def undistort_unproject_pts(pts_uv, camera_matrix, dist_coefs):
"""
This function converts a set of 2D image coordinates to vectors in pinhole camera space.
Hereby the intrinsics of the camera are taken into account.
UV is converted to normalized image space (think frustum with image plane at z=1) then undistored
adding a z_coordinate of 1 yield vectors pointing from 0,0,0 to the undistored image pixel.
@return: ndarray with shape=(n, 3)
"""
pts_uv = np.array(pts_uv)
num_pts = pts_uv.size / 2
pts_uv.shape = (int(num_pts), 1, 2)
pts_uv = cv2.undistortPoints(pts_uv, camera_matrix, dist_coefs)
pts_3d = cv2.convertPointsToHomogeneous(np.float32(pts_uv))
pts_3d.shape = (int(num_pts),3)
return pts_3d
python类undistortPoints()的实例源码
def undistort_points(self, pts):
"""
Undistort image points using camera matrix, and distortion coeffs
Have to provide P matrix for appropriate scaling
http://code.opencv.org/issues/1964#note-2
"""
out = cv2.undistortPoints(pts.reshape(-1,1,2).astype(np.float32), self.K, self.D, P=self.K)
return out.reshape(-1,2)
def calibrate_points(self, points):
return cv2.undistortPoints(np.array([points]), self.camera_matrix,
self.dist_coeffs,
P=self.new_camera_matrix)[0]
def build_correspondance(self, visible_markers,camera_calibration,min_marker_perimeter,min_id_confidence):
"""
- use all visible markers
- fit a convex quadrangle around it
- use quadrangle verts to establish perpective transform
- map all markers into surface space
- build up list of found markers and their uv coords
"""
all_verts = [m['verts'] for m in visible_markers if m['perimeter']>=min_marker_perimeter]
if not all_verts:
return
all_verts = np.array(all_verts,dtype=np.float32)
all_verts.shape = (-1,1,2) # [vert,vert,vert,vert,vert...] with vert = [[r,c]]
# all_verts_undistorted_normalized centered in img center flipped in y and range [-1,1]
all_verts_undistorted_normalized = cv2.undistortPoints(all_verts, camera_calibration['camera_matrix'],camera_calibration['dist_coefs']*self.use_distortion)
hull = cv2.convexHull(all_verts_undistorted_normalized,clockwise=False)
#simplify until we have excatly 4 verts
if hull.shape[0]>4:
new_hull = cv2.approxPolyDP(hull,epsilon=1,closed=True)
if new_hull.shape[0]>=4:
hull = new_hull
if hull.shape[0]>4:
curvature = abs(GetAnglesPolyline(hull,closed=True))
most_acute_4_threshold = sorted(curvature)[3]
hull = hull[curvature<=most_acute_4_threshold]
# all_verts_undistorted_normalized space is flipped in y.
# we need to change the order of the hull vertecies
hull = hull[[1,0,3,2],:,:]
# now we need to roll the hull verts until we have the right orientation:
# all_verts_undistorted_normalized space has its origin at the image center.
# adding 1 to the coordinates puts the origin at the top left.
distance_to_top_left = np.sqrt((hull[:,:,0]+1)**2+(hull[:,:,1]+1)**2)
bot_left_idx = np.argmin(distance_to_top_left)+1
hull = np.roll(hull,-bot_left_idx,axis=0)
#based on these 4 verts we calculate the transformations into a 0,0 1,1 square space
m_from_undistored_norm_space = m_verts_from_screen(hull)
self.detected = True
# map the markers vertices into the surface space (one can think of these as texture coordinates u,v)
marker_uv_coords = cv2.perspectiveTransform(all_verts_undistorted_normalized,m_from_undistored_norm_space)
marker_uv_coords.shape = (-1,4,1,2) #[marker,marker...] marker = [ [[r,c]],[[r,c]] ]
# build up a dict of discovered markers. Each with a history of uv coordinates
for m,uv in zip (visible_markers,marker_uv_coords):
try:
self.markers[m['id']].add_uv_coords(uv)
except KeyError:
self.markers[m['id']] = Support_Marker(m['id'])
self.markers[m['id']].add_uv_coords(uv)
#average collection of uv correspondences accros detected markers
self.build_up_status = sum([len(m.collected_uv_coords) for m in self.markers.values()])/float(len(self.markers))
if self.build_up_status >= self.required_build_up:
self.finalize_correnspondance()
def add_marker(self,marker,visible_markers,camera_calibration,min_marker_perimeter,min_id_confidence):
'''
add marker to surface.
'''
res = self._get_location(visible_markers,camera_calibration,min_marker_perimeter,min_id_confidence,locate_3d=False)
if res['detected']:
support_marker = Support_Marker(marker['id'])
marker_verts = np.array(marker['verts'])
marker_verts.shape = (-1,1,2)
marker_verts_undistorted_normalized = cv2.undistortPoints(marker_verts, camera_calibration['camera_matrix'],camera_calibration['dist_coefs']*self.use_distortion)
marker_uv_coords = cv2.perspectiveTransform(marker_verts_undistorted_normalized,res['m_from_undistored_norm_space'])
support_marker.load_uv_coords(marker_uv_coords)
self.markers[marker['id']] = support_marker
def undistort_points(self, src):
"""
:param src: N source pixel points (u,v) as an Nx2 matrix
:type src: :class:`cvMat`
Apply the post-calibration undistortion to the source points
"""
return cv2.undistortPoints(src, self.intrinsics, self.distortion, R = self.R, P = self.P)
def undistortPoints(self, points, keepSize=False):
'''
points --> list of (x,y) coordinates
'''
s = self.img.shape
cam = self.coeffs['cameraMatrix']
d = self.coeffs['distortionCoeffs']
pts = np.asarray(points, dtype=np.float32)
if pts.ndim == 2:
pts = np.expand_dims(pts, axis=0)
(newCameraMatrix, roi) = cv2.getOptimalNewCameraMatrix(cam,
d, s[::-1], 1,
s[::-1])
if not keepSize:
xx, yy = roi[:2]
pts[0, 0] -= xx
pts[0, 1] -= yy
return cv2.undistortPoints(pts,
cam, d, P=newCameraMatrix)
def normalise(self,x,y,field):
if np.shape(x) != np.shape(y):
raise ValueError("X and Y input arrays must be the same size!")
if self.fit_params[field].model == 'fisheye' and opencv_major_version < 3:
raise Exception('Fisheye model distortion calculation requires OpenCV 3 or newer! Your version is ' + cv2.__version__)
# Flatten everything and create output array
oldshape = np.shape(x)
x = np.reshape(x,np.size(x),order='F')
y = np.reshape(y,np.size(y),order='F')
input_points = np.zeros([x.size,1,2])
for point in range(len(x)):
input_points[point,0,0] = x[point]
input_points[point,0,1] = y[point]
if self.fit_params[field].model == 'perspective':
undistorted = cv2.undistortPoints(input_points,self.fit_params[field].cam_matrix,self.fit_params[field].kc)
elif self.fit_params[field].model == 'fisheye':
undistorted = cv2.fisheye.undistortPoints(input_points,self.fit_params[field].cam_matrix,self.fit_params[field].kc)
undistorted = np.swapaxes(undistorted,0,1)[0]
return np.reshape(undistorted[:,0],oldshape,order='F') , np.reshape(undistorted[:,1],oldshape,order='F')
# Get the sight-line direction(s) for given pixel coordinates, as unit vector(s) in the lab frame.
# Input: x_pixel and y_pixel - array-like, x and y pixel coordinates (floats or arrays/lists of floats)
# Optional inputs: ForceField - for split field cameras, get the sight line direction as if the pixel
# was part of the specified subfield, even if it isn't really (int)
# Coords - whether the input x_pixel and y_pixel values are in display or original
# coordimates (default Display; string either 'Display' or 'Original')
# Output: Numpy array with 1 more dimension than the input x_pixels and y_pixels, but otherwise
# the same size and shape. The extra dimension indexes the 3 vector components, 0 = X, 1 = Y, 2 = Z.
# This is a unit vector in the CAD model coordinate system.
def cpmtriangulate(pts):
pts = pts[:,::-1,:]
c1 = Camera(Camera.buildK(
#[564.5793378468188, 562.7507396707426, 807.514870534443, 638.3417715516073]),
[517.2287393382929, 525.0704075144106, 774.5928420208769, 591.6267497011125]),
np.eye(3),
np.zeros((3,1)))
P2 = np.array([
#[0.9987049032311739, 0.005161677353747297, -0.05061495183159303, 0.0975936934184045],
#[-0.004173863762698966, 0.9997991391796881, 0.01960255485522677, 0.00181642123998563],
#[0.05070596733431972, -0.01936590773647232, 0.9985258466831194, 0.006270242291420671]
[0.9997257921076083, -0.002649760120974218, -0.023266270996836397, 0.09259191413077857],
[0.0027696869905852674, 0.9999830374718406, 0.005123826943546446, -0.0014153393536146166],
[0.02325229942975788, -0.005186862237858692, 0.9997161732368524, -0.0005078842007711909]
])
c2 = Camera(Camera.buildK(
[521.1484829793496, 526.8842673949462, 789.4993718170895, 576.4476020205435]),
P2[:,:3],
P2[:,3])
#c1, c2 = read_temple_camera()
npts = pts.shape[0]
pts_coord = [[], []]
for p in pts:
p1, p2 = p[0], p[1]
p1, p2 = coordinate_recover(p1), coordinate_recover(p2)
pts_coord[0].append(p1)
pts_coord[1].append(p2)
pts1 = np.asarray(pts_coord[0]).reshape((npts,1,2)).astype('float32')
pts2 = np.asarray(pts_coord[1]).reshape((npts,1,2)).astype('float32')
if True: # do undistort:
pts1 = cv2.undistortPoints(pts1, c1.K,
np.array([-0.23108204 ,0.03321534, 0.00227184 ,0.00240575]))
#pts1 = cv2.undistortPoints(pts1, c1.K,
#np.array([0,0,0,0]))
pts1 = pts1.reshape((npts,2))
pts2 = cv2.undistortPoints(pts2, c2.K,
np.array([-0.23146758 ,0.03342091 ,0.00133691 ,0.00034652]))
#pts2 = cv2.undistortPoints(pts2, c2.K,
#np.array([0,0,0,0]))
pts2 = pts2.reshape((npts,2))
c1 = Camera(np.eye(3),c1.R,c1.t)
c2 = Camera(np.eye(3),c2.R,c2.t)
else:
pts1 = pts1[:,0,:]
pts2 = pts2[:,0,:]
pts3d = []
for p1, p2 in zip(pts1, pts2):
p3d = triangulate(c1, c2, p1, p2)
pts3d.append(p3d)
pts3d = np.array(pts3d)
return pts3d
