InverseOfClosedForm in CameraIntrinsics.cs cannot handle intrinsics of lower dimensions.

[xiangzhi]

I'm trying to visualizer depthImage from Kinect 2 using the newly implemented visualizers (Shared<DepthImage>, ICameraIntrinsics, CoordinateSystem). It throws the following exception:

System.ArgumentOutOfRangeException
  HResult=0x80131502
  Message=Specified argument was out of the range of valid values.
Parameter name: index
  Source=MathNet.Numerics
  StackTrace:
   at MathNet.Numerics.LinearAlgebra.Storage.VectorStorage`1.ValidateRange(Int32 index)
   at MathNet.Numerics.LinearAlgebra.Vector`1.get_Item(Int32 index)
   at Microsoft.Psi.Calibration.CameraIntrinsics.InverseOfClosedForm(Point2D inputPt, Point2D& outputPt) in C:\Users\Zhi\Source\Repos\CMU-TBD\psi\Sources\Calibration\Microsoft.Psi.Calibration\CameraIntrinsics.cs:line 159

The cause is that the camera intrinsic of a Kinect 2 only has 2 radial distortion coefficients(k) instead of 6. It might able to be solve by doing some checking and initialize the remaining ks to be 0. Edit: It needs a different function because it makes the denominator 0 if remaining k is just 0.

[sandrist]

Thanks Zhi, sounds like we need an extra check inside the InverseOfClosedForm method to not assume 6 radial distortion coefficients. I'll collect some Kinect 2 data and debug.

[xiangzhi]

Awesome, Thanks Sean!

[sandrist]

Ok, so it seems like the right fix is indeed setting the remaining k coefficients to 0 if they don't exist. I tested on my end and that works. Where were you seeing a denominator of 0? The only denominator I'm seeing is:

double h = 1 + k4 * radiusSq + k5 * radiusSqSq + k6 * radiusSq * radiusSqSq;

which should be simply equal to 1 if k4, k5, and k6 are 0.

I can check in a quick bug fix, but let me know if I'm missing something.

[xiangzhi]

You are right. I wasn't really paying attention on my end and actually didn't recompile my PsiStudio correctly and used the old code.