DUST : An On-Orbit Star-Tracker Benchmark for RSO Detection and Attitude Estimation

The DUST (Dual-Use Star Tracker) dataset is an openly accessible collection of on-orbit, wide-field near-infrared imagery acquired by the Fast Auroral Imager (FAI) aboard the CASSIOPE spacecraft between January and August 2023. The dataset comprises 1,378 astrometrically calibrated images containing 4,237 manually verified resident space object (RSO) instances across 160 transits, alongside catalog-matched stars, spacecraft ephemeris and attitude data, and detailed background characterization metrics. Images were captured at 1 Hz cadence with a 26° field of view and include challenging real-world effects such as dense stellar backgrounds, stray light, lens flare, motion blur, and spacecraft-induced jitter. RSOs are annotated in both YOLO (object detection) and MOT (multi-object tracking) formats, while star detections include pixel centroids, right ascension and declination, magnitudes, and astrometric match confidence. DUST enables research in RSO detection, multi-object tracking, star–RSO discrimination, and spacecraft attitude estimation using star-tracker-class imagery. It bridges the gap between synthetic benchmarks and restricted operational datasets by providing realistic, fully annotated, and reproducible on-orbit data. Code used for dataset generation, preprocessing, annotation, analysis, and figure production is provided alongside the data to support transparency and full reproducibility. This work was supported by the Natural Sciences and Engineering Research Council of Canada Discovery Grant (grant number: RGPIN-2025-06284), DND/NSERC Discovery Grant Supplement (Application ID: DGDND-2025-06284) and the Canadian Space Agency Flights and Fieldwork for the Advancement of Science and Technology (FAST) program (grant number: 23FAYORA06) in collaboration with Magellan Aerospace and Defence Research and Development Canada.