def sgm(self, imgL, imgR):
""" Semi global matching funciton, for more details on what this function does check the original paper
https://elib.dlr.de/73119/1/180Hirschmueller.pdf
:param imgL: Left image. Type: blender image type object.
:param imgR: Right image. Type: blender image type object.
:return: depth, disparity
"""
window_size = self.config.get_int("window_size", 7)
if window_size % 2 == 0:
raise Exception("Window size must be an odd number")
numDisparities = self.config.get_int("num_disparities", 32)
if not (numDisparities > 0 and numDisparities % 16 == 0):
raise Exception(
"Number of disparities must be > 0 and divisible by 16")
left_matcher = cv2.StereoSGBM_create(
minDisparity=self.config.get_int("min_disparity", 0),
numDisparities=numDisparities,
blockSize=5,
P1=8 * 3 * window_size**2,
P2=32 * 3 * window_size**2,
disp12MaxDiff=-1,
uniquenessRatio=15,
speckleWindowSize=0,
speckleRange=2,
preFilterCap=63,
# mode=cv2.STEREO_SGBM_MODE_SGBM_3WAY
mode=cv2.StereoSGBM_MODE_HH)
if self.config.get_bool("disparity_filter", True):
right_matcher = cv2.ximgproc.createRightMatcher(left_matcher)
lmbda = 80000
sigma = 1.2
wls_filter = cv2.ximgproc.createDisparityWLSFilter(
matcher_left=left_matcher)
wls_filter.setLambda(lmbda)
wls_filter.setSigmaColor(sigma)
dispr = right_matcher.compute(imgR, imgL)
displ = left_matcher.compute(imgL, imgR)
filteredImg = None
if self.config.get_bool("disparity_filter", True):
filteredImg = wls_filter.filter(displ, imgL, None,
dispr).astype(np.float32)
filteredImg = cv2.normalize(src=filteredImg,
dst=filteredImg,
beta=0,
alpha=255,
norm_type=cv2.NORM_MINMAX)
disparity_to_be_written = filteredImg if self.config.get_bool(
"disparity_filter", True) else displ
disparity = np.float64(np.copy(disparity_to_be_written)) / 16.0
# Crop and resize, due to baseline, a part of the image on the left can't be matched with the one on the right
disparity = resize(disparity[:, numDisparities:],
(self.width, self.height))
# Triangulation
depth = (1.0 / disparity) * self.baseline * self.focal_length
# Clip from depth map to 25 meters
depth[depth > self.depth_max] = self.depth_max
depth[depth < 0] = 0.0
if self.config.get_bool("depth_completion", True):
depth = fill_in_fast(depth, self.depth_max)
return depth, disparity_to_be_written
def sgm(self, imgL, imgR):
window_size = self.config.get_int("window_size", 7)
if window_size % 2 == 0:
raise Exception("Window size must be an odd number")
numDisparities = self.config.get_int("num_disparities", 32)
if not (numDisparities > 0 and numDisparities % 16 == 0):
raise Exception(
"Number of disparities must be > 0 and divisible by 16")
left_matcher = cv2.StereoSGBM_create(
minDisparity=self.config.get_int("min_disparity", 0),
numDisparities=numDisparities,
blockSize=5,
P1=8 * 3 * window_size**2,
P2=32 * 3 * window_size**2,
disp12MaxDiff=-1,
uniquenessRatio=15,
speckleWindowSize=0,
speckleRange=2,
preFilterCap=63,
# mode=cv2.STEREO_SGBM_MODE_SGBM_3WAY
mode=cv2.StereoSGBM_MODE_HH)
if self.config.get_bool("disparity_filter", True):
right_matcher = cv2.ximgproc.createRightMatcher(left_matcher)
lmbda = 80000
sigma = 1.2
wls_filter = cv2.ximgproc.createDisparityWLSFilter(
matcher_left=left_matcher)
wls_filter.setLambda(lmbda)
wls_filter.setSigmaColor(sigma)
dispr = right_matcher.compute(imgR, imgL)
dispr = np.int16(dispr)
displ = left_matcher.compute(imgL, imgR)
displ = np.int16(displ)
filteredImg = None
if self.config.get_bool("disparity_filter", True):
filteredImg = wls_filter.filter(displ, imgL, None,
dispr).astype(np.float32)
filteredImg = cv2.normalize(src=filteredImg,
dst=filteredImg,
beta=0,
alpha=255,
norm_type=cv2.NORM_MINMAX)
disparity = np.float64(filteredImg) / 16.0 if self.config.get_bool("disparity_filter", True) else \
np.float64(displ) / 16.0
# Crop and resize, due to baseline, a part of the image on the left can't be matched with the one on the right
disparity = resize(disparity[:, numDisparities:],
(self.width, self.height))
# Triangulation
depth = (1.0 / disparity) * self.baseline * self.focal_length
# Clip from depth map to 25 meters
depth[depth > self.depth_max] = self.depth_max
depth[depth < 0] = 0.0
if self.config.get_bool("depth_completion", True):
depth = fill_in_fast(depth, self.depth_max)
disparity = np.int16(disparity)
return depth, disparity