def executeAll(self):
""" Run the complete extraction procedure. """
# Check if the average image is too white and skip it
if np.average(self.compressed[2]) > self.config.white_avg_level:
log.debug("[" + self.filename + "] frames are all white")
return
t = time.time()
# Extract points of the fireball
event_points = self.findPoints()
log.debug("[" + self.filename + "] time for thresholding and subsampling: " + str(time.time() - t) + "s")
# Skip the event if not points where found
if len(event_points) == 0:
log.debug("[" + self.filename + "] nothing found, not extracting anything 1")
return
t = time.time()
# Find lines in 3D space and store them to line_list
line_list = Grouping3D.find3DLines(event_points, time.time(), self.config)
log.debug("[" + self.filename + "] Time for finding lines: " + str(time.time() - t) + "s")
if line_list is None:
log.debug("[" + self.filename + "] no lines found, not extracting anything")
return
t = time.time()
# Find the parameters of the line in the 3D point cloud
coeff = Grouping3D.findCoefficients(line_list)
log.debug("[" + self.filename + "] Time for finding coefficients: " + str(time.time() - t) + "s")
if len(coeff) == 0:
log.debug("[" + self.filename + "] nothing found, not extracting anything 2")
return
t = time.time()
# Extract video clips from the raw data
clips = self.extract(coeff)
log.debug("[" + self.filename + "] Time for extracting: " + str(time.time() - t) + "s")
t = time.time()
# Save the extracted clips
self.save(clips)
log.debug("[" + self.filename + "] Time for saving: " + str(time.time() - t) + "s")
def findPoints(self):
"""Threshold and subsample frames and return as list of points.
Return:
(y, x, z): [tuple] Coordinates of points that form the fireball, where Z is the frame number.
"""
# Threshold and subsample frames
length, x, y, z = Grouping3D.thresholdAndSubsample(self.frames, self.compressed, \
self.config.min_level, self.config.min_pixels, self.config.k1, self.config.j1, self.config.f)
# Return empty list if there is no points
if length == 0:
return []
# Return list with the number of occurence of each frame number (Z axis)
freq = stats.itemfreq(z)
# Reject the image if there are too little event frames
if len(freq) <= self.config.min_frames:
return []
# Calculate a threshold based on factors and median number of points on the images per frame
outlier_threshold = self.config.max_per_frame_factor * np.median(
freq[:, 1])
if outlier_threshold > self.config.max_points_per_frame:
outlier_threshold = self.config.max_points_per_frame
# Remove all outliers (aka. frames with a strong flare)
for frameNum, count in freq:
if count >= outlier_threshold:
indices = np.where(z != frameNum)
y = y[indices]
x = x[indices]
z = z[indices]
# Randomize points if there are too many in total
if len(z) > self.config.max_points:
indices = np.random.choice(len(z),
self.config.max_points,
replace=False)
y = y[indices]
x = x[indices]
z = z[indices]
# Sort by frame number
indices = np.argsort(z)
y = y[indices]
x = x[indices]
z = z[indices]
# Do a quick test to check if the points form a viable solution
if not self.testPoints(y, x, z):
return []
# Convert to python list
event_points = np.squeeze(np.dstack((y, x, z))).tolist()
return event_points
def extract(self, coefficients):
""" Determinate window size and crop out frames.
Arguments:
coefficients: [list] linear coefficients for each detected meteor
Return:
clips: [list] Cropped frames in format [frames, size and position]
"""
clips = []
# [first point, slope of XZ, slope of YZ, first frame, last frame]
for point, slopeXZ, slopeYZ, firstFrame, lastFrame in coefficients:
slopeXZ = float(slopeXZ)
slopeYZ = float(slopeYZ)
firstFrame = int(firstFrame)
lastFrame = int(lastFrame)
diff = lastFrame - firstFrame
# Extrapolate before first detected point
before_frames = math.ceil(diff * self.config.before)
# Make sure at least 4 frames before are taken
if before_frames < 4:
before_frames = 4
firstFrame = firstFrame - before_frames
if firstFrame < 0:
firstFrame = 0
# Extrapolate after last detected point
after_frames = math.ceil(diff * self.config.after)
# Make sure at least 4 frames after are taken
if after_frames < 4:
after_frames = 4
lastFrame = lastFrame + after_frames
if lastFrame >= self.frames.shape[0]:
lastFrame = self.frames.shape[0] - 1
# Cut of the fireball from raw video frames
length, cropouts, sizepos = Grouping3D.detectionCutOut(self.frames, self.compressed,
point.astype(np.uint16), slopeXZ, slopeYZ, firstFrame, lastFrame, self.config.f, \
self.config.limitForSize, self.config.minSize, self.config.maxSize)
# Shorten the output arrays to their maximum sizes
cropouts = cropouts[:length]
sizepos = sizepos[:length]
clips.append([cropouts, sizepos])
return clips
def testPoints(self, pointsy, pointsx, pointsz):
""" Quick test if points are interesting (ie. something is detected).
Arguments:
pointsy: 1D numpy array with Y coords of points
pointsx: 1D numpy array with X coords of points
pointsz: 1D numpy array with Z coords of points (aka. frame number)
Return:
[bool] True if video should be further checked for fireballs, False otherwise
"""
# Check if there are enough points
if(len(pointsy) < self.config.min_points):
return False
# Check how many points are close to each other (along the time line)
count = Grouping3D.testPoints(self.config.gap_threshold, pointsy, pointsx, pointsz)
return count >= self.config.min_points
def testPoints(self, pointsy, pointsx, pointsz):
""" Quick test if points are interesting (ie. something is detected).
Arguments:
pointsy: 1D numpy array with Y coords of points
pointsx: 1D numpy array with X coords of points
pointsz: 1D numpy array with Z coords of points (aka. frame number)
Return:
[bool] True if video should be further checked for fireballs, False otherwise
"""
# Check if there are enough points
if (len(pointsy) < self.config.min_points):
return False
# Check how many points are close to each other (along the time line)
count = Grouping3D.testPoints(self.config.gap_threshold, pointsy,
pointsx, pointsz)
return count >= self.config.min_points
def executeAll(self):
""" Run the complete extraction procedure. """
# Check if the average image is too white and skip it
if np.average(self.compressed[2]) > self.config.white_avg_level:
log.debug("[" + self.filename + "] frames are all white")
return
t = time.time()
# Extract points of the fireball
event_points = self.findPoints()
log.debug("[" + self.filename +
"] time for thresholding and subsampling: " +
str(time.time() - t) + "s")
# Skip the event if not points where found
if len(event_points) == 0:
log.debug("[" + self.filename +
"] nothing found, not extracting anything 1")
return
t = time.time()
# Find lines in 3D space and store them to line_list
line_list = Grouping3D.find3DLines(event_points, time.time(),
self.config)
log.debug("[" + self.filename + "] Time for finding lines: " +
str(time.time() - t) + "s")
if line_list is None:
log.debug("[" + self.filename +
"] no lines found, not extracting anything")
return
t = time.time()
# Find the parameters of the line in the 3D point cloud
coeff = Grouping3D.findCoefficients(line_list)
log.debug("[" + self.filename + "] Time for finding coefficients: " +
str(time.time() - t) + "s")
if len(coeff) == 0:
log.debug("[" + self.filename +
"] nothing found, not extracting anything 2")
return
t = time.time()
# Extract video clips from the raw data
clips = self.extract(coeff)
log.debug("[" + self.filename + "] Time for extracting: " +
str(time.time() - t) + "s")
t = time.time()
# Save the extracted clips
self.save(clips)
log.debug("[" + self.filename + "] Time for saving: " +
str(time.time() - t) + "s")
def findPoints(self):
"""Threshold and subsample frames and return as list of points.
Return:
(y, x, z): [tuple] Coordinates of points that form the fireball, where Z is the frame number.
"""
# Threshold and subsample frames
length, x, y, z = Grouping3D.thresholdAndSubsample(self.frames, self.compressed, \
self.config.min_level, self.config.min_pixels, self.config.k1, self.config.j1, self.config.f)
# Return empty list if there is no points
if length == 0:
return []
# Return list with the number of occurence of each frame number (Z axis)
freq = stats.itemfreq(z)
# Reject the image if there are too little event frames
if len(freq) <= self.config.min_frames:
return []
# Calculate a threshold based on factors and median number of points on the images per frame
outlier_threshold = self.config.max_per_frame_factor * np.median(freq[:, 1])
if outlier_threshold > self.config.max_points_per_frame:
outlier_threshold = self.config.max_points_per_frame
# Remove all outliers (aka. frames with a strong flare)
for frameNum, count in freq:
if count >= outlier_threshold:
indices = np.where(z != frameNum)
y = y[indices]
x = x[indices]
z = z[indices]
# Randomize points if there are too many in total
if len(z) > self.config.max_points:
indices = np.random.choice(len(z), self.config.max_points, replace=False)
y = y[indices]
x = x[indices]
z = z[indices]
# Sort by frame number
indices = np.argsort(z)
y = y[indices]
x = x[indices]
z = z[indices]
# Do a quick test to check if the points form a viable solution
if not self.testPoints(y, x, z):
return []
# Convert to python list
event_points = np.squeeze(np.dstack((y, x, z))).tolist()
return event_points
def extract(self, coefficients):
""" Determinate window size and crop out frames.
Arguments:
coefficients: [list] linear coefficients for each detected meteor
Return:
clips: [list] Cropped frames in format [frames, size and position]
"""
clips = []
# [first point, slope of XZ, slope of YZ, first frame, last frame]
for point, slopeXZ, slopeYZ, firstFrame, lastFrame in coefficients:
slopeXZ = float(slopeXZ)
slopeYZ = float(slopeYZ)
firstFrame = int(firstFrame)
lastFrame = int(lastFrame)
diff = lastFrame - firstFrame
# Extrapolate before first detected point
before_frames = math.ceil(diff*self.config.before)
# Make sure at least 4 frames before are taken
if before_frames < 4:
before_frames = 4
firstFrame = firstFrame - before_frames
if firstFrame < 0:
firstFrame = 0
# Extrapolate after last detected point
after_frames = math.ceil(diff*self.config.after)
# Make sure at least 4 frames after are taken
if after_frames < 4:
after_frames = 4
lastFrame = lastFrame + after_frames
if lastFrame >= self.frames.shape[0]:
lastFrame = self.frames.shape[0] - 1
# Cut of the fireball from raw video frames
length, cropouts, sizepos = Grouping3D.detectionCutOut(self.frames, self.compressed,
point.astype(np.uint16), slopeXZ, slopeYZ, firstFrame, lastFrame, self.config.f, \
self.config.limitForSize, self.config.minSize, self.config.maxSize)
# Shorten the output arrays to their maximum sizes
cropouts = cropouts[:length]
sizepos = sizepos[:length]
clips.append([cropouts, sizepos])
return clips