypotheses. This thesis presents and evaluates new heuristic methods and tools for efficiently extracting usable tracks (linkages between observed points) from "deep stacks", allowing "deep stacks" to be used with existing MHT-based tools. Though our application in this case is on solar system object search, these methods could potentially be adapted to other problems in computer vision or machine learning, in which one wishes to identify an unknown number of objects moving ac- cording to a linear motion model given a large set of noisy but sparse data points. Similarly, many of the methods and filters presented here could be applied to any multiple hypothesis tracking method.