OrbitTrack: Advanced RSO detection and tracking from wide field-of-view on-orbit images
Bibliographic record
Abstract
The growing congestion in Low Earth Orbit, driven by increased space exploration and satellite deployments, has made Space Situational Awareness more critical than ever. To address this challenge cost-effectively, we propose OrbitTrack, a deep learning-based approach for detecting and tracking RSO using wide field-of-view cameras such as commercial off-the-shelf star trackers. Our method is specifically designed to process the Fast Auroral Imager dataset from the CASSIOPE satellite, which presents unique difficulties such as low frame rates, low resolution, and short-exposure images. Additionally, the satellite’s spin-stabilized nature causes both stars and RSOs to follow curved trajectories, complicating their detection and tracking. OrbitTrack employs YOLOX-Nano, a lightweight real-time object detector, leveraging consecutive images to capture trajectory-based features for more accurate detection. For tracking, we introduce a training-free association mechanism inspired by Simple Online and Realtime Tracking but enhanced to use RSO velocity and center distances to predict future positions and associate tracklets. This method overcomes the limitations of traditional Kalman filter and IoU-based approaches, which struggle with low frame rates, and eliminates the need for hyperparameters such as covariance matrices and IoU thresholds. The system efficiently processes 640 × 640 images in 43 ms and 768 × 768 images in 55 ms on a standard CPU, making it suitable for real-time applications. In experiments, OrbitTrack demonstrates a precision of 97.2 %, recall of 74.2 %, MOTA of 71.4, IDF1 of 81.9, and an ID switch rate of just 0.7 over a stack of three consecutive frames. This represents a significant improvement over the rule-based state-of-the-art method, increasing precision by 10.2 % and recall by 11.2 %. The method’s efficiency and accuracy position OrbitTrack as a promising tool for enhancing Near-Earth space monitoring using existing hardware.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".