Develop Reference

SVM always gives the same prediction

I have a page and I am trying to predict letters on the page by using SVM. However, the prediction result always the same. As a result, I am unable to read the page.
This is my code to detect the letters from the line images(I separated lines before):
def letters(img):
rect_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (2,2))
morph_letters = cv2.morphologyEx(img, cv2.MORPH_CROSS, rect_kernel)
cv2.imshow("morph", morph_letters)
cnts = cv2.findContours(morph_letters.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
cnts = imutils.grab_contours(cnts)
lengthOfContours = len(cnts)
print("Number of counturs: ", lengthOfContours)
points = []
rects = []
for (i, c) in enumerate(cnts):
area = cv2.contourArea(c)
if area > 80:
print("area "+str(area))
(x, y, w, h) = cv2.boundingRect(c)
x = x - 3
y = y - 10
w = w + 4
h = h + 12
cv2.rectangle(img_letters, (x, y), (x + w, y + h), (0, 255, 0), 2)
rect = [x, y, w+x, h+y]
leftPoint = x
#print("leftPoint: ", leftPoint)
points.append(leftPoint)
points.sort()
rects.append([i,rect])
rects.sort(key=lambda x: x[1])
#print("rects: ", rects)
if i == lengthOfContours-1:
for i in range(len(rects)):
found_letter = thresh_adaptive[rects[i][1][1]:rects[i][1][3], rects[i][1][0]:rects[i][1][2]]
prediction(found_letter)
And this is my prediction code. It takes the letter image from def letters function, name: found_letter. However, the result is always the same, l.
def prediction(word_img):
pick = open('model-letters.sav', 'rb')
#pickle.dump(model, pick)
model = pickle.load(pick)
pick.close()
#prediction = model.predict(x_test)
#accuracy = model.score(x_test, y_test)
categories = ['a','b','c','d','e','f','g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v','w','x','y','z','A','B','C','D','E','F','G','H','I','J', 'K','L','M','N','O','P','Q','R','S','T','U','V','W','X','Y','Z','1','2','3','4', '5','6','7','8','9','0']
img = word_img
try:
img = cv2.resize(img, (50,50))
img = np.array(img).flatten()
img = img.reshape(1, -1)
print("working")
prediction = model.predict(img)
letter = categories[prediction[0]]
print("Prediction ", prediction)
print("Prediction letter: ", letter)
#print("Accuracy: ",accuracy)
letterImg = img
cv2.imshow("letter", letterImg)
cv2.waitKey(0)
fh = open("letters.txt", "a")
fh.write(letter)
fh.close()
except:
print("error")
And this is my for loop to gives all the line images to def letters(img) function.
index = -1
for i in range(30,-1,-1):
index = index + 1
dir = "images-lines2/line" + str(i) + ".png"
img_orj = cv2.imread(dir)
img_orj = cv2.resize(img_orj, None, fx=1, fy=1.5, interpolation=cv2.INTER_AREA)
img_letters = img_orj.copy()
gray = cv2.cvtColor(img_orj, cv2.COLOR_BGR2GRAY)
img_lines = img_orj.copy()
blur_gaus = cv2.GaussianBlur(gray,(3,3),0)
kernel = np.ones((3,3), np.uint8)
thresh_adaptive = cv2.adaptiveThreshold(blur_gaus,255,cv2. ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY_INV,7,11)
letter_img = letters(thresh_adaptive)
If you want to see cv2.imshow("letter", letterImg), this is the
output
Any help is appreciated.

Extract text from pdf file

I want to extract certain words from pdf file, such as the words in the first column in the image, but the first letter of them is not recognized, perhaps because it is close to the edge of the table.
Is there a suggestion to extract the text correctly from a file as the displayed image??.
I used python tesseract lib.
image = cv2.imread('page' + str(j) + '.jpg')
rows,cols,_ = image.shape
image = cv2.resize(image, (np.int32(cols/2),np.int32(rows/2)))
# convert to gray and binarize
gray_img = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
binary_img = cv2.adaptiveThreshold(gray_img, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 9, 9)
# note: erosion and dilation works on white forground
binary_img = cv2.bitwise_not(binary_img)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3,3))
dilated_img = cv2.morphologyEx(binary_img, cv2.MORPH_DILATE ,kernel,iterations=2)
cnts, _ = cv2.findContours(dilated_img, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
rect_cnts = [] # Rectangular contours
for cnt in cnts:
rect_cnts.append(cnt)
rect_cnts = sorted(rect_cnts, key=cv2.contourArea, reverse=True)[:1]
for rec in rect_cnts:
# find bounding rectangle of biggest contour
box = cv2.boundingRect(rec)
x,y,w,h = box[:]
img = image[y:y+h,x:x+w]
cv2.imwrite('ROI_{}.png'.format(ROI_number), img)
ROI_number += 1
if j < len(d):
d[j] = pytesseract.image_to_data(img, output_type=Output.DICT,config = config, lang='eng') # extract words from image - store extracted words in d
else:
d.append(pytesseract.image_to_data(img, output_type=Output.DICT,config = config, lang='eng'))
n_boxes = len(d[j]['level'])
for i in range(n_boxes):
text = d[j]['text'][i]
text = text.replace("-", "").replace("/", "").replace("=", "").replace(",", "").replace(".", "").upper()
print(text)
for word in wordMap:
if text[len(text) - 1 : len(text)] == ',':
print("----- detect text ------")
print(text[len(text) - 1 : len(text)])
if word in text[0 : len(text) - 1]:
if not(word in im_list):
im_list.append(word)
else:
if word in text:
print("----- detect text ------")
print(text[len(text) - 1 : len(text)])
if not(word in im_list):
im_list.append(word)

My Automated Python script loses so much performance over time

This is my code, i know its long and messy code, but basically. At start its pretty fast and stuff, but after while you can notice really huge performance jumps, after 6 hours, i came back. and one click of keyboard button took like 10 seconds, can anybody help me find why is it slowing so much.
If you dont understand the code because i know its very messy and no comments. Its basically while loop that runs arround pressings buttons, and when it encounters pokemon, it reads its name and if its the one i want, it notifies me. Or if it errors, it notifies me too.
If i dont want it, it kills it
while active:
gc.collect()
gc.collect()
gc.collect()
time.sleep(0.1)
def waitclick():
while True:
if keyboard.read_key() == "f6":
global active
active = False
thread = threading.Thread(target=waitclick).start()
im = numpy.asarray(pyautogui.screenshot(region=hpcheck))
# Define the blue colour we want to find - remember OpenCV uses BGR ordering
color = [175, 65, 63]
X, Y = numpy.where(numpy.all(im == color, axis=2))
x = list(X)
if lenght:
cnt = 0
if len(x) <= lenght * (3 / 4):
while len(x) <= lenght* (5 / 6):
if cnt > 20:
pyautogui.click(2119, 793)
pyautogui.keyDown('4')
time.sleep(0.01)
pyautogui.keyUp('4')
pyautogui.click(2050, 1120)
pyautogui.click(1720, 1030)
break
cnt += 1
time.sleep(1)
pyautogui.moveTo(potion[0],potion[1], random.uniform(0.5,1))
pyautogui.click(potion[0],potion[1])
time.sleep(0.01)
pyautogui.click(potion[0],potion[1])
time.sleep(0.1)
pyautogui.moveTo(mainpoke[0],mainpoke[1],random.uniform(0.5,1))
pyautogui.click(mainpoke[0],mainpoke[1])
pyautogui.click(mainpoke[0], mainpoke[1])
im = Image.fromarray(im)
im.save("health.png")
im.close()
im = numpy.asarray(pyautogui.screenshot(region=(hpcheck)))
# Define the blue colour we want to find - remember OpenCV uses BGR ordering
color = [175, 65, 63]
X, Y = numpy.where(numpy.all(im == color, axis=2))
x = list(X)
lenght = len(x)
else:
lenght = len(x)
key = keys[0] if steps % 2 == 0 else keys[1]
pyautogui.keyDown(key)
pixels = []
battle = False
timeup = False
def wait():
global stepsleft
global steps
global battle
counter = 0
for i in range(stepsleft):
if battle:
break
counter += 1
time.sleep(onetile-(random.uniform(0.001,0.002)))
stepsleft = stepsleft - counter
if stepsleft == 0:
steps += 1
stepsleft = tiles
global timeup
timeup = True
thread = threading.Thread(target=wait).start()
while True:
if timeup:
break
im = numpy.asarray(pyautogui.screenshot(region=(check)))
if list(im[0, 0]) == [230, 230, 230]:
battle = True
break
pyautogui.keyUp(key)
stepcount += 1
im = numpy.asarray(pyautogui.screenshot(region=(check)))
# im = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
if stepcount > 15:
if fails > 100:
call = client.calls.create(twiml='<Response><Say>Ahoy, World!</Say></Response>',
to='+421949847318', from_='+18596961895')
payload = {
"content": f"Step Fail, continuing."
}
r = requests.post(f"https://discord.com/api/v9/channels/{channel_id}/messages",
data=payload,
headers=header)
gc.collect()
fails = 0
stepfails += 1
if stepfails > 10:
break
pyautogui.click(2119, 793)
pyautogui.keyDown('4')
time.sleep(0.01)
pyautogui.keyUp('4')
pyautogui.click(2050,1120)
pyautogui.click(1720, 1030)
time.sleep(1)
fails += 1
if battle == True:
breaks += 1
stepcount = 0
c = 0
e = False
while True:
if c > 600:
e = True
break
else:
im = numpy.asarray(pyautogui.screenshot(region=(namecheck)))
if list(im[0, 0]) == [254, 254, 254]:
break
c += 1
if e:
pyautogui.keyDown('4')
time.sleep(0.01)
pyautogui.keyUp('4')
pyautogui.click(2050, 1120)
time.sleep(1)
else:
encounters += 1
im = numpy.asarray(pyautogui.screenshot(region=(mon)))
im = cv2.resize(im, None, fx=1.2, fy=1.2, interpolation=cv2.INTER_CUBIC)
im = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
kernel = numpy.ones((1, 1), numpy.uint8)
im = cv2.threshold(cv2.bilateralFilter(im, 5, 75, 75), 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
im = cv2.dilate(im, kernel, iterations=1)
im = cv2.erode(im, kernel, iterations=1)
im = ~im
text = pytesseract.image_to_string(im)
text = text.strip().replace("\n","")
ime = Image.fromarray(im)
ime.save(f"Wild {text}.png")
ime.close()
if text == "":
ime = Image.fromarray(im)
ime.save("fail.png")
ime.close()
im = numpy.asarray(pyautogui.screenshot(region=(mon)))
im = cv2.resize(im, None, fx=1.6, fy=1.6, interpolation=cv2.INTER_CUBIC)
im = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
kernel = numpy.ones((1, 1), numpy.uint8)
im = cv2.threshold(cv2.bilateralFilter(im, 5, 75, 75), 0, 255,
cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
im = cv2.dilate(im, kernel, iterations=1)
im = cv2.erode(im, kernel, iterations=1)
im = ~im
text = pytesseract.image_to_string(im).strip()
print(">" + text + "<")
if text == "":
ime = Image.fromarray(im)
ime.save("fail2.png")
ime.close()
im = numpy.asarray(pyautogui.screenshot(region=mon))
print(">" + text + "<")
im = cv2.resize(im, None, fx=2, fy=2, interpolation=cv2.INTER_CUBIC)
im = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
kernel = numpy.ones((1, 1), numpy.uint8)
im = cv2.threshold(cv2.bilateralFilter(im, 5, 75, 75), 0, 255,
cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
im = cv2.dilate(im, kernel, iterations=1)
im = cv2.erode(im, kernel, iterations=1)
im = ~im
text = pytesseract.image_to_string(im).strip()
if text == "":
im = numpy.asarray(pyautogui.screenshot(region=check))
if list(im[0, 0]) != [230, 230, 230]:
pass
else:
Image.fromarray(im)
ime.save("fail3.png")
ime.close()
print(">" + text + "<")
payload = {
"content": f"Go look at this bro, i failed again."
}
r = requests.post(f"https://discord.com/api/v9/channels/{channel_id}/messages", data=payload,
headers=header)
active = False
call = client.calls.create(twiml='<Response><Say>Ahoy, World!</Say></Response>',
to='+421949847318', from_='+18596961895')
gc.collect()
break
if text.replace("[S]","").replace("[E]","").replace("\n","") not in pokemonlist:
ime = Image.fromarray(im)
ime.save("fail.png")
ime.close()
im = numpy.asarray(pyautogui.screenshot(region=mon))
im = cv2.resize(im, None, fx=1.6, fy=1.6, interpolation=cv2.INTER_CUBIC)
im = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
kernel = numpy.ones((1, 1), numpy.uint8)
im = cv2.threshold(cv2.bilateralFilter(im, 5, 75, 75), 0, 255,
cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
im = cv2.dilate(im, kernel, iterations=1)
im = cv2.erode(im, kernel, iterations=1)
im = ~im
text = pytesseract.image_to_string(im).strip().replace("\n","")
print(">" + text + "<")
if text.replace("[S]","").replace("\n","").replace("[E]","") not in pokemonlist:
ime = Image.fromarray(im)
ime.save("fail2.png")
ime.close()
im = numpy.asarray(pyautogui.screenshot(region=mon))
im = cv2.resize(im, None, fx=2, fy=2, interpolation=cv2.INTER_CUBIC)
im = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
kernel = numpy.ones((1, 1), numpy.uint8)
im = cv2.threshold(cv2.bilateralFilter(im, 5, 75, 75), 0, 255,
cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
im = cv2.dilate(im, kernel, iterations=1)
im = cv2.erode(im, kernel, iterations=1)
im = ~im
text = pytesseract.image_to_string(im).strip()
print(">" + text + "<")
if text.replace("[S]","").replace("\n","").replace("[E]","") not in pokemonlist:
ime = Image.fromarray(im)
ime.save("fail3.png")
ime.close()
print(">" + text + "<")
payload = {
"content": f"Go look at this bro, i failed again."
}
r = requests.post(f"https://discord.com/api/v9/channels/{channel_id}/messages", data=payload,
headers=header)
active = False
call = client.calls.create(twiml='<Response><Say>Ahoy, World!</Say></Response>',
to='+421949847318', from_='+18596961895')
gc.collect()
break
encounter_time = datetime.datetime.utcnow()
times.append(encounter_time)
if not ram or encounters < 5:
ram = process.memory_info().rss
if encounters < 15:
recentspeed = "None yet"
else:
recentspeed = str(round(
(15 / (encounter_time - times[encounters - 15]).total_seconds()) * 60, 2)) + "/min"
print(
f"Pokemon: {text} #{encounters}\nSpeed: {round((encounters / (encounter_time - start).total_seconds()) * 60, 2)}/min \nRecent Speed: {recentspeed}\nRam usage is {process.memory_info().rss} Starting usage was {process.memory_info().rss-ram} Garbage is {gc.get_count()}")
payload = {
"content": f"{'-'*20}\nPokemon: {text} #{encounters}\nSpeed: {round((encounters / (encounter_time - start).total_seconds()) * 60, 2)}/min \nRecent Speed: {recentspeed}\n Ram Change: `{(process.memory_info().rss-ram)//1000000}MB`\nRam usage: `{process.memory_info().rss//1000000}MB`\nStarting usage: `{ram//1000000}MB`\nGarbage: {gc.get_count()}\nCPU: {psutil.cpu_percent()}"
}
r = requests.post(f"https://discord.com/api/v9/channels/{logs_id}/messages", data=payload,
headers=header)
gc.collect()
#json
with open("encounters.json", "r") as a_file:
json_object = json.load(a_file)
a_file.close()
try:
json_object[text] += 1
except:
json_object[text] = 1
with open("encounters.json", "w") as a_file:
json.dump(json_object, a_file)
a_file.close()
c = 0
e = False
while True:
if c > 500:
pyautogui.click(2119, 793)
pyautogui.keyDown('4')
time.sleep(0.01)
pyautogui.keyUp('4')
pyautogui.click(2050, 1120)
pyautogui.click(1720, 1030)
time.sleep(1)
e = True
break
im = numpy.asarray(pyautogui.screenshot(region=battlecheck))
if list(im[0, 0]) == [240, 211, 92]:
break
c += 1
if e == True:
continue
if (text.startswith("[S]") if shinies else text.replace("[E]","") in shinywanted) or text.replace("[S]","").replace("[E]","") in wanted:
time.sleep(2)
end = datetime.datetime.utcnow()
done = end - start
timed = humanize.naturaldelta(done)
print("Done! in " + timed)
im = numpy.asarray(pyautogui.screenshot(region=win))
ime = Image.fromarray(im)
ime.save(f"{text}.png")
ime.close()
# File
files = {
"file": (f"{text}.png", open(f"{text}.png", 'rb'))
# The picture that we want to send in binary
}
# Optional message to send with the picture
payload = {
"content": f"Found {text} in {timed} of searching. It took {encounters} encounters"
}
active = False
r = requests.post(f"https://discord.com/api/v9/channels/{channel_id}/messages", data=payload,
headers=header, files=files)
gc.collect()
call = client.calls.create(twiml='<Response><Say>Ahoy, World!</Say></Response>',to='+421949847318',from_='+18596961895')
time.sleep(25)
if remote:
pyautogui.click(3804, 15)
break
playsound('ringtone.mp3')
elif text.startswith("[E]") or text.replace("[S]","").replace("[E]","") in avoid:
run()
c = 0
while True:
if c > 300:
pyautogui.click(2119, 793)
pyautogui.click(1720, 1030)
pyautogui.keyDown('4')
time.sleep(0.01)
pyautogui.keyUp('4')
pyautogui.click(2050, 1120)
time.sleep(1)
break
else:
im = numpy.asarray(pyautogui.screenshot(region=check))
if list(im[0, 0]) != [230, 230, 230]:
break
c += 1
else:
#time.sleep(random.randint(50, 70) / 100)
kill(text.replace("[S]","").replace("[E]",""))
c = 0
time.sleep(1)
while True:
if c > 750:
pyautogui.click(2119, 793)
pyautogui.click(1720, 1030)
pyautogui.keyDown('4')
time.sleep(0.01)
pyautogui.keyUp('4')
pyautogui.click(2050, 1120)
time.sleep(1)
break
else:
im = numpy.asarray(pyautogui.screenshot(region=check))
if list(im[0, 0]) != [230, 230, 230]:
break
c += 1

The code is barely legible, split it in functions or classes, and avoid global variables. Chances are you are overflowing the RAM of your computer or something like that, if you are appending information to a variable, remember deleting it after if it grows to much or storing it in a database-like system.
Something similar (or worst) applies to threads if you create too many without deleting them at some point.

how to resolve this error which generates in matching two images in image stitching code

while doing image stitching code in google colab I got an error can anyone help how to resolve it?
ERROR: File "<ipython-input-12-7c271414318b>", line 58
def filter_matches(self, matches, ratio = 0.75):
^
IndentationError: unindent does not match any outer indentation level
CODE:
import os
import sys
import cv2
import math
import numpy as np
import utils
from numpy import linalg
class AlignImagesRansac(object):
def _init_(self, image_dir, key_frame, output_dir, img_filter=None):
'''
image_dir: 'C:\Users\Hamza Ahmed\Desktop\auto' containing all images
key_frame: 'C:\Users\Hamza Ahmed\Desktop\auto\c.jpg' of the base image
output_dir: 'C:\Users\Hamza Ahmed\Desktop\auto' where to save output images
optional:
img_filter = 'JPG'; None->Take all images
'''
self.key_frame_file = os.path.split(key_frame)[-1]
self.output_dir = output_dir
# Open the directory given in the arguments
self.dir_list = []
try:
self.dir_list = os.listdir(image_dir)
if img_filter:
# remove all files that doen't end with .[image_filter]
self.dir_list = filter(lambda x: x.find(img_filter) > -1, self.dir_list)
try: #remove Thumbs.db, is existent (windows only)
self.dir_list.remove('Thumbs.db')
except ValueError:
pass
except:
print >> sys.stderr, ("Unable to open directory: %s" % image_dir)
sys.exit(-1)
self.dir_list = map(lambda x: os.path.join(image_dir, x), self.dir_list)
self.dir_list = filter(lambda x: x != key_frame, self.dir_list)
base_img_rgb = cv2.imread(key_frame)
if base_img_rgb == None:
raise IOError("%s doesn't exist" %key_frame)
# utils.showImage(base_img_rgb, scale=(0.2, 0.2), timeout=0)
# cv2.destroyAllWindows()
final_img = self.stitchImages(base_img_rgb, 0)
def filter_matches(self, matches, ratio = 0.75):
filtered_matches = []
for m in matches:
if len(m) == 2 and m[0].distance < m[1].distance * ratio:
filtered_matches.append(m[0])
return filtered_matches
def imageDistance(self, matches):
sumDistance = 0.0
for match in matches:
sumDistance += match.distance
return sumDistance
def findDimensions(self, image, homography):
base_p1 = np.ones(3, np.float32)
base_p2 = np.ones(3, np.float32)
base_p3 = np.ones(3, np.float32)
base_p4 = np.ones(3, np.float32)
(y, x) = image.shape[:2]
base_p1[:2] = [0,0]
base_p2[:2] = [x,0]
base_p3[:2] = [0,y]
base_p4[:2] = [x,y]
max_x = None
max_y = None
min_x = None
min_y = None
for pt in [base_p1, base_p2, base_p3, base_p4]:
hp = np.matrix(homography, np.float32) * np.matrix(pt, np.float32).T
hp_arr = np.array(hp, np.float32)
normal_pt = np.array([hp_arr[0]/hp_arr[2], hp_arr[1]/hp_arr[2]], np.float32)
if ( max_x == None or normal_pt[0,0] > max_x ):
max_x = normal_pt[0,0]
if ( max_y == None or normal_pt[1,0] > max_y ):
max_y = normal_pt[1,0]
if ( min_x == None or normal_pt[0,0] < min_x ):
min_x = normal_pt[0,0]
if ( min_y == None or normal_pt[1,0] < min_y ):
min_y = normal_pt[1,0]
min_x = min(0, min_x)
min_y = min(0, min_y)
return (min_x, min_y, max_x, max_y)
def stitchImages(self, base_img_rgb, round=0):
if ( len(self.dir_list) < 1 ):
return base_img_rgb
# print base_img_rgb.channels()
# if(image.channels()==1)
# { /* Grayscale */ }
# else if (image.channels==4)
# { /* ARGB or RGBA image */
base_img = cv2.GaussianBlur(cv2.cvtColor(base_img_rgb,cv2.COLOR_BGR2GRAY), (5,5), 0)
# Use the SIFT feature detector
detector = cv2.SIFT()
# Find key points in base image for motion estimation
base_features, base_descs = detector.detectAndCompute(base_img, None)
# Create new key point list
# key_points = []
# for kp in base_features:
# key_points.append((int(kp.pt[0]),int(kp.pt[1])))
# utils.showImage(base_img, key_points, scale=(0.2, 0.2), timeout=0)
# cv2.destroyAllWindows()
# Parameters for nearest-neighbor matching
FLANN_INDEX_KDTREE = 1 # bug: flann enums are missing
flann_params = dict(algorithm = FLANN_INDEX_KDTREE,
trees = 5)
matcher = cv2.FlannBasedMatcher(flann_params, {})
print ("Iterating through next images...")
closestImage = None
# TODO: Thread this loop since each iteration is independent
# Find the best next image from the remaining images
for next_img_path in self.dir_list:
print ("Reading %s..." % next_img_path)
if ( self.key_frame_file in next_img_path ):
print ("\t Skipping %s..." % self.key_frame_file)
continue
# Read in the next image...
next_img_rgb = cv2.imread(next_img_path)
next_img = cv2.GaussianBlur(cv2.cvtColor(next_img_rgb,cv2.COLOR_BGR2GRAY), (5,5), 0)
# if ( next_img.shape != base_img.shape ):
# print "\t Skipping %s, bad shape: %s" % (next_img_path, next_img.shape)
# continue
print ("\t Finding points...")
# Find points in the next frame
next_features, next_descs = detector.detectAndCompute(next_img, None)
matches = matcher.knnMatch(next_descs, trainDescriptors=base_descs, k=2)
print ("\t Match Count: ", len(matches))
matches_subset = self.filter_matches(matches)
print ("\t Filtered Match Count: ", len(matches_subset))
distance = self.imageDistance(matches_subset)
print ("\t Distance from Key Image: ", distance)
averagePointDistance = distance/float(len(matches_subset))
print ("\t Average Distance: ", averagePointDistance)
kp1 = []
kp2 = []
for match in matches_subset:
kp1.append(base_features[match.trainIdx])
kp2.append(next_features[match.queryIdx])
p1 = np.array([k.pt for k in kp1])
p2 = np.array([k.pt for k in kp2])
H, status = cv2.findHomography(p1, p2, cv2.RANSAC, 5.0)
print ('%d / %d inliers/matched' % (np.sum(status), len(status)))
inlierRatio = float(np.sum(status)) / float(len(status))
# if ( closestImage == None or averagePointDistance < closestImage['dist'] ):
if ( closestImage == None or inlierRatio > closestImage['inliers'] ):
closestImage = {}
closestImage['h'] = H
closestImage['inliers'] = inlierRatio
closestImage['dist'] = averagePointDistance
closestImage['path'] = next_img_path
closestImage['rgb'] = next_img_rgb
closestImage['img'] = next_img
closestImage['feat'] = next_features
closestImage['desc'] = next_descs
closestImage['match'] = matches_subset
print ("Closest Image: ", closestImage['path'])
print ("Closest Image Ratio: ", closestImage['inliers'])
self.dir_list = filter(lambda x: x != closestImage['path'], self.dir_list)
# utils.showImage(closestImage['img'], scale=(0.2, 0.2), timeout=0)
# cv2.destroyAllWindows()
H = closestImage['h']
H = H / H[2,2]
H_inv = linalg.inv(H)
if ( closestImage['inliers'] > 0.1 ): # and
(min_x, min_y, max_x, max_y) = self.findDimensions(closestImage['img'], H_inv)
# Adjust max_x and max_y by base img size
max_x = max(max_x, base_img.shape[1])
max_y = max(max_y, base_img.shape[0])
move_h = np.matrix(np.identity(3), np.float32)
if ( min_x < 0 ):
move_h[0,2] += -min_x
max_x += -min_x
if ( min_y < 0 ):
move_h[1,2] += -min_y
max_y += -min_y
print ("Homography: \n", H)
print ("Inverse Homography: \n", H_inv)
print ("Min Points: ", (min_x, min_y))
mod_inv_h = move_h * H_inv
img_w = int(math.ceil(max_x))
img_h = int(math.ceil(max_y))
print ("New Dimensions: ", (img_w, img_h))
# Warp the new image given the homography from the old image
base_img_warp = cv2.warpPerspective(base_img_rgb, move_h, (img_w, img_h))
print ("Warped base image")
# utils.showImage(base_img_warp, scale=(0.2, 0.2), timeout=5000)
# cv2.destroyAllWindows()
next_img_warp = cv2.warpPerspective(closestImage['rgb'], mod_inv_h, (img_w, img_h))
print ("Warped next image")
# utils.showImage(next_img_warp, scale=(0.2, 0.2), timeout=5000)
# cv2.destroyAllWindows()
# Put the base image on an enlarged palette
enlarged_base_img = np.zeros((img_h, img_w, 3), np.uint8)
print ("Enlarged Image Shape: ", enlarged_base_img.shape)
print ("Base Image Shape: ", base_img_rgb.shape)
print ("Base Image Warp Shape: ", base_img_warp.shape)
# enlarged_base_img[y:y+base_img_rgb.shape[0],x:x+base_img_rgb.shape[1]] = base_img_rgb
# enlarged_base_img[:base_img_warp.shape[0],:base_img_warp.shape[1]] = base_img_warp
# Create a mask from the warped image for constructing masked composite
(ret,data_map) = cv2.threshold(cv2.cvtColor(next_img_warp, cv2.COLOR_BGR2GRAY),
0, 255, cv2.THRESH_BINARY)
enlarged_base_img = cv2.add(enlarged_base_img, base_img_warp,
mask=np.bitwise_not(data_map),
dtype=cv2.CV_8U)
# Now add the warped image
final_img = cv2.add(enlarged_base_img, next_img_warp,
dtype=cv2.CV_8U)
# utils.showImage(final_img, scale=(0.2, 0.2), timeout=0)
# cv2.destroyAllWindows()
# Crop off the black edges
final_gray = cv2.cvtColor(final_img, cv2.COLOR_BGR2GRAY)
_, thresh = cv2.threshold(final_gray, 1, 255, cv2.THRESH_BINARY)
contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
print ("Found %d contours..." % (len(contours)))
max_area = 0
best_rect = (0,0,0,0)
for cnt in contours:
x,y,w,h = cv2.boundingRect(cnt)
# print "Bounding Rectangle: ", (x,y,w,h)
deltaHeight = h-y
deltaWidth = w-x
area = deltaHeight * deltaWidth
if ( area > max_area and deltaHeight > 0 and deltaWidth > 0):
max_area = area
best_rect = (x,y,w,h)
if ( max_area > 0 ):
print ("Maximum Contour: ", max_area)
print ("Best Rectangle: ", best_rect)
final_img_crop = final_img[best_rect[1]:best_rect[1]+best_rect[3],
best_rect[0]:best_rect[0]+best_rect[2]]
# utils.showImage(final_img_crop, scale=(0.2, 0.2), timeout=0)
# cv2.destroyAllWindows()
final_img = final_img_crop
# Write out the current round
final_filename = "%s/%d.JPG" % (self.output_dir, round)
cv2.imwrite(final_filename, final_img)
return self.stitchImages(final_img, round+1)
else:
return self.stitchImages(base_img_rgb, round+1)
# ----------------------------------------------------------------------------
#if _name_ == '_main_':
# if ( len(args) < 4 ):
# print >> sys.stderr, ("Usage: %s <image_dir> <key_frame> <output>" % args[0])
# sys.exit(-1)
#AlignImagesRansac(sys.args[1:])

Opencv(Python) memory usage issue

I've a problem with my software in Python. It's a big while cicle where I took a intel realsense (USB camera) stream. Using opencv I make a couple of findContours and I send the results of contours to another software.
The problem is that there is a memory consuption. In fact the RAM usage increase every 2-3 seconds by 0.1%.
II don't know what to do...
This is the code (sorry if it's not beautifull but I'm testing a lot of things)
import numpy as np
import random
import socket
import cv2
import time
import math
import pickle
import httplib, urllib
from xml.etree import ElementTree as ET
import logging
logging.basicConfig(level=logging.INFO)
try:
import pyrealsense as pyrs
except:
print("No pyralsense Module installed!")
#funzione per registrare gli eventi del mouse
def drawArea(event,x,y, flag, param):
global fx,fy,ix,iy
if event == cv2.EVENT_LBUTTONDOWN:
ix,iy = x,y
elif event == cv2.EVENT_LBUTTONUP:
fx,fy = x,y
def RepresentsInt(s):
try:
int(s)
return True
except ValueError:
return False
quit = False
read = False
while read == False:
file = open('default.xml', 'r')
tree = ET.parse(file)
root = tree.getroot()
for child in root:
if child.tag == "intel":
intel = int(child[0].text)
elif child.tag == "output":
portOut = int(child[2].text)
elif child.tag =="source":
video_source = child.text
file.close()
root.clear()
ix,iy = -1,-1
fx,fy = -1,-1
timeNP = 10
last = time.time()
smoothing = 0.9
fps_smooth = 30
#video_source = video_source.split(",")
read = True
if RepresentsInt(video_source):
video_source = int(video_source)
if intel == 1:
pyrs.start()
dev = pyrs.Device(video_source)
master = 1
address = ('', 3333)
broadSockListe = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
broadSockListe.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
broadSockListe.setsockopt(socket.SOL_SOCKET, socket.SO_BROADCAST, 1)
broadSockListe.bind(('',3333))
while True:
if master == 0:
datas, address = broadSockListe.recvfrom(1024)
if str(datas) == "8000":
separator = ":"
seq = (address[0],"8081")
masterAddr = separator.join(seq)
IP = str([l for l in (
[ip for ip in socket.gethostbyname_ex(socket.gethostname())[2] if not ip.startswith("127.")][:1], [
[(s.connect(('8.8.8.8', 53)), s.getsockname()[0], s.close()) for s in
[socket.socket(socket.AF_INET, socket.SOCK_DGRAM)]][0][1]]) if l][0][0])
params = separator.join(("addUnit",IP,str(portOut),"camera","generalList.xml"))
params = urllib.urlencode({"Python":params})
headers = {}
conn = httplib.HTTPConnection(masterAddr)
conn.request("POST",masterAddr ,params, headers)
params = separator.join(("masterIP",address[0],str(portOut)+"/","default.xml"))
params = urllib.urlencode({"Python":params})
headers = {}
myip = IP + ":8081"
conn = httplib.HTTPConnection(myip)
#eseguo una post al mio server
conn.request("POST", myip, params, headers)
broadSockListe.close()
#imposto master a 1 per dire che l'ho registrato e posso partire col programma
master = 1
read = False
while read == False:
'''# leggo le varie impostazioni dal file default
file = open('default.xml','r+')
tree = ET.parse(file)
root = tree.getroot()
for child in root:
if child.tag == "modifica" and child.text == "1":
child.text = "0"
tree.write('default.xml')
root.clear()
file.close()'''
read = True
prev,prevprev,dirX,dirY = 0,0,0,0
spostamento = 15
UDP_IP = ["", ""]
UDP_PORT = ["", ""]
UDP_IP[0] = "127.0.0.1"
UDP_PORT[0] = 3030
IP_left = "127.0.0.1"
IP_right = "127.0.0.1"
sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
sock.bind(("",portOut))
message = ""
sep = "-"
font = cv2.FONT_HERSHEY_SIMPLEX
kernel = cv2.getStructuringElement(cv2.MORPH_CROSS, (3, 3))
#rettangoli = [x,y,width,height,angle,box, area, contours]
rettangoli = []
cnt = 0
letto = 0
while True:
now = time.time()
if letto < now - 2 or letto == 0 or now < letto:
letto = now
print(now)
read = False
while read == False:
file = open('default.xml', 'r')
tree = ET.parse(file)
root = tree.getroot()
for child in root:
if child.tag == "output":
UDP_IP[1] = child[0].text
UDP_PORT[1] = int(child[1].text)
if child.tag == "effects":
erode = int(child[0].text)
erodePos = int(child[1].text)
erode2 = int(child[2].text)
erodePos2 = int(child[3].text)
dilate1 = int(child[4].text)
dilatePos1= int(child[5].text)
dilate2 = int(child[6].text)
dilatePos2 = int(child[7].text)
blur = int(child[8].text)
blurPos = int(child[9].text)
if child.tag == "intel":
val1Min = int(child[1].text)
val1Max = int(child[2].text)
val2Min = int(child[3].text)
val2Max = int(child[4].text)
val3Min = int(child[5].text)
val3Max = int(child[6].text)
if child.tag == "modifica":
if child.text == "1":
break
#definisco dimensioni per collisioni
if child.tag == "size":
blobSize= int(child[0].text)
dimBordoBlob= int(child[1].text)
if child.tag == "visualizza":
visualizza= child.text
if child.tag == "feedback":
SFB = int(child.text)
root.clear()
file.close()
read = True
dev.wait_for_frame()
c = dev.colour
c = cv2.cvtColor(c, cv2.COLOR_RGB2BGR)
d = dev.depth * dev.depth_scale * -60
d = d[5:485, 25:635]
d = cv2.applyColorMap(d.astype(np.uint8), cv2.COLORMAP_HSV)
c = cv2.resize(c, (320 ,240), interpolation=cv2.INTER_AREA)
d = cv2.resize(d, (320,240), interpolation=cv2.INTER_AREA)
#trasformo i colori in HSV per filtrarli
frame = cv2.cvtColor(d, cv2.COLOR_BGR2HSV)
lower_red = np.array([val1Min, val2Min, val3Min])
upper_red = np.array([val1Max, val2Max, val3Max])
frame = cv2.inRange(frame, lower_red, upper_red)
dimensions = frame.shape
widthStream = dimensions[1]
heightStream = dimensions[0]
roomFrame = np.zeros(( heightStream,widthStream, 3), np.uint8)
roomFrame[:] = (0, 0, 0)
fgmask = frame
halfheight = int(heightStream / 2)
halfwidth = int(widthStream / 2)
for i in range(0, 15):
if erode >= 1 and erodePos == i:
fgmask = cv2.erode(fgmask, kernel, iterations=erode)
if dilate1 >= 1 and dilatePos1 == i:
fgmask = cv2.dilate(fgmask, kernel, iterations=dilate1)
if erode2 >= 1 and erodePos2 == i:
fgmask = cv2.erode(fgmask, kernel, iterations=erode2)
if dilate2 >= 1 and dilatePos2 == i:
fgmask = cv2.dilate(fgmask, kernel, iterations=dilate2)
if blur == 1 and blurPos == 1:
fgmask = cv2.GaussianBlur(fgmask, (5, 5), 0)
if ix > fx:
temp = fx
fx = ix
ix = temp
if iy > fy:
temp = fy
fy = iy
iy = temp
if cnt == 0:
ix,iy = 1,1
fx,fy = widthStream-1,heightStream-1
fgmask, contours, hierarchy = cv2.findContours(fgmask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
rettangoli = []
for cont in contours:
rect = cv2.minAreaRect(cont)
box = cv2.boxPoints(rect)
box = np.int0(box)
width = rect[1][0]
height = rect[1][1]
angle = rect[2]
if width > height:
angle = 180 + angle
else:
angle = 270 + angle
x, y, w, h = cv2.boundingRect(cont)
centerX = int(w / 2 + x)
centerY = int(h / 2 + y)
M = cv2.moments(cont)
area = int(M['m00'])
if area > blobSize:
if ix < centerX < fx and iy < centerY < fy:
cv2.drawContours(fgmask, [cont], 0, (100, 100, 100), dimBordoBlob)
cv2.drawContours(fgmask, [cont], 0, (255, 255, 255), -1)
rettangoli.append([centerX, centerY, w, h, angle, box, area, cont])
indice = 0
fgmask, contours, hierarchy = cv2.findContours(fgmask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_TC89_KCOS)
if intel == 1:
fgmask = cv2.cvtColor(fgmask, cv2.COLOR_GRAY2RGB)
rettangoli = []
for cont in contours:
rect = cv2.minAreaRect(cont)
box = cv2.boxPoints(rect)
box = np.int0(box)
width = rect[1][0]
height = rect[1][1]
angle = rect[2]
if width > height:
angle = 180 + angle
else:
angle = 270 + angle
x, y, w, h = cv2.boundingRect(cont)
centerX = int(w / 2 + x)
centerY = int(h / 2 + y)
M = cv2.moments(cont)
indice += 1
if M['m00'] > blobSize:
if ix < centerX < fx and iy < centerY < fy:
rettangoli.append([centerX, centerY, w, h, angle, box, int(M['m00']), cont])
cv2.drawContours(roomFrame, [cont], 0, (255, 255, 255), -1)
for rett in rettangoli:
seq = (message,np.array_str(rett[7]))
message = sep.join(seq)
temp = 0
while temp < len(UDP_IP):
sock.sendto(bytes(message), (UDP_IP[temp], UDP_PORT[temp]))
temp += 1
message = ""
if SFB == 1:
cv2.imshow("Camera Intel", roomFrame)
if cv2.waitKey(1) & 0xFF == ord('r'):
break
if cv2.waitKey(1) & 0xFF == ord('q'):
quit = True
break
name = "color.jpeg"
cv2.imwrite(name, c)
name = "bn.jpeg"
cv2.imwrite(name, roomFrame)
if intel == 0:
cap.release()
cv2.destroyAllWindows()

You are creating new objects in your while loop. Take now for example, you create a variable and then you assign a new object to it that only lives in that loop. If you declare the variables before your loop the same object will be overwritten instead of re-created.
By just declaring the variables ahead of time with name = None you will be able to make sure you reuse these variables.
I hope this works for you.