Automatic astrometric calibration of low-cost all-sky cameras

Introduction to the show: ident: Astrophysics. The day in astrophysics: observations, models and what the data forces.

Vera: I'm Vera.

Jocelyn: And I'm Jocelyn, here with Subrahmanyan.

Vera: We're looking at the paper: "Automatic astrometric calibration of low-cost all-sky cameras".

Jocelyn: Here's the gist of the paper. All-sky cameras require reliable astrometric calibration despite imperfect camera alignment and variable observing conditions The procedure presents an open-source method combining a compact fisheye model, exact rotation for camera tilt, and automatic star matching This work provides an open workflow for calibrating low-cost all-sky cameras with an eight-parameter fisheye model, automatic star matching and single-frame initialisation The authors present two complementary calibration workflows: single-frame calibration for rapid initialization and multi-frame calibration for greater precision The base model comprises eight parameters, including two tilt angles, a rotation about the optical axis, the coordinates of the optical center, a focal scale, and two radial coefficients from Kannala & Brandt (2006) The extended model adds an optional two-parameter Brown–Conrady term to represent decentering distortion Star detection is performed using DAOStarFinder, and catalogue stars are matched to image detections using a procedure involving two passes, fitting a seed model robustly to the pairs accepted in pass 1 Parameter estimation minimizes the residual in pixel space by fitting both base and extended models using a trust-region reflective least-squares solver Performance on Clear Frames On nine cloud-free, moonless frames, the base model yields a median residual of 0.70 px across 4172 test matches, which is reduced to 0.60 px by two optional decentering parameters The extended model reaches a median residual of 0.60 px (0.53–0.71 px), showing an improvement of 0.10 px in the overall median The residuals grow towards the horizon, with about 2.6 px below 10◦ altitude Treatment of Camera Tilt The exact rotation model substantially improves on both a free zenith pixel and an altitude-only correction for a camera tilted by 3.7◦ The base model, which represents the tilt as an exact rotation of sky directions, reduces the median residual to 0.70 px overall and 1.0 px below 30◦ The comparison shows that altitude-only approximations or models representing tilt only through the zenith pixel are insufficient for a camera tilted by this amount Reusability and Limitations A calibration obtained from clear, moonless frames can be reused on moonlit frames, predicting stars with residuals comparable to those on the clear frames Direct calibration from moonlit or partly cloudy images is also possible in many cases, although the quality gate accepted one incorrect solution in this sample The main limitations are the larger residuals near the horizon and the small reference sample from a single camera and lens The overall residual is dominated by well-detected stars above 30◦, while low-altitude scatter can reflect centroid errors, extinction, refraction, obstruction, blending and model inadequacy The procedure provides the astrometric calibration for the first Lumaria station and will be applied to subsequent units The distinguishing contribution is the joint demonstration of explicit camera-tilt correction, automatic one-shot calibration without a prior model, and robustness to tested lunar illumination and partial cloud cover A calibration can therefore be obtained from clear, moonless observations and reused while the camera geometry remains unchanged and enough stars remain visible for reliable matching The implementation is released as ascal (Gonz´alez Fern´andez 2026) References Allande Stars.

Vera: 2026a, Allande Stars web site, https://allandestars.com Antu˜na-S´anchez, J.

Jocelyn: C., Rom´an, R., Bosch, J.

Vera: L., et al.

Paper summary: Jocelyn: We've covered the summary of "Automatic astrometric calibration of low-cost all-sky cameras," focusing on how this open-source procedure combines a compact fisheye model with exact rotation for tilt and automatic star matching to initialize calibration from a single frame.

Vera: It’s about providing a practical way to get astrometric calibration for all-sky cameras that are often installed by non-experts, without needing complex instruments on site.

Subrahmanyan: The authors present this as a procedure that should recover the camera orientation automatically and use a compact model while quantifying how well the calibration predicts positions on subsequent frames.

Jocelyn: They also built in an eight-parameter camera model, which includes two tilt angles, rotation about the optical axis, the coordinates of the optical center, a focal scale, and two radial coefficients from Kannala and Brandt two thousand six <ref:2609.15724#pg1>.

Vera: And they also introduced an optional two-parameter Brown–Conrady term to represent decentering distortion if you want to push for better accuracy than the base model provides.

Subrahmanyan: The methodology involves using DAOStarFinder for star detection and matching catalogue stars to image detections using a procedure involving two passes, fitting a seed model robustly to the pairs accepted in pass one.

Jocelyn: Parameter estimation is done by minimizing the residual in pixel space by fitting both base and extended models using a trust-region reflective least-squares solver.

Vera: The paper’s evaluation was done on nine cloud-free, moonless frames obtained with a Raspberry Pi camera, which is a very specific and relevant piece of hardware for low-cost setups.

Subrahmanyan: Those nine frames resulted in four thousand one hundred seventy-two test matches, and the results showed the base model median residual was zero point seven zero pixels across those matches.

Jocelyn: The paper’s main contribution seems to be the joint demonstration of explicit camera-tilt correction, automatic one-shot calibration without a prior model, and robustness to tested lunar illumination and partial cloud cover.

Conclusion: Vera: So looking at the "Automatic astrometric calibration of low-cost all-sky cameras," it seems like the authors are really focusing on making this work usable for real-world amateur or monitoring applications.

Jocelyn: The title itself points to the practical nature of their work, suggesting they’re targeting those who need reliable star positions without heavy equipment.

Subrahmanyan: For me, what this means is that we are getting a tool that can provide an initial calibration for these low-cost systems quickly, which is a big step toward making all-sky monitoring more feasible.

Vera: It’s about moving away from needing expensive instruments just to get a usable astrometric baseline for things like weather monitoring or space debris detection.

Jocelyn: The procedure allows someone with just a Raspberry Pi camera and some basic site coordinates to generate an initial calibration without needing a prior model setup.

Subrahmanyan: The paper's main implication is showing that you don't need a massive amount of data upfront to get started; single-frame initialization is possible, and the procedure can be reused with minimal change.

Vera: But we have to remember the caveats they laid out, specifically those larger residuals near the horizon and how low-altitude scatter can affect your measurements.

Jocelyn: They also admit that this calibration relies on clear, moonless observations to get its best results, but it’s still useful for later use under different conditions.

Subrahmanyan: Overall, this work provides a procedure for obtaining astrometric calibration for the first Lumaria station and will be applied to subsequent units as they come online.

Vera: The real value here is the combination of explicit camera-tilt correction and automatic one-shot calibration without needing a pre-existing model.

Jocelyn: And the open-source release, ascal, means this tool is available for others to use and build upon.