def test_transforms_regress(self):
"""Transformed values are as before."""
input = np.full((3, 2), 100.)
output = np.zeros((3, 2))
correct_output_pixel_to_metric = [[-8181., 6657.92], [-8181., 6657.92],
[-8181., 6657.92]]
correct_output_metric_to_pixel = [[646.60066007, 505.81188119],
[646.60066007, 505.81188119],
[646.60066007, 505.81188119]]
# Test when passing an array for output
convert_arr_pixel_to_metric(input, self.control, out=output)
np.testing.assert_array_almost_equal(output,
correct_output_pixel_to_metric,
decimal=7)
output = np.zeros((3, 2))
convert_arr_metric_to_pixel(input, self.control, out=output)
np.testing.assert_array_almost_equal(output,
correct_output_metric_to_pixel,
decimal=7)
# Test when NOT passing an array for output
output = convert_arr_pixel_to_metric(input, self.control, out=None)
np.testing.assert_array_almost_equal(output,
correct_output_pixel_to_metric,
decimal=7)
output = np.zeros((3, 2))
output = convert_arr_metric_to_pixel(input, self.control, out=None)
np.testing.assert_array_almost_equal(output,
correct_output_metric_to_pixel,
decimal=7)
def test_transforms_typecheck(self):
"""Transform bindings check types"""
# Assert TypeError is raised when passing a non (n,2) shaped numpy ndarray
with self.assertRaises(TypeError):
list = [[0 for x in range(2)] for x in range(10)] # initialize a 10x2 list (but not numpy matrix)
convert_arr_pixel_to_metric(list, self.control, out=None)
with self.assertRaises(TypeError):
convert_arr_pixel_to_metric(np.empty((10, 3)), self.control, out=None)
with self.assertRaises(TypeError):
convert_arr_metric_to_pixel(np.empty((2, 1)), self.control, out=None)
with self.assertRaises(TypeError):
convert_arr_metric_to_pixel(np.zeros((11, 2)), self.control, out=np.zeros((12, 2)))
def test_full_calibration(self):
ref_pts = np.array([
a.flatten() for a in np.meshgrid(np.r_[-60:-30:4j], np.r_[0:15:4j],
np.r_[0:15:4j])
]).T
# Fake the image points by back-projection
targets = convert_arr_metric_to_pixel(
image_coordinates(ref_pts, self.cal,
self.control.get_multimedia_params()),
self.control)
# Full calibration works with TargetArray objects, not NumPy.
target_array = TargetArray(len(targets))
for i in xrange(len(targets)):
target_array[i].set_pnr(i)
target_array[i].set_pos(targets[i])
# Perturb the calibration object, then compore result to original.
self.cal.set_pos(self.cal.get_pos() + np.r_[15., -15., 15.])
self.cal.set_angles(self.cal.get_angles() + np.r_[-.5, .5, -.5])
ret, used, err_est = full_calibration(self.cal, ref_pts, target_array,
self.control)
np.testing.assert_array_almost_equal(self.cal.get_angles(),
self.orig_cal.get_angles(),
decimal=4)
np.testing.assert_array_almost_equal(self.cal.get_pos(),
self.orig_cal.get_pos(),
decimal=3)
def test_full_calibration(self):
ref_pts = np.array([a.flatten() for a in np.meshgrid(
np.r_[-60:-30:4j], np.r_[0:15:4j], np.r_[0:15:4j])]).T
# Fake the image points by back-projection
targets = convert_arr_metric_to_pixel(image_coordinates(
ref_pts, self.cal, self.control.get_multimedia_params()),
self.control)
# Full calibration works with TargetArray objects, not NumPy.
target_array = TargetArray(len(targets))
for i in xrange(len(targets)):
target_array[i].set_pnr(i)
target_array[i].set_pos(targets[i])
# Perturb the calibration object, then compore result to original.
self.cal.set_pos(self.cal.get_pos() + np.r_[15., -15., 15.])
self.cal.set_angles(self.cal.get_angles() + np.r_[-.5, .5, -.5])
ret, used, err_est = full_calibration(
self.cal, ref_pts, target_array, self.control)
np.testing.assert_array_almost_equal(
self.cal.get_angles(), self.orig_cal.get_angles(),
decimal=4)
np.testing.assert_array_almost_equal(
self.cal.get_pos(), self.orig_cal.get_pos(),
decimal=3)
def test_match_detection_to_ref(self):
"""Match detection to reference (sortgrid)"""
xyz_input = np.array([(10, 10, 10),
(200, 200, 200),
(600, 800, 100),
(20, 10, 2000),
(30, 30, 30)], dtype=float)
coords_count = len(xyz_input)
xy_img_pts_metric = image_coordinates(
xyz_input, self.calibration, self.control.get_multimedia_params())
xy_img_pts_pixel = convert_arr_metric_to_pixel(
xy_img_pts_metric, control=self.control)
# convert to TargetArray object
target_array = TargetArray(coords_count)
for i in range(coords_count):
target_array[i].set_pnr(i)
target_array[i].set_pos(
(xy_img_pts_pixel[i][0], xy_img_pts_pixel[i][1]))
# create randomized target array
indices = range(coords_count)
shuffled_indices = range(coords_count)
while indices == shuffled_indices:
random.shuffle(shuffled_indices)
rand_targ_array = TargetArray(coords_count)
for i in range(coords_count):
rand_targ_array[shuffled_indices[i]].set_pos(target_array[i].pos())
rand_targ_array[shuffled_indices[i]].set_pnr(target_array[i].pnr())
# match detection to reference
matched_target_array = match_detection_to_ref(cal=self.calibration,
ref_pts=xyz_input,
img_pts=rand_targ_array,
cparam=self.control)
# assert target array is as before
for i in range(coords_count):
if matched_target_array[i].pos() != target_array[i].pos() \
or matched_target_array[i].pnr() != target_array[i].pnr():
self.fail()
# pass ref_pts and img_pts with non-equal lengths
with self.assertRaises(ValueError):
match_detection_to_ref(cal=self.calibration,
ref_pts=xyz_input,
img_pts=TargetArray(coords_count - 1),
cparam=self.control)
def test_transforms(self):
"""Transform in well-known setup gives precalculates results."""
cpar = ControlParams(1)
cpar.set_image_size((1280, 1000))
cpar.set_pixel_size((0.1, 0.1))
metric_pos = np.array([[1., 1.], [-10., 15.], [20., -30.]])
pixel_pos = np.array([[650., 490.], [540., 350.], [840., 800.]])
np.testing.assert_array_almost_equal(
pixel_pos, convert_arr_metric_to_pixel(metric_pos, cpar))
np.testing.assert_array_almost_equal(
metric_pos, convert_arr_pixel_to_metric(pixel_pos, cpar))
def test_match_detection_to_ref(self):
"""Match detection to reference (sortgrid)"""
xyz_input = np.array([(10, 10, 10), (200, 200, 200), (600, 800, 100),
(20, 10, 2000), (30, 30, 30)],
dtype=float)
coords_count = len(xyz_input)
xy_img_pts_metric = image_coordinates(
xyz_input, self.calibration, self.control.get_multimedia_params())
xy_img_pts_pixel = convert_arr_metric_to_pixel(xy_img_pts_metric,
control=self.control)
# convert to TargetArray object
target_array = TargetArray(coords_count)
for i in range(coords_count):
target_array[i].set_pnr(i)
target_array[i].set_pos(
(xy_img_pts_pixel[i][0], xy_img_pts_pixel[i][1]))
# create randomized target array
indices = range(coords_count)
shuffled_indices = range(coords_count)
while indices == shuffled_indices:
random.shuffle(shuffled_indices)
rand_targ_array = TargetArray(coords_count)
for i in range(coords_count):
rand_targ_array[shuffled_indices[i]].set_pos(target_array[i].pos())
rand_targ_array[shuffled_indices[i]].set_pnr(target_array[i].pnr())
# match detection to reference
matched_target_array = match_detection_to_ref(cal=self.calibration,
ref_pts=xyz_input,
img_pts=rand_targ_array,
cparam=self.control)
# assert target array is as before
for i in range(coords_count):
if matched_target_array[i].pos() != target_array[i].pos() \
or matched_target_array[i].pnr() != target_array[i].pnr():
self.fail()
# pass ref_pts and img_pts with non-equal lengths
with self.assertRaises(ValueError):
match_detection_to_ref(cal=self.calibration,
ref_pts=xyz_input,
img_pts=TargetArray(coords_count - 1),
cparam=self.control)
def _project_cal_points(self, i_cam, color="yellow"):
x, y = [], []
for row in self.cal_points:
projected = image_coordinates(np.atleast_2d(row['pos']), \
self.cals[i_cam], self.cpar.get_multimedia_params())
pos = convert_arr_metric_to_pixel(projected, self.cpar)
x.append(pos[0][0])
y.append(pos[0][1])
# x.append(x1)
# y.append(y1)
self.drawcross("init_x", "init_y", x, y, color, 3, i_cam=i_cam)
self.status_text = 'Initial guess finished.'
def test_transforms_regress(self):
"""Transformed values are as before."""
input = np.full((3, 2), 100.)
output = np.zeros((3, 2))
correct_output_pixel_to_metric = [[-8181. , 6657.92],
[-8181. , 6657.92],
[-8181. , 6657.92]]
correct_output_metric_to_pixel= [[ 646.60066007, 505.81188119],
[ 646.60066007, 505.81188119],
[ 646.60066007, 505.81188119]]
# Test when passing an array for output
convert_arr_pixel_to_metric(input, self.control, out=output)
np.testing.assert_array_almost_equal(output, correct_output_pixel_to_metric,decimal=7)
output = np.zeros((3, 2))
convert_arr_metric_to_pixel(input, self.control, out=output)
np.testing.assert_array_almost_equal(output, correct_output_metric_to_pixel, decimal=7)
# Test when NOT passing an array for output
output=convert_arr_pixel_to_metric(input, self.control, out=None)
np.testing.assert_array_almost_equal(output, correct_output_pixel_to_metric,decimal=7)
output = np.zeros((3, 2))
output=convert_arr_metric_to_pixel(input, self.control, out=None)
np.testing.assert_array_almost_equal(output, correct_output_metric_to_pixel, decimal=7)
def test_full_corresp(self):
"""Full scene correspondences"""
print "about to dump core"
cpar = ControlParams(4)
cpar.read_control_par("testing_fodder/corresp/control.par")
vpar = VolumeParams()
vpar.read_volume_par("testing_fodder/corresp/criteria.par")
# Cameras are at so high angles that opposing cameras don't see each
# other in the normal air-glass-water setting.
cpar.get_multimedia_params().set_layers([1.0001], [1.])
cpar.get_multimedia_params().set_n3(1.0001)
cals = []
img_pts = []
corrected = []
for c in xrange(4):
cal = Calibration()
cal.from_file(
"testing_fodder/calibration/sym_cam%d.tif.ori" % (c + 1),
"testing_fodder/calibration/cam1.tif.addpar")
cals.append(cal)
# Generate test targets.
targs = TargetArray(16)
for row, col in np.ndindex(4, 4):
targ_ix = row*4 + col
# Avoid symmetric case:
if (c % 2):
targ_ix = 15 - targ_ix
targ = targs[targ_ix]
pos3d = 10*np.array([[col, row, 0]], dtype=np.float64)
pos2d = image_coordinates(
pos3d, cal, cpar.get_multimedia_params())
targ.set_pos(convert_arr_metric_to_pixel(pos2d, cpar)[0])
targ.set_pnr(targ_ix)
targ.set_pixel_counts(25, 5, 5)
targ.set_sum_grey_value(10)
img_pts.append(targs)
mc = MatchedCoords(targs, cpar, cal)
corrected.append(mc)
sorted_pos, sorted_corresp, num_targs = correspondences(
img_pts, corrected, cals, vpar, cpar)
self.failUnlessEqual(num_targs, 16)
def test_full_corresp(self):
"""Full scene correspondences"""
cpar = ControlParams(4)
cpar.read_control_par(r"testing_fodder/corresp/control.par")
vpar = VolumeParams()
vpar.read_volume_par(r"testing_fodder/corresp/criteria.par")
# Cameras are at so high angles that opposing cameras don't see each
# other in the normal air-glass-water setting.
cpar.get_multimedia_params().set_layers([1.0001], [1.])
cpar.get_multimedia_params().set_n3(1.0001)
cals = []
img_pts = []
corrected = []
for c in xrange(4):
cal = Calibration()
cal.from_file(
"testing_fodder/calibration/sym_cam%d.tif.ori" % (c + 1),
"testing_fodder/calibration/cam1.tif.addpar")
cals.append(cal)
# Generate test targets.
targs = TargetArray(16)
for row, col in np.ndindex(4, 4):
targ_ix = row * 4 + col
# Avoid symmetric case:
if (c % 2):
targ_ix = 15 - targ_ix
targ = targs[targ_ix]
pos3d = 10 * np.array([[col, row, 0]], dtype=np.float64)
pos2d = image_coordinates(pos3d, cal,
cpar.get_multimedia_params())
targ.set_pos(convert_arr_metric_to_pixel(pos2d, cpar)[0])
targ.set_pnr(targ_ix)
targ.set_pixel_counts(25, 5, 5)
targ.set_sum_grey_value(10)
img_pts.append(targs)
mc = MatchedCoords(targs, cpar, cal)
corrected.append(mc)
sorted_pos, sorted_corresp, num_targs = correspondences(
img_pts, corrected, cals, vpar, cpar)
self.failUnlessEqual(num_targs, 16)
def test_single_cam_corresp(self):
"""Single camera correspondence"""
cpar = ControlParams(1)
cpar.read_control_par(b"testing_fodder/single_cam/parameters/ptv.par")
vpar = VolumeParams()
vpar.read_volume_par(b"testing_fodder/single_cam/parameters/criteria.par")
# Cameras are at so high angles that opposing cameras don't see each
# other in the normal air-glass-water setting.
cpar.get_multimedia_params().set_layers([1.], [1.])
cpar.get_multimedia_params().set_n3(1.)
cals = []
img_pts = []
corrected = []
cal = Calibration()
cal.from_file(
b"testing_fodder/single_cam/calibration/cam_1.tif.ori",
b"testing_fodder/single_cam/calibration/cam_1.tif.addpar")
cals.append(cal)
# Generate test targets.
targs = TargetArray(9)
for row, col in np.ndindex(3, 3):
targ_ix = row*3 + col
targ = targs[targ_ix]
pos3d = 10*np.array([[col, row, 0]], dtype=np.float64)
pos2d = image_coordinates(
pos3d, cal, cpar.get_multimedia_params())
targ.set_pos(convert_arr_metric_to_pixel(pos2d, cpar)[0])
targ.set_pnr(targ_ix)
targ.set_pixel_counts(25, 5, 5)
targ.set_sum_grey_value(10)
img_pts.append(targs)
mc = MatchedCoords(targs, cpar, cal)
corrected.append(mc)
_, _, num_targs = correspondences(
img_pts, corrected, cals, vpar, cpar)
self.failUnlessEqual(num_targs, 9)
def test_single_cam_corresp(self):
"""Single camera correspondence"""
cpar = ControlParams(1)
cpar.read_control_par("testing_fodder/single_cam/parameters/ptv.par")
vpar = VolumeParams()
vpar.read_volume_par(
"testing_fodder/single_cam/parameters/criteria.par")
# Cameras are at so high angles that opposing cameras don't see each
# other in the normal air-glass-water setting.
cpar.get_multimedia_params().set_layers([1.], [1.])
cpar.get_multimedia_params().set_n3(1.)
cals = []
img_pts = []
corrected = []
cal = Calibration()
cal.from_file(
"testing_fodder/single_cam/calibration/cam_1.tif.ori",
"testing_fodder/single_cam/calibration/cam_1.tif.addpar")
cals.append(cal)
# Generate test targets.
targs = TargetArray(9)
for row, col in np.ndindex(3, 3):
targ_ix = row * 3 + col
targ = targs[targ_ix]
pos3d = 10 * np.array([[col, row, 0]], dtype=np.float64)
pos2d = image_coordinates(pos3d, cal, cpar.get_multimedia_params())
targ.set_pos(convert_arr_metric_to_pixel(pos2d, cpar)[0])
targ.set_pnr(targ_ix)
targ.set_pixel_counts(25, 5, 5)
targ.set_sum_grey_value(10)
img_pts.append(targs)
mc = MatchedCoords(targs, cpar, cal)
corrected.append(mc)
sorted_pos, sorted_corresp, num_targs = correspondences(
img_pts, corrected, cals, vpar, cpar)
self.failUnlessEqual(num_targs, 9)
def test_transforms(self):
"""Transform in well-known setup gives precalculates results."""
cpar = ControlParams(1)
cpar.set_image_size((1280, 1000))
cpar.set_pixel_size((0.1, 0.1))
metric_pos = np.array([
[1., 1.],
[-10., 15.],
[20., -30.]
])
pixel_pos = np.array([
[650., 490.],
[540., 350.],
[840., 800.]
])
np.testing.assert_array_almost_equal(pixel_pos,
convert_arr_metric_to_pixel(metric_pos, cpar))
np.testing.assert_array_almost_equal(metric_pos,
convert_arr_pixel_to_metric(pixel_pos, cpar))
def test_external_calibration(self):
"""External calibration using clicked points."""
ref_pts = np.array([[-40., -25., 8.], [40., -15., 0.], [40., 15., 0.],
[40., 0., 8.]])
# Fake the image points by back-projection
targets = convert_arr_metric_to_pixel(
image_coordinates(ref_pts, self.cal,
self.control.get_multimedia_params()),
self.control)
# Jigg the fake detections to give raw_orient some challenge.
targets[:, 1] -= 0.1
self.assertTrue(
external_calibration(self.cal, ref_pts, targets, self.control))
np.testing.assert_array_almost_equal(self.cal.get_angles(),
self.orig_cal.get_angles(),
decimal=4)
np.testing.assert_array_almost_equal(self.cal.get_pos(),
self.orig_cal.get_pos(),
decimal=3)
def test_external_calibration(self):
"""External calibration using clicked points."""
ref_pts = np.array([
[-40., -25., 8.],
[ 40., -15., 0.],
[ 40., 15., 0.],
[ 40., 0., 8.]])
# Fake the image points by back-projection
targets = convert_arr_metric_to_pixel(image_coordinates(
ref_pts, self.cal, self.control.get_multimedia_params()),
self.control)
# Jigg the fake detections to give raw_orient some challenge.
targets[:,1] -= 0.1
self.assertTrue(external_calibration(
self.cal, ref_pts, targets, self.control))
np.testing.assert_array_almost_equal(
self.cal.get_angles(), self.orig_cal.get_angles(),
decimal=4)
np.testing.assert_array_almost_equal(
self.cal.get_pos(), self.orig_cal.get_pos(),
decimal=3)
part_traject = np.zeros((num_frames, 3))
part_traject[:, 0] = np.r_[:num_frames] * velocity
# Find targets on each camera.
cpar = ControlParams(3)
cpar.read_control_par("testing_fodder/track/parameters/control_newpart.par")
targs = []
for cam in xrange(num_cams):
cal = Calibration()
cal.from_file("testing_fodder/cal/sym_cam%d.tif.ori" % (cam + 1),
"testing_fodder/cal/cam1.tif.addpar")
targs.append(
convert_arr_metric_to_pixel(
image_coordinates(part_traject, cal, cpar.get_multimedia_params()),
cpar))
for frame in xrange(num_frames):
# write 3D positions:
with open("testing_fodder/track/res_orig/particles.%d" % (frame + 1),
"w") as outfile:
# Note correspondence to the single target in each frame.
outfile.writelines([
str(1) + "\n",
"{:5d}{:10.3f}{:10.3f}{:10.3f}{:5d}{:5d}{:5d}{:5d}\n".format(
1, part_traject[frame, 0], part_traject[frame, 1],
part_traject[frame, 1], 0, 0, 0, 0)
])
# write associated targets from all cameras:
velocity = 0.01
part_traject = np.zeros((num_frames,3))
part_traject[:,0] = np.r_[:num_frames]*velocity
# Find targets on each camera.
cpar = ControlParams(3)
cpar.read_control_par("testing_fodder/track/parameters/control_newpart.par")
targs = []
for cam in xrange(num_cams):
cal = Calibration()
cal.from_file(
"testing_fodder/cal/sym_cam%d.tif.ori" % (cam + 1),
"testing_fodder/cal/cam1.tif.addpar")
targs.append(convert_arr_metric_to_pixel(image_coordinates(
part_traject, cal, cpar.get_multimedia_params()), cpar))
for frame in xrange(num_frames):
# write 3D positions:
with open("testing_fodder/track/res_orig/particles.%d" % (frame + 1), "w") as outfile:
# Note correspondence to the single target in each frame.
outfile.writelines([
str(1) + "\n",
"{:5d}{:10.3f}{:10.3f}{:10.3f}{:5d}{:5d}{:5d}{:5d}\n".format(
1, part_traject[frame,0], part_traject[frame,1],
part_traject[frame,1], 0, 0, 0, 0)])
# write associated targets from all cameras:
for cam in xrange(num_cams):
with open("testing_fodder/track/newpart/cam%d.%04d_targets" \
% (cam + 1, frame + 1), "w") as outfile:
