def _register(self, frame, gray, username):
"""Detects faces in frames and uses them to train a face recognition algorithm.
Face data is associated with the given username.
:param frame: an 8 bit, 3-channel UMat object
:param gray: an 8-bit, 1-channel ndarray
:param username: name of registering user
"""
if self.samples < self.sample_size:
# Find all faces and store only the location of the largest (closest to the camera)
faces = self._findFaces(frame)
max_area = 0
x = 0
y = 0
w = 0
h = 0
for (startX, startY, endX, endY) in faces:
c_w = endX - startX
c_h = endY - startY
if c_w * c_h > max_area:
x = startX
y = startY
w = c_w
h = c_h
maxArea = w * h
# Resize and add face image to list for training
if faces:
self.samples += 1
gray = cv2.UMat(gray, [y, y + h], [x, x + w])
gray = cv2.resize(gray, (100, 100))
self.sample_images.append(gray)
else:
# Finished collecting face data
# Associate registering user id with training data
db = DBHelper()
user_id = db.getIDByUsername(username)
id_array = [user_id] * self.sample_size
for i in range(self.sample_size):
self.sample_images[i] = cv2.UMat.get(self.sample_images[i])
# Update or create new face recognizer
if Path('./training_data/recognizer.yml').is_file():
self.recognizer.update(self.sample_images, np.array(id_array))
else:
self.recognizer.train(self.sample_images, np.array(id_array))
self.recognizer.write('./training_data/recognizer.yml')
# registration complete
self.reg_complete = True
self.rec_trained = True
# reset variables before detection begins
self._reset()
return frame
