I am working on a scanner that combines thermal images with 3D scans (see this post) To align the thermal cameras, photograph a heated calibration board with holes. I use the hole location to try to align the thermal cameras.
My question is: How do I correct alignment in RC? I have limited information on the thermal camera optics, but I figure it should be easy enough to find the correct position of the cameras based on these points. as long as they are defined in multiple images.
Features are detected on the thermal images, but rather limited. I’d like to force RC to only use these CPs in alignment. I cannot however get a consistent workflow / result when I align the images and thus cannot quantify my alignment quality. There seem to be multiple solutions to this alignment, and often times RC ignores my control points completely. Changing the CP weight does not seem to help.
I have tried feeding in an approximate position of the cameras and using that as a draft for orientation and position, but with limited succes. Does anyone have suggestions for a more robust workflow to force correct alignment?
Regards,
Bart
Hi Bart,
what do you mean by feeding in? Manually creating an xml / xmp file for camera location? That’s quite clever I think. 
From what I have gathered reading through posts with alignment problems, I figure that RC strictly needs some self detected features to be able to operate at all.
Hi Götz,
Thank you again for the reply. I had already considered using an XMP file to manually input alignment. The problem is that my design is modular and therefore not precisely located relative to the ‘normal’ cameras. A calibration is therefore needed to guarantee good relative alignment.
It seems that RC can come up with multiple solutions to the alignment problems based on image features. Some of them make sense, others don’t. The problem is that I don’t have exact lens and camera parameters. They thus have to be estimated by RC, making it harder to find the one ‘true’ solution.
Since the thermal and ‘normal’ cameras contain such different information. I have my doubts about the matches it makes based on image features. Best would be if it ONLY considered CPs when positioning the cameras. One of my thermal images had 0 features and is still aligned somewhat correctly, so RC is considering CPs somehow. It is just hard for me to figure out what to do to create a reliable calibration method. I guess I should at least measure lens characteristics on my thermal sensors more thoroughly.
I am almost at the end of my project, so I will not invest more time in improving the alignment for now, but this will definitely be in my recommendation for the person that continues with my work.
Hi bart,
also thank you!
The info on RC using images with 0 features but CPs is quite valuable!