Rough Texture Result From Meshing

[jasonddsmith]

I am trying to mesh together a point cloud from a 3D scanner, and I keep getting this horrible rough texture on every surface that I can't figure out how to get rid of.

I have tried virtually every configuration in the compute normals menu.The best quality seems to be Plane, Octree (r=1.54) and -Barycenter, but no matter what settings I use I still get the rough texture.

I have also played with all the settings in the Poisson reconstruction, but 10 octrees with the default settings of 1, 5, 4 in each box seem to work as well as anything else. 11 octrees improves the texture slightly, but there is no further improvement at 12. The image is of 11 octrees. The smooth areas is where there aren't enough points and the poisson reconstruction sealed it up. It could be an issue with my scans, but every single scan I try has the same effect.

Any insight or help would be greatly appreciated!

[daniel]

In fact this rough texture is mainly due to the noise in your cloud. You can help Poisson a little by computing smooth normals but it won't change that much the result, especially since you mesh at a high level (= high level of precision). At high levels Poisson will actually use all the points and therefore the cloud noise will arise more.

And I doubt the barycenter option will give you such good normals everywhere (especially near the paws, the jaws and the tail). This is surely why you see these kind of bubbles or weird shapes. But to get clean normals on such a cloud is hard (at least you'll need a very clean cloud). How did you acquire this cloud? Do you have the individual scans?

One solution would be to compute very clean and smooth normals and then mesh at a low depth level (but you'll lose a lot of details).
The other solution is to first clean the cloud (use the SOR filter for instance). And of course you can also apply the mesh smoothing algorithm afterwards.

Don't hesitate to share the cloud with me so that I can try to find the best process.

[jasonddsmith]

Thanks so much for the insight David. I tried a lower level for meshing and that helped a lot, although like you said it looses some detail. It might be that my scanner just can't get high enough quality scans to mesh at 10 or 11.

I have the individual scans, I can send them over, but they were somewhat preliminary. We are working on getting some higher quality scans. This cloud came from about 4 scans that were merged together and cleaned, both manually and with the SOR filter. I have tried mesh smoothing as well and I think there is more detail loss than from meshing at a lower level. But using a little of both might be the way to go.

We really were curious to know if it was a scan issue or a CC issue, and it seems clear its in the scans. We will keep trying to up the quality.

Do you think trying to remove noise using the SOR filter, even if there will be some detail loss, is a good idea? I always seem to lose a few points at the core of the cloud when I filter to remove outlying points. Getting a scan with no noise may not be a feasible thing, as we are working on a $99 laser scanner for Kickstarter,so passable, not exceptional quality is the name of the game here.

[daniel]

Using individual scans can mainly help you computing very clean normals (as they are all oriented along the same direction for a single scan). This will help Poisson a little bit.

But you should definitely clean the noise as much as possible. The SOR filter performs well on smooth / curvy surfaces. But it doesn't handle well the right or acute angles. Therefore you will indeed lose some high frequency details (but that's always the same issue with noise: it's almost impossible to remove it without removing ALL the high frequency information as they are basically the same if we don't have any information about the expected shape of the cloud).