Session
Session 8: Frank J. Redd Student Competition
Abstract
We present Tetra, a star identification algorithm that uses the minimum possible computation time and number of database accesses to solve the calibrationless lost-in-space problem. To solve the calibrationless lost-in-space problem, a star tracker must determine its attitude with no a priori knowledge, not even lens parameters such as the field-of-view or distortions. Tetra is based on a directly-addressed hash table data structure, which enables it to identify star patterns with a single database access. We prove a tight bound on Tetra's false positive rate and empirically compare Tetra's runtime, centroiding error sensitivity, and field-of-view error sensitivity with earlier lost-in-space algorithms: Pyramid and Nondimensional Star ID. We also compare Tetra with hash table based modifications of Pyramid’s cross-referencing step and Nondimensional Star ID's database lookup, which improve Pyramid and Nondimensional Star ID's runtimes by an order of magnitude without otherwise impacting their performance.
Presentation
TETRA: Star Identification with Hash Tables
We present Tetra, a star identification algorithm that uses the minimum possible computation time and number of database accesses to solve the calibrationless lost-in-space problem. To solve the calibrationless lost-in-space problem, a star tracker must determine its attitude with no a priori knowledge, not even lens parameters such as the field-of-view or distortions. Tetra is based on a directly-addressed hash table data structure, which enables it to identify star patterns with a single database access. We prove a tight bound on Tetra's false positive rate and empirically compare Tetra's runtime, centroiding error sensitivity, and field-of-view error sensitivity with earlier lost-in-space algorithms: Pyramid and Nondimensional Star ID. We also compare Tetra with hash table based modifications of Pyramid’s cross-referencing step and Nondimensional Star ID's database lookup, which improve Pyramid and Nondimensional Star ID's runtimes by an order of magnitude without otherwise impacting their performance.
