Alexandre Marcireau, International Centre for Neuromorphic Systems, Western Sydney University; Saeed Afshar, WSU; Nicholas Owen Ralph, Western Sydney University; Imogen Jones, International Centre for Neuromorphic Systems, Western Sydney University; Gregory Cohen, Western Sydney University
Keywords: Event-based, Neuromorphic, Ground-based SSA, Noise reduction, Atmospheric artifacts
Abstract:
Detecting the unexpected is essential in Space Situational Awareness (SSA) systems. A robust Space Situational Awareness (SSA) system must be able to detect and correctly classify un-catalogued debris, spacecraft, and orbital anomalies in space. However, optical and atmospheric artifacts can readily lead to misclassifications. This is because, much like astronomy, SSA lacks strong priors on novel observations. Examples of misclassification of observations have a long history in Astronomy, but none is perhaps more famous or more relevant to this work than Galileo’s conjecture that comets are not physical objects but atmospheric phenomena.
In the new field of Event-based Space Situational Awareness (EBSSA), Neuromorphic sensors, which are modeled on operation of the human visual system, have been shown to be highly effective for ground observations of Resident Space Objects (RSO). Neuromorphic event-based sensors excel at detecting fast-moving objects and unexpected changes in luminance, such as glints. Unlike conventional cameras, event-based sensors do not capture frames but detect luminance changes with independent pixels, allowing for an extremely high dynamic range and temporal resolution enabling unique opportunities for space imaging and novel approaches to RSO detection and tracking, high-speed adaptive optics, satellite identification, and real-time in-frame astrometry. With recent advances in sensor sensitivity, quality, and resolution, the Neuromorphic sensor has become a serious contender in SSA applications.
While EBSSA provide significant advantages, the event-based sensing paradigm also introduces novel challenges not present conventional frame-based SSA or terrestrial event-based sensing. The challenge in event-based space imaging is often the extraction of high-speed and/or faint point sources from an event stream that can contain spurious or noise detection events. Such detections maybe external to the camera and due to atmospheric objects such as insects, bats, and planes. The lack of absolute luminance information in an event-based sensor can make it more challenging to differentiate between RSOs and atmospheric objects when their trajectories are similar from the observer’s perspective.
However, noise detections are not limited to atmospheric artefact but may also be due to noise events generated by the sensor itself. While event-based sensors have dramatically improved SNRs, it still is desirable to operate the system as close to the noise floor as possible to detect ever fainter objects at the highest tolerable noise level. This creates a strong motivation to develop algorithmic and observational methodologies that allow the system to go deeper into the noise to extract more information while rejecting spurious detections.
To address these issues, a binocular telescope can be used in conjunction with Neuromorphic sensors. By using two cameras and two telescopes attached to the same mount one meter apart, we are able to capture stereo event-streams allowing us to detect the presence of parallax and identify atmospheric objects. Our angular resolution, 3.5 arcseconds per pixel, allows the detection of parallax for objects within the atmosphere. Correlating the sensors outputs allows us to filter atmospheric objects as well as spurious detections while preserving isolated changes caused by faint satellites. We test several correlation algorithms which leverage the inherent high temporal resolution and the spatial sparsity of Neuromorphic sensors. The presented algorithmic and observational methodologies allow us to improve the differentiation of atmospheric objects and sensor noise from genuine detections of potentially un-catalogued objects.
Date of Conference: September 19-22, 2023
Track: SDA Systems & Instrumentation