In this background we first look at what is required to allow a photograph to orient (have its 3D position and angles computed). This section applies to control assigned to points (and not to camera stations).
Note that the Preference setting for the minimum number of control points accepted is in Preferences - Processing and Cameras, Control Orient Inverse Camera, Thresholds section for minimum number of control points, Basic minimum'. We call this value M (default is 4 control points and mathematically it can’t be lower than 3).
There are three basic rules for a photo to orient:
• The photo should share 6 reference points with another photo, and one reference point with two other photos if the project has 3 or more photos, or
• The photo should have M control points marked, or
• The photo should share 6 reference points with another photo that has M control (of which at least one of these has to be one of the M marked control points).
In the case where no photos have M control points marked, then rule 1 above applies, and the control are used post-orientation to align the model.
There are two different methods that PhotoModeler uses to orient a photo: a) resection - which needs M or more control marked on the photo, and b) relative orientation which needs 6 or more shared references between photos. In addition to that orientation (finding angles and relative position), a scaling step is usually needed to place camera stations at the correct distance – which needs a known 3d point marked on the photo being scaled (either a control or a point referenced on two other photos). The 3 rules above follow from these three processes (resection, relative orientation, and scaling).
One way to think about the process is this: how do you orient the first photo, and how do you orient a photo relative to the photos already oriented? If you have a photo with M control points the algorithm can start by using resection. If you have no control points (or less than M on any photo) the algorithm sets the first photo to an arbitrary orientation (at 0,0,0 and pointing down Z axis), and then the algorithm tries to orient other photos relative to the starting one.
If the project has at least one photo with M or more control points marked on it, then:
• this photo will orient using the M+ control points (and it will be in the coordinate system of the control points),
If the project has more than one photo and there are no control points in the project, then
• all orientation is done using the relative orientation method
If the project has more than one photo, and no photo has M or more control marked on it, then:
• all photos in the project must share at least six referenced points with one of the other photos (plus one referenced point with a third photo), and
• there should be at least M control points across the whole project that are assigned to points that will be 3D (are referenced on two or more photos).
If the project has more than one photo, and at least one photo has M or more control points, then
• all photos with M+ control photo will be solved by resection and the project will be in the coordinate system of the control (these photos do not need to share points), and
• each photo without sufficient control points will need to share 6 or more control or non-control points with one or more of the controlled photos, and share one or more 3D points for scale (sharing a control point or a point referenced on 2 or more previously referenced photos)
Note: the minimum of 'M' points is increased in the case of Inverse Camera. Every photo that has Inverse Camera turned on needs at minimum M control points marked on that photo plus one control point for every parameter to be solved (e.g. M+1 (usually 5) control marked on a photo with inverse camera focal length).
When control points are involved, there are three setups that work:
• a single photo project with M or more control points marked,
• a multi-photo project with M or more control points marked on one or more photos, with remaining photos sharing six or more control or non-control references and one 3D point, or
• a multi-photo project with no one photo with M control points marked but at least M control points marked across the whole project.