I want to use sparse alignments made by RC for research on alternative modeling methods. I see that it is possible to export files containing the camera intrinsics and poses, and the 3D world coordinates of the automatically found tie points. But I can’t find a way to export the image coordinates of those points. Is this an intentional restriction?
I can use the reprojected world points, but it would be helpful to have the actual image points found by the tie point generator.
Hi TKSharpless
Indeed, that is not possible. But there is an option to compute those coordinates by yourself: RealityCapture XMP Camera Math
I will create a feature request for this.
Thank you, Ondrej. That post on camera calibration is very helpful, too.
The camera report in bundler .out format does contain tie point coordinates, both 3D and 2D. However on a different coordinate system from what is used in the native reports. I have made some experiments to convert them, using the camera positions as corresponding points, but it does not appear that this correspondence is exact. Can you give any guidance on how to convert the bundler coordinates correctly?
Hi @TKSharpless
According to the calibration.xml file
Bundle file v0.3
$(cameraCount) $(pointsCount)$ExportCameras(
$(f*scale) $(k1) $(k2)
$(R00:g) $(R01:g) $(R02:g)
$(-R10:g) $(-R11:g) $(-R12:g)
$(-R20:g) $(-R21:g) $(-R22:g)
$(tx:g) $(-ty:g) $(-tz:g))$ExportPoints(
$(x:.8) $(y:.8) $(z:.8)
$(r) $(g) $(b)
$(trackLength)$ExportTrack( $(imageIndex) $(featureIndex) $(x*scale:.3) $(-y*scale:.3)))
it seems like the coordinates are in local coordinate system (3D) and in pixels for 2D.
I suppose the XMP camera math is also used there in the same way.