SPIN: Cleaning, Monitoring, and Querying Image Streams Generated by Ground-Based Telescopes

With the increasing number of satellites, spacecrafts, and debris around the earth, space situational awareness (SSA) becomes critical to national security and space safety. According to the recent data from CelesTrak (celestrak.com), there are over 13,000 satellites in orbit, and over 20,500 satellites have decayed since 1957. Among the satellites in orbit, there are just under 3,500 satellites that are both functioning in their correct orbit, compared to nearly 10,000 classed as debris. With the increasing number of debris, the chance of collision increases, too. In 2009, the defunct Russian communications satellite crashed into an operational Iridium spacecraft, creating a new debris cloud comprising about 700 objects (http://www.tomsguide.com/us/Space-Satellite-Collision-Sibera,news-3477.html). There is an urgent need to monitor all these space objects to predict and avoid collisions.

We consider one of the important image sources: the images continuously captured by ground-based telescopes. Ground-based telescopes are much cheaper to obtain, operate, and maintain than space-based ones. They can be installed in multiple locations around the world and used to simultaneously observe the same space object from different locations. It is possible to utilize images of the same space entity taken from different times, or at the same time from different locations, to gain reliable information for detecting anomalies and improving situational awareness. There are two important challenges.

• Large and Noisy Data. Many ground-based telescopes can be installed around the world due to their low cost, each of which can generate enormous amounts of image data. A typical image in FITS format captured by a 17-inch telescope has 1.4 Megabytes(MB). They are generated at a rate of at least one image per minute for a continuous period, e.g., eight night hours, producing gigabytes of data per monitoring period. The number of space objects of interest can be hundreds to thousands, which end up with a huge amount of data. However, because of the changing meteorological conditions or equipment setup, the quality of the images from ground-based telescopes varies from time to time. As a result, not all the raw images contain useful information due to the uncertain image quality, which needs to be cleaned before going to the image repository.

• Stream Monitoring. In addition to collecting good-quality images for large-scale offline analysis, one of the most important applications is to monitor specific space objects online, i.e., an anomaly needs to be detected as fast as possible. A system needs to analyze images at a fairly frequent rate, e.g., a few images per minute, to create an "image stream". This stream is processed and mined to detect possible anomalies, e.g., missing monitored objects or emerging new objects around the monitored ones.

We develop the SPIN (SPace Image cleaning and moNitoring) system to investigate the challenges in space situational awareness using images generated from ground-based telescopes. This system addresses the above challenges of large and streaming image processing. Our research focuses on several problems. (1) How to efficiently identify space images of bad quality so we can skip them? (2) How to efficiently process a stream of images to accurately identify the observed key objects and possible anomalies? And (3) how to query the space image repository to find relevant image sequences?

Figure 1: System Framework

This project is sponsored by AFRL.

Some developed algorithms/tools/scripts:

1. Fast raw object feature extraction algorithm

2. Higher-level feature extraction algorithms

3. Image quality labeling tool

4. Image sequence visualization tool

5. Indexing and query processing scripts