Non-rigid star pattern recognition for preparation of IOD’s observations

Abstract

The invention of electro-optical devices at the beginning of the 21st century was really a rebirth in the geodetic astronomy. Today, the digital cameras with relatively high geometric and radiometric accuracy have opened a new insight in the satellite attitude determination and the study of the Earth's surface geometry and physics of its interior, i.e., the computation of astronomical coordinates and the vertical deflection components. In the automatic star detection, high precision and reliability in extraction of the star's centers from the captured images and corresponding them with the astronomical coordinates is the most important point. In this article, the probabilistic method has been applied for the star matching. The registration is treated as a Maximum Likelihood estimation problem with the motion constraint over the velocity field such that the catalogue coordinates set moves coherently to align with the pixels coordinates set. The motion coherence has been constrained to the matching problem and derives a solution of regularized ML estimation through the vibrational approach, which leads to an elegant kernel form. In this way, the EM algorithm has been applied for the penalized ML optimization with deterministic annealing. This method finds correspondence between stars coordinates in catalogue and image without making any prior assumption of the transformation model except the motion coherence. This method can estimate the gnomonic transformations between the catalogue and the image and is shown to be accurate and robust in the presence of image noise and outliers. The result of evaluation by proposed algorithm on the image taken by the TZK2-D camera, indicated that the point matching is achieved by standard deviation less than 0.001 pixel.