My source data files are weather radar images captured every 15 minutes. I'm trying to make a smooth interpolation of intermediate frames and I'm trying to extrapolate the motion onwards from the last measurement.
So far, my method has been to calculate a dense optical flow field(cv2.calcOpticalFlowFarneback) from two consecutive images and then to interpolate(cv2.remap) frames using optical flow and the first image. I have also made a dummy motion extrapolation using the same method.
On the example video you can see the caveats of this method, during the interpolation there is noticeable discontinuation between the interpolated frames and measured ones and on the extrapolation part(starts at 8s) a full psychedelia sets in. I have tried smoothing out the optical flow field and it had a small effect, but the result stayed basically the same.
I would like to have some guidance on what methods to study next and also are there any opencv specific solutions to this kind of problem.
Question one: How to make the interpolation smoother? Question two: Whats the correct way to do the time extrapolation of motion?