This repository contains the star identification and attitude determination algorithm presented in the paper below:
TETRA: Star Identification with Hash Tables
Initial validation scripts to prove out propagation using Hipparcos / Gaia DR3 documentation taking RA/DEC and apprent motion with UT Time. Makes sure that angular space of catalog is correct:
Some amount of optimization of number of stars, database general size (Total number of stars, apparent magnitude limits, and angular distance limits) were attempted to get a 100Hz runtime star tracker to be possible. This runtime varies depending on the number of hash checks required for each pattern and if the pattern we are checking actually has returns. Future optimzation includes some sort of hashing function with low utilization and high density of patterns, and definitely needs some smarter way to look up patterns quickly. Another interesting idea is to apply this to satellite ephemeris to create an "artificial sky" with beacons similar to gps.
As far as actual runtime is concerned, centroiding with a simple least squares of a 2d gaussian fit to a 5x5, 9x9 or other kernal seems reasonable for reducing the shear number of stars:
This keeps the density of first pass "lost in space" acquisition stars low, with very reasonable (< 1 pixel) centroid error low even under simulated motion:
This paper builds on popular methods such as Triangle/Pyramid/ISA approaches by Mortari, Salmaan, and Christan. Paper is implemented in C by author and in Python by ESA. This method was used by Pedrotty and co for Seeker / R5 calibration alogirthm validation, but not flown (they used a classic triangle / K-vector method written by J. Christan's group). This repo implements this paper in C++ using the OpenCV, Eigen, and Ceres libraries. Catalog generation only currently.
Purpose:
- General Attitude estimation through star locations
- Allows for mostly sensor agnostic approach (CCD,CMOS, IR, Neuromorphic, Multi-Camera)
- Fast lost-in-space initialization
- Low real time compute
- Capable of using multiple catalogs
- Agnostic to number of stars (star patterns can be 4+ depending on desired capabilities)
Tradeoffs:
- Somewhat larger memory requirements (hash table and general catalog info required)
- High compute required to create catalog
- Assumption: Catalogs of star positions and ISA's (Intra Star Angles) are stored on rad hard EE-PROM type devices.
- If ISA catalogs are lost from memory, options for recompiling must limited. Would take on scale of minutes to re-initialize.
Challenges:
- Hash does not neccessarily capture star pattern density in local region.
- Initial star candidates in catalog are brightest stars, must limit nearby stars according to camera iFOV (instantaneous (pixel) field of view).
Per ESA Tetra documentation: https://github.com/esa/tetra3 https://tetra3.readthedocs.io/en/latest/api_docs.html#tetra3-a-fast-lost-in-space-plate-solver-for-star-trackers "‘hip_main’ and ‘tyc_main’ are available from https://cdsarc.u-strasbg.fr/ftp/cats/I/239/"
COTS Star Tracker repo used newer 311 version (Major cleanups): https://cdsarc.cds.unistra.fr/viz-bin/cat/I/311 https://www.aanda.org/articles/aa/pdf/2007/41/aa8357-07.pdf
Steps:
- Query with Vizier: https://vizier.cfa.harvard.edu/viz-bin/VizieR-3?-source=I/311/hip2
- Preferences (Left column): HTML Table dropdown --> Tab-Seperated-Values
This mimics COTS Star Tracker outputs. Unclear why pm generated values were unused.
Not implemented. Just researching.
Amazing repos here (generally a really cool developer, this guy figured out the chinese rocket lunar collision):
https://github.com/Bill-Gray/star_cats
Need to implement some of these. Other possibilities:
https://github.com/Nerenisa/UCAC4-UCAC5
https://github.com/jobovy/gaia_tools
Good tooling in astroquery, but a lot of work needed for cleaning up: https://arxiv.org/pdf/1804.09366.pdf
TODO: Implement in plantUML or equivilent.
- Catalog Generation
- Lost In Space Phase (LISP)
- Tracking Optical-Flow Phase (TOP)
This will simulate the camera optics (generally) and render star field images based on a number of star catalogs and ephemeris data.
I have this running in docker to generate images: https://github.com/nasa/giant
Warning: There are a few tricks with OpenGL to force a headless container to render, need to write this up.
More documentation here: https://aliounis.github.io/giant_documentation/
Plan to validate using similar approach as Pedrotty against Astronometry.net algorithm (ISA Triangle).
Cool discussion here: https://astronomy.stackexchange.com/questions/33575/how-the-heck-does-astrometry-net-work
Some python implemented in the COTS repo: https://github.com/nasa/COTS-Star-Tracker/blob/master/py_src/tools/astrometry_output_comparison.py