The needed realism is difficult to quantify for the general case
due to its dependence on the choice of CV algorithm and camera
hardware. However, the required level of realism of the virtual images
is bounded since it does not need to exceed the resolution of
the real image data used by the system. Furthermore, in many applications,
there is no detailed knowledge of the CV algorithm being
evaluated. A solution to this issue could be to validate the method
experimentally for each type of system to be evaluated.