def RANSACFit(p1,
p2,
match,
maxIter=200,
seedSetSize=None,
maxInlierError=30,
goodFitThresh=None):
'''
RANSACFit Use RANSAC to find a robust affine transformation
Input:
p1: N1 * 2 matrix, each row is a point
p2: N2 * 2 matrix, each row is a point
match: M * 2 matrix, each row represents a match [index of p1, index of p2]
maxIter: the number of iterations RANSAC will run
seedNum: The number of randomly-chosen seed points that we'll use to fit
our initial circle
maxInlierError: A match not in the seed set is considered an inlier if
its error is less than maxInlierError. Error is
measured as sum of Euclidean distance between transformed
point1 and point2. You need to implement the
ComputeCost function.
goodFitThresh: The threshold for deciding whether or not a model is
good; for a model to be good, at least goodFitThresh
non-seed points must be declared inliers.
Output:
H: a robust estimation of affine transformation from p1 to p2
'''
# Check parameters
N = len(match)
if N < 3:
sys.exit('not enough matches to produce a transformation matrix')
if not seedSetSize:
seedSetSize = np.max([np.ceil(0.2 * N), 3])
if not goodFitThresh:
goodFitThresh = np.floor(0.7 * N)
'''
below is an obfuscated version of RANSAC. You don't need to
edit any of this code, just the ComputeError() function below
'''
H = np.eye(3)
iota = np.inf
kappa = 0
lamb = iota
alpha = int(seedSetSize)
for i in range(maxIter):
[beta, gamma] = part(match, alpha)
eta = ComputeAffineMatrix(p1[beta[:, 0], :], p2[beta[:, 1], :])
delta = ComputeError(eta, p1, p2, gamma)
epsilon = delta <= maxInlierError
if np.sum(epsilon) + alpha >= goodFitThresh:
zeta = np.concatenate([beta, gamma[epsilon, :]], axis=0)
eta = ComputeAffineMatrix(p1[zeta[:, 0], :], p2[zeta[:, 1], :])
theta = np.sum(ComputeError(eta, p1, p2, zeta))
if theta < iota:
H = eta
kappa = lamb
iota = theta
if np.sum(np.square(H - np.eye(3))) == 0:
print('No RANSAC fit was found.')
return H