class __impl:
def __init__(self, vs, imgOutput):
self.__vs = vs
self.__imgOutput = imgOutput
self.image = None
self.logger = Logger()
self.state = State()
self.tesseract = PyTessBaseAPI(psm=PSM.SINGLE_CHAR,
oem=OEM.LSTM_ONLY,
lang="digits")
self.filter = Filter()
self.signalThresholdY = 160
self.LAPPatternSesibility = 5
self.recordStamp = time.strftime(self.logger.timeFormat)
self.recordNum = 0
self.recordFolder = None
self.cntNum = 0
if (self.state.RecordImage):
root = 'record'
if not os.path.isdir(root):
os.mkdir(root)
self.recordFolder = os.path.join(root, self.recordStamp)
if not os.path.isdir(self.recordFolder):
os.mkdir(self.recordFolder)
def showImg(self, window, image):
if self.__imgOutput:
cv2.imshow(window, image)
def warmup(self):
time.sleep(2.0)
self.tesserOCR(np.zeros((1, 1, 3), np.uint8))
def tesserOCR(self, image):
self.tesseract.SetImage(Image.fromarray(image))
return self.tesseract.GetUTF8Text(
), self.tesseract.AllWordConfidences()
def dominantColor(self, img, clusters=2):
data = np.reshape(img, (-1, 3))
data = np.float32(data)
criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 10,
1.0)
flags = cv2.KMEANS_RANDOM_CENTERS
_, _, centers = cv2.kmeans(data, 1, None, criteria, 10, flags)
return centers[0].astype(np.int32)
def analyzeRect(self, image, warped, box, x, y):
# find amount of color blue in warped area, assuming over X% is the lap signal
if (self.getAmountOfColor(warped, Colors.lower_blue_color,
Colors.upper_blue_color, True) > 0.1):
self.logger.info("Rundensignal")
self.state.setCurrentSignal(Signal.LAP)
return "Rundensignal"
def analyzeSquare(self, image, warped, box, x, y):
#dominantColor, percent, _ = self.dominantColor(warped, 3)
# dominantColor = self.dominantColor(
# cv2.cvtColor(warped, cv2.COLOR_BGR2HSV), 3)
""" color = 'k'
# find amount of color black in warped area, assuming over X% is a numeric signal
if ((dominantColor <= 70).all()):
color = 'Black'
elif ((dominantColor >= 180).all()):
color = 'White'
if (color): """
resizedWarp = cv2.resize(warped,
None,
fx=2.0,
fy=2.0,
interpolation=cv2.INTER_CUBIC)
# gray
optimized = cv2.cvtColor(resizedWarp, cv2.COLOR_BGR2GRAY)
# blur
optimized = cv2.GaussianBlur(optimized, (5, 5), 0)
# binary image
optimized = cv2.threshold(optimized, 0, 255,
cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
# binary inversion if dominant color is black
""" if (color == 'Black'):
optimized = cv2.bitwise_not(optimized) """
# now check the frame (1px) of the image.. there shouldn't be any noise since its a clean signal background
h, w = optimized.shape[0:2]
clean = optimized[0, 0]
for iFrame in range(0, 2):
for iHeight in range(h):
if not (optimized[iHeight, iFrame] == clean) or not (
optimized[iHeight, w - 1 - iFrame] == clean):
return False
for iWidth in range(w):
# or not(optimized[h - iFrame, iWidth])
if not (optimized[iFrame, iWidth] == clean):
return False
# cv2.imwrite("records/opt/" + str(self.cntNum) + ".jpg", optimized)
output, confidence = self.tesserOCR(optimized)
# if the resulting text is below X% confidence threshold, we skip it
if not output or confidence[0] < 95:
return False
# clean up output from tesseract
output = output.replace('\n', '')
output = output.replace(' ', '')
if output.isdigit() and 0 < int(output) < 10:
""" self.showImg("opt " + str(self.cntNum),
np.hstack((resizedWarp, cv2.cvtColor(optimized, cv2.COLOR_GRAY2BGR)))) """
if y <= self.signalThresholdY:
self.logger.info('Stop Signal OCR: ' + output + ' X: ' +
str(x) + ' Y: ' + str(y) +
' Confidence: ' + str(confidence[0]) +
'%') # + ' DC: ' + str(dominantColor))
self.state.setStopSignal(int(output))
return 'S: ' + output
elif self.state.StopSignalNum != 0:
self.logger.info('Info Signal OCR: ' + output + ' X: ' +
str(x) + ' Y: ' + str(y) +
' Confidence: ' + str(confidence[0]) +
'%') # + ' DC: ' + str(dominantColor))
self.state.setCurrentSignal(Signal.UPPER, int(output))
return 'I: ' + output
def getAmountOfColor(self,
img,
lowerColor,
upperColor,
convert2hsv=True):
if (convert2hsv):
img = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
# create mask from color range
maskColor = cv2.inRange(img, lowerColor, upperColor)
# get ratio of active pixels
ratio_color = cv2.countNonZero(maskColor) / (img.size)
return ratio_color
# color picker for manual debugging
def pick_color(self, event, x, y, flags, param):
if event == cv2.EVENT_LBUTTONDOWN:
pixel = self.image[y, x]
color = np.array([pixel[0], pixel[1], pixel[2]])
self.logger.info(pixel)
# capture frames from the camera
def capture(self, savedImg=None):
if (savedImg is not None):
image = savedImg
else:
image = self.__vs.read()
if (self.state.InvertCamera):
image = imutils.rotate(image, angle=180)
self.image = image
if (self.state.RecordImage):
self.recordNum += 1
cv2.imwrite(
os.path.join(self.recordFolder,
str(self.recordNum) + ".jpg"), image)
return
if (self.state.Approaching == Signal.UPPER
or self.state.Approaching == Signal.LOWER):
self.findNumberSignal(image)
elif (self.state.Approaching == Signal.LAP):
self.findLapSignal(image)
def findLapSignal(self, image):
contourImage = image.copy()
blur = cv2.GaussianBlur(image, (3, 3), 0)
hsv = cv2.cvtColor(blur, cv2.COLOR_BGR2HSV)
self.image = hsv
mask = cv2.inRange(hsv, Colors.lower_blue_color,
Colors.upper_blue_color)
cnts = imutils.grab_contours(
cv2.findContours(mask.copy(), cv2.RETR_LIST,
cv2.CHAIN_APPROX_SIMPLE))
if len(cnts) > 0:
# transform all contours to rects
rects = [cv2.boundingRect(cnt) for cnt in cnts]
# now iterate all of the rects, trying to find an approximated sibiling shifted in Y-direction
for rect in rects:
(x, y, w, h) = rect
cv2.rectangle(contourImage, (x, y), (x + w, y + h),
(0, 0, 255), 2)
# try to match the pattern from a given rect in all rects
counterPart = [
counterRect for counterRect in rects
if (counterRect != rect and x - 5 <= counterRect[0] <=
x + 5 and 2 * -(h + 5) <= y - counterRect[1] <= 2 *
(h + 5) and w - 5 <= counterRect[2] <= w + 5)
and h - 5 <= counterRect[3] <= h + 5
]
if (counterPart):
(x, y, w, h) = counterPart[0]
cv2.rectangle(contourImage, (x, y), (x + w, y + h),
(0, 255, 0), 2)
self.logger.info('LAP Signal')
self.state.captureLapSignal()
break
self.showImg(
'contourImage',
np.hstack(
(hsv, contourImage, cv2.cvtColor(mask,
cv2.COLOR_GRAY2BGR))))
cv2.setMouseCallback('contourImage', self.pick_color)
def findNumberSignal(self, image):
image_height = np.size(image, 0)
image_width = np.size(image, 1)
contourImage = image.copy()
# focus only on the part of the image, where a signal could occur
# image = image[int(image.shape[0] * 0.2):int(image.shape[0] * 0.8), 0:int(image.shape[1]*0.666)]
mask = self.filter.autoCanny(image, 2, 3)
# get a list of contours in the mask, chaining to just endpoints
cnts = imutils.grab_contours(
cv2.findContours(mask.copy(), cv2.RETR_LIST,
cv2.CHAIN_APPROX_SIMPLE))
# only proceed if at least one contour was found
if len(cnts) > 0:
# loop contours
for self.cntNum, cnt in enumerate(cnts):
rect = cv2.minAreaRect(cnt)
_, _, angle = rect
# approximate shape
approx = cv2.approxPolyDP(cnt,
0.02 * cv2.arcLength(cnt, True),
True)
# the rectangle must not have a too big rotation (+/-10)
# and more than 3 connecting points
if len(approx) >= 3 and (-90 <= angle <= -80
or angle >= -10):
box = cv2.boxPoints(rect)
box = np.int0(box)
(x, y, w, h) = cv2.boundingRect(approx)
# limit viewing range
if (y <= image_height * 0.2 or x >= image_width * 0.8):
continue
if (w <= 5 or h <= 5):
continue
# we are in approaching mode, thus we only care for the lower signals <= threshold
if ((self.state.Approaching == Signal.UPPER
and y >= self.signalThresholdY)
and not self.state.Standalone):
continue
elif ((self.state.Approaching == Signal.LOWER
and y <= self.signalThresholdY)
and not self.state.Standalone):
continue
sideRatio = w / float(h)
absoluteSizeToImageRatio = (
100 / (image_width * image_height)) * (w * h)
# calculate area of the bounding rectangle
rArea = w * float(h)
# calculate area of the contour
cArea = cv2.contourArea(cnt)
if (cArea):
rectContAreaRatio = (100 / rArea) * cArea
else:
continue
# cv2.drawContours(contourImage, [box], 0, (255, 0, 0), 1)
result = None
# is the rectangle sideways, check for lap signal
# if (h*2 < w and y <= self.signalThresholdY and rectContAreaRatio >= 80):
#result = self.analyzeRect(image, four_point_transform(image, [box][0]), box, x, y)
# find all contours looking like a signal with minimum area (1%)
if absoluteSizeToImageRatio >= 0.01:
# is it approx a square, or standing rect? then check for info or stop signal
if 0.2 <= sideRatio <= 1.1:
# transform ROI
if (sideRatio <= 0.9):
coords, size, angle = rect
size = size[0] + 8, size[1] + 4
coords = coords[0] + 1, coords[1] + 1
rect = coords, size, angle
box = cv2.boxPoints(rect)
box = np.int0(box)
""" cv2.drawContours(
contourImage, [box], 0, (0, 255, 0), 1) """
warp = four_point_transform(image, [box][0])
result = self.analyzeSquare(
image, warp, box, x, y)
if (result):
if (self.__imgOutput):
color = None
if (y >= self.signalThresholdY):
color = (0, 0, 255)
else:
color = (255, 0, 0)
cv2.drawContours(contourImage, [box], 0, color,
1)
cv2.drawContours(contourImage, [cnt], -1,
color, 2)
""" M = cv2.moments(cnt)
cX = int(M["m10"] / M["m00"])
cY = int(M["m01"] / M["m00"])
cv2.putText(contourImage, str(
self.cntNum), (cX - 30, cY - 30), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1) """
self.logger.debug(
"[" + str(self.cntNum) + "] SideRatio: " +
str(sideRatio) + " AreaRatio: " +
str(rectContAreaRatio) + " ContArea: " +
str(cArea) + " RectArea: " + str(rArea) +
" AbsSize: " +
str(absoluteSizeToImageRatio) +
" CntPoints: " + str(len(approx)) +
" Size: " + str(w) + "x" + str(h))
""" if (self.__imgOutput): # hsv img output
hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
cv2.namedWindow('contourImage')
cv2.setMouseCallback('contourImage', self.pick_color)
# self.showImg("hsv", hsv) """
self.showImg(
"contourImage",
np.hstack((contourImage, cv2.cvtColor(mask,
cv2.COLOR_GRAY2BGR))))
