Rachel Oliver, Air Force Institute of Technology; Paul Rickershauser, Air Force Institute of Technology; Brian McReynolds, U.S. Air Force Academy
Keywords: Event-Based Sensors, Neuromorphic Sensors, Occultation, Dark Object Detection, Space Domain Awareness
Abstract:
Event-based vision sensors (EVS) are poised to revolutionize Space Domain Awareness (SDA) due to their exceptional change detection capabilities. Their unique architecture, featuring independently operating pixels that record binary events only when a threshold is exceeded, enables efficient sensing with minimal resource requirements. While EVS offer significant advantages in data production and power consumption, they also outperform traditional sensors in terms of dynamic range and temporal sensitivity. As such, EVS hold great promise for space-based applications, which often face stringent constraints on power, processing, and communication bandwidth. Due to their high dynamic range, EVS also pose a unique opportunity for an electro-optical sensor, regardless of sensor location, to observe dim or eclipsed objects passing in front of a bright source. In this paper, we present a simulation-based study on the feasibility of using commercial off-the-shelf (COTS) EVS for detecting and tracking satellites in the presence of a bright lunar background.
With traditional optical sensors, the collection of photons from a bright source can overexpose the image resulting in a maximum digital number readout from the hardware, saturation. The relationship between pixels further exacerbates the problem as the exposure is typically controlled by a universal exposure time that applies to all pixels. Choosing a shorter exposure time to not overexpose the bright object yields underexposed areas in other parts of the scene. In contrast, EVS only reads out logarithmic changes in the photocurrent from a memorized value that is unique to each pixel. There is no equivalent to a standard exposure time applied. This architecture ensures the sensor remains sensitive to change long after a traditional sensor would be saturated. Therefore, dimmer objects in the scene, such as a satellite in front of the moon, are no longer hidden within a saturated pixel. As the satellite passes between the transmitted photon flux from the moon and the sensor, the induced current on the sensor should drop, generating OFF events.
Leveraging this unique aspect of EVS still has challenges; since the change detection relies on logarithmic changes to the photocurrent, detectability is dependent on the ability for a logarithmic change to occur. When the induced current is large, a larger change in the current must occur on any pixel for an event to be detected versus a lower induced current. For objects occluding larger than a pixel, this issue is inconsequential, because the individual pixel incident photon flux should drop to near-zero photons per second during the occlusion. However, most RSOs will be subpixel, only partially occluding the photon flux irradiating the pixel. Therefore, to detect this smaller change in photon flux, the induced current on the pixel must be lowered to enable sensitivity to smaller changes. We propose two possible solutions to lower the incident photon flux per pixel and thus enable sensitivity to smaller differences in flux: increase the spatial resolution of the optical system which decreases the subtended area on the moon, or apply a filter to the optical system.
To further explore these ideas on detectability of occluding objects, we plan to update a simulation previously built to recreate EVS data streams for SDA applications. This simulation estimates irradiated photon flux information to estimate the EVS time series output. It currently propagates point sources with a split-step propagation method through the atmosphere and then applies state of the art techniques to capture the proper number and frequency of events due to true signals and noise sources. The circuit simulation captures the delay in the temporal decay of the current signal which may prove to limit the detectability of the occluding RSOs. Using this circuit simulation foundation, we simulate a bright multi-pixel background with a subpixel-occluding RSO moving through the field of view. These simulations show that reducing the background photon flux improves detectability significantly. While not the only factor impacting detectability, we show the geometry between the observer and occulting RSO is a primer driver of the pixel photon flux and, therefore, a key factor in the logarithmic difference required to generate events. Since the speed at which the induced current decays is proportional to the quantity of current, filtering techniques show a theoretical lower limit to their improvement of event production as the pixel becomes too slow to respond to an occlusion.
In summary, by leveraging a low light EVS simulator, this paper explores the possibility of capturing satellites occluding the moon. The research’s quantification of the limits of occlusion sensing is important because it addresses the feasibility of a new paradigm of optical based sensing for SDA. While normal optical sensing requires lighting conditions in which the observed RSO can reflect light from the sun to the sensor, occlusion sensing promises the opportunity to leverage times when the satellites cannot reflect light. Considering large portions of night time imaging put low earth orbit satellites in the umbra of the Earth, occlusion imaging could prove to be a valuable SDA technique. This initial feasibility study will lead to future studies that evaluate the value of occlusion techniques. For example, when connected with EVS orbital fitting techniques such as the Advanced Uni-sensor Rapid Orbit Reconstruction Algorithm and Sensing (AURORAS) algorithm, this research’s simulation can assess if an occlusion observation provides enough information to fit an orbit.
Date of Conference: September 16-19, 2025
Track: Space Domain Awareness