Automatic normals computation in CC 2.6.2 beta

[Christian]

Hi daniel,

i´m wondering if there is a reason for changing the normals computation of .ptx files in CC 2.6.2 beta. In the last version CC 2.6.1 we can choose if normals should compute at loading time or not. Is there a reason for this change?

I personal find it very useful because everything is ready to go and i don´t need to compute normals afterwards. I know it take time but it is ok for me. After i merged a bunch of clouds i can´t get useful normals anymore. The only way to get accurate normals is to compute them at the beginning of the work or i´m doing something wrong. Thats why i find it useful. But i love this tool, it´s great!

Thanks


Christian

[daniel]

Because:

• it simplifies the code a lot :D. More seriously this way it also works for FARO files now. And it will work seamlessly for the to-be-supported structured files (E57, etc.)
• now we also remember the grid structure (for PTX, FARO or DP files for now). And this grid structure can be used to enhance the standard way of computing the normals (i.e. with the octree) as computing the normals with the octree is generally cleaner/nicer. The issue was generally to orient the normals proprerly and this is exactly what the grid structure is better/faster for.
• and last but not least, it is even better to compute the normals on the merged cloud (if you have several structured clouds/scans).

Therefore to get very clean/good normals, you should definitely consider first merging the scans (the grid structure should always be preserved by merging operations - apart if there's a bug of course ;). And then compute the normals with the octree (there's a new 'auto' button to help you set the right radius). And of course use the grid structure(s) to orient the normals. You'll get much better results (for a few more clicks, that's true).

Some users were also confused about this 'normals' dialog. Or simply annoyed (when they don't need normals ;).

[Laserbrain]

Hi Daniel,

Thanks so much for implementing these features!

Sorry for being dense, but I'm unclear on the correct steps to compute normals for a FARO file.

Which 'surface approximation' setting is best?

for 'neighbors' you say use ocotree, Auto? Not scan grids?

for 'orientation' use scan grids.

Thanks

[daniel]

Generally you should prefer the quadric model by default. Otherwise the LS plane if you need to go a little faster.

And yes it's always nicer with the octree (the auto radius should hopefully be sufficient in most cases).

And the grid structure of Faro or PTX files is mostly useful to orient the normals.

Computing the normals with the grid structure appeared to be counter-productive (it's generally not faster and the result is not as clean). But you can still test this once just to be sure this applies to your data.

And I'm always interested in some feedback!