class Worker:
def __init__(self,
NOMEROFF_NET_DIR,
MASK_RCNN_DIR,
MASK_RCNN_LOG_DIR,
load_model="latest",
options_detector="latest",
text_detector_module="eu",
load_text_detector="latest"):
sys.path.append(NOMEROFF_NET_DIR)
# Initialize npdetector with default configuration file.
self.nnet = Detector(MASK_RCNN_DIR, MASK_RCNN_LOG_DIR)
self.nnet.loadModel(load_model)
self.rectDetector = RectDetector()
self.optionsDetector = OptionsDetector()
self.optionsDetector.load(options_detector)
# Initialize text detector.
self.textDetector = TextDetector.get_static_module(
text_detector_module)()
self.textDetector.load(load_text_detector)
def detect(self, img_path):
# Detect numberplate
img = mpimg.imread(img_path)
NP = self.nnet.detect([img])
# Generate image mask.
cv_img_masks = filters.cv_img_mask(NP)
# Detect points.
arrPoints = self.rectDetector.detect(cv_img_masks)
zones = self.rectDetector.get_cv_zonesBGR(img, arrPoints)
# find standart
regionIds, stateIds, countLines = self.optionsDetector.predict(zones)
regionNames = self.optionsDetector.getRegionLabels(regionIds)
# find text with postprocessing by standart
textArr = self.textDetector.predict(zones)
textArr = textPostprocessing(textArr, regionNames)
return textArr
def loadModel():
global nnet, textDetector, rectDetector, optionsDetector
try:
# Initialize npdetector with default configuration file.
nnet = Detector(MASK_RCNN_DIR, MASK_RCNN_LOG_DIR)
nnet.loadModel("latest")
rectDetector = RectDetector()
optionsDetector = OptionsDetector()
optionsDetector.load("latest")
# Initialize text detector.
textDetector = TextDetector.get_static_module("eu")()
textDetector.load("latest")
return True
except:
return False
import os
import numpy as np
import sys
import matplotlib.image as mpimg
import tensorflow as tf
from NomeroffNet import filters, RectDetector, TextDetector, OptionsDetector, Detector, textPostprocessing, \
textPostprocessingAsync
os.environ['CUDA_VISIBLE_DEVICES'] = '-1'
if __name__ == "__main__":
print('npdetector with default configuration file')
nnet = Detector()
nnet.loadModel("latest")
print('Initialize Rect Detector')
rectDetector = RectDetector()
print('Initialize Options Detector')
optionsDetector = OptionsDetector()
optionsDetector.load("latest")
print('Initialize text detector')
textDetector = TextDetector({
"eu_ua_2004_2015": {
"for_regions": ["eu_ua_2015", "eu_ua_2004"],
"model_path": "latest"
},
"eu_ua_1995": {
OPTIONS_MODEL_PATH = os.path.join(NOMEROFF_NET_DIR, "models/numberplate_options_2019_03_05.h5")
# If you use gpu version tensorflow please change model to gpu version named like *-gpu.pb
mode = "cpu" if "NN_MODE" not in os.environ else os.environ["NN_MODE"] if os.environ["NN_MODE"]=="gpu" else "cpu"
OCR_NP_UKR_TEXT = os.path.join(NOMEROFF_NET_DIR, "models/anpr_ocr_ua_12-{}.h5".format(mode))
OCR_NP_EU_TEXT = os.path.join(NOMEROFF_NET_DIR, "models/anpr_ocr_eu_2-{}.h5".format(mode))
OCR_NP_RU_TEXT = os.path.join(NOMEROFF_NET_DIR, "models/anpr_ocr_ru_3-{}.h5".format(mode))
sys.path.append(NOMEROFF_NET_DIR)
# Import license plate recognition tools.
from NomeroffNet import filters, RectDetector, TextDetector, OptionsDetector, Detector, textPostprocessing, textPostprocessing
# Initialize npdetector with default configuration file.
nnet = Detector(MASK_RCNN_DIR, MASK_RCNN_LOG_DIR)
nnet.loadModel(MASK_RCNN_MODEL_PATH)
rectDetector = RectDetector()
optionsDetector = OptionsDetector()
optionsDetector.load(OPTIONS_MODEL_PATH)
# Initialize text detector.
textDetector = TextDetector({
"eu_ua_2004_2015": {
"for_regions": ["eu_ua_2015", "eu_ua_2004"],
"model_path": OCR_NP_UKR_TEXT
},
"eu": {
"for_regions": ["eu", "eu_ua_1995"],
"model_path": OCR_NP_EU_TEXT
import sys
import glob
import matplotlib.image as mpimg
import matplotlib.pyplot as plt
# NomeroffNet path
NOMEROFF_NET_DIR = os.path.abspath('../../')
sys.path.append(NOMEROFF_NET_DIR)
# Import license plate recognition tools.
from NomeroffNet import Detector
from NomeroffNet import filters
# load model
nnet = Detector()
nnet.loadModel(NOMEROFF_NET_DIR)
# Walking through the ./examples/images/ directory and checking each of the images for license plates.
rootDir = '../images/*'
imgs = [mpimg.imread(img_path) for img_path in glob.glob(rootDir)]
cv_imgs_masks = nnet.detect_mask(imgs)
for img, cv_img_masks in zip(imgs, cv_imgs_masks):
# Generate splashs.
splashs = filters.color_splash(img, cv_img_masks)
for splash in splashs:
plt.imshow(splash)
plt.axis("off")
plt.show()
warnings.filterwarnings('ignore')
# change this property
NOMEROFF_NET_DIR = os.path.abspath('./nomeroff-net')
# specify the path to Mask_RCNN if you placed it outside Nomeroff-net project
MASK_RCNN_DIR = os.path.join(NOMEROFF_NET_DIR, 'Mask_RCNN')
MASK_RCNN_LOG_DIR = os.path.join(NOMEROFF_NET_DIR, 'logs')
sys.path.append(NOMEROFF_NET_DIR)
# Import license plate recognition tools.
from NomeroffNet import filters, RectDetector, TextDetector, OptionsDetector, Detector, textPostprocessing
nnet = Detector(MASK_RCNN_DIR, MASK_RCNN_LOG_DIR)
nnet.loadModel('latest')
rectDetector = RectDetector()
optionsDetector = OptionsDetector()
optionsDetector.load('latest')
textDetector = TextDetector.get_static_module("kz")()
textDetector.load("latest")
def read_number_plates(url):
global graph, sess
with urlopen(url) as file:
img = mpimg.imread(file, 0)
def main(ss):
trigger = datetime.time(hour=10, minute=0, second=0, microsecond=0)
# specify the path to Mask_RCNN if you placed it outside Nomeroff-net project
NOMEROFF_NET_DIR = os.path.abspath('../')
MASK_RCNN_DIR = os.path.join(NOMEROFF_NET_DIR, 'Mask_RCNN')
MASK_RCNN_LOG_DIR = os.path.join(NOMEROFF_NET_DIR, 'logs')
MASK_RCNN_MODEL_PATH = os.path.join(
NOMEROFF_NET_DIR, "models/mask_rcnn_numberplate_0700.h5")
OPTIONS_MODEL_PATH = os.path.join(
NOMEROFF_NET_DIR, "models/numberplate_options_2019_03_05.h5")
# If you use gpu version tensorflow please change model to gpu version named like *-gpu.pb
mode = "gpu"
OCR_NP_RU_TEXT = os.path.join(NOMEROFF_NET_DIR,
"models/anpr_ocr_ru_3-{}.h5".format(mode))
sys.path.append(NOMEROFF_NET_DIR)
# Import license plate recognition tools.
from NomeroffNet import filters, RectDetector, TextDetector, OptionsDetector, Detector, textPostprocessing
# Initialize npdetector with default configuration file.
nnet = Detector(MASK_RCNN_DIR, MASK_RCNN_LOG_DIR)
# Load weights in keras format.
nnet.loadModel(MASK_RCNN_MODEL_PATH)
# Initialize rect detector with default configuration file.
rectDetector = RectDetector()
# Initialize text detector.
# Also you may use gpu version models.
textDetector = TextDetector(
{"ru": {
"for_regions": ["ru"],
"model_path": OCR_NP_RU_TEXT
}})
# Initialize train detector.
optionsDetector = OptionsDetector()
optionsDetector.load(OPTIONS_MODEL_PATH)
script_dir = os.path.dirname(os.path.realpath(__file__))
main_directory = os.path.join(script_dir, "data")
if not os.path.exists(main_directory):
os.makedirs(main_directory)
car_snapshot = len(os.listdir(main_directory))
print(car_snapshot)
caffe_net_filepath = os.path.join(script_dir,
'MobileNetSSD_deploy.caffemodel')
caffe_net_filepath = os.path.abspath(caffe_net_filepath)
proto_filepath = os.path.join(script_dir,
'MobileNetSSD_deploy.prototxt.txt')
proto_filepath = os.path.abspath(proto_filepath)
car_net = cv2.dnn.readNetFromCaffe(proto_filepath, caffe_net_filepath)
CLASSES = [
"background", "aeroplane", "bicycle", "bird", "boat", "bottle", "bus",
"car", "cat", "chair", "cow", "diningtable", "dog", "horse",
"motorbike", "person", "pottedplant", "sheep", "sofa", "train",
"tvmonitor"
]
COLORS = np.random.uniform(0, 255, size=(len(CLASSES), 3))
limit_for_confidence = 0.7
number_cascade = cv2.CascadeClassifier(
'haarcascade_russian_plate_number.xml')
src0 = 'rtsp://...' # outside camera
src1 = 'rtsp://...' # inside camera
inc_capture = VideoCaptureAsync(src0).start()
out_capture = VideoCaptureAsync(src1).start()
time.sleep(3.0)
application = MyWindow(sock=ss)
application.show()
flag_cnt = False
flag_outside = False
start_save = True
clear_log_files = dt.month # clear logging every month
cnt = 0
file = open(script_dir + "/found_numbers.txt", 'a')
while True:
# clear history every month
if dt.month != clear_log_files:
clear_log_files = dt.month
with open(script_dir + "/found_numbers.txt", 'w'):
pass
for the_file in os.listdir(main_directory):
file_path = os.path.join(main_directory, the_file)
try:
if os.path.isfile(file_path):
os.unlink(file_path)
except Exception as e:
print(e)
# getting the frame from incoming camera and frame preprocessing
try:
_, frame = inc_capture.read()
except AttributeError:
print("Incoming camera isn't responding")
break
frame = resize(frame, width=720)
# Crop the ROI
height_frame, width_frame = frame.shape[:2]
frame = frame[int(0.3 * height_frame):height_frame, 0:width_frame]
# if we get the frame in grayscale format (night mode enable) - convert it to bgr
try:
frame = cv2.cvtColor(frame, cv2.COLOR_GRAY2BGR)
except cv2.error:
pass
# getting the frame from outcoming camera and frame preprocessing
try:
_, shot = out_capture.read()
except AttributeError:
print("Outcoming camera isn't responding")
break
shot = resize(shot, width=720)
# Crop the ROI
height_shot, width_shot = shot.shape[:2]
shot = shot[int(0.3 * height_shot):height_shot, 0:width_shot]
# if we get the frame in grayscale format (night mode enable) - convert it to bgr
try:
shot = cv2.cvtColor(shot, cv2.COLOR_GRAY2BGR)
except cv2.error:
pass
copy_frame = frame.copy()
copy_shot = shot.copy()
# Pass the blob through the network and obtain the detections and predictions
(h, w) = frame.shape[:2]
blob = cv2.dnn.blobFromImage(cv2.resize(frame, (300, 300)), 0.007843,
(300, 300), 127.5)
car_net.setInput(blob)
detections = car_net.forward()
(H, W) = shot.shape[:2]
blobbed = cv2.dnn.blobFromImage(cv2.resize(shot, (300, 300)), 0.007843,
(300, 300), 127.5)
car_net.setInput(blobbed)
discoveries = car_net.forward()
number_plate = ''
# analyze incoming camera frame
for ind in np.arange(0, detections.shape[2]):
# extract the confidence (i.e., probability) associated with the predictions
confidence = detections[0, 0, ind, 2]
# Filter out weak detections by ensuring the confidence is greater than
# the minimum confidence
if confidence > limit_for_confidence:
# Extract the index of the class labels from detections, then compute
# the (x, y)-coordinates of the bounding box for the object
idx = int(detections[0, 0, ind, 1])
box = detections[0, 0, ind, 3:7] * np.array([w, h, w, h])
(startX, startY, endX, endY) = box.astype("int")
# Draw the bounding boxes for everything car
if CLASSES[idx] == 'car':
label = "{}: {:.2f}%".format(CLASSES[idx],
confidence * 100)
car_frame = frame[startY:endY, startX:endX]
car_number = car_frame.copy()
# looking for a license plate on car frame
plaques = []
plaques = number_cascade.detectMultiScale(car_number,
scaleFactor=1.3,
minNeighbors=4)
# if any license plate has been found - try to recognize its
if len(plaques) > 0 and flag_cnt is False:
start = time.time()
try:
car_number = resize(car_number, width=1080)
except cv2.error:
print('cv2 error')
except ZeroDivisionError:
print('zero division')
try:
NP = nnet.detect([car_number])
# Generate image mask.
cv_img_masks = filters.cv_img_mask(NP)
# Detect points.
arrPoints = rectDetector.detect(cv_img_masks)
zones = rectDetector.get_cv_zonesBGR(
car_number, arrPoints)
# find standart
# Added a classifier (isHiddenIds) for determining the fact of hide text of number,
# in order not to recognize a deliberately damaged license plate image.
regionIds, isHiddenIds = optionsDetector.predict(
zones)
regionNames = optionsDetector.getRegionLabels(
regionIds)
# find text with postprocessing by standart
textArr = textDetector.predict(zones, regionNames)
textArr = textPostprocessing(textArr, regionNames)
number_plate = ''.join(textArr)
for (xx, yy, ww, hh) in plaques:
cv2.rectangle(
copy_frame, (startX + xx, startY + yy),
(startX + xx + ww, startY + yy + hh),
(0, 0, 255), 2)
# if license plate has been recognized - compare the number with white numbers
if len(number_plate) > 0:
print(number_plate)
# if not flag_cnt:
flag, show = check_whitelist(number_plate)
if not flag:
sending_flag = send_message(b'2', ss)
if show is not None:
application.ui.carNumber.setText(
'Номер:' + number_plate)
application.ui.carNumber.adjustSize()
if sending_flag:
application.ui.barrierStatus.setText(
'Шлагбаум: открыт')
application.ui.barrierStatus.adjustSize(
)
flag_cnt = True
else:
application.ui.barrierStatus.setText(
'Команда \nне принята')
application.ui.barrierStatus.adjustSize(
)
except ZeroDivisionError:
print('zero')
except RecursionError:
print('Recursion Error')
finish = time.time()
print(finish - start,
'at:',
datetime.datetime.now(),
sep=' ')
y = startY - 15 if startY - 15 > 15 else startY + 15
cv2.putText(copy_frame, label, (startX, y),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, COLORS[idx], 2)
# analyze outcoming camera frame
for index in np.arange(0, discoveries.shape[2]):
# extract the confidence (i.e., probability) associated with the predictions
confidence = discoveries[0, 0, index, 2]
# Filter out weak detections by ensuring the confidence is greater than
# the minimum confidence
if confidence > limit_for_confidence:
# Extract the index of the class labels from detections, then compute
# the (x, y)-coordinates of the bounding box for the object
idx = int(discoveries[0, 0, index, 1])
box = discoveries[0, 0, index, 3:7] * np.array([W, H, W, H])
(startX, startY, endX, endY) = box.astype("int")
# Draw the bounding boxes for everything car
if CLASSES[idx] == 'car':
# heightShot, widthShot = shot.shape[:2]
label = "{}: {:.2f}%".format(CLASSES[idx],
confidence * 100)
car_shot = shot[startY:endY, startX:endX]
car_area = car_shot.copy()
height_car, width_car = car_area.shape[:2]
# cv2.rectangle(frame, (startX, startY), (endX, endY),
# COLORS[idx], 2)
y = startY - 15 if startY - 15 > 15 else startY + 15
cv2.putText(copy_shot, label, (startX, y),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, COLORS[idx], 2)
if not flag_cnt:
if height_car > 0.3 * H or width_car > 0.3 * W:
if not flag_outside:
sending_flag = send_message(b'2', ss)
if sending_flag:
application.ui.outCar.setText(
"Авто на выезде: да")
application.ui.outCar.adjustSize()
application.ui.barrierStatus.setText(
'Шлагбаум: открыт')
application.ui.barrierStatus.adjustSize()
flag_outside = True
else:
application.ui.barrierStatus.setText(
'Команда \nне принята')
application.ui.barrierStatus.adjustSize()
flag_cnt = True
if start_save:
if len(number_plate) > 0:
number_with_time = str(number_plate) + " at: " + str(dt.now())
file.write("%s\n" % str(number_with_time))
car_snapshot += 1
cv2.imwrite(main_directory + "/" + str(car_snapshot) + ".jpg",
frame)
# time.sleep(5.0)
print("The frame has been saved")
cv2.imshow('Incoming', copy_frame)
cv2.imshow('Outcoming', copy_shot)
time.sleep(0.2)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
if close_flag:
break
# while flag_cnt is True:
if flag_cnt:
cnt += 1
# time.sleep(1.0)
if cnt > 100:
sending_flag = send_message(b'1', ss)
if sending_flag:
application.ui.barrierStatus.setText('Шлагбаум: закрыт')
application.ui.outCar.setText("Авто на выезде: нет")
application.ui.outCar.adjustSize()
application.ui.barrierStatus.adjustSize()
flag_outside = False
flag_cnt = False
else:
application.ui.barrierStatus.setText(
'Команда \nне принята')
application.ui.barrierStatus.adjustSize()
cnt = 0
# release all cameras
inc_capture.__exit__()
out_capture.__exit__()
file.close()
cv2.destroyAllWindows()