A Discrete HMM-Based Feature Sequence Model Approach for Star Identification

This paper presents a novel star pattern recognition algorithm based on a discrete hidden Markov model (HMM) for autonomous spacecraft attitude determination in the lost-in-space mode. A two-layer structure is proposed to build an HMM-based star pattern for every guide star. The hidden layer describes the unique geometric distributions of stars in the field of view via the transitions among hidden states, and the observation layer uses discrete and rounded feature sequences consisting of the angular distances of star pairs as outputs. An HMM-based star pattern with maximum probabilities in generating the sequences obtained in the recognition process is uniquely matched. Experiments of synthesized and real star images are tested for the HMM algorithm. Results show that the proposed algorithm has a fast average identification time and is highly robust toward star positions, magnitude noise, and false stars. In the simulation, the HMM algorithm exhibits 99.65% identification rate with 0.4 pixel standard deviations of positional noises, 0.323-Mv magnitude noises, and 1.5-ms average identification time. Moreover, the identification rate is 99.11% and the average identification time is 1.2 ms when testing real star images.