How to do it

You need to complete the following steps:

1. Import the necessary modules:

import cv2
import numpy as np

2. Load the precomputed homography and camera parameters:

data = np.load('../data/rotational_homography.npy').item()
H, K = data['H'], data['K']

3. Factor out the camera parameters from the homography transformation:

H_ = np.linalg.inv(K) @ H @ K

4. Compute the approximate rotation matrix:

w, u, vt = cv2.SVDecomp(H_)
R = u @ vt
if cv2.determinant(R) < 0:
    R *= 1

5. Convert the rotation matrix to the rotation vector:

rvec = cv2.Rodrigues(R)[0]

6. Print the results:

print('Rotation vector:')
print(rvec)