I have a setup where a few cameras are pointed at an object on a turntable. I ONLY need to scan the tops of these objects and while some of them have plenty of detail to track, others do not.
I can use an object with a lot of texture detail/tracking points to get a really good alignment for fixed XMP data (all cameras remaine in the same position and they are using prime [fixed] lenses). I bring the next set in and the camera positions look perfect, the point cloud looks good to my eyes... but when I build the mesh, it's a lumpy disaster. It has the basic shape but is extremely messy.
I have also measured the automated turntable rotations numerous times and it seems to be exact every time. I mean... I set a needle on the turntable and a needle off the turntable - point to point - and after a 360 degree rotation it returned to the exact same point multiple times in a row... I mean like dead on.
I tried locked, fixed and draft position. The locked and fixed positions generate the exact same results as far as I can tell. The draft seems to work partially but creates multiple components, which forces me to manually create control points (which is one thing I'm trying to avoid). Actually as far as I can tell, the draft works the same as using no XML data at all - so I'm not exactly sure what the draft option actually does.
I can also use "the other" photogrammetry software and I don't know how it does it, but it gets the alignment perfect 100% of the time - no matter the object (in my case anyway) but I haven't figured out how to get a rig setup from that "other software" into RC. I can bring the alignment over with images from a Bundler Out file but not the camera position data. Or now that I think of it... maybe I should try to just replace the images in the Bundler Out folder for each set? Though that requires me to rename all images to just 1.jpg, 2.jpg, 3.jpg, etc... and hope I'm renaming each camera to the correct name.
Many of my scans work fine but there are a lot that create multiple components, forcing me to spend a lot of time manually creating control points (which may seem fine for one-off scans but when you have a lot of things to scan and hard deadlines, it's a nightmare). And then sometimes it just throws cameras all over the place - out in space randomly. How "the other software" gets alignments perfect literally 100% of the time (for my scans anyway) is beyond me. RC just seems to have a lot of trouble with alignment for some reason... but I feel like there has to be a way to get the XMP workflow to work with a precise, automated turntable... By the way, there is no other data behind the subject - so it just tracks points on the subject itself - this is how I have to do it in my case but I mean, if I have an XMP data set that I can get to work, this should be fine even with less tracking points... I would assume.
Anyway... if anyone has any input on this or has gotten it to work successfully, I'd love to hear it.
Please sign in to leave a comment.