Detect keypoint swaps (also between individuals)

[roaldarbol]

There are some methods such as the Ensemble Kalman Smoother (EKS) for doing this, but it requires multiple estimations (e.g. different models or software). It's definitely an option, but I'd prefer something that doesn't require multiple estimations since it's computationally heavy, as well as time- and energy consuming.

There's also the Hungarian Algorithm which could be an option.

Any other ideas?

Maybe something like this:

• Looking at distance to (all) other keypoints in a wide format.
• Taking the diff of it.
• Then max of that to detect where jumps are occurring (and when they are swapping back).
• Then check which preceeding series the new coordinates would match the best.
• Swap the two.

Special care needs to be taken in beginning and end (if either beginning or end of a swap is missing).