def subimage(self, image, center, theta, width, height):
print "theta is:", theta
theta = theta
image = libvision.cv2_to_cv(image)
output_image = cv.CreateImage((int(width), int(height)), image.depth, image.nChannels)
mapping = np.array([[np.cos(theta), -np.sin(theta), center[0]], [np.sin(theta), np.cos(theta), center[1]]])
map_matrix_cv = cv.fromarray(mapping)
print mapping
cv.GetQuadrangleSubPix(image, output_image, map_matrix_cv)
return output_image
def subimage(self, image, center, theta, width, height):
print "theta is:", theta
theta = theta
image = libvision.cv2_to_cv(image)
output_image = cv.CreateImage((int(width), int(height)), image.depth, image.nChannels)
mapping = np.array([[np.cos(theta), -np.sin(theta), center[0]],
[np.sin(theta), np.cos(theta), center[1]]])
map_matrix_cv = cv.fromarray(mapping)
print mapping
cv.GetQuadrangleSubPix(image, output_image, map_matrix_cv)
return output_image
def process_frame(self, frame):
# This is equivalent to the old routine, but it isn't actually necessary
#height, width, depth = libvision.cv_to_cv2(frame).shape
#self.debug_frame = np.zeros((height, width, 3), np.uint8)
# Debug numpy is CV2
self.debug_frame = libvision.cv_to_cv2(frame)
# CV2 Transforms
self.numpy_frame = self.debug_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 5)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
(self.frame1, self.frame2, self.frame3) = cv2.split(self.numpy_frame)
# Change the frame number to determine what channel to focus on
self.numpy_frame = self.frame2
# Thresholding
self.numpy_frame = cv2.adaptiveThreshold(self.numpy_frame,
255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
#kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
#kernel = np.ones((2,2), np.uint8)
self.numpy_frame = cv2.erode(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel)
self.adaptive_frame = self.numpy_frame.copy()
# Find contours
contours, hierarchy = cv2.findContours(self.numpy_frame,
cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
self.raw_bins = []
self.failed_bins = []
if len(contours) > 1:
cnt = contours[0]
cv2.drawContours(self.numpy_frame, contours, -1, (255, 255, 255), 3)
for h, cnt in enumerate(contours):
#hull = cv2.convexHull(cnt)
rect = cv2.minAreaRect(cnt)
box = cv2.cv.BoxPoints(rect)
box = np.int0(box)
# test aspect ratio & area, create bin if matches
(x, y), (w, h), theta = rect
if w > 0 and h > 0:
area = h * w
if self.min_area < area < self.max_area:
approx = cv2.approxPolyDP(
cnt, 0.01 * cv2.arcLength(cnt, True), True)
aspect_ratio = float(h) / w
# Depending on the orientation of the bin, "width" may be flipped with height, thus needs 2 conditions for each case
if ((1/self.ratio_range[1]) < aspect_ratio < (1/self.ratio_range[0]) or self.ratio_range[0] < aspect_ratio < self.ratio_range[1]):
new_bin = Bin(tuple(box[0]), tuple(
box[1]), tuple(box[2]), tuple(box[3]))
new_bin.id = self.recent_id
new_bin.area = area
self.recent_id += 1
self.raw_bins.append(new_bin)
for bin in self.raw_bins:
self.match_bins(bin)
self.sort_bins()
self.draw_bins()
#populate self.output with infos
self.output.found = False
if len(self.confirmed) > 0:
self.output.found = True
self.return_output()
print self
self.return_output()
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(self.adaptive_frame)
# svr.debug("processed", self.numpy_to_cv)
# svr.debug("adaptive", self.adaptive_to_cv)
# svr.debug("debug", self.debug_to_cv)
self.debug_stream("debug", self.debug_frame)
self.debug_stream("processed", self.numpy_frame)
self.debug_stream("adaptive", self.adaptive_frame)
def process_frame(self, frame):
# Debug numpy is CV2
self.debug_frame = libvision.cv_to_cv2(frame)
# CV2 Transforms
self.numpy_frame = self.debug_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 5)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
(self.frame1, self.frame2, self.frame3) = cv2.split(self.numpy_frame)
# Change the frame number to determine what channel to focus on
self.numpy_frame = self.frame2
# Thresholding
self.buoy_frame = cv2.adaptiveThreshold(self.numpy_frame, 255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (4, 4))
self.buoy_frame = cv2.erode(self.numpy_frame, kernel)
self.buoy_frame = cv2.dilate(self.numpy_frame, kernel2)
self.buoy_adaptive = self.buoy_frame.copy()
# Find contours
contours, hierarchy = cv2.findContours(self.numpy_frame, cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
self.raw_led = []
self.raw_buoys = []
if len(contours) > 1:
cnt = contours[0]
cv2.drawContours(self.numpy_frame, contours, -1, (255, 255, 255),
3)
for h, cnt in enumerate(contours):
approx = cv2.approxPolyDP(cnt, 0.01 * cv2.arcLength(cnt, True),
True)
center, radius = cv2.minEnclosingCircle(cnt)
x, y = center
if len(approx) > 12:
if (radius > 30):
new_buoy = Buoy(int(x), int(y), int(radius), "unknown")
new_buoy.id = self.recent_id
self.recent_id += 1
self.raw_buoys.append(new_buoy)
cv2.drawContours(self.numpy_frame, [cnt], 0,
(0, 0, 255), -1)
self.raw_buoys.append(new_buoy)
for buoy1 in self.raw_buoys[:]:
for buoy2 in self.raw_buoys[:]:
if buoy1 is buoy2:
continue
if buoy1 in self.raw_buoys and buoy2 in self.raw_buoys and \
math.fabs(buoy1.centerx - buoy2.centerx) > self.mid_sep and \
math.fabs(buoy1.centery - buoy2.centery) > self.mid_sep:
if buoy1.area < buoy2.area:
self.raw_buoys.remove(buoy1)
elif buoy2.area < buoy1.area:
self.raw_buoys.remove(buoy2)
for buoy in self.raw_buoys:
self.match_buoys(buoy)
self.sort_buoys()
self.draw_buoys()
self.return_output()
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(self.adaptive_frame)
svr.debug("processed", self.numpy_to_cv)
svr.debug("adaptive", self.adaptive_to_cv)
svr.debug("debug", self.debug_to_cv)
def process_frame(self, frame):
# This is equivalent to the old routine, but it isn't actually necessary
#height, width, depth = libvision.cv_to_cv2(frame).shape
#self.debug_frame = np.zeros((height, width, 3), np.uint8)
# Debug numpy is CV2
self.debug_frame = libvision.cv_to_cv2(frame)
# CV2 Transforms: denoise and convert to hsv
self.numpy_frame = self.debug_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 5)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
# Separate the channels convenience later
(self.frame1, self.frame2, self.frame3) = cv2.split(self.numpy_frame)
# Change the frame number to determine what channel to focus on
self.numpy_frame = self.frame3
# Thresholding
self.numpy_frame = cv2.adaptiveThreshold(
self.numpy_frame, 255, cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV, self.adaptive_thresh_blocksize,
self.adaptive_thresh)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
#kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
#kernel = np.ones((2,2), np.uint8)
self.numpy_frame = cv2.erode(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel)
# capture frames representing the effect of the adaptive threshold
self.adaptive_frame = self.numpy_frame.copy()
# Find contours of every shape present after threshold
contours, hierarchy = cv2.findContours(self.numpy_frame,
cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
self.raw_bins = []
# if there are enough contours for at least one bin
if len(contours) > 1:
cv2.drawContours(self.numpy_frame, contours, -1, (255, 255, 255),
3)
for h, cnt in enumerate(contours):
#hull = cv2.convexHull(cnt)
rect = cv2.minAreaRect(cnt)
box = cv2.cv.BoxPoints(rect)
box = np.int0(box)
# test aspect ratio & area, create bin if matches
(x, y), (w, h), theta = rect
if w > 0 and h > 0:
area = h * w
if self.min_area < area < self.max_area:
#approximate raw contour points to a simpler polygon with less points
approx = cv2.approxPolyDP(
cnt, 0.01 * cv2.arcLength(cnt, True), True)
aspect_ratio = float(h) / w
# Depending on the orientation of the bin, "width" may be flipped with height, thus needs 2 conditions for each case
if 2 <= len(approx) < 12 and (
.4 < aspect_ratio < .6
or 1.8 < aspect_ratio < 2.2):
new_bin = Bin(tuple(box[0]), tuple(box[1]),
tuple(box[2]), tuple(box[3]))
new_bin.id = self.recent_id
new_bin.area = area
# print "new bin created with slope: ", new_bin.line_slope
#print -theta
# if theta != 0:
# new_bin.theta = np.pi*(-theta)/180
# else:
# new_bin.theta = 0
self.recent_id += 1
self.raw_bins.append(new_bin)
for bin in self.raw_bins:
self.match_bins(bin)
self.sort_bins()
self.draw_bins()
self.return_output()
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(self.adaptive_frame)
# svr.debug("processed", self.numpy_to_cv)
# svr.debug("adaptive", self.adaptive_to_cv)
# svr.debug("debug", self.debug_to_cv)
self.debug_stream("debug", self.debug_frame)
self.debug_stream("processed", self.numpy_frame)
self.debug_stream("adaptive", self.adaptive_frame)
for bin in self.confirmed:
print type(bin.patch)
if (bin.patch.shape[1] != 0) and (bin.patch.shape[0] != 0):
self.debug_stream("Patch" + str(bin.id), bin.patch)
# svr.debug("Patch"+str(bin.id),libvision.cv2_to_cv(bin.patch))
print bin.id
def process_frame(self, frame):
self.numpy_frame = libvision.cv_to_cv2(frame)
self.debug_frame = self.numpy_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 7)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
(rf1, rf2, rf3) = cv2.split(self.numpy_frame)
# RF2-inverted for red
# RF1 for green
rBinary = rf2
# rBinary = cv2.bitwise_not(rBinary)
gBinary = rf1
# Adaptive Threshold
rBinary = cv2.adaptiveThreshold(rBinary, 255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
gBinary = cv2.adaptiveThreshold(gBinary, 255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.Gadaptive_thresh_blocksize,
self.Gadaptive_thresh)
rFrame = rBinary.copy()
gFrame = gBinary.copy()
# Morphology
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
rBinary = cv2.erode(rBinary, kernel)
rBinary = cv2.dilate(rBinary, kernel)
gBinary = cv2.erode(gBinary, kernel)
gBinary = cv2.dilate(gBinary, kernel)
gray = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2GRAY)
edges = cv2.Canny(gray, 150, 200, apertureSize=3)
lines = cv2.HoughLines(edges, 1, np.pi / 180, 275)
for rho, theta in lines[0]:
a = np.cos(theta)
b = np.sin(theta)
x0 = a * rho
y0 = b * rho
x1 = int(x0 + 1000 * (-b)) # Here i have used int() instead of rounding the decimal value, so 3.8 --> 3
y1 = int(y0 + 1000 * (a)) # But if you want to round the number, then use np.around() function, then 3.8 --> 4.0
x2 = int(x0 - 1000 * (-b)) # But we need integers, so use int() function after that, ie int(np.around(x))
y2 = int(y0 - 1000 * (a))
cv2.line(self.debug_frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
rFrame = libvision.cv2_to_cv(rFrame)
gFrame = libvision.cv2_to_cv(gFrame)
self.debug_frame = libvision.cv2_to_cv(self.debug_frame)
# svr.debug("Rframe", rFrame)
# svr.debug("Gframe", gFrame)
svr.debug("debug", self.debug_frame)
def process_frame(self, frame):
self.numpy_frame = libvision.cv_to_cv2(frame)
self.debug_frame = self.numpy_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 7)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
(rf1, rf2, rf3) = cv2.split(self.numpy_frame)
# RF2-inverted for red
# RF1 for green
rBinary = rf2
# rBinary = cv2.bitwise_not(rBinary)
gBinary = rf1
# Adaptive Threshold
rBinary = cv2.adaptiveThreshold(rBinary, 255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
gBinary = cv2.adaptiveThreshold(gBinary, 255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.Gadaptive_thresh_blocksize,
self.Gadaptive_thresh)
rFrame = rBinary.copy()
gFrame = gBinary.copy()
# Morphology
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
rBinary = cv2.erode(rBinary, kernel)
rBinary = cv2.dilate(rBinary, kernel)
gBinary = cv2.erode(gBinary, kernel)
gBinary = cv2.dilate(gBinary, kernel)
gray = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2GRAY)
edges = cv2.Canny(gray, 150, 200, apertureSize=3)
lines = cv2.HoughLines(edges, 1, np.pi / 180, 275)
for rho, theta in lines[0]:
a = np.cos(theta)
b = np.sin(theta)
x0 = a * rho
y0 = b * rho
x1 = int(
x0 + 1000 * (-b)
) # Here i have used int() instead of rounding the decimal value, so 3.8 --> 3
y1 = int(
y0 + 1000 * (a)
) # But if you want to round the number, then use np.around() function, then 3.8 --> 4.0
x2 = int(
x0 - 1000 * (-b)
) # But we need integers, so use int() function after that, ie int(np.around(x))
y2 = int(y0 - 1000 * (a))
cv2.line(self.debug_frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
rFrame = libvision.cv2_to_cv(rFrame)
gFrame = libvision.cv2_to_cv(gFrame)
self.debug_frame = libvision.cv2_to_cv(self.debug_frame)
# svr.debug("Rframe", rFrame)
# svr.debug("Gframe", gFrame)
svr.debug("debug", self.debug_frame)
def process_frame(self, frame):
# frame directors
#self.debug_frame -- Frame containing helpful debug information
# Debug numpy in CV2
raw_frame = libvision.cv_to_cv2(frame)
self.debug_frame = raw_frame
# CV2 blur
blur_frame = cv2.medianBlur(self.debug_frame, 5)
hsv_blur_frame =
# collect brightly colored areas
frame1 = self.adaptive_threshold(blur_frame, 0,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
# collect shadowes under colored areas
frame2 = self.adaptive_threshold(blur_frame, 1,
self.shadow_thresh_blocksize,
self.shadow_thresh)
# use composite as the adaptive threshold
adaptive_frame = cv2.add(frame1, frame2*0)
frame = adaptive_frame
#self.debug_stream("help", <frame>)
# morphology
sequence = ([-self.erode_factor, self.erode_factor]*1
+[self.bloom_factor, -self.bloom_factor]*1)
despeckled_frame = self.morphology(frame, sequence)
frame = despeckled_frame
self.debug_stream("despeckled", despeckled_frame)
# collect edges
#a = 800
# TODO: ROI_edge detection
edge_frame = self.ROI_edge_detection(raw_frame, frame, True)
#edge_frame = cv2.Canny(frame, 150, 250, apertureSize=3)
# collect buoy candidates using hough circles
self.raw_circles = []
self.raw_buoys = []
self.raw_circles = cv2.HoughCircles(
edge_frame,
cv2.cv.CV_HOUGH_GRADIENT,
self.inv_res_ratio,
self.center_sep,
np.array([]),
self.upper_canny_thresh,
self.acc_thresh,
self.min_radius,
self.max_radius,
)
# create a new buoy object for every circle that is detected
if self.raw_circles is not None and len(self.raw_circles[0] > 0):
#print self.confirmed
for circle in self.raw_circles[0]:
(x, y, radius) = circle
new_buoy = Buoy(x, y, radius, "unknown", self.next_id)
self.next_id += 1
self.raw_buoys.append(new_buoy)
self.match_buoys(new_buoy)
# sort buoys among confirmed/canditates
self.sort_buoys()
# self.debug_frame= cv2.add(<HUD_FRAME>,cv2.cvtColor(<annotated_frame>, cv2.COLOR_GRAY2BGR) )
# perform color detection
if self.confirmed is not None and len(self.confirmed) > 0:
# vvv start color detection
for buoy in self.confirmed:
# draw a cirle around the confirmed bouy
cv2.circle(self.debug_frame, (int(buoy.centerx), int(buoy.centery)),
int(buoy.radius) + 10, (255, 255, 255), 5)
# attain hue from a pixel on the buoy
color_pick_point = ( int(buoy.centerx), int(buoy.centery - buoy.radius/2) )
_c = color_pick_point
# ^^offset a couple pixels upward for some reason
colorHue = np.mean(self.hsv_frame[_c[1]-buoy.radius/2 : _c[1]+buoy.radius/2,
_c[0]-buoy.radius/2 : _c[0]+buoy.radius/2,
0])
if BUOY_COLOR_PRINTS:
print("buoy%d has a hue of %d" %(buoy.id,int(colorHue)))
# note: color wraps around at 180. Range is 0->180
if (colorHue >= 0 and colorHue < 45) or colorHue >= 95: # 105->180->45
cv2.putText(self.debug_frame,str(buoy.id)+"RED", (int(buoy.centerx), int(buoy.centery)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255))
buoy.color = "red"
elif (colorHue >= 80 and colorHue < 95): # green is hardest to detect
cv2.putText(self.debug_frame,str(buoy.id)+"GRE", (int(buoy.centerx), int(buoy.centery)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255))
#if buoy.color != "red" and buoy.color != "yellow":
#print "switched from ", buoy.color
buoy.color = "green"
else: #yellow is about 50->80
cv2.putText(self.debug_frame,str(buoy.id)+"YEL", (int(buoy.centerx), int(buoy.centery)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255))
buoy.color = "yellow"
cv2.putText(self.debug_frame,"HUE="+str(int(colorHue)), (int(buoy.centerx), int(buoy.centery-20)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255))
# ^^^ end color detection
# debug frames
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
#self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(adaptive_frame)
#svr.debug("processed", self.numpy_to_cv)
svr.debug("adaptive", self.adaptive_to_cv)
svr.debug("debug", self.debug_to_cv)
# generate vision output
self.output.buoys = []
if self.confirmed is not None and len(self.confirmed) > 0:
for buoy in self.confirmed:
buoy.theta = buoy.centerx #<- a rough approximation
buoy.phi = buoy.centery #<- a rough approximation
buoy.id = buoy.id
self.output.buoys.append(buoy)
# publish output
#print ("%d buoys currently confirmed." % len(self.confirmed))
if self.output.buoys:
self.return_output()
return self.output
def process_frame(self, frame):
numpy_frame = libvision.cv_to_cv2(frame)
svr.debug("Original", frame)
numpy_frame = cv2.medianBlur(numpy_frame, 7)
debug_frame = numpy_frame.copy()
numpy_frame = cv2.cvtColor(numpy_frame, cv2.COLOR_BGR2HSV)
(h, s, v) = cv2.split(numpy_frame)
binary = h
binary = cv2.adaptiveThreshold(binary, 255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
# Morphology
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
binary = cv2.dilate(binary, kernel)
binary = cv2.dilate(binary, kernel)
# Hough Transform
raw_lines = cv2.HoughLinesP(binary,
rho=1,
theta=math.pi / 180,
threshold=self.hough_threshold,
minLineLength=50,
maxLineGap=10)
if raw_lines is None:
raw_lines = []
else:
raw_lines = raw_lines[0]
def slope(line):
""" Determine the slope [in degrees] of a line """
(x1, y1, x2, y2) = line
p1 = (x1, y1)
p2 = (x2, y2)
leftx = min(x1, x2)
if p1[0] == leftx:
left = p1
right = p2
else:
right = p1
left = p2
slope = (right[1]-left[1]) / (right[0]-left[0])
return slope
def angle(line):
sl = slope(line)
return math.degrees(math.atan2(sl, 1))
def length(line):
(x1, y1, x2, y2) = line
return ((x2-x1)**2 + (y2-y1)**2) ** .5
def center(line):
""" Determine the center of a line """
(x1, y1, x2, y2) = line
p1 = (x1, y1)
p2 = (x2, y2)
leftx = min(x1, x2)
if p1[0] == leftx:
left = p1
right = p2
else:
right = p1
left = p2
centerx = int(left[0] + length(line)/2*math.cos(math.atan2(slope(line), 1)))
centery = int(left[1] + length(line)/2*math.sin(math.atan2(slope(line), 1)))
return (centerx, centery)
def is_vertical(line):
return 60 <= abs(angle(line)) <= 90
def is_horizontal(line):
return 0 <= abs(angle(line)) <= 30
def get_avg_endpoints(lines):
lefts = []
rights = []
for line in lines:
(x1, y1, x2, y2) = line
p1 = (x1, y1)
p2 = (x2, y2)
leftx = min(x1, x2)
if p1[0] == leftx:
left = p1
right = p2
else:
right = p1
left = p2
lefts.append(left)
rights.append(right)
return (average_pts(lefts), average_pts(rights))
def get_med_endpoints(lines):
lefts = []
rights = []
for line in lines:
(x1, y1, x2, y2) = line
p1 = (x1, y1)
p2 = (x2, y2)
leftx = min(x1, x2)
if p1[0] == leftx:
left = p1
right = p2
else:
right = p1
left = p2
lefts.append(left)
rights.append(right)
return (bad_median(lefts, .25), bad_median(rights, .75))
def average_pts(pts):
num = len(pts)
if num == 0:
return None
avg_x = sum(x for (x, y) in pts) / num
avg_y = sum(y for (x, y) in pts) / num
return (int(avg_x), int(avg_y))
def median_pts(pts):
num = len(pts)
if num == 0:
return None
pts = sorted(pts, key=lambda x: x[0])
return pts[num//2]
def bad_median(pts, val=.5):
num = len(pts)
if num == 0:
return None
pts = sorted(pts, key=lambda x: x[0])
return pts[int(num*val)]
def get_normal_vec(line):
sl = slope(line)
return line
h_lines = []
v_lines = []
for line in raw_lines:
if is_horizontal(line):
h_lines.append(line)
elif is_vertical(line):
v_lines.append(line)
else:
(x1, y1, x2, y2) = line
cv2.line(debug_frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
if h_lines:
self.seen_crossbar = False
self.crossbar_depth = None
for line in h_lines:
(x1, y1, x2, y2) = line
cv2.line(debug_frame, (x1, y1), (x2, y2), (0, 0, 255), 2)
for line in v_lines:
(x1, y1, x2, y2) = line
cv2.line(debug_frame, (x1, y1), (x2, y2), (255, 0, 0), 2)
h_centers = [center(line) for line in h_lines]
v_centers = sorted([center(line) for line in v_lines], key=lambda x: x[0])
h_avg_center = median_pts(h_centers)
v_avg_center = average_pts(v_centers)
if h_avg_center:
cv2.circle(debug_frame, h_avg_center, 5, (0, 0, 0), -1)
if v_avg_center:
cv2.circle(debug_frame, v_avg_center, 5, (0, 0, 0), -1)
split_pt = None
for i in range(len(v_centers)):
if i < len(v_centers)-1 and v_centers[i+1][0] - v_centers[i][0] > 40:
split_pt = i+1
break
left_pole_center = None
right_pole_center = None
if split_pt:
left_centers = v_centers[:split_pt]
right_centers = v_centers[split_pt:]
avg_left = average_pts(left_centers)
avg_right = average_pts(right_centers)
left_pole_center = avg_left
right_pole_center = avg_right
elif v_avg_center and h_avg_center and h_avg_center[0] - v_avg_center[0] > 60:
left_pole_center = v_avg_center
cv2.circle(debug_frame, v_avg_center, 5, (0, 0, 0), -1)
elif v_avg_center and h_avg_center and h_avg_center[0] - v_avg_center[0] < -60:
right_pole_center = v_avg_center
cv2.circle(debug_frame, v_avg_center, 5, (0, 0, 0), -1)
else:
avg_endpoints = get_med_endpoints(h_lines)
lefts = avg_endpoints[0]
rights = avg_endpoints[1]
if lefts:
cv2.circle(debug_frame, lefts, 5, (0, 0, 0), -1)
left_pole_center = (lefts[0], lefts[1] - 80)
if rights:
cv2.circle(debug_frame, rights, 5, (0, 0, 0), -1)
right_pole_center = (rights[0], rights[1] - 80)
if left_pole_center:
self.left_pole = left_pole_center[0]
cv2.circle(debug_frame, left_pole_center, 5, (0, 0, 0), -1)
if right_pole_center:
self.right_pole = right_pole_center[0]
cv2.circle(debug_frame, right_pole_center, 5, (0, 0, 0), -1)
# median_slope_h = np.median(list(slope(line) for line in h_lines))
# average_slope_v = None if len(v_lines) == 0 else sum(slope(line) for line in v_lines) / len(v_lines)
# center_horiz =
# points = []
# for x1, y1, x2, y2 in raw_lines:
# points.append((x1, y1))
# points.append((x2, y2))
# if points:
# rect = cv2.minAreaRect(np.array(points))
# box = cv2.cv.BoxPoints(rect)
# box = np.int0(box)
# # test aspect ratio & area, create bin if matches
# (x, y), (w, h), theta = rect
# cv2.drawContours(debug_frame, [box], 0, (0, 0, 255), 2)
binary = libvision.cv2_to_cv(binary)
svr.debug("Binary", binary)
debug_frame = libvision.cv2_to_cv(debug_frame)
svr.debug("Debug", debug_frame)
def process_frame(self, frame):
# frame types:
#self.debug_frame -- Frame containing helpful debug information
# Debug numpy in CV2
raw_frame = libvision.cv_to_cv2(frame)
self.debug_frame = raw_frame
# CV2 blur
blur_frame = cv2.medianBlur(self.debug_frame, 5)
# collect brightly colored areas
frame1 = self.adaptive_threshold(blur_frame, 4,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
# collect shadowes under colored areas
frame2 = self.adaptive_threshold(blur_frame, 1,
self.shadow_thresh_blocksize,
self.shadow_thresh)
# use composite as the adaptive threshold
adaptive_frame = cv2.add(frame1, frame2*0)
frame = adaptive_frame
# morphology
sequence = ([-self.erode_factor, self.erode_factor]*1
+[self.bloom_factor, -self.bloom_factor]*1)
despeckled_frame = self.morphology(frame, sequence)
frame = despeckled_frame
self.debug_stream("despeckled", despeckled_frame)
# collect edges
# ROI_edge detection
edge_frame = self.ROI_edge_detection(raw_frame, frame, self.edge_threshold, 0, True)
# collect buoy candidates using hough circles
self.raw_circles = []
self.raw_buoys = []
self.raw_circles = cv2.HoughCircles(
image =edge_frame,
method =cv2.cv.CV_HOUGH_GRADIENT,
dp =self.inv_res_ratio,
minDist =self.center_sep,
param1 =self.upper_canny_thresh,
param2 =self.acc_thresh,
minRadius=self.min_radius,
maxRadius=self.max_radius,
)
if self.raw_circles is not None:
self.raw_circles = np.round(self.raw_circles[:,0]).astype(int)
# create a new buoy object for every circle that is detected
#print(self.raw_circles)
if self.raw_circles is not None:
#print self.confirmed
for circle in self.raw_circles:
(x, y, radius) = circle
new_buoy = Buoy(x, y, radius, "unknown", self.next_id)
self.next_id += 1
self.raw_buoys.append(new_buoy)
self.match_buoys(new_buoy)
cv2.circle(self.debug_frame, (x, y),
int(radius), (0, 255, 0), 5)
# sort buoys among confirmed/canditates
self.sort_buoys()
# self.debug_frame= cv2.add(<HUD_FRAME>,cv2.cvtColor(<annotated_frame>, cv2.COLOR_GRAY2BGR) )
# perform color detection
self.hsv_frame = cv2.cvtColor(raw_frame, cv2.COLOR_BGR2HSV)[:,:,:]
if self.confirmed is not None and len(self.confirmed) > 0:
# vvv start color detection
for buoy in self.confirmed:
self.debug_frame = self.detect_buoy(buoy,self.debug_frame,self.hsv_frame)
"""
# draw a cirle around the confirmed bouy
cv2.circle(self.debug_frame, (int(buoy.centerx), int(buoy.centery)),
int(buoy.radius) + 10, (255, 255, 255), 5)
# attain hue from a pixel on the buoy
color_pick_point = ( int(buoy.centerx), int(buoy.centery - buoy.radius/2) )
_c = color_pick_point
# ^^offset a couple pixels upward for some reason
(total_height, total_width, _) = self.hsv_frame.shape
colorHue = np.mean(self.hsv_frame[in_range(_c[1]-buoy.radius/2,0,total_width)
: in_range(_c[1]+buoy.radius/2, 0, total_width),
in_range(_c[0]-buoy.radius/2, 0, total_height)
: in_range(_c[0]+buoy.radius/2, 0, total_height),
0])
print(_c[0],_c[1], buoy.radius/2)
print(buoy.centery-20, buoy.centerx)
if BUOY_COLOR_PRINTS:
print("buoy%d has a hue of %d" %(buoy.id,int(colorHue)))
# note: color wraps around at 180. Range is 0->180
if (colorHue >= 0 and colorHue < 45) or colorHue >= 95: # 105->180->45
cv2.putText(self.debug_frame,str(buoy.id)+"RED", (int(buoy.centerx), int(buoy.centery)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255))
buoy.color = "red"
elif (colorHue >= 80 and colorHue < 95): # green is hardest to detect
cv2.putText(self.debug_frame,str(buoy.id)+"GRE", (int(buoy.centerx), int(buoy.centery)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255))
#if buoy.color != "red" and buoy.color != "yellow":
#print "switched from ", buoy.color
buoy.color = "green"
else: #yellow is about 50->80
cv2.putText(self.debug_frame,str(buoy.id)+"YEL", (int(buoy.centerx), int(buoy.centery)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255))
buoy.color = "yellow"
#print(buoy.centerx)
cv2.putText(self.debug_frame,"HUE="+str(int(colorHue)), (int(buoy.centerx), int(buoy.centery-20)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255))
cv2.putText(self.debug_frame,"last_seen="+str(int(buoy.lastseen)), (int(buoy.centerx), int(buoy.centery-40)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255))
cv2.putText(self.debug_frame,"candidate="+str(int(buoy in self.candidates)), (int(buoy.centerx), int(buoy.centery-60)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255))
# ^^^ end color detection
"""
# debug frames
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
#self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(adaptive_frame)
#svr.debug("processed", self.numpy_to_cv)
svr.debug("adaptive", self.adaptive_to_cv)
svr.debug("debug", self.debug_to_cv)
# generate vision output
FOV_x = 71.0
FOV_y = 40.0
x_resolution = frame.shape[1]
y_resolution = frame.shape[0]
self.output.buoys = []
if self.confirmed is not None and len(self.confirmed) > 0:
for buoy in self.confirmed:
buoy.theta = (buoy.centerx - x_resolution/2.0) / (x_resolution/2.0) * (FOV_x/2.0) #<- a rough approximation
buoy.phi = -(buoy.centery - y_resolution/2.0) / (y_resolution/2.0) * (FOV_y/2.0) #<- a rough approximation
buoy.id = buoy.id
self.output.buoys.append(buoy)
# publish output
#print ("%d buoys currently confirmed." % len(self.confirmed))
if self.output.buoys:
self.return_output()
return self.output
def process_frame(self, frame):
# This is equivalent to the old routine, but it isn't actually necessary
#height, width, depth = libvision.cv_to_cv2(frame).shape
#self.debug_frame = np.zeros((height, width, 3), np.uint8)
# Debug numpy is CV2
self.debug_frame = libvision.cv_to_cv2(frame)
# CV2 Transforms
self.numpy_frame = self.debug_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 5)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
(self.frame1, self.frame2, self.frame3) = cv2.split(self.numpy_frame)
# Change the frame number to determine what channel to focus on
self.numpy_frame = self.frame2
# Thresholding
self.numpy_frame = cv2.adaptiveThreshold(
self.numpy_frame, 255, cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV, self.adaptive_thresh_blocksize,
self.adaptive_thresh)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (4, 4))
# kernel = np.ones((2,2), np.uint8)
self.numpy_frame = cv2.erode(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel2)
self.adaptive_frame = self.numpy_frame.copy()
# Find contours
contours, hierarchy = cv2.findContours(self.numpy_frame,
cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
self.raw_buoys = []
if len(contours) > 0:
cnt = contours[0]
cv2.drawContours(self.numpy_frame, contours, -1, (255, 255, 255),
3)
for h, cnt in enumerate(contours):
approx = cv2.approxPolyDP(cnt, 0.01 * cv2.arcLength(cnt, True),
True)
center, radius = cv2.minEnclosingCircle(cnt)
x, y = center
if len(approx) > 12:
if (radius > 30):
new_buoy = Buoy(int(x), int(y), int(radius), "unknown")
new_buoy.id = self.recent_id
self.recent_id += 1
self.raw_buoys.append(new_buoy)
cv2.drawContours(self.numpy_frame, [cnt], 0,
(0, 0, 255), -1)
self.raw_buoys.append(new_buoy)
for buoy1 in self.raw_buoys[:]:
for buoy2 in self.raw_buoys[:]:
if buoy1 is buoy2:
continue
if buoy1 in self.raw_buoys and buoy2 in self.raw_buoys and \
math.fabs(buoy1.centerx - buoy2.centerx) > self.mid_sep and \
math.fabs(buoy1.centery - buoy2.centery) > self.mid_sep:
if buoy1.radius < buoy2.radius:
self.raw_buoys.remove(buoy1)
elif buoy2.radius < buoy1.radius:
self.raw_buoys.remove(buoy2)
for buoy in self.raw_buoys:
self.match_buoys(buoy)
self.sort_buoys()
self.draw_buoys()
self.return_output()
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(self.adaptive_frame)
svr.debug("processed", self.numpy_to_cv)
svr.debug("adaptive", self.adaptive_to_cv)
svr.debug("debug", self.debug_to_cv)
# Convert to output format
self.output.buoys = []
if self.raw_buoys is not None and len(self.raw_buoys) > 0:
for buoy in self.raw_buoys:
x = buoy.centerx
y = buoy.centery
buoy = Container()
buoy.theta = x
buoy.phi = y
buoy.id = 1
self.output.buoys.append(buoy)
if self.output.buoys:
self.return_output()
return self.output
def process_frame(self, frame):
# This is equivalent to the old routine, but it isn't actually necessary
#height, width, depth = libvision.cv_to_cv2(frame).shape
#self.debug_frame = np.zeros((height, width, 3), np.uint8)
# Debug numpy is CV2
self.debug_frame = libvision.cv_to_cv2(frame)
# CV2 Transforms: denoise and convert to hsv
self.numpy_frame = self.debug_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 5)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
# Separate the channels convenience later
(self.frame1, self.frame2, self.frame3) = cv2.split(self.numpy_frame)
# Change the frame number to determine what channel to focus on
self.numpy_frame = self.frame3
# Thresholding
self.numpy_frame = cv2.adaptiveThreshold(self.numpy_frame,
255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
#kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
#kernel = np.ones((2,2), np.uint8)
self.numpy_frame = cv2.erode(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel)
# capture frames representing the effect of the adaptive threshold
self.adaptive_frame = self.numpy_frame.copy()
# Find contours of every shape present after threshold
contours, hierarchy = cv2.findContours(self.numpy_frame,
cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
self.raw_bins = []
# if there are enough contours for at least one bin
if len(contours) > 1:
cv2.drawContours(self.numpy_frame, contours, -1, (255, 255, 255), 3)
for h, cnt in enumerate(contours):
#hull = cv2.convexHull(cnt)
rect = cv2.minAreaRect(cnt)
box = cv2.cv.BoxPoints(rect)
box = np.int0(box)
# test aspect ratio & area, create bin if matches
(x, y), (w, h), theta = rect
if w > 0 and h > 0:
area = h * w
if self.min_area < area < self.max_area:
#approximate raw contour points to a simpler polygon with less points
approx = cv2.approxPolyDP(
cnt, 0.01 * cv2.arcLength(cnt, True), True)
aspect_ratio = float(h) / w
# Depending on the orientation of the bin, "width" may be flipped with height, thus needs 2 conditions for each case
if 2 <= len(approx) < 12 and (.4 < aspect_ratio < .6 or 1.8 < aspect_ratio < 2.2):
new_bin = Bin(tuple(box[0]), tuple(
box[1]), tuple(box[2]), tuple(box[3]))
new_bin.id = self.recent_id
new_bin.area = area
# print "new bin created with slope: ", new_bin.line_slope
#print -theta
# if theta != 0:
# new_bin.theta = np.pi*(-theta)/180
# else:
# new_bin.theta = 0
self.recent_id += 1
self.raw_bins.append(new_bin)
for bin in self.raw_bins:
self.match_bins(bin)
self.sort_bins()
self.draw_bins()
self.return_output()
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(self.adaptive_frame)
# svr.debug("processed", self.numpy_to_cv)
# svr.debug("adaptive", self.adaptive_to_cv)
# svr.debug("debug", self.debug_to_cv)
self.debug_stream("debug", self.debug_frame)
self.debug_stream("processed", self.numpy_frame)
self.debug_stream("adaptive", self.adaptive_frame)
for bin in self.confirmed:
print type(bin.patch)
if (bin.patch.shape[1] != 0) and (bin.patch.shape[0] != 0):
self.debug_stream("Patch" + str(bin.id), bin.patch)
# svr.debug("Patch"+str(bin.id),libvision.cv2_to_cv(bin.patch))
print bin.id
def process_frame(self, frame):
# This is equivalent to the old routine, but it isn't actually necessary
#height, width, depth = libvision.cv_to_cv2(frame).shape
#self.debug_frame = np.zeros((height, width, 3), np.uint8)
# Debug numpy is CV2
self.debug_frame = libvision.cv_to_cv2(frame)
# CV2 Transforms
self.numpy_frame = self.debug_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 5)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
(self.frame1, self.frame2, self.frame3) = cv2.split(self.numpy_frame)
# Change the frame number to determine what channel to focus on
self.numpy_frame = self.frame2
# Thresholding
self.numpy_frame = cv2.adaptiveThreshold(
self.numpy_frame, 255, cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV, self.adaptive_thresh_blocksize,
self.adaptive_thresh)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
#kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
#kernel = np.ones((2,2), np.uint8)
self.numpy_frame = cv2.erode(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel)
self.adaptive_frame = self.numpy_frame.copy()
# Find contours
contours, hierarchy = cv2.findContours(self.numpy_frame,
cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
self.raw_bins = []
self.failed_bins = []
if len(contours) > 1:
cnt = contours[0]
cv2.drawContours(self.numpy_frame, contours, -1, (255, 255, 255),
3)
for h, cnt in enumerate(contours):
#hull = cv2.convexHull(cnt)
rect = cv2.minAreaRect(cnt)
box = cv2.cv.BoxPoints(rect)
box = np.int0(box)
# test aspect ratio & area, create bin if matches
(x, y), (w, h), theta = rect
if w > 0 and h > 0:
area = h * w
if self.min_area < area < self.max_area:
approx = cv2.approxPolyDP(
cnt, 0.01 * cv2.arcLength(cnt, True), True)
aspect_ratio = float(h) / w
# Depending on the orientation of the bin, "width" may be flipped with height, thus needs 2 conditions for each case
if ((1 / self.ratio_range[1]) < aspect_ratio <
(1 / self.ratio_range[0]) or self.ratio_range[0] <
aspect_ratio < self.ratio_range[1]):
new_bin = Bin(tuple(box[0]), tuple(box[1]),
tuple(box[2]), tuple(box[3]))
new_bin.id = self.recent_id
new_bin.area = area
self.recent_id += 1
self.raw_bins.append(new_bin)
for bin in self.raw_bins:
self.match_bins(bin)
self.sort_bins()
self.draw_bins()
#populate self.output with infos
self.output.found = False
if len(self.confirmed) > 0:
self.output.found = True
self.return_output()
print self
self.return_output()
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(self.adaptive_frame)
# svr.debug("processed", self.numpy_to_cv)
# svr.debug("adaptive", self.adaptive_to_cv)
# svr.debug("debug", self.debug_to_cv)
self.debug_stream("debug", self.debug_frame)
self.debug_stream("processed", self.numpy_frame)
self.debug_stream("adaptive", self.adaptive_frame)
def process_frame(self, frame):
# This is equivalent to the old routine, but it isn't actually necessary
#height, width, depth = libvision.cv_to_cv2(frame).shape
#self.debug_frame = np.zeros((height, width, 3), np.uint8)
# Debug numpy is CV2
self.debug_frame = libvision.cv_to_cv2(frame)
# CV2 Transforms
self.numpy_frame = self.debug_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 5)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
(self.frame1, self.frame2, self.frame3) = cv2.split(self.numpy_frame)
# Change the frame number to determine what channel to focus on
self.numpy_frame = self.frame3
# Thresholding
self.numpy_frame = cv2.adaptiveThreshold(self.numpy_frame,
255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh
)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
#kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
#kernel = np.ones((2,2), np.uint8)
self.numpy_frame = cv2.erode(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel)
self.adaptive_frame = self.numpy_frame.copy()
# Find contours
contours, hierarchy = cv2.findContours(self.numpy_frame,
cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
self.raw_bins = []
if len(contours) > 1:
cnt = contours[0]
cv2.drawContours(self.numpy_frame, contours, -1, (255, 255, 255), 3)
for h, cnt in enumerate(contours):
#hull = cv2.convexHull(cnt)
rect = cv2.minAreaRect(cnt)
box = cv2.cv.BoxPoints(rect)
box = np.int0(box)
# test aspect ratio & area, create bin if matches
(x, y), (w, h), theta = rect
if w > 0 and h > 0:
area = h * w
if self.min_area < area < self.max_area:
aspect_ratio = float(h) / w
# Depending on the orientation of the bin, "width" may be flipped with height, thus needs 2 conditions for each case
if .4 < aspect_ratio < .6 or 1.8 < aspect_ratio < 2.2:
new_bin = Bin(tuple(box[0]), tuple(
box[1]), tuple(box[2]), tuple(box[3]))
new_bin.id = self.recent_id
new_bin.area = area
new_bin.theta = -theta
self.recent_id += 1
self.raw_bins.append(new_bin)
# Removes bins that have centers too close to others (to prevent bins inside bins)
for bin1 in self.raw_bins[:]:
for bin2 in self.raw_bins[:]:
if bin1 is bin2:
continue
if bin1 in self.raw_bins and bin2 in self.raw_bins and \
math.fabs(bin1.midx - bin2.midx) < self.mid_sep and \
math.fabs(bin1.midy - bin2.midy) < self.mid_sep:
if bin1.area < bin2.area:
self.raw_bins.remove(bin1)
elif bin2.area < bin1.area:
self.raw_bins.remove(bin2)
for bin in self.raw_bins:
self.match_bins(bin)
self.sort_bins()
self.draw_bins()
self.return_output()
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(self.adaptive_frame)
svr.debug("processed", self.numpy_to_cv)
svr.debug("adaptive", self.adaptive_to_cv)
svr.debug("debug", self.debug_to_cv)
def process_frame(self, frame):
self.numpy_frame = libvision.cv_to_cv2(frame)
self.debug_frame = self.numpy_frame.copy()
self.test_frame = self.numpy_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 7)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
(rf1, rf2, rf3) = cv2.split(self.numpy_frame)
Rbinary = rf3
Gbinary = rf1
# Adaptive Threshold
Rbinary = cv2.adaptiveThreshold(Rbinary, 255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
Gbinary = cv2.adaptiveThreshold(Gbinary, 255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.Gadaptive_thresh_blocksize,
self.Gadaptive_thresh)
# Morphology
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
Rbinary = cv2.erode(Rbinary, kernel)
Rbinary = cv2.dilate(Rbinary, kernel)
Gbinary = cv2.erode(Gbinary, kernel)
Gbinary = cv2.dilate(Gbinary, kernel)
Rframe = cv2.cvtColor(Rbinary, cv2.COLOR_GRAY2RGB)
Gframe = cv2.cvtColor(Gbinary, cv2.COLOR_GRAY2RGB)
# Hough Transform
raw_linesG = cv2.HoughLines(Gbinary,
rho=1,
theta=math.pi / 180,
threshold=self.hough_thresholdG)
# Get vertical lines
vertical_linesG = []
for line in raw_linesG[0]:
rho = line[0]
theta = line[1]
if theta < self.vertical_thresholdG or \
theta > math.pi - self.vertical_thresholdG:
vertical_linesG.append((abs(rho), theta))
# Group vertical lines
vertical_line_groupsG = [] # A list of line groups which are each a line list
for line in vertical_linesG:
group_found = False
for line_group in vertical_line_groupsG:
if line_group_accept_test(line_group, line, self.max_range):
line_group.append(line)
group_found = True
if not group_found:
vertical_line_groupsG.append([line])
# Average line groups into lines
vertical_linesG = []
for line_group in vertical_line_groupsG:
rhos = map(lambda line: line[0], line_group)
angles = map(lambda line: line[1], line_group)
line = (sum(rhos) / len(rhos), circular_average(angles, math.pi))
vertical_linesG.append(line)
# Get horizontal lines
horizontal_lines = []
for line in raw_linesG[0]:
rho = line[0]
theta = line[1]
dist_from_horizontal = (math.pi / 2 + line[1]) % math.pi
if dist_from_horizontal < self.horizontal_threshold or \
dist_from_horizontal > math.pi - self.horizontal_threshold:
horizontal_lines.append((abs(line[0]), line[1]))
# Group horizontal lines
horizontal_line_groups = [] # A list of line groups which are each a line list
print "Horizontal lines: ",
for line in horizontal_lines:
group_found = False
for line_group in horizontal_line_groups:
if line_group_accept_test(line_group, line, self.max_range):
line_group.append(line)
group_found = True
if not group_found:
horizontal_line_groups.append([line])
if len(horizontal_line_groups) is 1:
self.seen_crossbar = True
rhos = map(lambda line: line[0], horizontal_line_groups[0])
angles = map(lambda line: line[1], horizontal_line_groups[0])
line = (sum(rhos) / len(rhos), circular_average(angles, math.pi))
horizontal_lines = [line]
else:
self.seen_crossbar = False
horizontal_lines = []
self.left_pole = None
self.right_pole = None
Rframe = libvision.cv2_to_cv(Rframe)
Gframe = libvision.cv2_to_cv(self.debug_frame)
Rbinary = libvision.cv2_to_cv(Rbinary)
self.debug_frame = libvision.cv2_to_cv(self.debug_frame)
self.test_frame = libvision.cv2_to_cv(self.test_frame)
Gbinary = libvision.cv2_to_cv(Gbinary)
if len(vertical_linesG) is 2:
roi = cv.GetImageROI(frame)
width = roi[2]
height = roi[3]
self.left_pole = round(min(vertical_linesG[0][0], vertical_linesG[1][0]), 2) - width / 2
self.right_pole = round(max(vertical_linesG[0][0], vertical_linesG[1][0]), 2) - width / 2
# TODO: If one pole is seen, is it left or right pole?
# Calculate planar distance r (assuming we are moving perpendicular to
# the hedge)
if self.left_pole and self.right_pole:
theta = abs(self.left_pole - self.right_pole)
self.r = 3 / math.tan(math.radians(theta / 2))
else:
self.r = None
if self.r and self.seen_crossbar:
bar_phi = (-1 * horizontal_lines[0][0] + Gframe.height / 2) / (Gframe.height / 2) * 32
self.crossbar_depth = self.r * math.atan(math.radians(bar_phi))
else:
self.crossbar_depth = None
# Line Finding on Red pvc
# Hough Transform
line_storage = cv.CreateMemStorage()
raw_linesR = cv.HoughLines2(Rbinary, line_storage, cv.CV_HOUGH_STANDARD,
rho=1,
theta=math.pi / 180,
threshold=self.hough_thresholdR,
param1=0,
param2=0
)
# Get vertical lines
vertical_linesR = []
for line in raw_linesR:
if line[1] < self.vertical_thresholdR or \
line[1] > math.pi - self.vertical_thresholdR:
vertical_linesR.append((abs(line[0]), line[1]))
# Group vertical lines
vertical_line_groupsR = [] # A list of line groups which are each a line list
for line in vertical_linesR:
group_found = False
for line_group in vertical_line_groupsR:
if line_group_accept_test(line_group, line, self.max_range):
line_group.append(line)
group_found = True
if not group_found:
vertical_line_groupsR.append([line])
# Average line groups into lines
vertical_linesR = []
for line_group in vertical_line_groupsR:
rhos = map(lambda line: line[0], line_group)
angles = map(lambda line: line[1], line_group)
line = (sum(rhos) / len(rhos), circular_average(angles, math.pi))
vertical_linesR.append(line)
'''
for red_line in vertical_linesR:
print "Red Line:", red_line[0],", ",red_line[1]
for green_line in vertical_linesG:
print "Green Line:", green_line[0],", ",green_line[1]
'''
for red_line in vertical_linesR:
for green_line in vertical_linesG[:]:
if math.fabs(green_line[0] - red_line[0]) < self.GR_Threshold0 and \
math.fabs(green_line[1] - red_line[1]) < self.GR_Threshold1:
vertical_linesG.remove(green_line)
for red_line in vertical_linesR:
print "New Red Line:", red_line[0], ", ", red_line[1]
for green_line in vertical_linesG:
print "New Green Line:", green_line[0], ", ", green_line[1]
if len(vertical_linesR) is 0:
print "No Red Found"
self.left_pole = None
self.right_pole = None
if len(vertical_linesR) is 2:
roi = cv.GetImageROI(frame)
width = roi[2]
height = roi[3]
self.left_pole = round(min(vertical_linesR[0][0], vertical_linesR[1][0]), 2) - width / 2
self.right_pole = round(max(vertical_linesR[0][0], vertical_linesR[1][0]), 2) - width / 2
# TODO: If one pole is seen, is it left or right pole?
# Calculate planar distance r (assuming we are moving perpendicular to
# the hedge)
if self.left_pole and self.right_pole:
theta = abs(self.left_pole - self.right_pole)
self.r = 3 / math.tan(math.radians(theta / 2))
else:
self.r = None
for i in range(len(vertical_linesR[:])):
if vertical_linesR[i][1] > math.pi / 2:
vertical_linesR[i] = (vertical_linesR[i][0], -(math.pi - vertical_linesR[i][1]))
print "Line changed to ", vertical_linesR[i]
for line in vertical_linesR:
print line
if line[1] > math.pi / 2:
line = (line[0], math.pi - line[1])
print "Line changed to ", line
libvision.misc.draw_lines(Gframe, vertical_linesG)
libvision.misc.draw_lines(Gframe, horizontal_lines)
libvision.misc.draw_lines(Rframe, vertical_linesR)
# there was a merge error, these 3 lines conflicted b/c your copy out of date
for line in vertical_linesR:
roi = cv.GetImageROI(frame)
width = roi[2]
height = roi[3]
x = line[0] * math.cos(line[1])
y = line[0] * math.sin(line[1])
cv.Circle(Rframe, (int(x), int(y)), 5, (0, 255, 0), -1, 8, 0)
if x > width or y > width or x < 0 or y < 0:
print "Lost point ", x
svr.debug("Original", self.test_frame)
svr.debug("Red", Rframe)
svr.debug("Green", Gframe)
def process_frame(self, frame):
# This is equivalent to the old routine, but it isn't actually necessary
#height, width, depth = libvision.cv_to_cv2(frame).shape
#self.debug_frame = np.zeros((height, width, 3), np.uint8)
inv_res_ratio = 2
center_sep = 100
upper_canny_thresh = 40 # 40
acc_thresh = 10 # 20, 50 with green settings
min_radius = 3
max_radius = 50
# Debug numpy is CV2
debug_frame = libvision.cv_to_cv2(frame)
svr.debug("original", frame)
# CV2 Transforms
numpy_frame = debug_frame.copy()
numpy_frame = cv2.medianBlur(numpy_frame, 5)
numpy_frame = cv2.cvtColor(numpy_frame, cv2.COLOR_BGR2HSV)
# Kernel for erosion/dilation
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
# Split into HSV frames
(h_frame, s_frame, v_frame) = cv2.split(numpy_frame)
# Run inverse adaptive thresh on saturation channel
s_adapt_thresh = cv2.adaptiveThreshold(s_frame,
255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
47,
10)
# Erode and dilate the value frame
s_eroded = cv2.erode(s_adapt_thresh, kernel)
s_dilated = cv2.dilate(s_eroded, kernel)
# Threshold the value frame
_, v_thresh = cv2.threshold(v_frame, 250, 255, cv2.THRESH_BINARY)
# Erode and dilate the value frame
v_eroded = cv2.erode(v_thresh, kernel)
v_dilated = cv2.dilate(v_eroded, kernel)
s_contours = s_dilated.copy()
# Find contours on the dilated saturation channel
s_cnt, hy = cv2.findContours(
s_dilated,
cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE
)
v_contours = v_dilated.copy()
# Find contours on the dilated
v_cnt, hy = cv2.findContours(
v_dilated,
cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE
)
if len(s_contours) > 0:
cv2.drawContours(s_contours, s_cnt, -1, (255, 255, 255), 3)
if len(v_contours) > 0:
cv2.drawContours(v_contours, v_cnt, -1, (255, 255, 255), 3)
s_circles = cv2.HoughCircles(
s_contours,
cv2.cv.CV_HOUGH_GRADIENT,
inv_res_ratio,
center_sep,
np.array([]),
upper_canny_thresh,
acc_thresh,
min_radius,
max_radius,
)
v_circles = cv2.HoughCircles(
v_contours,
cv2.cv.CV_HOUGH_GRADIENT,
inv_res_ratio,
center_sep,
np.array([]),
upper_canny_thresh,
acc_thresh,
min_radius,
max_radius,
)
for circle in s_circles[0]:
(x, y, radius) = circle
cv2.circle(debug_frame, (int(x), int(y)), int(radius) + 10, (0, 255, 0), 5)
# for circle in v_circles[0]:
# (x, y, radius) = circle
# cv2.circle(debug_frame, (int(x), int(y)), int(radius) + 10, (0, 0, 255), 5)
# debug_to_cv = libvision.cv2_to_cv(v_circles)
# svr.debug("v_frame", debug_to_cv)
# debug_to_cv = libvision.cv2_to_cv(s_circles)
# svr.debug("s_frame", debug_to_cv)
debug_to_cv = libvision.cv2_to_cv(debug_frame)
svr.debug("debug_frame", debug_to_cv)
def process_frame(self, frame):
self.numpy_frame = libvision.cv_to_cv2(frame)
self.debug_frame = self.numpy_frame.copy()
self.test_frame = self.numpy_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 7)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
(rf1, rf2, rf3) = cv2.split(self.numpy_frame)
Rbinary = rf3
Gbinary = rf1
# Adaptive Threshold
Rbinary = cv2.adaptiveThreshold(Rbinary, 255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
Gbinary = cv2.adaptiveThreshold(Gbinary, 255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.Gadaptive_thresh_blocksize,
self.Gadaptive_thresh)
# Morphology
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
Rbinary = cv2.erode(Rbinary, kernel)
Rbinary = cv2.dilate(Rbinary, kernel)
Gbinary = cv2.erode(Gbinary, kernel)
Gbinary = cv2.dilate(Gbinary, kernel)
Rframe = cv2.cvtColor(Rbinary, cv2.COLOR_GRAY2RGB)
Gframe = cv2.cvtColor(Gbinary, cv2.COLOR_GRAY2RGB)
# Hough Transform
raw_linesG = cv2.HoughLines(Gbinary,
rho=1,
theta=math.pi / 180,
threshold=self.hough_thresholdG)
# Get vertical lines
vertical_linesG = []
if raw_linesG is None:
raw_linesG = []
if len(raw_linesG) > 0:
for line in raw_linesG[0]:
rho = line[0]
theta = line[1]
if theta < self.vertical_thresholdG or theta > (
math.pi - self.vertical_thresholdG):
vertical_linesG.append((rho, theta))
# Group vertical lines
vertical_line_groupsG = [
] # A list of line groups which are each a line list
for line in vertical_linesG:
#print "Green Line Grouping Possibility:", line[0], ", ", line[1]
group_found = False
for line_group in vertical_line_groupsG:
if line_group_accept_test(line_group, line, self.max_range):
line_group.append(line)
group_found = True
if not group_found:
vertical_line_groupsG.append([line])
# Average line groups into lines
vertical_linesG = []
for line_group in vertical_line_groupsG:
rhos = map(lambda line: line[0], line_group)
angles = map(lambda line: line[1], line_group)
line = (sum(rhos) / len(rhos), circular_average(angles, math.pi))
vertical_linesG.append(line)
# Get horizontal lines
horizontal_lines = []
if len(raw_linesG) > 0:
for line in raw_linesG[0]:
rho = line[0]
theta = line[1]
dist_from_horizontal = (math.pi / 2 + line[1]) % math.pi
if dist_from_horizontal < self.horizontal_threshold or dist_from_horizontal > math.pi - self.horizontal_threshold:
horizontal_lines.append((abs(line[0]), line[1]))
# Group horizontal lines
horizontal_line_groups = [
] # A list of line groups which are each a line list
for line in horizontal_lines:
group_found = False
for line_group in horizontal_line_groups:
if line_group_accept_test(line_group, line, self.max_range):
line_group.append(line)
group_found = True
if not group_found:
horizontal_line_groups.append([line])
if len(horizontal_line_groups) is 1:
self.seen_crossbar = True
rhos = map(lambda line: line[0], horizontal_line_groups[0])
angles = map(lambda line: line[1], horizontal_line_groups[0])
line = (sum(rhos) / len(rhos), circular_average(angles, math.pi))
horizontal_lines = [line]
else:
self.seen_crossbar = False
horizontal_lines = []
self.left_pole = None
self.right_pole = None
Rframe = libvision.cv2_to_cv(Rframe)
Gframe = libvision.cv2_to_cv(self.debug_frame)
Rbinary = libvision.cv2_to_cv(Rbinary)
self.debug_frame = libvision.cv2_to_cv(self.debug_frame)
self.test_frame = libvision.cv2_to_cv(self.test_frame)
Gbinary = libvision.cv2_to_cv(Gbinary)
if len(vertical_linesG) is 2:
roi = cv.GetImageROI(frame)
width = roi[2]
height = roi[3]
self.left_pole = round(
min(vertical_linesG[0][0], vertical_linesG[1][0]),
2) - width / 2
self.right_pole = round(
max(vertical_linesG[0][0], vertical_linesG[1][0]),
2) - width / 2
# TODO: If one pole is seen, is it left or right pole?
# Calculate planar distance r (assuming we are moving perpendicular to
# the hedge)
if self.left_pole and self.right_pole:
theta = abs(self.left_pole - self.right_pole)
self.r = 3 / math.tan(math.radians(theta / 2))
else:
self.r = None
if self.r and self.seen_crossbar:
bar_phi = (-1 * horizontal_lines[0][0] +
Gframe.height / 2) / (Gframe.height / 2) * 32
self.crossbar_depth = self.r * math.atan(math.radians(bar_phi))
else:
self.crossbar_depth = None
# Line Finding on Red pvc
# Hough Transform
line_storage = cv.CreateMemStorage()
raw_linesR = cv.HoughLines2(Rbinary,
line_storage,
cv.CV_HOUGH_STANDARD,
rho=1,
theta=math.pi / 180,
threshold=self.hough_thresholdR,
param1=0,
param2=0)
# Get vertical lines
vertical_linesR = []
for line in raw_linesR:
if line[1] < self.vertical_thresholdR or \
line[1] > math.pi - self.vertical_thresholdR:
vertical_linesR.append((abs(line[0]), line[1]))
# Group vertical lines
vertical_line_groupsR = [
] # A list of line groups which are each a line list
for line in vertical_linesR:
group_found = False
for line_group in vertical_line_groupsR:
if line_group_accept_test(line_group, line, self.max_range):
line_group.append(line)
group_found = True
if not group_found:
vertical_line_groupsR.append([line])
# Average line groups into lines
vertical_linesR = []
for line_group in vertical_line_groupsR:
rhos = map(lambda line: line[0], line_group)
angles = map(lambda line: line[1], line_group)
line = (sum(rhos) / len(rhos), circular_average(angles, math.pi))
vertical_linesR.append(line)
'''
for red_line in vertical_linesR:
print "Red Line:", red_line[0],", ",red_line[1]
for green_line in vertical_linesG:
print "Green Line:", green_line[0],", ",green_line[1]
'''
for red_line in vertical_linesR:
for green_line in vertical_linesG[:]:
if math.fabs(green_line[0] - red_line[0]) < self.GR_Threshold0 and \
math.fabs(green_line[1] - red_line[1]) < self.GR_Threshold1:
vertical_linesG.remove(green_line)
for red_line in vertical_linesR:
print "New Red Line:", red_line[0], ", ", red_line[1]
for green_line in vertical_linesG:
print "New Green VLine:", green_line[0], ", ", green_line[1]
for green_line in horizontal_lines:
print "New Green HLine:", green_line[0], ", ", green_line[1]
if len(vertical_linesR) is 0:
print "No Red Found"
self.left_pole = None
self.right_pole = None
if len(vertical_linesR) is 2:
roi = cv.GetImageROI(frame)
width = roi[2]
height = roi[3]
self.left_pole = round(
min(vertical_linesR[0][0], vertical_linesR[1][0]),
2) - width / 2
self.right_pole = round(
max(vertical_linesR[0][0], vertical_linesR[1][0]),
2) - width / 2
# TODO: If one pole is seen, is it left or right pole?
# Calculate planar distance r (assuming we are moving perpendicular to
# the hedge)
if self.left_pole and self.right_pole:
theta = abs(self.left_pole - self.right_pole)
self.r = 3 / math.tan(math.radians(theta / 2))
else:
self.r = None
for i in range(len(vertical_linesR[:])):
if vertical_linesR[i][1] > math.pi / 2:
vertical_linesR[i] = (vertical_linesR[i][0],
-(math.pi - vertical_linesR[i][1]))
print "Line changed to ", vertical_linesR[i]
for line in vertical_linesR:
print line
if line[1] > math.pi / 2:
line = (line[0], math.pi - line[1])
print "Line changed to ", line
libvision.misc.draw_lines(Gframe, vertical_linesG)
libvision.misc.draw_lines(Gframe, horizontal_lines)
libvision.misc.draw_lines(Rframe, vertical_linesR)
# there was a merge error, these 3 lines conflicted b/c your copy out of date
for line in vertical_linesR:
roi = cv.GetImageROI(frame)
width = roi[2]
height = roi[3]
x = line[0] * math.cos(line[1])
y = line[0] * math.sin(line[1])
cv.Circle(Rframe, (int(x), int(y)), 5, (0, 255, 0), -1, 8, 0)
if x > width or y > width or x < 0 or y < 0:
print "Lost point ", x
svr.debug("Original", self.test_frame)
svr.debug("Red", Rframe)
svr.debug("Green", Gframe)
svr.debug("Green Binary", Gbinary)
def process_frame(self, frame):
# Debug numpy is CV2
self.debug_frame = libvision.cv_to_cv2(frame)
# CV2 Transforms
self.numpy_frame = self.debug_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 5)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
(self.frame1, self.frame2, self.frame3) = cv2.split(self.numpy_frame)
# Change the frame number to determine what channel to focus on
self.numpy_frame = self.frame2
# Thresholding
self.buoy_frame = cv2.adaptiveThreshold(self.numpy_frame,
255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (4, 4))
self.buoy_frame = cv2.erode(self.numpy_frame, kernel)
self.buoy_frame = cv2.dilate(self.numpy_frame, kernel2)
self.buoy_adaptive = self.buoy_frame.copy()
# Find contours
contours, hierarchy = cv2.findContours(self.numpy_frame,
cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
self.raw_led = []
self.raw_buoys = []
if len(contours) > 1:
cnt = contours[0]
cv2.drawContours(
self.numpy_frame, contours, -1, (255, 255, 255), 3)
for h, cnt in enumerate(contours):
approx = cv2.approxPolyDP(
cnt, 0.01 * cv2.arcLength(cnt, True), True)
center, radius = cv2.minEnclosingCircle(cnt)
x, y = center
if len(approx) > 12:
if (radius > 30):
new_buoy = Buoy(int(x), int(y), int(radius), "unknown")
new_buoy.id = self.recent_id
self.recent_id += 1
self.raw_buoys.append(new_buoy)
cv2.drawContours(
self.numpy_frame, [cnt], 0, (0, 0, 255), -1)
self.raw_buoys.append(new_buoy)
for buoy1 in self.raw_buoys[:]:
for buoy2 in self.raw_buoys[:]:
if buoy1 is buoy2:
continue
if buoy1 in self.raw_buoys and buoy2 in self.raw_buoys and \
math.fabs(buoy1.centerx - buoy2.centerx) > self.mid_sep and \
math.fabs(buoy1.centery - buoy2.centery) > self.mid_sep:
if buoy1.area < buoy2.area:
self.raw_buoys.remove(buoy1)
elif buoy2.area < buoy1.area:
self.raw_buoys.remove(buoy2)
for buoy in self.raw_buoys:
self.match_buoys(buoy)
self.sort_buoys()
self.draw_buoys()
self.return_output()
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(self.adaptive_frame)
svr.debug("processed", self.numpy_to_cv)
svr.debug("adaptive", self.adaptive_to_cv)
svr.debug("debug", self.debug_to_cv)
def process_frame(self, frame):
# This is equivalent to the old routine, but it isn't actually necessary
#height, width, depth = libvision.cv_to_cv2(frame).shape
#self.debug_frame = np.zeros((height, width, 3), np.uint8)
# Debug numpy is CV2
self.debug_frame = libvision.cv_to_cv2(frame)
# CV2 Transforms
self.numpy_frame = self.debug_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 5)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
(self.frame1, self.frame2, self.frame3) = cv2.split(self.numpy_frame)
# Change the frame number to determine what channel to focus on
self.numpy_frame = self.frame2
# Thresholding
self.numpy_frame = cv2.adaptiveThreshold(self.numpy_frame,
255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh
)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (4, 4))
# #kernel = np.ones((2,2), np.uint8)
self.numpy_frame = cv2.erode(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel2)
self.adaptive_frame = self.numpy_frame.copy()
# # Find contours
contours, hierarchy = cv2.findContours(self.numpy_frame,
cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
self.raw_led = []
self.raw_buoys = []
if len(contours) > 1:
cnt = contours[0]
cv2.drawContours(self.numpy_frame, contours, -1, (255, 255, 255), 3)
for h, cnt in enumerate(contours):
approx = cv2.approxPolyDP(cnt, 0.01 * cv2.arcLength(cnt, True), True)
center, radius = cv2.minEnclosingCircle(cnt)
x, y = center
if len(approx) > 10:
if (radius > 17):
new_buoy = Buoy(int(x), int(y), int(radius))
new_buoy.id = self.recent_id
self.recent_id += 1
self.raw_buoys.append(new_buoy)
cv2.drawContours(self.numpy_frame, [cnt], 0, (0, 0, 255), -1)
for buoy in self.raw_buoys:
self.match_buoys(buoy)
self.sort_buoys()
self.draw_buoys()
self.return_output()
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(self.adaptive_frame)
svr.debug("processed", self.numpy_to_cv)
svr.debug("adaptive", self.adaptive_to_cv)
svr.debug("debug", self.debug_to_cv)
def process_frame(self, frame):
# This is equivalent to the old routine, but it isn't actually necessary
#height, width, depth = libvision.cv_to_cv2(frame).shape
#self.debug_frame = np.zeros((height, width, 3), np.uint8)
# Debug numpy is CV2
self.debug_frame = libvision.cv_to_cv2(frame)
# CV2 Transforms
self.numpy_frame = self.debug_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 5)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
(self.frame1, self.frame2, self.frame3) = cv2.split(self.numpy_frame)
# Change the frame number to determine what channel to focus on
self.numpy_frame = self.frame2
# Thresholding
self.numpy_frame = cv2.adaptiveThreshold(self.numpy_frame,
255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (4, 4))
# kernel = np.ones((2,2), np.uint8)
self.numpy_frame = cv2.erode(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel2)
self.adaptive_frame = self.numpy_frame.copy()
# Find contours
contours, hierarchy = cv2.findContours(self.numpy_frame,
cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
self.raw_buoys = []
if len(contours) > 0:
cnt = contours[0]
cv2.drawContours(
self.numpy_frame, contours, -1, (255, 255, 255), 3)
for h, cnt in enumerate(contours):
approx = cv2.approxPolyDP(
cnt, 0.01 * cv2.arcLength(cnt, True), True)
center, radius = cv2.minEnclosingCircle(cnt)
x, y = center
if len(approx) > 12:
if (radius > 30):
new_buoy = Buoy(int(x), int(y), int(radius), "unknown")
new_buoy.id = self.recent_id
self.recent_id += 1
self.raw_buoys.append(new_buoy)
cv2.drawContours(
self.numpy_frame, [cnt], 0, (0, 0, 255), -1)
self.raw_buoys.append(new_buoy)
for buoy1 in self.raw_buoys[:]:
for buoy2 in self.raw_buoys[:]:
if buoy1 is buoy2:
continue
if buoy1 in self.raw_buoys and buoy2 in self.raw_buoys and \
math.fabs(buoy1.centerx - buoy2.centerx) > self.mid_sep and \
math.fabs(buoy1.centery - buoy2.centery) > self.mid_sep:
if buoy1.radius < buoy2.radius:
self.raw_buoys.remove(buoy1)
elif buoy2.radius < buoy1.radius:
self.raw_buoys.remove(buoy2)
for buoy in self.raw_buoys:
self.match_buoys(buoy)
self.sort_buoys()
self.draw_buoys()
self.return_output()
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(self.adaptive_frame)
svr.debug("processed", self.numpy_to_cv)
svr.debug("adaptive", self.adaptive_to_cv)
svr.debug("debug", self.debug_to_cv)
# Convert to output format
self.output.buoys = []
if self.raw_buoys is not None and len(self.raw_buoys) > 0:
for buoy in self.raw_buoys:
x = buoy.centerx
y = buoy.centery
buoy = Container()
buoy.theta = x
buoy.phi = y
buoy.id = 1
self.output.buoys.append(buoy)
if self.output.buoys:
self.return_output()
return self.output
def process_frame(self, frame):
# Debug numpy is CV2
self.debug_frame = libvision.cv_to_cv2(frame)
# CV2 Transforms
self.numpy_frame = self.debug_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 5)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
self.hsv_frame = self.numpy_frame
(self.frame1, self.frame2, self.frame3) = cv2.split(self.numpy_frame)
# Change the frame number to determine what channel to focus on
self.numpy_frame = self.frame2
# Thresholding
self.numpy_frame = cv2.adaptiveThreshold(
self.numpy_frame, 255, cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV, self.adaptive_thresh_blocksize,
self.adaptive_thresh)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (4, 4))
self.numpy_frame = cv2.erode(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel2)
self.adaptive_frame = self.numpy_frame.copy()
self.numpy_frame = cv2.Canny(self.adaptive_frame,
100,
250,
apertureSize=3)
self.raw_circles = []
self.raw_buoys = []
self.raw_circles = cv2.HoughCircles(
self.numpy_frame,
cv2.cv.CV_HOUGH_GRADIENT,
self.inv_res_ratio,
self.center_sep,
np.array([]),
self.upper_canny_thresh,
self.acc_thresh,
self.min_radius,
self.max_radius,
)
if self.raw_circles is not None and len(self.raw_circles[0] > 0):
for circle in self.raw_circles[0]:
(x, y, radius) = circle
new_buoy = Buoy(x, y, radius, "unknown", self.next_id)
self.next_id += 1
self.raw_buoys.append(new_buoy)
self.match_buoys(new_buoy)
self.sort_buoys()
if self.confirmed is not None and len(self.confirmed) > 0:
for buoy in self.confirmed:
cv2.circle(self.debug_frame,
(int(buoy.centerx), int(buoy.centery)),
int(buoy.radius) + 10, (255, 255, 255), 5)
colorHue = self.hsv_frame[buoy.centery + buoy.radius / 2,
buoy.centerx][0]
if (colorHue >= 0 and colorHue < 45) or colorHue >= 300:
cv2.putText(self.debug_frame,
str(buoy.id) + "RED",
(int(buoy.centerx), int(buoy.centery)),
cv2.FONT_HERSHEY_SIMPLEX, 2, (0, 0, 255))
buoy.color = "red"
elif (colorHue >= 70 and colorHue < 180):
cv2.putText(self.debug_frame,
str(buoy.id) + "GRE",
(int(buoy.centerx), int(buoy.centery)),
cv2.FONT_HERSHEY_SIMPLEX, 2, (0, 0, 255))
if buoy.color != "red" and buoy.color != "yellow":
print "switched from ", buoy.color
buoy.color = "green"
else:
cv2.putText(self.debug_frame,
str(buoy.id) + "YEL",
(int(buoy.centerx), int(buoy.centery)),
cv2.FONT_HERSHEY_SIMPLEX, 2, (0, 0, 255))
buoy.color = "yellow"
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(self.adaptive_frame)
svr.debug("processed", self.numpy_to_cv)
svr.debug("adaptive", self.adaptive_to_cv)
svr.debug("debug", self.debug_to_cv)
# Convert to output format
self.output.buoys = []
if self.confirmed is not None and len(self.confirmed) > 0:
for buoy in self.confirmed:
buoy.theta = buoy.centerx
buoy.phi = buoy.centery
buoy.id = buoy.id
self.output.buoys.append(buoy)
if self.output.buoys:
self.return_output()
return self.output
def process_frame(self, frame):
# Debug numpy is CV2
self.debug_frame = libvision.cv_to_cv2(frame)
# CV2 Transforms
self.numpy_frame = self.debug_frame.copy()
self.numpy_frame = cv2.medianBlur(self.numpy_frame, 5)
self.numpy_frame = cv2.cvtColor(self.numpy_frame, cv2.COLOR_BGR2HSV)
self.hsv_frame = self.numpy_frame
(self.frame1, self.frame2, self.frame3) = cv2.split(self.numpy_frame)
# Change the frame number to determine what channel to focus on
self.numpy_frame = self.frame2
# Thresholding
self.numpy_frame = cv2.adaptiveThreshold(self.numpy_frame,
255,
cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV,
self.adaptive_thresh_blocksize,
self.adaptive_thresh)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (4, 4))
self.numpy_frame = cv2.erode(self.numpy_frame, kernel)
self.numpy_frame = cv2.dilate(self.numpy_frame, kernel2)
self.adaptive_frame = self.numpy_frame.copy()
self.numpy_frame = cv2.Canny(self.adaptive_frame, 100, 250, apertureSize=3)
self.raw_circles = []
self.raw_buoys = []
self.raw_circles = cv2.HoughCircles(
self.numpy_frame,
cv2.cv.CV_HOUGH_GRADIENT,
self.inv_res_ratio,
self.center_sep,
np.array([]),
self.upper_canny_thresh,
self.acc_thresh,
self.min_radius,
self.max_radius,
)
if self.raw_circles is not None and len(self.raw_circles[0] > 0):
for circle in self.raw_circles[0]:
(x, y, radius) = circle
new_buoy = Buoy(x, y, radius, "unknown", self.next_id)
self.next_id += 1
self.raw_buoys.append(new_buoy)
self.match_buoys(new_buoy)
self.sort_buoys()
if self.confirmed is not None and len(self.confirmed) > 0:
for buoy in self.confirmed:
cv2.circle(self.debug_frame, (int(buoy.centerx), int(buoy.centery)),
int(buoy.radius) + 10, (255, 255, 255), 5)
colorHue = self.hsv_frame[buoy.centery + buoy.radius/2,buoy.centerx][0]
if (colorHue >= 0 and colorHue < 45) or colorHue >= 300:
cv2.putText(self.debug_frame,str(buoy.id)+"RED", (int(buoy.centerx), int(buoy.centery)), cv2.FONT_HERSHEY_SIMPLEX, 2, (0, 0, 255))
buoy.color = "red"
elif (colorHue >= 70 and colorHue < 180):
cv2.putText(self.debug_frame,str(buoy.id)+"GRE", (int(buoy.centerx), int(buoy.centery)), cv2.FONT_HERSHEY_SIMPLEX, 2, (0, 0, 255))
if buoy.color != "red" and buoy.color != "yellow":
print "switched from ", buoy.color
buoy.color = "green"
else:
cv2.putText(self.debug_frame,str(buoy.id)+"YEL", (int(buoy.centerx), int(buoy.centery)), cv2.FONT_HERSHEY_SIMPLEX, 2, (0, 0, 255))
buoy.color = "yellow"
self.debug_to_cv = libvision.cv2_to_cv(self.debug_frame)
self.numpy_to_cv = libvision.cv2_to_cv(self.numpy_frame)
self.adaptive_to_cv = libvision.cv2_to_cv(self.adaptive_frame)
svr.debug("processed", self.numpy_to_cv)
svr.debug("adaptive", self.adaptive_to_cv)
svr.debug("debug", self.debug_to_cv)
# Convert to output format
self.output.buoys = []
if self.confirmed is not None and len(self.confirmed) > 0:
for buoy in self.confirmed:
buoy.theta = buoy.centerx
buoy.phi = buoy.centery
buoy.id = buoy.id
self.output.buoys.append(buoy)
if self.output.buoys:
self.return_output()
return self.output