When running AT using control points there are 2 options. The manual says:
What is the difference between accurately adjusted and rigidly registered and when would you use one or the other?
Use control points for adjustment - This is used with accurate survey control, when used it weights the positions of the photos where the control is marked positioning them exactly at this control location.
Use control points for rigid registration - This is used when the control is not survey accurate, i.e. if control coordinates are extracted from lower resolution scene to position (spatially reference) imagery captured from the ground. When used the photos exterior orientation is first computed then the block is positioned over the control using a network least square adjustment distributing the error across all the controls
Michael Barkasi
Application Engineer
Reality Modeling
That was my initial thoughts as I have worked with other photogrammetric software which have similar approaches. However I recently ran AT independently on two sets of photos (no control) and the relative orientation results were acceptable; when I combined the photos into one block the results were very poor. I used Google to get a few control points ( in-accurate) and added them to the two sets of photos. I reran AT using the 'control points for adjustment' and the 'rigid registration' options and the results were completely different. The 'control points for adjustment' (accurate) worked but the 'rigid registration (in-accurate) was very bad. I thought it would be the reverse seeing as how the control was in-accurate, hence my question. I have always used control points from ground survey in previous projects so I have never used the 'rigid registration' option. I will do some additional tests
Farouk Rohoman