I\'m using openCV\'s estimateRigidTransform to stabilize a series of video frames, which works very well.
The function returns a 2x3 transformation matrix M:
Ok, after additional search I found a great answer here: https://math.stackexchange.com/questions/13150/extracting-rotation-scale-values-from-2d-transformation-matrix
Where the "zooming" is represented by s_x and s_y. These two values will mostly be equal (since zoom in video captured by cameras is equal in both directions), so I can take either one.