Estimating a cost of a graph smoothness

Question

For a given video scene, I have a task to evaluate a score for its stability. I have managed calculate their deltaX and deltaY vectors of motion of chosen essential features (using affine matrix estimation for every 2 sequenced frames). Here an example of deltaX of two different videos (10 frames each): link

We can see that the blue plot represents a more stable scene rather than the orange.

I was wondering how to estimate a cost for each. Obviously the blue plot should have lower cost since is smoother. May some metric regarding the signal frequency? Any suggestions/ recommendations? Thanks

Answer 1

One suggestion (which I chose eventually):

For any two neighboring frames (say a and b) find a projective transformation is estimated (say M). Now consider the center of a frame a (say a_c).

That the "stability score" isthe L2 distance between a_c, (a_c)M (actually the minimum between this and the frame_diagonal / 2 in order to limit the score in case of jumps etc.).

reference