Can't Get Chunks to Align Properly

[kingsapo]

Hi, I'm attempting to capture both sides of an object and merge the chunks together. I've got the object propped on a stand made of museum putty for both captures. Camera alignment goes great, however when I try to merge the two chunks of the object the program seems to think the turntable is the focus and overlays it on the two chunks, rather than correctly aligning the object itself. I have tried various combinations of capturing with a marker pattern and the bare turntable, and nothing seems to work. I thought doing one side with the marker pattern and the other side with the bare turntable would work, but sadly this still results in the turntable being aligned instead of the object. There is also a version where both sides use the marker pattern, but the colors and shapes have been slightly changed on one side. I thought doing this would show the program that they are two separate things and not overlay them, but no matter what I try the turntable seems to be the focus.

I've uploaded the project if that helps. In the project, there are paris of numbered chunks that should go together, so the chunks that start with 0 and 1 should be aligned, 2 and 3 should be aligned, etc. Any guidance on what I might be doing wrong would be greatly appreciated. Thank you!

Project Link

[Alexey Pasumansky]

Hello kingsapo,

Have you tried to apply the masks to the background (turntable pattern) for any two subsets representing the two sides of the object and align the images in the single chunk?

[kingsapo]

Hi Alexey,

Sorry for taking so long to get back to you, and thank you for the suggestion, that worked great! I was actually able to avoid masking the turntable by using two different marker patterns under the object for each side, then aligning them in one chunk like you said. For anyone wondering, the rest of my procedure involved duplicating the chunk, aligning the two chunks (using camera based alignment, so that the two chunks could be merged later), removing half of the images from each chunk, building the dense cloud for each chunk, deleting the putty from the dense cloud for each chunk, and finally merging the two chunks. This results in one chunk with both sides of the object combined floating with nothing underneath it, which is exactly what I wanted.

Thanks again!

[Alexey Pasumansky]

In principle there are some ways to automate the masking:
- using the image of the background that is common for the multiple image in the selected subset,
- using rough mesh model, generated for the separate chunks (providing that unnecessary elements of the background/turntable are removed).

Usually for the better accuracy of the produced model it is suggested to align all the images in the same chunk, and it is possible by adding the masks to cover all the static elements for the turntable scenario. Or using the background suppression feature during the alignment.