def getNewTalks(self):
talkContainers = SoupStrainer(attrs = {'class':re.compile('talkMedallion')})
for talk in BeautifulSoup(self.html, parseOnlyThese = talkContainers):
link = URLTED+talk.dt.a['href']
title = cleanHTML(talk.dt.a['title'])
pic = resizeImage(talk.find('img', attrs = {'src':re.compile('.+?\.jpg')})['src'])
yield {'url':link, 'Title':title, 'Thumb':pic}
def getTalks(self):
talkContainer = SoupStrainer(attrs = {'class':re.compile('box clearfix')})
for talk in BeautifulSoup(self.html, parseOnlyThese = talkContainer):
title = talk.h4.a.string
link = URLTED+talk.dt.a['href']
pic = resizeImage(talk.find('img', attrs = {'src':re.compile('.+?\.jpg')})['src'])
yield {'url':link, 'Title':title, 'Thumb':pic}
def getNewTalks(self):
talkContainers = SoupStrainer(attrs = {'class':re.compile('talkMedallion')})
for talk in BeautifulSoup(self.html, parseOnlyThese = talkContainers):
link = URLTED+talk.dt.a['href']
title = cleanHTML(talk.dt.a['title'])
pic = resizeImage(talk.find('img', attrs = {'src':re.compile('.+?\.jpg')})['src'])
yield {'url':link, 'Title':title, 'Thumb':pic}
def detect_traffic_sign(self, frame):
img, tf_sign = None, ''
hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
mask = cv2.inRange(hsv, self.low_range, self.high_range)
mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, self.kernel)
mask = cv2.dilate(mask, self.kernel, iterations=1)
contours = cv2.findContours(mask, cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)[-2]
for c in contours:
area = cv2.contourArea(c)
if (area > 400):
x, y, w, h = cv2.boundingRect(c)
if (abs(w - h) < (w + h) / 10):
img = frame[y:y + h, x:x + w]
imagePredict = util.resizeImage(img)
imagePredict = imagePredict / 255.0
start = time.time()
predicted = self.model.predict(np.array([imagePredict]))
predicted = round(predicted[0][0], 2)
if 0.0 <= predicted <= 0.2:
tf_sign = 'straight'
if 0.8 <= predicted <= 1.0:
tf_sign = 'turn'
print("{}\t{}\t{}".format(predicted, tf_sign,
time.time() - start))
return img, tf_sign
def getTalks(self):
talkContainer = SoupStrainer(attrs = {'class':re.compile('box clearfix')})
for talk in BeautifulSoup(self.html, parseOnlyThese = talkContainer):
title = talk.h4.a.string
link = URLTED+talk.dt.a['href']
pic = resizeImage(talk.find('img', attrs = {'src':re.compile('.+?\.jpg')})['src'])
yield {'url':link, 'Title':title, 'Thumb':pic}
def get_talks(self, url):
url = url + "?page=%s"
page_index = 1
# Have to know when to stop paging, see condition for loop exit below.
found_titles = set()
found_on_last_page = 0
html = self.get_HTML(url % (page_index))
while True:
containers = SoupStrainer('dl', {'class':re.compile('talkMedallion')})
found_on_this_page = 0
for talk in BeautifulSoup(html, parseOnlyThese=containers):
a_tag = talk.dt.a
title = a_tag['title'].strip()
if title not in found_titles:
found_titles.add(title)
found_on_this_page += 1
link = a_tag['href']
img_tag = a_tag.find('img', {'src':re.compile('http://images\.ted\.com/')})
img = img_tag['src']
yield title, URLTED + link, resizeImage(img)
# Results on last page == results on (last page + 1), _not_ 0 as you might hope.
# The second clause allows us to skip looking at last page + 1 if the last page contains
# fewer results than that before it; which is usually but not always the case.
if found_on_this_page and found_on_this_page >= found_on_last_page:
page_index += 1
found_on_last_page = found_on_this_page
html = self.get_HTML(url % (page_index))
else:
break
def getFavoriteTalks(self, userID, url = URLFAVORITES):
if userID:
html = self.get_HTML(url + userID)
talkContainer = SoupStrainer(attrs = {'class':re.compile('box clearfix')})
for talk in BeautifulSoup(html, parseOnlyThese = talkContainer):
title = talk.ul.a.string
link = URLTED+talk.dt.a['href']
pic = resizeImage(talk.find('img', attrs = {'src':re.compile('.+?\.jpg')})['src'])
yield {'url':link, 'Title':title, 'Thumb':pic}
else:
self.logger('invalid user object')
def getFavoriteTalks(self, userID, url=URLFAVORITES):
if userID:
html = self.get_HTML(url + userID)
talkContainer = SoupStrainer(attrs={'class':re.compile('col clearfix')})
for talk in BeautifulSoup(html, parseOnlyThese=talkContainer):
title = talk.a['title']
link = URLTED + talk.a['href']
img = resizeImage(talk.a.img['src'])
yield title, link, img
else:
self.logger('invalid user object')
def getFavoriteTalks(self, user, url = URLFAVORITES):
"""user must be TedTalks().User object with .id attribute"""
if user.id is not None:
html = getHTML(url+user.id)
talkContainer = SoupStrainer(attrs = {'class':re.compile('box clearfix')})
for talk in BeautifulSoup(html, parseOnlyThese = talkContainer):
title = talk.h4.a.string
link = URLTED+talk.dt.a['href']
pic = resizeImage(talk.find('img', attrs = {'src':re.compile('.+?\.jpg')})['src'])
yield {'url':link, 'Title':title, 'Thumb':pic}
else:
print '[%s] %s invalid user object' % (pluginName, __name__)
def getFavoriteTalks(self, user, url = URLFAVORITES):
"""user must be TedTalks().User object with .id attribute"""
if user.id is not None:
html = getHTML(url+user.id)
talkContainer = SoupStrainer(attrs = {'class':re.compile('box clearfix')})
for talk in BeautifulSoup(html, parseOnlyThese = talkContainer):
title = talk.h4.a.string
link = URLTED+talk.dt.a['href']
pic = resizeImage(talk.find('img', attrs = {'src':re.compile('.+?\.jpg')})['src'])
yield {'url':link, 'Title':title, 'Thumb':pic}
else:
print '[%s] %s invalid user object' % (pluginName, __name__)
def getTalks(self):
# themes loaded with a json call. Why are they not more consistant?
from simplejson import loads
# search HTML for the link to tedtalk's "api". It is easier to use regex here than BS.
jsonUrl = URLTED+re.findall('DataSource\("(.+?)"', self.html)[0]
# make a dict from the json formatted string from above url
talksMarkup = loads(getHTML(jsonUrl))
# parse through said dict for all the metadata
for markup in talksMarkup['resultSet']['result']:
talk = BeautifulSoup(markup['markup'])
link = URLTED+talk.dt.a['href']
title = cleanHTML(talk.dt.a['title'])
pic = resizeImage(talk.find('img', attrs = {'src':re.compile('.+?\.jpg')})['src'])
yield {'url':link, 'Title':title, 'Thumb':pic}
def getTalks(self):
# themes loaded with a json call. Why are they not more consistant?
from simplejson import loads
# search HTML for the link to tedtalk's "api". It is easier to use regex here than BS.
jsonUrl = URLTED+re.findall('DataSource\("(.+?)"', self.html)[0]
# make a dict from the json formatted string from above url
talksMarkup = loads(getHTML(jsonUrl))
# parse through said dict for all the metadata
for markup in talksMarkup['resultSet']['result']:
talk = BeautifulSoup(markup['markup'])
link = URLTED+talk.dt.a['href']
title = cleanHTML(talk.dt.a['title'])
pic = resizeImage(talk.find('img', attrs = {'src':re.compile('.+?\.jpg')})['src'])
yield {'url':link, 'Title':title, 'Thumb':pic}
def preprocess(self, boxlabel_df, sampleFiles, shrink):
# First create a list with required length
self.dataStore = [None] * len(sampleFiles)
sampleCount = len(sampleFiles)
for fileIndex in range(sampleCount):
print("Preprocess: {}/{} \r".format(fileIndex, sampleCount),
end="",
file=sys.stderr)
fileName = sampleFiles.iloc[fileIndex].TrainImageId
# Load and resize image
img = sd_util.loadImage(
os.path.join(sd_util.TRAIN_IMAGES, fileName))
if shrink != 1:
img = sd_util.resizeImage(img, self.IMAGE_H, self.IMAGE_W)
img = img.reshape((self.IMAGE_H, self.IMAGE_W, 1))
img -= 88 # The average grayscale of images is 88
selectedRows = boxlabel_df[boxlabel_df.ImageId == fileName]
defectClasses = [None] * sd_util.NUM_CLASS
margins = [self.IMAGE_W, 0, self.IMAGE_H, 0]
for classIndex in range(sd_util.NUM_CLASS):
if selectedRows.iloc[classIndex].hasBox:
encodedBoxes = selectedRows.iloc[classIndex].EncodedBoxes
rects = sd_util.encoded2rects(encodedBoxes)
temp = np.zeros(
(len(rects), 4)) # Convert it from tuple to np matrix
temp[:, :] = rects
rects = temp
if shrink != 1:
rects /= shrink
defectClasses[classIndex] = rects
if np.min(rects[:, 0]) < margins[0]:
margins[0] = np.min(rects[:, 0])
if np.min(rects[:, 1]) < margins[1]:
margins[1] = np.min(rects[:, 1])
if np.max(rects[:, 2]) > margins[2]:
margins[2] = np.max(rects[:, 2])
if np.max(rects[:, 3]) > margins[3]:
margins[3] = np.max(rects[:, 3])
self.dataStore[fileIndex] = (img, defectClasses, margins)
def identifyCharacter(image1_path):
# read image
image = cv2.imread(image1_path)
# Use technology similar to QR code to correct image angulation
warped = FixPerspective.fixPerspective(imageOpenCV=image)
# Remove colored pixels to compare only black ones
warped = util.cleanImage(warped)
# resize image
image = util.resizeImage(warped)
# show the original and scanned images
print("STEP 3: Apply perspective transform")
cv2.imshow("Original", image)
cv2.waitKey(0)
# image to ascii art
ascii1 = ImageToAscii.handle_image_conversion(image=image)
print("ascii da imagem de entrada")
print ascii1
# find most similar key
# load KNN table
filehandler = open("matrix-KNN.pickle", 'r')
matrix = pickle.load(filehandler)
smallerDistance = float('inf') # a very large number
result = None
for case in matrix:
distance = Levenshtein.levenshtein(ascii1,
case[0],
signficance=case[1])
if distance < smallerDistance:
smallerDistance = distance
result = case[1]
print("distance = ", smallerDistance)
print("result = ", result)
def main():
'''
Steps:
Read all images of "testeCases/" folder catalogging by their names
i.g: image A.png contain the best case of an "A" char, so save it into the KNN
KNN will be a ordered matrix nX2
n == number of tests char
2 = ascii art + corresponding char
Do not read from "testeCases/unknown/". This images will be use to test the KNN
'''
# Get all image paths
folder = "../testeCases"
paths = [os.path.join(folder, nome) for nome in os.listdir(folder)]
files = [arq for arq in paths if os.path.isfile(arq)]
pngs = [arq for arq in files if arq.lower().endswith(".png")]
# convert each image to ascii and store into matrix
matrix = []
for imagePath in pngs:
# read image
image = cv2.imread(imagePath)
# resize image to compare always image of same sizes
image = util.resizeImage(image)
# Convert to ascii and build matrix
asciiImage = ImageToAscii.handle_image_conversion(image=image)
imageName = imagePath.replace(folder + '/', '').replace('.png', '')
matrix.append([asciiImage, imageName])
# save matrix
filehandler = open("matrix-KNN.pickle", 'w')
pickle.dump(matrix, filehandler)
def test_resizeImage(self):
self.assertEqual("feefifofum_389x292.jpg", util.resizeImage("feefifofum_132x99.jpg"))
self.assertEqual("feefifofum_389x292.jpg", util.resizeImage("feefifofum_123x321.jpg"))
contours = cv2.findContours(mask,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE)[-2]
# contours = imutils.grab_contours(contours)
img = None
tf_sign = 'None'
for c in contours:
area = cv2.contourArea(c)
# cv2.drawConatours()
# if area > 150:
if(area > 150 and area<600):
x,y,w,h = cv2.boundingRect(c)
if (abs(w-h) < (w+h)/10):
img = frame[y:y+h,x:x+w]
if img is not None:
# cv2.imshow('imge',img)
imagePredict = util.resizeImage(img)
imagePredict = imagePredict / 255.0
predicted = model.predict(np.array([imagePredict]))
# print (predicted)
if predicted[0][0] <= 0.2:
# print('left')
tf_sign = 'left'
# rospy.sleep(1)
if predicted[0][0] > 0.8 and predicted[0][0] <= 1.0:
# print('right'
tf_sign = 'right'
publish_traffic.publish(tf_sign)
if(config.get_debug() == True):
cv2.imshow('output', mask)
# cv2.imshow('hsv', hsv)
else:
def test_resizeImage(self):
self.assertEqual('feefifofum_389x292.jpg',
util.resizeImage('feefifofum_132x99.jpg'))
self.assertEqual('feefifofum_389x292.jpg',
util.resizeImage('feefifofum_123x321.jpg'))
