Adaptive integration method based on residual accumulating surface for event-based star sensors

To improve star spot imaging quality and centroid extraction accuracy of event-based star sensors under high-dynamic conditions, and to address the issues of star trailing caused by conventional fixed-time integration, unstable image publication rate of adaptive-time integration, and the lack of continuous response levels in binary event-integrated images, this paper proposes a residual accumulating surface adaptive integration method for event-based star sensors. First, a progressive spatiotemporal denoising method is designed, which employs rapid spatial-domain preprocessing combined with spatiotemporal clustering refinement to suppress background noise events. Subsequently, an integration strategy combining a fixed publication rate with an adaptive event count is proposed, in which the number of integrated events is dynamically adjusted through elliptical morphology analysis of star spots, thereby suppressing star trailing while maintaining a stable a stable event-frame publication rate. A residual accumulating surface is then constructed using an exponential-decay recursive mechanism to continuously encode the event-triggering frequency, thereby forming a relative pseudo-grayscale response jointly determined by the event-triggering frequency and temporal recency. Finally, the residual accumulating surface is converted into a pseudo-grayscale star image via piecewise linear mapping, and sub-pixel centroid localization is accomplished through Gaussian fitting. Experiments on a real star event stream dataset demonstrate that, under a focal-plane image-motion velocity of 2272.7 pixels/s (equivalent to approximately 44°/s for a configuration with a 20° field of view and a 1024 × 1024 detector), the proposed method achieves a mean centroid extraction error of approximately 1.56 pixels and a mean star-pair angular distance error of approximately 3.35 arcsec, representing reductions of 58.3% and 19.1%, respectively, compared to the fixed 5 ms integration method. High centroid extraction accuracy is consistently maintained across the wide dynamic range of focal-plane image-motion velocities from 284 to 2273 pixels/s. The proposed method effectively suppresses star trailing and forms a star-spot pseudo-grayscale response distribution suitable for weighted centroid estimation, providing high-precision star-spot centroids for subsequent star identification and attitude determination.