Yifan Zhou, University of Liverpool; Lee Devlin, University of Liverpool; Gemma Cook, University of Liverpool; Simon Maskell, University of Liverpool; Jordi Barr, Defence Science and Technology Laboratory
Keywords: Particle filter, Optical imaging, Detection, Tracking, Uncertainty
Abstract:
In this paper we present experimental results of a fully automatic, low cost system to discover and track unknown objects in geosynchronous orbit (GSO) using an optical telescope. This study is motivated by occurrences when there is limited occupation time on a narrow field of view (FoV) telescope per night and the precise observation time is unknown due to how the telescope is operated.
Within telescope images we are able to identify points and streaks. The former are objects travelling at the same rate the telescope rotates, while the latter are objects moving at different rates. By tasking a telescope to observe objects in a specific orbit we can infer that points in images are objects at that altitude the time the frame was taken. In the pursuit of space domain awareness, we wish to survey the sky to gather information in a wide field by tasking a telescope for a number of observations over a period of time. However, to generate a mosaic image from individual frames is not ideal compared to using a wide FoV telescope. This is because the observation times are sparsely distributed in hours which could miss the chance of observing some GSO objects with large inclination.
An alternative approach relies on a tracker that can initialise an accurate track rapidly and represent the uncertainty precisely, so the telescope can be according controlled to tract the target.
This approach comes with both novelties and challenges. These include, 1) instead of using historical data to initialise tracks, the whole process is a closed-loop without human intervention; 2) the tasks of the telescope are based on the previous observations rather than a pre-defined coverage; 3) a challenge comes in the form of a necessity to deal with the uncertainty changes which influence the task of next observations. 4) extracting useful detections from telescope images with noisy background.
We have built a system utilising these principles. The proposed system includes three components and each is described as follows.
Component one relies on the Liverpool telescope (LT) located on the island of La Palma which is a unstaffed robotic telescope that we can task observations. We have implemented a program that can submit jobs to the LT as a response to our sensor management system.
Component two is the imaging processor that extracts data from LT images. Because the GSO objects are moving relative to the observer on earth (with or without inclination), when the telescope take a picture of objects in the GEO/GSO orbit, most of objects in the GEO/GSO orbit are dots or short streaks and the background stars are equally long streaks. The dot detections can be extracted to initialise or update tracks. Although right ascension and declination of each pixel can be interpreted given FITS file, the image distortion effect (from telescope itself or atmosphere) introduces noises. We therefore use the streak detections (comparing with the star map) to reduce the measurement noise.
Component three is the back-end algorithm that processes the detections from component two, initialise and maintain the tracks of objects and plan the next observations. Considering the problem is highly non-linear and the propagation time interval is large, particle filters which are designed for non-linear problems will be used to estimate the object states with more sensible uncertainty representations. We will also present the filter tuning to ensure that fewer observations are required while we gather more days of detections.
Date of Conference: September 27-20, 2022
Track: Optical Systems & Instrumentation