A Spatial–Temporal Star Spots Extraction for High Dynamics Star Sensors

Abstract

In this article, we consider the problem of extracting star spots for traditional star sensors with short exposure time under high dynamic conditions (>5°/s). Under this setting, the signal-to-noise ratio (SNR) of star spots decreases visibly, resulting in less star information being presented, making most of star spots undetectable. We propose a spatial-temporal star spots extraction for this task. According to the constraints of angular distance and kinematic of stars in multi-frame images, a detection method based on spatial-temporal filtering is first proposed to detect star spots with low SNR. It establishes a star spot observation by using the correlation between its predicted image and measured image, which fuses the measurement information of the star and the motion information of the star sensor, and detects star spots based on the maximum a posteriori (MAP). Then, according to the imaging characteristics that the imaging region of a star spot can be regarded as a rectangle, an extraction method based on hypothesis-verification is developed to completely extract pixels of the star spot. It establishes the classification probability of a pixel by the multi-layer Gaussian mixture model, and extract pixels according to the maximum likelihood estimation (MLE) of the star image. Experiments demonstrate the extraction capability and accuracy of the method. In conclusion, there is a promising application value for the method to improve the attitude estimation accuracy and update rate of star sensor under high dynamic conditions.