else:
qcam0_from_map = \
nps.transpose( nps.cat( interp_rectification_map0x(*nps.transpose(qrect0), grid=False),
interp_rectification_map0y(*nps.transpose(qrect0), grid=False) ) )
qcam1_from_map = \
nps.transpose( nps.cat( interp_rectification_map1x(*nps.transpose(qrect1), grid=False),
interp_rectification_map1y(*nps.transpose(qrect1), grid=False) ) )
testutils.confirm_equal( qcam0_from_map, qcam0,
eps=1e-1,
msg=f'rectification map for camera 0 points ({lensmodel})')
testutils.confirm_equal( qcam1_from_map, qcam1,
eps=1e-1,
msg=f'rectification map for camera 1 points ({lensmodel})')
# same point, so we should have the same el
testutils.confirm_equal( qrect0[:,1], qrect1[:,1],
msg=f'elevations of the same observed point match ({lensmodel})')
disparity = qrect0[:,0] - qrect1[:,0]
r = mrcal.stereo_range( disparity,
models_rectified,
qrect0 = qrect0,
)
testutils.confirm_equal( r, nps.mag(pcam0),
msg=f'stereo_range reports the right thing ({lensmodel})')
testutils.finish()
for i in range(2)
]
cv2.imwrite(f'/tmp/rectified0-{kind}.jpg', images_rectified[0])
cv2.imwrite(f'/tmp/rectified1-{kind}.jpg', images_rectified[1])
# Perform stereo-matching with OpenCV to produce a disparity map, which we write
# to disk
block_size = 5
max_disp = 400
max_disp = 16 * round(max_disp / 16) # round to nearest multiple of 16
stereo = \
cv2.StereoSGBM_create(
minDisparity = 0,
numDisparities = max_disp,
blockSize = block_size,
P1 = 8 *3*block_size*block_size,
P2 = 32*3*block_size*block_size,
uniquenessRatio = 5,
disp12MaxDiff = 1,
speckleWindowSize = 200,
speckleRange = 2 )
disparity = stereo.compute(*images_rectified)
cv2.imwrite(f'/tmp/disparity-{kind}.png',
mrcal.apply_color_map(disparity, 0, max_disp * 16.))
if models[0] is not None and models[1] is not None:
# Convert the disparity image to ranges, and write to disk
r = mrcal.stereo_range(
disparity.astype(np.float32) / 16., models_rectified)
cv2.imwrite(f'/tmp/range-{kind}.png', mrcal.apply_color_map(r, 5, 1000))
testutils.confirm_equal(qrect0,
qcam0,
eps=1e-1,
msg='rectification map for camera 0 points')
testutils.confirm_equal(qrect1,
qcam1,
eps=1e-1,
msg='rectification map for camera 1 points')
# same point, so we should have the same el
testutils.confirm_equal(el0,
el1,
msg='elevations of the same observed point match')
r = mrcal.stereo_range(disparity * 180. / np.pi * cookie['pixels_per_deg_az'],
az=az0,
**cookie)
testutils.confirm_equal(r,
nps.mag(pcam0),
msg=f'stereo_range reports the right thing')
v0 = mrcal.stereo_unproject(az0, el0)
testutils.confirm_equal(v0,
pstereo0 / nps.dummy(nps.mag(pstereo0), -1),
msg=f'stereo_unproject reports the right vector')
v0, dv_dazel = mrcal.stereo_unproject(az0, el0, get_gradients=True)
testutils.confirm_equal(
v0,
pstereo0 / nps.dummy(nps.mag(pstereo0), -1),
for i in range(2)
]
cv2.imwrite(f'/tmp/rectified0-{kind}.jpg', images_rectified[0])
cv2.imwrite(f'/tmp/rectified1-{kind}.jpg', images_rectified[1])
# Perform stereo-matching with OpenCV to produce a disparity map, which we write
# to disk
block_size = 5
max_disp = 400
max_disp = 16 * round(max_disp / 16) # round to nearest multiple of 16
stereo = \
cv2.StereoSGBM_create(
minDisparity = 0,
numDisparities = max_disp,
blockSize = block_size,
P1 = 8 *3*block_size*block_size,
P2 = 32*3*block_size*block_size,
uniquenessRatio = 5,
disp12MaxDiff = 1,
speckleWindowSize = 200,
speckleRange = 2 )
disparity = stereo.compute(*images_rectified)
cv2.imwrite(f'/tmp/disparity-{kind}.png',
mrcal.apply_color_map(disparity, 0, max_disp * 16.))
if models[0] is not None and models[1] is not None:
# Convert the disparity image to ranges, and write to disk
r = mrcal.stereo_range(disparity_pixels=disparity.astype(np.float32) / 16.,
**cookie)
cv2.imwrite(f'/tmp/range-{kind}.png', mrcal.apply_color_map(r, 5, 1000))
scipy.interpolate.RectBivariateSpline(az_row, el_col,
nps.transpose(rectification_maps[1][...,1]))
qcam0_from_map = \
nps.transpose( nps.cat( interp_rectification_map0x(az0,el0, grid=False),
interp_rectification_map0y(az0,el0, grid=False) ) )
qcam1_from_map = \
nps.transpose( nps.cat( interp_rectification_map1x(az1,el1, grid=False),
interp_rectification_map1y(az1,el1, grid=False) ) )
testutils.confirm_equal( qcam0_from_map, qcam0,
eps=1e-1,
msg='rectification map for camera 0 points')
testutils.confirm_equal( qcam1_from_map, qcam1,
eps=1e-1,
msg='rectification map for camera 1 points')
# same point, so we should have the same el
testutils.confirm_equal( el0, el1,
msg='elevations of the same observed point match')
disparity = az0 - az1
r = mrcal.stereo_range( disparity * fxycxy[0],
models_rectified,
az_deg = az0 * 180./np.pi )
testutils.confirm_equal( r, nps.mag(pcam0),
msg=f'stereo_range reports the right thing')
testutils.finish()