To work, autoastrom requires some initial estimate of the
centre of the CCD frame. You can provide this with a WCS
component in the NDF, or with a FITS (NDF-) extension, or with
approximate astrometry given on the command line. By
default, autoastrom searches for a WCS component, then a
FITS extension, and fails if it finds neither. Because
autoastrom ultimately sits on top of the AST library, it
is able to make sense of a large variety of embedded FITS
information, and get its initial astrometry from such
pointing information.
You control this process using the --obsdata option.
If you wish to insert an initial approximate WCS component in the NDF, you can best do this using GAIA.
The initial calibration does not have to be particularly
accurate; it need only be accurate enough that autoastrom is
able to make a query to a catalogue server that will return a
set of catalogue objects that has a substantial overlap with
the imaged area of sky. The query covers the region of sky
which maps to the four corners of the image, based on the
initial astrometry, plus a small extra margin.
The default image scale is 1 arcsec per pixel. You will
need to specify the scale if the actual scale is significantly
different from this, as errors in this scale result in
substantial errors in the patch of sky requested from the
server. It is better to choose too small a scale than too
large -- that way, the area which autoastrom searches for
will tend to lie within the observed area, rather than
greatly overlap it.
autoastrom assumes that north is along
the 
-axis. The matching algorithms are generally
insensitive to this angle. However, the only case where you need
to specify this is if the CCD is significantly longer than
it is wide, since in this case a wrong orientation would
cause autoastrom to search for a patch of sky which was a
different shape from the patch covered by the CCD.
Similarly, you should specify any image inversion if you
know it, but it is not always necessary.