I am doing human scans, with a similar issue. On particularly difficult subjects that have very bland clothing, I get say 10% of the cameras that won't align. After the initial solve, I try to right-click "Align Selected Camera" on these NA cameras, and they still won't align.
Is there any other way to deal with this? Is this one of the situations that coded targets are meant to help with specifically? Can one load a camera XML from a previous solve, with this initial positioning being a 'suggestion' of where to start? Could one hack the "ground control" system to allow this?