Conor Benson, University of Colorado Boulder; Marcus Holzinger, University of Colorado Boulder
Keywords: event cameras, photometry, non-resolved object characterization
Abstract:
Event cameras (ECs) detect relative changes in brightness at the pixel level, outputting an asynchronous stream of pixel events with microsecond precision [1,2]. The EC architecture provides high spatial and temporal resolution as well as low data rates compared to high-speed, frame-based cameras and photon counters. While ECs have seen particular interest in recent years for “uncued” astrometric applications including fast initial orbit determination from event “streaks” (e.g. [3]), ECs also show clear promise for photometric characterization [4]. Unlike frame-based cameras with prescribed exposure times, event cameras dynamically react to brightness changes as they occur, rapidly triggering events. This enables “uncued” photometric studies for space objects with no a priori knowledge: for example, those with unknown, potentially rapid spin rates, unexpected glints, and/or high-frequency vibration. However, only limited studies of photometric characterization with ECs have been conducted (e.g. [4]). An object’s event signature is driven by its time-varying mean brightness as well as photon shot noise and atmospheric turbulence. To properly interpret and best exploit event data for non-resolved object characterization, we require a deep fundamental understanding of how stochastic brightness variations map to events through the complex event pixel circuit dynamics.
Through our modeling and observations, we show that the total event rate across an object’s event “point spread function” (PSF) is closely correlated to the object’s apparent magnitude, allowing for absolute photometric characterization from event data alone. Comparing our observed event rate – apparent magnitude curve for sidereally tracked stars to the first order event pixel model [4], we find that the model agrees qualitatively but generates much higher event rates (roughly 1 – 2 orders of magnitude). This is likely because the first order model assumes each photon impulsively increases the pixel voltage, resulting in fast voltage variation and excessive event rates. These findings motivate development of higher order event pixel models where there is lag between photon arrivals and voltage change. Observing objects simultaneously with our event and exposure-based cameras, we show that event-based curves closely track traditional exposure-based light curves, not their time derivatives. This demonstrates that photon shot noise and atmospheric turbulence events, which track the near-instantaneous mean brightness, dominate events due to mean brightness change. Overall, these findings greatly inform our interpretation of event data for photometric characterization.
Applying these findings to event data collected using the University of Colorado campus observatory, we conduct event-based frequency analysis for a number of well-known, rapidly rotating satellites from LEO to GEO. For NASA’s earth-observing SMAP and GPM satellites, we accurately extract the known rotation period of each object’s fast-rotating reflector, ~4 s and ~1.9 s respectively. We also observe Relay 2, launched in 1964 with a 173 rpm spin rate, showing that its spin rate has decreased to ~77 rpm, potentially due to dissipative eddy current torques. For the formerly spin-stabilized Brazilsat B1 in GEO, we detect ~5 ms glints that repeat every ~1.8 s, commensurate with the satellite’s last observed spin rate in 2015. Finally, we present our ground-based support of NASA’s recently launched Advanced Composite Solar Sail System (ACS3) during the spacecraft’s attitude anomaly. From event-based data, we extract the two fundamental periods of ACS3’s time-varying, non-principal axis spin state. We show that ACS3 spun up rapidly from September to October 2024 due to solar torques, reaching a peak spin rate of ~40 deg/s. Our event-derived spin rate time history closely tracks downlinked ACS3 attitude telemetry.
Overall, the model comparison and observation analysis presented in this work improve our understanding of event pixel dynamics and facilitate further exploitation and modeling of event data for “uncued” photometric and astrometric applications. Our findings demonstrate absolute photometric characterization from event cameras which will drive development of future SSA optical sensor systems.
[1] P. Lichtsteiner, C. Posch, and T. Delbruck, A 128×128 120 dB 15 μs latency asynchronous temporal contrast vision sensor, IEEE J. Solid-State Circuits, Vol. 43(2), 2008.
[2] G. Gallego et al., Event-based Vision: A Survey, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 44(1), 2020.
[3] S. Fedeler, B. Schmachtenberger, J. Brew, and Z. Watson, Laboratory and Field Evaluation of Event Cameras for Space Situational Awareness (AAS 25-378), 35th AAS/AIAA Space Flight Mechanics Meeting, Kaua’i, HI, 2025.
[4] C.J. Benson and M.J. Holzinger, Simulation and Analysis of Event Camera Data for Non-resolved Objects. Proceedings of AMOS, 2022.
Date of Conference: September 16-19, 2025
Track: Satellite Characterization