Page 1 of 1

Point cloud falling apart after rotating view

Posted: Tue Nov 13, 2018 8:31 pm
by zero_hu
Hello everyone,

I have a problem with working with a point cloud. If I open it for the first time, it's fine. My view starts in "outer space", where the whole, 65M point cloud is only displayed as one pixel. I zoom in, and it's still fine. I can pan and zoom as much as I like, and I can see the correct cloud. But if I try to rotate the view, the whole point cloud falls apart, it rearranges itself in a straight line with some width. But I lose all real coordinate information, it's some arbitary, nonsense cloud arranged along a straight line.
If I save it immediately in CC's own .bin after loading, the .bin loads as a line too.
Even if I subsample the cloud to 1/10th of it's original density, it still falls apart if I try to rotate the view.
I had no problem when I imported the exact same point cloud in .obj format, I could work with it just fine. But I lost the color information, so I tried another formats to retain it. But this error prevents me doing it. (By the way, most of the formats both Photoscan and CloudCompare recognizes don't load correctly...)

Details:
Windows10, CloudCompare Stereo 2.9.1, file type is .las, it's a 65M 16bit RGB point cloud in WGS86 coordinate system.

Can you help me please?

Zero_HU

Re: Point cloud falling apart after rotating view

Posted: Tue Nov 13, 2018 11:11 pm
by daniel
Ok, this problem becomes a 'classic' now ;)

This is due to the fact that one dimension of your cloud has probably a totally different scale than the others. Which make the bounding-box (probably along Z) veeeeeeeery long. The initial zoom is computed based on the bounding-box diagonal. This is why you have a very small zoom at startup. But when you rotate, you see all the points stretched along the Z dimension... Check the 'bounding box' dimension in the cloud properties.

To fix this issue, it's better to fix it at the source (i.e. when the input file is created). Otherwise you can use the 'Edit > Scale/multiply' method to rescale the (Z) dimension to make it equivalent to the others. But in this case you may lose some accuracy...