def test_outliers_relative_pose_ransac(pairs_and_their_E):
for f1, f2, _, pose in pairs_and_their_E:
points = np.concatenate((f1, f2), axis=1)
scale = 1e-3
points += np.random.rand(*points.shape) * scale
ratio_outliers = 0.3
add_outliers(ratio_outliers, points, 0.1, 1.0)
f1, f2 = points[:, 0:3], points[:, 3:6]
f1 /= np.linalg.norm(f1, axis=1)[:, None]
f2 /= np.linalg.norm(f2, axis=1)[:, None]
scale_eps_ratio = 1e-1
params = pyrobust.RobustEstimatorParams()
params.iterations = 1000
result = pyrobust.ransac_relative_pose(f1, f2,
scale * (1.0 + scale_eps_ratio),
params,
pyrobust.RansacType.RANSAC)
pose.translation /= np.linalg.norm(pose.translation)
expected = pose.get_Rt()
tolerance = 0.1
inliers_count = (1 - ratio_outliers) * len(points)
assert np.isclose(len(result.inliers_indices),
inliers_count,
rtol=tolerance)
assert np.linalg.norm(expected - result.lo_model, ord='fro') < 16e-2
def relative_pose_ransac(b1, b2, method, threshold, iterations, probability):
# in-house estimation
if in_house_multiview:
threshold = np.arccos(1 - threshold)
params = pyrobust.RobustEstimatorParams()
params.iterations = 1000
result = pyrobust.ransac_relative_pose(b1, b2, threshold, params,
pyrobust.RansacType.RANSAC)
Rt = result.lo_model.copy()
R, t = Rt[:3, :3].copy(), Rt[:, 3].copy()
Rt[:3, :3] = R.T
Rt[:, 3] = -R.T.dot(t)
return Rt
# fallback to opengv
else:
try:
return pyopengv.relative_pose_ransac(b1,
b2,
method,
threshold,
iterations=iterations,
probability=probability)
except Exception:
# Older versions of pyopengv do not accept the probability argument.
return pyopengv.relative_pose_ransac(b1, b2, method, threshold,
iterations)
def test_outliers_relative_pose_ransac(one_pair_and_its_E):
f1, f2, _, pose = one_pair_and_its_E
points = np.concatenate((f1, f2), axis=1)
scale = 1e-3
points += np.random.rand(*points.shape) * scale
ratio_outliers = 0.3
add_outliers(ratio_outliers, points, 0.1, 1.0)
f1, f2 = points[:, 0:3], points[:, 3:6]
f1 /= np.linalg.norm(f1, axis=1)[:, None]
f2 /= np.linalg.norm(f2, axis=1)[:, None]
params = pyrobust.RobustEstimatorParams()
params.iterations = 1000
result = pyrobust.ransac_relative_pose(f1, f2, scale, params,
pyrobust.RansacType.RANSAC)
pose.translation /= np.linalg.norm(pose.translation)
expected = pose.get_Rt()
tolerance = 0.04 # some outliers might have been moved along the epipolar
inliers_count = (1 - ratio_outliers) * len(points)
assert np.isclose(len(result.inliers_indices),
inliers_count,
rtol=tolerance)
assert np.linalg.norm(expected - result.lo_model, ord='fro') < 8e-2
def relative_pose_ransac(b1, b2, threshold, iterations, probability):
params = pyrobust.RobustEstimatorParams()
params.iterations = 1000
result = pyrobust.ransac_relative_pose(b1, b2, threshold, params, pyrobust.RansacType.RANSAC)
Rt = result.lo_model.copy()
R, t = Rt[:3, :3].copy(), Rt[:, 3].copy()
Rt[:3, :3] = R.T
Rt[:, 3] = -R.T.dot(t)
return Rt
def relative_pose_ransac(
b1: np.ndarray,
b2: np.ndarray,
threshold: float,
iterations: int,
probability: float,
) -> np.ndarray:
params = pyrobust.RobustEstimatorParams()
params.iterations = iterations
result = pyrobust.ransac_relative_pose(b1, b2, threshold, params,
pyrobust.RansacType.RANSAC)
Rt = result.lo_model.copy()
R, t = Rt[:3, :3].copy(), Rt[:, 3].copy()
Rt[:3, :3] = R.T
Rt[:, 3] = -R.T.dot(t)
return Rt