Best way is to use Cubic mapping, as then there are no distortions in the images to be accounted for. But, could still be tricky to determine the correct offset to make the person appear to be at the appropriate distance, could try eyeballing it based on things in the image and see how that works (e.g. matching their location to some feature on the floor texture map at the required distance). I guess you could consider adding a dummy object which will be covered over in post by the pasted person. I don't know of a "formula" for it to determine it precisely, though.
One of those times when a full 3D person may be a better solution ;)