Analysis and design of algorithms for astrometric data reduction

In December 2013 Gaia satellite was launched by ESA (European Space Agency). Since the start of the nominal operations in mid 2014 the instrument is collecting all-sky images of our galaxy. During its five years of operation the satellite will detect a billion of celestial objects with the objective of calculating with accuracies up to tens of microsecond of arc, for each of them, the fundamental astrometric parameters: position, proper motion and parallaxes. The scientific instruments is made of two identical telescopes pointing two fields of view separated by an angle of 106.5° which share a common focal plane. The focal plane is made of 106 CCDs, of which 62 are devoted to astrometry. Each CCD works in TDI (time delayed integration) in synch with the scanning motion of the satellite, collecting small windows (12x18 pixels) around each observed star. The data processing is entrusted to DPAC (Data Processing and Analysis Consortium). One of the main systems composing the consortium is the Astrometric Instrument Model (AIM), that is devoted to the calibration of the stellar point spread functions (PSFs) that are subsequently used to evaluate the astrometric parameters. AIM is one of the processes that is run in DPCT data processing center in Turin with the supervision of the staff of INAF (Istituto Nazionale di AstroFisica) - OATo (Osservatorio Astronomico di Torino). In order to create a template of the stellar profiles to use in the astrometric parameter computation, oversampling is performed by overlapping multiple images belonging to "similar" exposures. In particular, stellar observations are considered on a daily basis and then grouped into classes . Each variable that is related to the shape of the stellar profiles (ex. CCD position, stellar magnitude and spectrum) introduces a further subdivision of the observations into classes. This procedure allows to merge (overlap) only images that share the same characteristics and, in a subsequent step, to use curve fitting to obtain customized PSF templates for each class of observations. The precession of the spin axis of the satellite results in a periodical vertical deformation, or smearing, of the stellar images, known as AC (across-scan) motion. AC motion is empirically computed and the value is quantized into seven levels, which correspond to a further set of possible configurations of the observations. The work of the thesis is focused in characterizing the speed of the star in the focal plane in order to attempt a compensation of the smearing effect by shrinking the observation windows. This process would allow to reduce the complexity of the calibration phase, by eliminating one variable from the system and cutting down the number of classes in which the whole dataset is subdivided. In addition, deleting AC motion from the stellar profile would allow, together with a proper image up-sampling procedure, to build PSF templates directly on the individual images, avoiding in this way to introduce spurious information by merging different observations. Finally, investigation on periodical variations on periodical variations on images can be performed by removing the main component, which is linked to the satellite precession. The whole work is implemented in Python. The project includes image processing techniques to shrink the stellar profiles and the computation of image descriptors to verify the effectiveness of the AC motion compensation.