I’m having an issue with an image set I’m trying to turn into a scanned 3D asset. The first alignment was a failure.
I then manually added 7 control points, with each control points being instanced at least 10 times, and with most image having at least 3 control points visible and some even 4.
Still, the alignment is a failure and I’m shown the “error in defined control points” warning.
(ALL control points have been carefully and manually placed by me)
Alignment settings are also default but I’ve also experimented with switching detector sensitivity to ultra, switch image overlap to low and high, change preselector feature to 20 000 and change max feature reprojection error to 2.
Here’s an imgur link where I share a couple of screenshots:
(You will notice only 1 set of 43 images but I actually have 4 sets of around 50 images from many different angles, I am doing the depth and mask workflow per image set angles, having all images imported from all sets does not make the alignment work any better)
I would start with the image measurements assigned to your control points. You don’t really need that much, and sometimes it is advised not to have too many of them. The human error may vary when placing a control point onto an image, so it may even brake the alignment. But that is not the problem here I assume.
Your problem is the background, not the control points. The background is static, while the object is not. This messes up the alignment. If you are going to rotate the object, make sure that the background is natural, one color only, with no features and no shadows. All these features are being recognised during the alignment so they end up in the point cloud too. If you have one static and one movable element, and both of those elements have strong detectable features, that will mess up the alignment.
You can retake the images with a natural background, retake them while going around the object, or try using masks which may be a bit complex with the shape of your object.