|
| Overview: |
My research is on large-scale multiple target tracking with sparse observation schedules. Specifically, I am examining the problem of asteroid tracking. |
| Proposal: |
Proposal Document: My proposal document in PDF format can be found here. Abstract: Detecting and tracking asteroids from observational data is an important, but extremely computationally expensive, task. The goal is to link together observations from different time steps into sets of observations that correspond to the same asteroid. These linkages can then be used to determine the orbits for new asteroids, attribute the observations to previously known asteroids, and assess the potential risk posed by the object. What sets this domain apart from previous large-scale multiple target tracking problems is the sheer scale of the problem and the sparse nature of the observation schedule. New sky surveys allow us to observe increasing faint objects, providing millions of true and noise observations in the process. Further, detections of a given object may be sparse, sporadic, and widely spaced. Each individual viewing will only cover a small fraction of the sky and there may be a significant gap in time before that area is observed again. This work deals with the computational issues inherent to large-scale target tracking and the development of techniques and algorithms that mitigate or eliminate these issues. I examine two fundamental issues to tracking in the asteroid domain: tracking with a large number of both true and noise observations and tracking under a sparse observation schedule. I propose new data structures, algorithms, and general approaches to efficiently and accurately deal with these issues. |
| Example: |
The figure below illustrates one representative problem of this work, track initiation. Observations from five different (equally spaced) time steps are shown on a single image with observations numbered according to their time step. The goal is to take the raw data and find sets of observations that follow a linear motion model. Thus you have to find five observations in order (1-5) that are equally spaced along a line.  (Click for answers) |
| Related Publications: |
|
