calculating normals

[im_thelumberjack]

I am working a lot with lidar dense point clouds at the moment, and currently using photoscan for the mesh generation stage.  My current workflow is to import the cloud data into cloud compare, do the cloud clean up and point normals and export out to photoscan.  I am currently hitting a snag though while importing the point clouds into photoscan.  The points will import quickly, but photoscan will then have to calculate point normals which can take a very long time, and typically limit how many points I can acceptably import.  Also I find that the normals often have multiple areas that are inverted which can be a pain.  For file types I have been trying laz,ply,pts, and e57, but they all go through the same import process.

Basically I want to know if there is a way that I can import the dense point cloud with out having to go through the calculate point normal phase, or if there is something I can do that will speed it up.

Thank you!

[benj_G]

if you have find any tips i'm interessed !!!!!
thanks

[Alexey Pasumansky]

How many points there are typically in the point clouds that you are importing to PhotoScan?

[benj_G]

For me, 300 million points for the biggest job site, on average 100 million points for the other. and it's a key points cloud

[Alexey Pasumansky]

Hello benj_G,

We'll try to reproduce the issue on a similar size point clouds. Which import format you are using?

[benj_G]

thanks for your interest and your reply ! I work with LAS but if you have any other most fast format I doesn't matter!
thanks

[SAV]

Hi benj_G,

As far as I know las and laz formats DO NOT store any point normal information, hence PhotoScan has to recompute them.

Save your points as a ASCII cloud in CloudCompare in *.pts format. Also make sure that you tick/enable columns title & number of points (separate line) in the export options. Set order to [PTS].
The pts files will be much larger (about 3-4 times), but PhotoScan should not need to compute all normals again.

All the best.

Regards,
SAV

[benj_G]

thanks for your reply !! I try your methode but I have some errors:

1- you need to calculate de normals in cloudcompare and it's slow too !
2- and when I export like you (in PTS) PhotoScan didn't import anything and exit with an error.

thanks again

[SAV]

Hi benj_G,

Sorry, my fault. I double checked, and you are right.
PhotoScan did not calculate any normals because the file wasn't correctly imported. The dense point cloud symbol showed up in the workspace pane, but normally it also shows the number of points, which was missing in my case.
It looks like you need to export from CloudCompare using SPACE as separator in order to be able to open/read *.pts files in PhotoScan. But it will automatically compute normals again. I haven't found a way to get around it, same as you. Which means you probably don't need to compute normals in CloudCompare if  PhotoScan is doing it again during import.

Would be nice if PhotoScan accepted point clouds without normals. #featurerequest

Benj_G, why do you want to import the point cloud into PhotoScan after cleaning it up with third party software packages? Maybe there is a different/easier workflow which would help you to achieve what you want. #thereisalwaysadifferentwaytodothings 

Regards,
SAV

[benj_G]

hi,
yes pleaaase and this request 

I work with classify lidar points and a lot of images (average 2500 tif's of phaseone) so I have already a dense cloud and I want to skip de generation of it.

[SAV]

Hi benj_G,

Have you tried CloudCompare to classify your lidar point clouds? CloudCompare is free & open-source and has two main algorithms/tools to automatically classify point clouds (detecting ground):
1. CSF: http://www.cloudcompare.org/doc/wiki/index.php?title=CSF_(plugin)
2. CANUPO: http://www.cloudcompare.org/doc/wiki/index.php?title=CANUPO_(plugin)

Alternatively, you can also classify based on properties of each lidar point itself (e.g., number of returns, scan angle rank, intensity, etc) or you can compute scalar field values for each point such as density, curvature etc. and classify and crop your cloud based on these.

All the best.

Regards,
SAV

[benj_G]

Hi,
Yes I try Cloudcompare but I prefer lastools ! I have some scripts and it's fully parallelized ! It's not free but it's really rocks

[benj_G]

Hi all,
So i try to avoid import of points cloud and this generation of normal by importing DEM.
So I import DEM created by Global Mapper but Photscan import nothing. I have an error "can't parse geo key directory".
Can you tell me what export parameter I need In globalmapper!

THANKS