Research map / Time series / Causal and generative time series

EBS-EKF: Accurate and High Frequency Event-based Star Tracking

CVPR 2025 · Time series › Causal and generative time series · approach: Generative temporal point processes · 7 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes EBS-EKF, an event-based star tracker combining a low-light circuit-derived centroiding model with a 3D extended Kalman filter, and evaluates it against a space-ready APS star tracker on a newly released real night-sky event dataset.

Problem

Event-based star tracking has only been evaluated in simulation with simplified signal models; its real-world accuracy on night-sky data, where low light and noise matter, has not been quantified.

Approach

Model the event likelihood of a star with a low-light EBS circuit model (intensity-dependent cutoff frequency) that shifts the positive-event centroid by a magnitude-dependent offset, approximate it with a Gaussian, and estimate 3D attitude and angular velocity with an asynchronous extended Kalman filter updated per event.

Key results

• EBS-EKF with offset achieves roughly 25-30 arcseconds across and 60-90 arcseconds about on most real tracks, versus 1-2 degrees (thousands of arcseconds) for ICP, Hough and 2D-KF.
• Analytical low-light model centroiding error is 0.16 ± 0.18 pixels; Gaussian approximation is 0.42 ± 0.48 pixels, both better than prior event centroiding methods.
• Tracking runs at up to 1 kHz update rate, and on a 7.5 deg/s track the method reports 80.4 arcseconds total difference versus 774.3 for 2D-KF.

Measured effects reported in the paper

Surprising findings

• Existing EBS star trackers drift by hundreds to thousands of arcseconds, requiring periodic absolute corrections, while 2D-KF is the closest competitor.
• Modeling the event centroid shift with star brightness reduces error spikes when bright stars enter the field of view.

Limitations

• Ground truth is bounded by the accuracy of the reference APS star tracker, which is itself rated at 5 arcseconds across and 50 arcseconds about.
• Terrestrial night-sky data contains atmospheric artifacts not present in space.
• Existing baselines were reimplemented by the authors without open-source code and rely on precomputed absolute measurements.

Experimental setup

Datasets: 14 real night-sky event tracks recorded with Prophesee EVK4-HD synchronized to Rocket Lab ST-16RT2 APS tracker LCD monitor simulated star sequences

Benchmarks: real night-sky event tracks vs APS tracker LCD monitor centroiding experiments

Metrics: mean squared difference across (pointing) arcseconds mean squared difference about (roll) arcseconds centroiding error pixels

Baselines: ICP (Chin et al.) Hough (Bagchi et al.) 2D-KF (Ng et al.) Rocket Lab APS star tracker

Builds on: Ng et al. asynchronous Kalman filter for event-based star tracking Astrometry.net DBSCAN

Closest papers by meaning

• Event-based Structure-from-Orbit (CVPR 2024 · similarity 0.80)
• ETAP: Event-based Tracking of Any Point (CVPR 2025 · similarity 0.78)
• Seeing Motion at Nighttime with an Event Camera (CVPR 2024 · similarity 0.78)
• Event Stream-based Visual Object Tracking: A High-Resolution Benchmark Dataset and A Novel Baseline (CVPR 2024 · similarity 0.77)
• Towards Realistic Earth-Observation Constellation Scheduling: Benchmark and Methodology (NeurIPS 2025 · similarity 0.77)
• Probabilistic Data-Driven Modelling of Astrophysical Transients: The Neural Process Family for Ultrafast and Class-Agnostic Light Curve Reconstruction (NeurIPS 2026 · similarity 0.77)