Hello,
I’m trying to replace laser scanned point cloud textures with the corresponding higher resolution image produced by the scanning process. The scanner produces an equirectangular panorama image, which I’m converting to cubemap format with an open source conversion script. The problem is that the resulting cubemap doesn’t quite align with the lower resolution cubemap produced by RealityCapture, when importing the laser scanned data. I’ve tried adjusting the scaling and offset of the equirectangular image before the cubemap conversion, but as the adjustments improve the alignment in one part of the image they start to introduce alignment errors in other regions. I’ve tried six different conversion scripts/tools and they all produce similar results compared to each other, but differ from the lower-resolution cubemap produced by RealityCapture. Has anyone faced this kind of problem, or would have any idea, what could be causing the misalignment?
Br,
Tuomas
Hi tuomas,
some indication of how big those alignment errors are would be helpful, ideally via screenshot.
I’ve never tried anything similar nor remember reading anything about it in the forum, so I think you’re on the fringes.
But my gut tells me that this probably won’t work as well as you intend it to. For one thing, have you made sure that the result you produce has absolutely the same geometry as the files produced by RC? Secondly, what you’re trying to do contradicts the basic good practice and intricate mechanisms of photogrammetry, which is what RC uses to achive the alignment. The only way I can imagine this could work is if you find a way to replace the images RC creates with your own. How to achieve that is completely beyond me though…
Fingers crossed!
Hi Götz,
indeed there might be a mismatch between the e57 and full resolution panoramic image and we are investigating this at the moment. For the alignment the idea was to use xmp-files of the rc’s generated lsp files (or should I say RC’s generated equirectangular images). If we would get these full resolution images to match RC’s generated ones we could just lock these full resolution images in place with xmp-files. This I already managed to do and there results were good. Texture quality of laser scanned data could be improved a lot this way. This misalignment is our only problem at the moment and we would not like to use basic photogrammetric way for alignment because we already know the “exact” places from the laser scan registrations.
I will try to produce some screenshots here.
Br,
Tuomas
Hi Tuomas,
ah interesting. So you are saying that you use the XMP workflow for the 6 LSPs but RC still moves them slightly?
Do you have enough scans to cover all the important surfaces at good angles? Because there seems to be a cut-off angle for texture projection at the flat end: https://support.capturingreality.com/hc/en-us/community/posts/360018945331-Controlling-Texture-Projection-cut-off-angle-on-a-mesh-