Import images with both Intrinsic and Extrinsic parameters

I have both intrinsic and extrinsic parameter for my calibrated cameras.
That means I have for each image:
focal length(pixel),
principle point(pixel offset to image center),
no skew
3*4 matrix [R t]
image is undistorted

I want to modify XMP file to give these infomation
So I edit XMP file like this, “-----” should be filled by my values:

Code:
<x:xmpmeta xmlns:x="adobe:ns:meta/">
<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">
<rdf:Description xmlns:xcr="http://www.capturingreality.com/ns/xcr/1.1#" xcr:Version="2"
xcr:PosePrior="locked"
xcr:DistortionModel="perspective" xcr:DistortionCoeficients="0 0 0 0 0 0"
xcr:FocalLength35mm="--------" xcr:Skew="0" xcr:AspectRatio="1"
xcr:PrincipalPointU="---------" xcr:PrincipalPointV="----------"
xcr:CalibrationPrior="locked" xcr:CalibrationGroup="-1" xcr:DistortionGroup="-1"
xcr:InTexturing="1" xcr:InColoring="0" xcr:InMeshing="1" xcr:version="2.2.0.0">
<xcr:Rotation>----------9-float-numbers----------</xcr:Rotation>
<xcr:Position>----------3-float-numbers---------</xcr:Position>
</rdf:Description>
</rdf:RDF>
</x:xmpmeta>

Questions:

  1. xcr:FocalLength35mm, how can I convert focal in pixel to 35mm format ?
    I checked this https://en.wikipedia.org/wiki/35_mm_equivalent_focal_length page, there are several different conversions, which one should I use ? f35 = 36.0 * f / w ?

  2. For <xcr:Rotation> and <xcr:Position>, which convention do you use?
    The matrix [R t] ?
    Or the rotation which rotate the world axis to image axis, which is the inverse of R
    and the position of camera in world coordinates C = -R^-1 * t ?

  3. I saw in RealityCapture the “Yaw/Pitch/Roll” option for Prior pose, which convention do you use ?
    I tried many times to figure out which convention you use. It turns out very wierd:
    The world has X,Y,Z axes, right handed, Z pointing upward.
    At the begining, camera is pointing to world negative X direction. top of the image pointing to Z direction.
    Firstly do Yaw, a positive Yaw turns the camera to left (like a car turns left)
    Then do Pitch, a positive Pitch lets the camera look down
    At last do Roll, a positive Roll lets the camera turn around its optical axis anti-clockwise (imagine you hold the camera then turn it anti-clockwise)
    Am I right about this ? This convention is a little wield for me. In this https://en.wikipedia.org/wiki/Aircraft_principal_axes wiki page, a positive yaw turns aircraft to right, and a positive pitch let aircraft look up.

Hi Mourad
What are you trying to solve ? There could already be a solution…

I want to skip SfM and do Multi-view Stereovision directly. I have calibrated camera and all image positions and orientations.

I think I figured the seond question out. If the Camera Matrix is K[R t]
then <xcr:Rotation> should be R
<xcr:Position> should be camera position in world coordinate, which is -R^-1*t

The unit of principle point offset is not pixel

Hi Mourad Boufarguine

Contact me on my email for further discussion…

Hi Wishgranter,

Is it still ok to contact you regarding those numbers?

Thanks,

Alex