Cloud to mesh - cannot seem to get any decent results

[mr_pablo]

I have quite a comprehensive scan that contains a bridge spanning a river (the river is not in the scan, just the bridge, and the two embankments)

It is quite detailed, and large, in terms of physical dimensions and points.

I am trying first by segmenting parts of the file, as I do not think my measly 8gb of RAM can handle much.

So I sectioned out a part of the bridge, and then used to "Normals" option to set the normals, then used the "Poisson Surface" plugin.

And It creates a huge horrible blob.

What settings should I be looking to use to even start getting somewhere close to a proper mesh?

I do not need it to be millimetre perfect, but right now, I can't even get it metre perfect!

[daniel]

I guess the normals are not properly computed. Can you post a snapshot of the cloud with and without its normals? (there shouldn't be any black normals).

[mr_pablo]

I'm a total CC newbie, apologies. How can I show you the view you want to see?

[daniel]

You can simply make a screen shot of the screen with the 'Print screen' key of your keyboard (or even ALT + Print Screen, to only grab the contents of the active window).

Otherwise there's a tool to export the view as an image in CC (Display > Render to file).

[mr_pablo]

When ever I try to use Poisson, it says my cloud must have normals. This data is from a high-end scanner, so I'd assume it has normals?

Using the compute normals tool, what kind of settings should I use? my cloud data is quite complex

[daniel]

Nope, sadly normals are not included / generated in general.

You have to compute them, and it's much easier if you have a structured cloud (i.e. FARO, PTX, DP, etc.). Then you just have to find the right radius (which is quite easy, you just have to select a radius that is small enough while getting enough points (e.g. between 6 and 12 in average).

If you don't have a structured cloud, then it's harder for CC to guess the right normal orientation. If it has a simple shape, you can generally use one of the simple heuristics (e.g. all the normals point in a given direction, etc.) or rely on the more generic 'Minimum Spanning Tree' algorithm (for which you might need to increase the 'knn' value (= k nearest neighbors). The bigger it is the more robust the result is ... generally ;)

[mr_pablo]

The data was originally form a FARO I believe, initially received in .POD format. The client then provided the same data in a .PT, which I was able to load into CC.

I'll try segmenting the scan up and trying to calculate the normals.

What settings would you recommend for a scan that contains detailed surfaces (remember, it is of a bridge, so there is a lot of detail)?

Regarding segmenting, is there a way to define multiple areas and create the segments in one go? And is there a way to batch export the segments?

[Tiagbor]

The data was originally form a FARO I believe, initially received in .POD format. The client then provided the same data in a .PT, which I was able to load into CC.

I'll try segmenting the scan up and trying to calculate the normals.

What settings would you recommend for a scan that contains detailed surfaces (remember, it is of a bridge, so there is a lot of detail)?

Regarding segmenting, is there a way to define multiple areas and create the segments in one go? And is there a way to batch export the segments?

Did you get a solution for the above problem? I am having the same issue with the point cloud of a bridge.