A High-Accuracy Star Centroid Extraction Method Based on Kalman Filter for Multi-Exposure Imaging Star Sensors

Abstract

A multi-exposure imaging approach proposed in earlier studies is used to increase star sensors’ attitude update rate by N times. Unfortunately, serious noises are also introduced in the star image due to multiple exposures. Therefore, a star centroid extraction method based on Kalman Filter is proposed in this paper. Firstly, star point prediction windows are generated based on centroids’ kinematic model. Secondly, the classic centroid method is used to calculate the coarse centroids of the star points within the prediction windows. Lastly, the coarse centroids are, respectively, processed by each Kalman Filter to filter image noises, and thus fine centroids are obtained. Simulations are conducted to verify the Kalman-Filter-based estimation model. Under noises with zero mean and ±0.4, ±1.0, and ±2.5 pixel maximum deviations, the coordinate errors after filtering are reduced to about 37.5%, 26.3%, and 20.7% of the original ones, respectively. In addition, experiments are conducted to verify the star point prediction windows. Among 100 star images, the average proportion of the number of effective star point objects obtained by the star point prediction windows in the total object number of each star image is calculated as only 0.95%. Both the simulated and experimental results demonstrate the feasibility and effectiveness of the proposed method.