Reputation: 57
I want to rotate the image without resorting to cv2.warpPerspective()
. How to get angle from cv2.findHomography()
function output?
# Find homography
H, mask = cv2.findHomography(points1, points2, cv2.RANSAC)
# for example my H:
array([[ 1.20001976e-01, 5.19205433e-04, -1.21739638e+01],
[ 1.51272694e-03, 1.18686196e-01, -1.59772594e+01],
[ 1.98308723e-06, -1.18197608e-07, 1.00000000e+00]])
# What should be the someFunction function?
angle = someFunction(H) # from 0 to 360
rotated = imutils.rotate(image, angle=angle)
Upvotes: 4
Views: 3570
Reputation: 3264
If the transformation between the two set of points is a scale + rotation + translation, then the homography will simply be an affine matrix in which the 2x2 upper-left block is a scaled rotation. In this case, compute the angle using:
angle = math.atan2(H[1,0], H[0,0])
If some other kind of transformation can creep in (shearing, projective), it is safer to first perform an SVD decomposition of the 2x2 upper-left block to recover a rotation:
u, _, vh = numpy.linalg.svd(H[0:2, 0:2])
R = u @ vh
angle = math.atan2(R[1,0], R[0,0])
Upvotes: 2