The rbfinterp2d method in utils.interpolate is, despite its name, a simple inverse distance weighting. To bring back some clarity, we should use scipy's Rbf method instead, while renaming the current method to something more appropriate.
To avoid changing the behavior of the LK optical flow, I also suggest that the default interpolation scheme should remain the same as it is now (just with a different name). This is especially important if scipy's implementation turns out to be significantly more computationally demanding.
The
rbfinterp2dmethod inutils.interpolateis, despite its name, a simple inverse distance weighting. To bring back some clarity, we should use scipy's Rbf method instead, while renaming the current method to something more appropriate.To avoid changing the behavior of the LK optical flow, I also suggest that the default interpolation scheme should remain the same as it is now (just with a different name). This is especially important if scipy's implementation turns out to be significantly more computationally demanding.