Grid Star Identification Improvement Using Optimization Approaches

The grid method is an all-sky star identification approach which tolerates high values of position and stellar magnitude noise compared to its state-of-the-art counterparts. Nevertheless, its performance is affected by sensor's radiometric bias and large values of position noise standard deviation (>1 pixel). In this paper, some optimization approaches are exploited to make the grid algorithm more robust against the above hazards. First, a statistical analysis of stars' population with respect to the stellar magnitudes is performed to define radiometric clusters which make the grid algorithm less sensitive to radiometric calibration bias. Second, by optimizing an objective function defined by the product of robustness factor and grid resolution, the optimal grid cell size is achieved to make the algorithm more robust against noisy sky field. Finally, using Bayesian decision theory, a method is developed for computing optimal static threshold for miss/correct matches classification. The proposed algorithm has been tested on a camera (with 14.6 degrees x 14.6 degrees field of view and 512x512 pixels resolution) against harsh conditions of both star position and stellar magnitude uncertainty. Sensor identification probability of 99.8% against 2 pixels position noise standard deviation has been obtained.