Different rotation of cameras in .xmp and .csv

Development Asset Creation
1

Hello everyone,

I am exporting cameras using XMP. And also export using Internal / External camera parameters. In my understanding, they should duplicate information, albeit in a different representation. And all the parameters in these two views converge, except for rotations.

in the XMP file: <xcr: Rotation> 0.982594073038806 0.185426733553445 -0.0112166890261935 -0.032113633044729 0.11008102707942 -0.993403685341366 -0.18296885582437 0.976472781986697 0.114> Rotation>

And in CSV:
heading (163.736597399311)
pitch (83.4154313461628)
roll (174.386496244756).

These are two different turns. I attach screenshots of export. Can you tell me what might be wrong?


Upd:
I noticed that if you change the values like this
heading = heading - 180
roll = roll - 180
Then the turns coincide, what could be the problem in the end?

 

 

 

 

 

2

Hello dear user, 
I will try to get your question answered, please be patient meanwhile.

3

Erik, thanks!

Perhaps this will be helpful in answering the question. The rotation received from CSV matches the rotation matrix from XMP if
heading = heading - 180
roll = roll - 180
and use YXZ. This can be checked using the calculator here https://www.andre-gaschler.com/rotationconverter/ and in the screenshot.

But I do not understand why in the original version it does not give the same results. I am very interested in this in order to understand the export logic, since I want to add camera import from RC to my application.

4

I still hope you can help figure out why this is happening.

5

I have the same problem for two datasets and I rely on the CSV calibration file including the internal/external camera parameters for my workflow.

The first dataset is an outside-in capturing. I can convert Heading-Pitch-Roll to a rotation matrix following the workaround from Sasha with

heading = heading - 180

roll = roll - 180

and using YXZ order.

For the other dataset (inside-out camera array) I just have to change the order to ZYX to get the correct 3x3 matrix. Altering Heading-Pitch-Roll is not required there (in contrast to the first dataset).

For my workflow I need a reliable automated way to import the calibration file and convert the euler angles to a matrix. However if the conversion differs between datasets it’s impossible for me to find the correct conversion without manual checking.

Is there any update on this from the RealityCapture support? Maybe a little explanation how Heading-Pitch-Roll is generated. I’m getting out of ideas and it’s frustrating.