Star Tracker — Spacecraft Attitude Determination from Real Satellite Imagery

Blind spacecraft-attitude estimation on real NASA TESS imagery, built with deep learning + classical geometry. Given a single image plus calibrated camera intrinsics (SIP, CRPIX, and de-rotated CD scale/shear), the pipeline identifies the star field and recovers all three attitude degrees of freedom without an initial pose or IMU prior.

🎯 Try the live interactive demo →

Walks through every pipeline stage with the actual numbers (detections, triangle, plate-solve, quality gate, final pose) on 16 real TESS images. Replay mode renders in ~5 seconds; Live mode runs the full RANSAC + plate-solve.

Six pipeline stages: from raw TESS frame to attitude quaternion.

Results in one table

End-to-end lost-in-space attitude on the full real-TESS test split. Thirteen frames without WCS calibration are reported as data exclusions; solve rate is measured over the 107 valid frames.

Metric	Value
Attitude solve rate	104 / 107 (97.2%)
Median angular error	8.98″ (0.44 px at 20.57″/px)
90th percentile	19.41″ (0.94 px)
Cross-boresight / roll median	2.74″ / 8.33″
Below one pixel	95 / 104 solved frames
Below one arcminute	104 / 104 solved frames
False locks	0 — three candidates honestly refused

The repository bundles a compact 16-frame replay/demo subset: 15/16 solved, 6.63″ median, 0 false locks. Machine-readable full-test metrics are in Results/full_test_metrics.json, with all 120 frame outcomes in Results/full_test_per_frame.csv.

Detector benchmark (centroid quality, --use-gt mode, 120 test images):

Detector	Parameters	Median error
U-Net	7.76 M	4.6″
HRNet	3.99 M	4.8″

The medians differ by only 0.2″ while HRNet uses half the parameters. This suggests that detector architecture is not the dominant accuracy bottleneck; a formal paired confidence interval remains future work.

What this is

A complete star-tracker pipeline that converts a single sky image into a camera attitude quaternion, given the camera’s calibrated intrinsics. Validated on real TESS satellite imagery (not simulations), in lost-in-space mode — no initial pose guess, no IMU prior, and no supplied catalog correspondences. RANSAC/Wahba initializes RA, Dec, and physical body roll; the label WCS is used only to supply camera calibration and to score the final result.

        PNG image (2136 × 2078) + camera intrinsics (SIP + CRPIX + CD)
                │
                ▼
   ┌────────────────────────────┐
   │  Stage 1 — CNN detection   │   U-Net / HRNet → ~480 sub-pixel centroids
   └────────────────────────────┘
                │
                ▼
   ┌────────────────────────────┐
   │  Stage 2 — Body vectors    │   SIP-corrected gnomonic projection
   └────────────────────────────┘
                │
                ▼
   ┌────────────────────────────┐
   │  Stage 3 — Triangle ID     │   RANSAC + Hipparcos pair-DB + 3-point Wahba SVD
   └────────────────────────────┘
                │
                ▼
   ┌────────────────────────────┐
   │  Stage 4 — Refine R        │   Wahba refit on all verified detections
   └────────────────────────────┘
                │
                ▼
   ┌────────────────────────────┐
   │  Stage 5 — Plate-solve     │   scipy least-squares over (RA, Dec, roll)
   └────────────────────────────┘
                │
                ▼
   ┌────────────────────────────┐
   │  Stage 6 — Quality gate    │   honest "no-solution" if residual > 2 px
   └────────────────────────────┘
                │
                ▼
       Attitude quaternion

Why I think this is interesting

Real satellite data, not synthetic. Most public star-tracker projects train and evaluate on procedurally generated star fields. TESS images carry real noise, real optical distortion, real diffraction spikes, and saturated bright stars.
Diagnosed a non-obvious failure mode. Initial linear pinhole geometry hid a 200–1000″ residual at the field corners (TESS uses a 6th-order SIP polynomial for its 12° FOV). The oracle-correspondence --use-gt benchmark looked excellent (4.6″ median) but the catalog-based end-to-end pipeline silently failed on every image. The fix was to apply SIP correction via astropy.wcs.sip_pix2foc before the projection — a small change with a large effect.
Two-pass refinement. RANSAC Wahba lock + scipy plate-solve. The first finds a rough constellation; the second turns that into sub-pixel attitude by minimizing pixel residuals through the SIP forward projection.
A quality gate. Wrong locks fail loudly rather than silently. Production star trackers can’t return wildly wrong attitudes; this pipeline refuses to answer when the post-solve median Euclidean per-star residual exceeds 2 px.
Interactive walkthrough. Streamlit app runs every stage live with real numbers, not just plots.

Try it

Requires Python ≥ 3.10.

git clone https://github.com/SpaceDevEngineer/star-tracker.git
cd star-tracker
python3 -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate
python3 -m pip install -r requirements.txt

First-run expectations. The first invocation builds and caches a Hipparcos pair-angle database (~60–90 s). RANSAC convergence is image-dependent: easy images finish in seconds, while harder fields can take several minutes. Replay mode avoids this wait and is the recommended way to explore the demo.

Interactive Streamlit walkthrough

streamlit run Code/Streamlit_app/pipeline_app.py

Opens at http://localhost:8501. Pick one of the 16 demo TESS images in the sidebar and press Run pipeline. Each stage materialises with the actual body vectors, triangle-angle table, plate-solve iterations, and a final pose comparison against the ground truth.

End-to-end batch evaluation

python3 Code/Star_ID/inference_full.py \
    --data-dir Data/dataset_tess_test \
    --model    Results/unet_run3/best_model.pt \
    --catalog  Data/hybrid/catalog_hipparcos_full.csv \
    --mag-limit 7.5 \
    --out-dir  Results/star_id_run

Writes per-image JSON artefacts to --out-dir and prints a summary table (solve rate, median / 90th-pct / max error) to stdout.

Generate the visualisations

python3 Code/Star_ID/visualize_inference.py \
    --data-dir Data/dataset_tess_test \
    --run-dir  Results/star_id_run \
    --out-dir  Results/star_id_run/viz

Produces per-image overlays (detections + projected catalog + match lines), per-star residual maps, and a project summary chart.

Repository layout

star-tracker/
├── README.md              ← this file
├── PROJECT_DESCRIPTION.md ← longer technical write-up
├── RESULTS.md             ← detailed results tables
├── requirements.txt
├── Code/
│   ├── Star_ID/
│   │   ├── inference_full.py       ← main lost-in-space pipeline
│   │   ├── triangle_id.py          ← RANSAC star identification
│   │   ├── normalize_result_conventions.py
│   │   ├── summarize_results.py       ← auditable result-table export
│   │   ├── visualize_inference.py  ← per-image diagnostic plots
│   │   └── visualize_pipeline.py   ← 6-panel pipeline trace
│   ├── Model_train_code/train.py   ← U-Net architecture + training
│   ├── HRNet_train/                ← HRNet architecture + training + inference
│   ├── Tess_Dataset/process_tess.py ← FITS → PNG + JSON labels (with SIP WCS)
│   └── Streamlit_app/pipeline_app.py
├── Data/
│   ├── dataset_tess_test/  ← 16 TESS demo images + labels (28 MB)
│   └── hybrid/             ← Hipparcos catalog (5 MB)
└── Results/
    ├── full_test_metrics.json       ← aggregate 120-frame summary
    ├── full_test_per_frame.csv      ← one auditable row per frame
    └── unet_run3/best_model.pt      ← trained U-Net weights (30 MB)

The trained HRNet weights, full 800-image training set, and FITS archives live with the research project; this repo carries only what’s needed for the demo and to read the code.

Tech stack

Languages & ML: Python, PyTorch (U-Net + HRNet, heatmap regression), NumPy, SciPy (SVD + nonlinear least-squares). Astronomy / geometry: Astropy (FITS, WCS, SIP polynomial), gnomonic projection, quaternion algebra. Algorithms: RANSAC, Wahba’s problem, chirality-filtered triangle matching, two-pass plate-solving. Tooling: Streamlit (interactive demo), Matplotlib (visualisations), SLURM (training).

Limitations and honest caveats

Calibration source. The evaluation reads SIP, CRPIX, and CD from each TESS WCS, then explicitly factors the rotational component out of CD. A flight implementation would estimate and store one stable intrinsic calibration per CCD.
Residual calibration error. Remaining corner-dependent distortion limits the corrected full-test median to 8.98″. A dedicated per-CCD residual map is the next accuracy lever.
Variable CPU latency. Pair-angle lookup, vectorized third-star search, and FOV-cone verification are implemented, but RANSAC convergence still varies from seconds to several minutes depending on the field.
Honest abstention. Three of 107 valid full-test frames were refused by the quality gate. All 104 published solutions remain below one arcminute.

Acknowledgements

NASA TESS mission for the public TICA Full-Frame-Image data. The Hipparcos catalog team for the reference star positions used in catalog matching.

Author

Temur Kuchkorov · Master’s thesis project, 2025–2026. Email · LinkedIn

License

MIT — see LICENSE.