Point cloud accuracy loss after applying transformation matrix

[Daniel Anton]

Hi Daniel,


First of all, congratulations for the versatile CloudCompare. It has become an essential tool for my research.

I am currently involved in massive point cloud processing and I decided to use CloudCompare for geo-referencing my scan data.
When applying a transformation matrix to any point cloud (apparently regardless the number of points) using CloudCompare v2.11 Alpha, the quality of the data decreases -the point clouds lose accuracy as in the screenshots attached (they are the same point cloud. Left: original; Right: transformed).

Issue after applying transformation matrix2.png (641.82 KiB) Viewed 5026 times

This does not happen when transforming the point cloud with the Translate/Rotate tool.


Thank you so much.


Best regards,
Daniel

[daniel]

What kind of transformation matrix do you input? Can you give us an example?

And does the cloud have large coordinates maybe?

[Buczer]

Hi Daniel !

I have the same problem, I'm losing the precision of the point cloud after the application of the transformation matrix. You can see below how my point cloud looks like before and after transformation.

before:

pc_raw.jpg (267.44 KiB) Viewed 4618 times

after transformation:

pc_transform.jpg (245.3 KiB) Viewed 4618 times

I want to transform my data from a local coordinate system to a national coordinate system. I further steps I'll want to process data and get the georeferenced output which will be ready to use in the field with GPS app.

Is there a way to do it without losing the precision ?

My transformation matrix look like this :

1.000000000000 0.000000000000 0.000000000000 607798.500000000000
0.000000000000 1.000000000000 0.000000000000 540818.937500000000
0.000000000000 0.000000000000 1.000000000000 111.369804382324
0.000000000000 0.000000000000 0.000000000000 1.000000000000

Thank in advance for your answer.

[daniel]

And when you input such a transformation matrix, CC doesn't ask you to apply a Global Shift? (if yes, then you should accept this temporary shift).

Which version of CloudCompare are you using?

[Buczer]

When I applied such a transformation matrix CC doesn't ask me to apply a Global Shift

I tried to do this in CC 2.11.3 and CC 2.12 beta, in both cases the same situation occurred.
I'm working on Windows OS. Input point clouds are stored in LAS/LAZ format.

I attached a sample data so you can see how it looks like.

You can try this transformation matrix :

1.000000000000 0.000000000000 0.000000000000 607798.500000000000
0.000000000000 1.000000000000 0.000000000000 540818.937500000000
0.000000000000 0.000000000000 1.000000000000 111.369804382324
0.000000000000 0.000000000000 0.000000000000 1.000000000000

Thanks for any suggestions.

[daniel]

Hum, that's weird. On my side, with version 2.12.0, I see:

globa_shift.jpg (66.03 KiB) Viewed 4539 times

Can you check that your 'File > Global Shift settings' dialog has the right values? (should be 10^4 for coordinates, and 10^6 for the diagonal).

[cragoras]

I'll revive this post because i have a similar problem
When importing the point cloud with local coordinates (<1000), due to its size, there is the automated query to apply a global shift to the coordinates. This query is done again after the transformation.

The parameters of these shifts are then, as i understand it, saved and should be attached to the coordinates after export. However, i think there is a processing error when the shift is attached in both processing steps. In this case, CloudCompare uses only the last one applied, which results in incorrect coordinates (my result, maybe someone can verify?)
The option to apply the shift during import and not during the following transformation results in a loss of point density , probably because the translations are very large ? Here i get just the same result then the others, somehow its "rastered"

Unbenannt.PNG (18.36 KiB) Viewed 4141 times

Only if no shift is applied during the import of the locally referenced point cloud, but during the transformation, correct results are obtained. There seems to be some kind of error i Cloud compare cause only the last shift is applied?

my matrix is
0.5694660473416330 -0.8220148544430920 0.000 404335.6047
0.8220148544430920 0.5694660473416330 0.000 5432579.8802
0.000 0.000 1.000 310.546
0.000 0.000 0.000 1.0000

[daniel]

Hi,

Which version are you using?

And could you provide us with the specific values that were used? (the first shift, then the transformation parameters, and last the second shift that was applied)?

[cragoras]

Hey,
i am using v2.9.1, cant use any newer versions cause i'm working on university computers. My pointcloud is approx 500Million points.
Global shift settings are 10^4 and 10^6.

I have not entered any specific shifts myself and jsut used the numbers cloud compare provided automatically.
Matrix for every case is
0.5694660473416330 -0.8220148544430920 0.000 404335.6047
0.8220148544430920 0.5694660473416330 0.000 5432579.8802
0.000 0.000 1.000 310.546
0.000 0.000 0.000 1.0000

Import shift is the same for both and i press YES

Shift1.PNG (19.13 KiB) Viewed 4103 times

then afterwards while applying the transformation there is this option

shift2 pre trafo.PNG (24.58 KiB) Viewed 4103 times

Now there are two cases

1) If i go NO for the transformation shift i get this which is just wrong

no for shift2.PNG (7.05 KiB) Viewed 4103 times

2) If i go YES for the Transformation shift i get this result for a random target which coordinates i know which is slightly wrong

yes for shift2.PNG (150.83 KiB) Viewed 4103 times

as i said, only NO for the shift at the beginning(first picture) and yes at transformation right here gives a good result

no_thenhere_yes.PNG (58.38 KiB) Viewed 4103 times

correct result

correct_result.PNG (171.31 KiB) Viewed 4103 times

Hope this was understandable
Thank you for your help

[daniel]

Version 2.9?! It has been released in 2016... I bet that's definitely a bug, and it has long been fixed since then.

Maybe you should make a request so that the university updates the SW.