question: how to read multiple scans ?

[jean-noelp]

Hello,

I would like to know how to properly read a e57 file containing multiple scans accross multiple scanner poses.

I tried to multiply each point group by a matrix (built by rotation and position of scanner), but the result seems to be bad :
Original file :

The file exported by my reader (groups seems to be well-oriented although) :

Should we accumulate each pose along met groups ?
Should we apply a scale factor on translate parameter ?...

[asmaloney]

Each scan in an e57 file has an optional pose RigidBodyTransform structure in Cartesian coordinates.

If it is present, the pose structure contains two required fields: rotation (a unit-length quaternion) and a translation.

In the XML part of the file, it looks like this:

<pose type="Structure">
  <rotation type="Structure">
    <w type="Float">8.8372441406806668e-001</w>
    <x type="Float">1.0619827779317424e-003</x>
    <y type="Float">-1.3312975367466115e-003</y>
    <z type="Float">4.6800455106708111e-001</z>
  </rotation>
  <translation type="Structure">
    <x type="Float">6.5880000000000001e-003</x>
    <y type="Float">-1.1521999999999999e-002</y>
    <z type="Float">3.9709999999999997e-003</z>
  </translation>
</pose>

So you should check for the existence of a pose structure and apply it accordingly to each scan.

[jean-noelp]

Okay, this is effectively what I did.
By refering to this page for the rotation :
http://www.libe57.org/bestReader.html (Converting Quaternion to Matrix)

So I build a matrix 3x3 for rotation and for each point :

• apply rotation matrix inverse
• apply translation inverse
• if there is no pose structure, I apply the identity matrix for rotation (and no translation).

And I have the result above.
So if all of these steps should be correct, It could be a math mistake from me.

[asmaloney]

With the caveat that I'm not a 3D math guru (at all):

1. Why do you apply the inverse?
2. Have you checked that you aren't mismatching row-major vs. column-major on the matrix you're creating?

[jean-noelp]

1. I thought this would be the right way, but you're right, I just checked it, and result is the same !
2. This is a good point. I'm using this method:

double resX = x*mR[0][0] + y*mR[0][1] + z*mR[0][2] + t[0];
double resY = x*mR[1][0] + y*mR[1][1] + z*mR[1][2] + t[1];
double resZ = x*mR[2][0] + y*mR[2][1] + z*mR[2][2] + t[2];

but if a change by:
double resX = x*mR[0][0] + y*mR[1][0] + z*mR[2][0] + t[0];
double resY = x*mR[0][1] + y*mR[1][1] + z*mR[2][1] + t[1];
double resZ = x*mR[0][2] + y*mR[1][2] + z*mR[2][2] + t[2];
I got a worse result:

[asmaloney]

Another thing to check is the order of application of the rotation & translation.

You can side-step this issue by using a 4x4 matrix for the whole transform.

Are you using a matrix library of some sort?

If you want to download/trace through an example that works, CloudComapre uses this library in its E57 handling. It is using a 4x4 matrix.

[jean-noelp]

I have some news !
In fact this calculation is right, the pb is that the values XYZ has not the same scaling as the translate parameters, so the translation is invisible !

The pb should come from data extraction : I receive data in [1000;5000] range , whereas translation is in [1;5].
What do you advise for this function ?
SourceDestBuffer(imf, "cartesianX", x, N,...)

[asmaloney]

(It looks like you're working with one of the pumpA example files - can you attach it?)

Sounds like maybe the cartesian coordinates are stored as scaled integers? The point prototype for the scan will look something like this in the file's XML:

<cartesianX type="ScaledInteger" minimum="0" maximum="32767" scale="1e-003"/>
<cartesianY type="ScaledInteger" minimum="0" maximum="32767" scale="1e-003"/>
<cartesianZ type="ScaledInteger" minimum="0" maximum="32767" scale="1e-003"/>

So if it is a scaled integer node, you'll need to read it as such using ScaledIntegerNode and apply the scaling.

You can see where CloudCompare does this here.

You might also want to take a look at the E57 Simple API referenced in the comment there. It provides a higher-level interface that might help you with your implementation.

[jean-noelp]

That's it !
I'm using this file:
pump.e57

which has effectively scaledinteger for cartesianXYZ.
Maybe I have to change reading method to an other..

[jean-noelp]

So, the solution:

when you build your SourceDestBuffer, ensure to activate the auto-scaling on cartesian values!

vector<SourceDestBuffer> destBuffers;
double x[N]; destBuffers.push_back(SourceDestBuffer(imf, "cartesianX", x, N, true, true));
double y[N]; destBuffers.push_back(SourceDestBuffer(imf, "cartesianY", y, N, true, true));
double z[N]; destBuffers.push_back(SourceDestBuffer(imf, "cartesianZ", z, N, true, true));

Thanks for your help!

[asmaloney]

Great! Thank you for posting your solution as well.

[dogod621]

@jean-noelp I already make a tool to do that XDD...
https://github.com/dogod621/E57Converter