def detect_eyes(self, gray_image):
detected_eyes = self.eye_cascade.detectMultiScale(gray_image)
eyes = []
for (x, y, w, h) in detected_eyes:
eye = EyeObject()
eye.set_from_points(dlib.point(x, y), dlib.point(x + w, y + h))
eyes.append(eye)
return eyes
def test_dlib_shape_to_np_array_results(self):
"""Converts a few example dlib shapes (i.e. dlib.full_object_detection objects) to numpy arrays, and checks if
the result is as expected"""
dummy_shapes = [
full_object_detection(rectangle(), [point(0, 0)]),
full_object_detection(
rectangle(1, 2, 3, 4),
[point(1, 2),
point(2, 3),
point(3, 4),
point(4, 1)]),
full_object_detection(rectangle(50, 100, 150, 200), [
point(50, 100),
point(100, 150),
point(150, 200),
point(200, 50)
])
]
exptected_results = [[[0, 0]], [[1, 2], [2, 3], [3, 4], [4, 1]],
[[50, 100], [100, 150], [150, 200], [200, 50]]]
for shape, result in zip(dummy_shapes, exptected_results):
self.assertEqual(
np.array_equal(utils.dlib_shape_to_np_array(shape), result),
True)
def get_farthest_vertex_from_point(point, dimensions):
ret_point = dlib.point(0, 0)
dist = get_euclidian_distance(point, ret_point)
vertices = [
dlib.point(0, dimensions.y),
dlib.point(dimensions.x, 0),
dlib.point(dimensions.x, dimensions.y)
]
for vertex in vertices:
new_dist = get_euclidian_distance(point, vertex)
if new_dist > dist:
dist = new_dist
ret_point = vertex
return ret_point
def encode_filter(filter_files):
images = []
faces = []
FACE_ALIGNMENT = FaceAlignment(LandmarksType._2D,
enable_cuda=True,
flip_input=False)
for i, filter_file in enumerate(filter_files):
images.append(skimage.io.imread(str(filter_file)))
faces.append(FACE_ALIGNMENT.get_landmarks(images[i]))
FACE_ALIGNMENT = None
face_recognition_model = face_recognition_models.face_recognition_model_location(
)
face_encoder = dlib.face_recognition_model_v1(face_recognition_model)
for i, face in enumerate(faces):
if face is None:
print('Warning: {} has no face.'.format(filter_files[i]))
continue
if len(face) > 1:
print('Warning: {} has more than one face.'.format(
filter_files[i]))
parts = []
for p in face[0]:
parts.append(dlib.point(p[0], p[1]))
raw_landmark_set = dlib.full_object_detection(rect, parts)
yield numpy.array(
face_encoder.compute_face_descriptor(images[i],
raw_landmark_set, 1))
def __init__(self, box, frame):
"""
Initializes Kalman filter
"""
# Person ID
self.id = 0
self.height = frame.shape[0]
self.width = frame.shape[1]
print(self.id, box)
# Coordinates of a bounding box
self.box = box
if box[0] < 0 or box[1] < 0 or box[2] < 0 or box[3] < 0:
print('---------------------')
print('Unexpected box', box)
print('---------------------')
# Number of detection matches
self.hits = 1
# Number of detection mismatches
self.misses = 0
self.tracker = dlib.correlation_tracker()
cent_x = int((box[0] + box[2]) / 2)
cent_y = int((box[1] + box[3]) / 2)
w = box[3] - box[1]
h = box[2] - box[0]
rect = dlib.centered_rect(dlib.point(cent_x, cent_y), h, w)
self.tracker.start_track(frame, rect)
def geometry_msgs_points_to_face_landmarks(geometry_msgs_points):
points = dlib.points()
[
points.append(dlib.point(int(p.x), int(p.y)))
for p in geometry_msgs_points
]
return points
def loadLandmarkFile(fileName):
inputFileTxt = open(fileName, "r")
lines = inputFileTxt.readlines()
lastFrameId = -1
landmarksList = []
for i in range(len(lines)):
line = lines[i].rstrip().split('\t')
rectStr = line[2].split(':')
rect = dlib.rectangle(left=int(rectStr[0]),
top=int(rectStr[1]),
right=int(rectStr[2]),
bottom=int(rectStr[3]))
parts = []
for j in range(len(line) - 5):
pointStr = line[j + 3].split(':')
parts.append(dlib.point(x=int(pointStr[0]), y=int(pointStr[1])))
shape = dlib.full_object_detection(rect, parts)
frameId = int(line[0])
if lastFrameId != frameId:
shapeList = []
landmarksList.append(shapeList)
shapeList.append(shape)
lastFrameId = frameId
print("Loaded landmarks for {} faces in {} frames.".format(
len(lines), len(landmarksList)))
return landmarksList
def parse_photos(self, graph, photos):
for photo in photos:
if 'tags' not in photo:
continue
photo['images'].sort(key=lambda x: x['height']*x['width'])
image_url = photo['images'][-1]
width, height = image_url['width'], image_url['height']
people = [
(
point(int(t['x']*width/100.0),
int(t['y']*height/100.0)),
t
)
for t in photo['tags']['data']
if t.get('x') and t.get('y')
]
faces = yield find_faces_url(image_url['source'], hash_face=True)
# go through the faces _we_ found and interpolate those results
# with the tags from the image
for face in faces:
face['tags'] = []
for p, tag in people:
if face['rect'].contains(p):
face['tags'].append(tag)
photo['faces'] = faces
return photos
def test_point_assignment():
p = point(27, 42)
p.x = 16
assert p.x == 16
assert p.y == 42
p.y = 31
assert p.x == 16
assert p.y == 31
def encode_one_face_local(image, faces):
for points in faces:
parts = []
for p in points:
parts.append(dlib.point(p[0], p[1]))
raw_landmark_set = dlib.full_object_detection(rect, parts)
yield numpy.array(
face_encoder.compute_face_descriptor(image, raw_landmark_set,
1))
def test_point():
p = point(27, 42)
assert repr(p) == "point(27, 42)"
assert str(p) == "(27, 42)"
assert p.x == 27
assert p.y == 42
ser = pickle.dumps(p, 2)
deser = pickle.loads(ser)
assert deser.x == p.x
assert deser.y == p.y
def perform_generic_hough_transform(self, img, record_hit):
assert(img.shape[0] == self.size)
assert(img.shape[1] == self.size)
cent = center(get_rect(img))
even_size = self.size - (self.size%2)
sqrt_2 = sqrt(2)
for r in range(img.shape[0]):
for c in range(img.shape[1]):
val = img[r][c]
if (val != 0):
x = c - cent.x
y = r - cent.y
# Now draw the curve in Hough space for this image point
for t in range(self.size):
theta = t*pi/even_size
radius = (x*cos(theta) + y*sin(theta))/sqrt_2 + even_size/2 + 0.5
rr = int(radius)
record_hit(point(t,rr), point(c,r), val)
def test_points():
ps = points()
ps.resize(5)
assert len(ps) == 5
for i in range(5):
assert ps[i].x == 0
assert ps[i].y == 0
ps.clear()
assert len(ps) == 0
ps.extend([point(1, 2), point(3, 4)])
assert len(ps) == 2
ser = pickle.dumps(ps, 2)
deser = pickle.loads(ser)
assert deser[0].x == 1
assert deser[0].y == 2
assert deser[1].x == 3
assert deser[1].y == 4
def substitute_points(dlib_points, mtcnn_points):
#face left, face right, nose, mouth left, mouth right
SUBSTITUTE = [0, 16, 30, 48, 54]
new_points = dlib.points()
len_mtcnn = len(mtcnn_points) // 2
for i in range(len_mtcnn):
dlib_points[SUBSTITUTE[i]] = dlib.point(
round(float(mtcnn_points[i])),
round(float(mtcnn_points[i + len_mtcnn])))
return dlib_points
def face_encodings(face_image,
landmarks,
known_face_locations=None,
num_jitters=1):
"""
Given an image, return the 128-dimension face encoding for each face in the image.
:param face_image: The image that contains one or more faces
:param known_face_locations: Optional - the bounding boxes of each face if you already know them.
:param num_jitters: How many times to re-sample the face when calculating encoding. Higher is more accurate, but slower (i.e. 100 is 100x slower)
:return: A list of 128-dimensional face encodings (one for each face in the image)
"""
if landmarks is None:
raw_landmarks = _raw_face_landmarks(face_image,
known_face_locations,
model="small")
else:
raw_landmarks = [
dlib.full_object_detection(parts=[
dlib.point(int(landmarks[1][0]), int(landmarks[1][1])),
dlib.point(int(landmarks[4][0]), int(landmarks[4][1])),
dlib.point(int(landmarks[0][0]), int(landmarks[0][1])),
dlib.point(int(landmarks[3][0]), int(landmarks[3][1])),
dlib.point(int(landmarks[2][0]), int(landmarks[2][1]))
],
rect=dlib.rectangle(
known_face_locations[0][3],
known_face_locations[0][0],
known_face_locations[0][1],
known_face_locations[0][2]))
]
return [
np.array(
face_encoder.compute_face_descriptor(face_image, raw_landmark_set,
num_jitters))
for raw_landmark_set in raw_landmarks
]
def usesmallbox(t, b, l, r, scale, scale_top):
h, w, _ = scaled_image.shape
fh = b - t
fw = r - l
t = max(0, t - int(scale_top*fh))
b = min(h, b + int(scale*fh))
l = max(0, l - int(scale*fw))
r = min(w, r + int(scale*fw))
facebb = self.fd(scaled_image[t:b, l:r, :])
if len(facebb) >= 1:
if len(facebb) > 1:
print("multiple faces detected, choosing one arbitrarily")
rect = facebb[0]
return dlib.translate_rect(rect, dlib.point(l, t))
else:
return None
def get_face_boxes(rgbImg, prev_bb, bbs_out):
"""A wrapper function for align.getAllFaceBoundingBoxes """
res = align.getAllFaceBoundingBoxes(rgbImg)
if not res:
return
local_left = res[0].left()
local_right = res[0].right()
local_top = res[0].top()
local_bottom = res[0].bottom()
padding = SERVER_FACE_SEARCH_PADDING
anchor_point = dlib.point(
y=max(prev_bb.top() - int(prev_bb.height() * padding), 0),
x=max(prev_bb.left() - int(prev_bb.width() * padding), 0))
if len(res) > 0:
bbs_out.append(
dlib.rectangle(left=local_left + anchor_point.x,
right=local_right + anchor_point.x,
top=local_top + anchor_point.y,
bottom=local_bottom + anchor_point.y))
return
def parse_photos(self, graph, photos):
for photo in photos:
if "tags" not in photo:
continue
photo["images"].sort(key=lambda x: x["height"] * x["width"])
image_url = photo["images"][-1]
width, height = image_url["width"], image_url["height"]
people = [
(point(int(t["x"] * width / 100.0), int(t["y"] * height / 100.0)), t)
for t in photo["tags"]["data"]
if t.get("x") and t.get("y")
]
faces = yield find_faces_url(image_url["source"], hash_face=True)
# go through the faces _we_ found and interpolate those results
# with the tags from the image
for face in faces:
face["tags"] = []
for p, tag in people:
if face["rect"].contains(p):
face["tags"].append(tag)
photo["faces"] = faces
return photos
def FakeHOGModel(_, face):
return full_object_detection(face, [point(i, i) for i in range(20)])
# crop the face from the image
faceOfs = (tl[0], tl[1])
faceImage = gray[tl[1]:br[1], tl[0]:br[0]]
if faceImage is not None and faceImage.shape[
0] > 0 and faceImage.shape[1] > 0:
# apply detector on gray face image (on the whole image)
rect = dlib.rectangle(0, 0, faceImage.shape[1], faceImage.shape[0])
landmarks = landmarkDetector(faceImage, rect)
if len(landmarks.parts()) == 68:
for p in landmarks.parts():
cv2.circle(frame, (faceOfs[0] + p.x, faceOfs[1] + p.y), 2,
(0, 0, 255), -1)
p = landmarks.parts()[36] # left eye
q = landmarks.parts()[45] # right eye
v = dlib.point(p.x - q.x, p.y - q.y)
u = dlib.point(v.y, -v.x)
angle = np.arctan2(u.y, u.x)
angle = 180 * angle / np.pi
angle -= 90
angle = -angle
print('angle', angle)
# set iCub's body with the body model seen
head.set((0, angle, 0, 0, 0, 0))
cv2.imshow('camera', frame)
if cv2.waitKey(1) == 27:
break
cv2.destroyAllWindows()
if (len(sys.argv) == 4):
with tf.Graph().as_default():
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.75)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options,
log_device_placement=False))
with sess.as_default():
pnet, rnet, onet = detect_face.create_mtcnn(sess, None)
capture = cv2.VideoCapture(sys.argv[1])
width = int(capture.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(capture.get(cv2.CAP_PROP_FRAME_HEIGHT))
fps = int(capture.get(cv2.CAP_PROP_FPS))
dlib_frame_dim = dlib.point(width, height)
fourcc = cv2.VideoWriter_fourcc(*'MJPG')
if not os.path.exists(sys.argv[2][0:sys.argv[2].rfind('/')]):
os.makedirs(sys.argv[2][0:sys.argv[2].rfind('/')])
out = cv2.VideoWriter(sys.argv[2], fourcc, fps, (width, height))
frame_count = 0
debug_file = open("empty.txt", 'w')
if DEBUG:
debug_file = open(
os.path.join(sys.argv[2][0:sys.argv[2].rfind('/')], "debug.txt"),
'a')
def get_mean_point(bounding_box):
x = int(round((bounding_box[0] + bounding_box[2]) / 2))
y = int(round((bounding_box[1] + bounding_box[3]) / 2))
return dlib.point(x, y)
print(str(path_to_image))
print(detected_faces)
left, top, right, bottom = get_points(detected_faces, 0)
face_box = dlib.rectangle(left, top, right, bottom)
#d = dlib.rectangle(1,2,3,4)
#print(str(path_to_image) + "----" + str(detected_faces[0]) + "----" + str(d))
# Get pose/landmarks of those faces
# Will be used as an input to the function that computes face encodings
# This allows the neural network to be able to produce similar numbers
#for faces of the same people, regardless of camera angle and/or face positioning in the image
shapes_faces = [shape_predictor(image, face_box)]
face_traits = dlib.points()
len_points = len(points) // 2
for i in range(len_points):
face_traits.append(dlib.point( round(float(points[i])) ,
round(float(points[i + len_points]))))
shapes_points = [dlib.full_object_detection(face_box, face_traits)]
#if ( str(path_to_image) == "pics/rakoruja.jpg"):
#print("\n\n--------------------------------------------")
#img = cv2.imread(path_to_image)
draw_facial_points(image, shapes_faces[0].parts(), 3)
cv2.imwrite(os.path.join("./", "OUTPUT-COLOURFUL" +
".jpg"), image)
cv2.waitKey(30)
for i in range(len(shapes_faces[0].parts())):
print(str(i) + " - " + str(shapes_faces[0].parts()[i]))
for f in glob.glob(os.path.join(faces_folder_path, "*.jpg")):
print("Processing file: {}".format(f))
img = dlib.load_rgb_image(f)
win.clear_overlay()
win.set_image(img)
dets = detector(img, 1)
print("Number of faces detected: {}".format(len(dets)))
for k, d in enumerate(dets):
print("Detection {}: Left: {} Top: {} Right: {} Bottom: {}".format(
k, d.left(), d.top(), d.right(), d.bottom()))
shape = predictor(img, d)
print("Part 0: {}, Part 1: {} ...".format(shape.part(0),
shape.part(1)))
for j in range(68):
x, y = shape.part(j).x, shape.part(j).y
win.add_overlay_circle(dlib.point(x, y), 1,
dlib.rgb_pixel(0, 0, 255))
win.add_overlay(shape)
win.add_overlay(dets)
dlib.hit_enter_to_continue()
def array_to_landmarks(arr, rect):
points = [dlib.point(x, y) for (x, y) in arr]
lm = dlib.full_object_detection(rect, points)
return lm
def test_point_init_kwargs():
p = point(y=27, x=42)
assert repr(p) == "point(42, 27)"
assert str(p) == "(42, 27)"
assert p.x == 42
assert p.y == 27
def test12(image, points):
size = image.shape
#2D image points. If you change the image, you need to change vector
image_points = np.array([
(points[0].x, points[0].y), # Nose tip
(points[1].x, points[1].y), # Chin
(points[2].x, points[2].y), # Left eye left corner
(points[3].x, points[3].y), # Right eye right corne
(points[4].x, points[4].y), # Left Mouth corner
(points[5].x, points[5].y) # Right mouth corner
], dtype="double")
# 3D model points.
model_points = np.array([
(0.0, 0.0, 0.0), # Nose tip
(0.0, -330.0, -65.0), # Chin
(-225.0, 170.0, -135.0), # Left eye left corner
(225.0, 170.0, -135.0), # Right eye right corne
(-150.0, -150.0, -125.0), # Left Mouth corner
(150.0, -150.0, -125.0) # Right mouth corner
])
# Camera internals
focal_length = size[1]
center = (size[1]/2, size[0]/2)
camera_matrix = np.array(
[[focal_length, 0, center[0]],
[0, focal_length, center[1]],
[0, 0, 1]], dtype = "double"
)
#print("Camera Matrix :\n {0}".format(camera_matrix))
dist_coeffs = np.zeros((4,1)) # Assuming no lens distortion
(success, rotation_vector, translation_vector) = cv2.solvePnP(model_points, image_points, camera_matrix, dist_coeffs, flags=cv2.SOLVEPNP_ITERATIVE)
#print("Rotation Vector:\n {0}".format(rotation_vector))
#print("Translation Vector:\n {0}".format(translation_vector))
rvec_matrix = cv2.Rodrigues(rotation_vector)[0]
proj_matrix = np.hstack((rvec_matrix, translation_vector))
eulerAngles = -cv2.decomposeProjectionMatrix(proj_matrix)[6]
pitch, yaw, roll = [math.radians(_) for _ in eulerAngles]
pitch = math.degrees(math.asin(math.sin(pitch)))
roll = -math.degrees(math.asin(math.sin(roll)))
yaw = math.degrees(math.asin(math.sin(yaw)))
# Project a 3D point (0, 0, 1000.0) onto the image plane.
# We use this to draw a line sticking out of the nose
(nose_end_point2D, jacobian) = cv2.projectPoints(np.array([(0.0, 0.0, 1000.0)]), rotation_vector, translation_vector, camera_matrix, dist_coeffs)
for p in image_points:
cv2.circle(image, (int(p[0]), int(p[1])), 3, (0,0,255), -1)
p1 = dlib.point(( int(image_points[0][0]), int(image_points[0][1])))
p2 = dlib.point(( int(nose_end_point2D[0][0][0]), int(nose_end_point2D[0][0][1])))
line = dlib.line(p1, p2)
win.add_overlay(line)
pitch_compliance = 0 if (abs(pitch) > 5) else (-20*(abs(pitch))+100)
roll_compliance = 0 if (abs(roll) > 8) else (-(100/8)*(abs(roll))+100)
yaw_compliance = 0 if (abs(yaw) > 5) else ((-20*abs(int(yaw)))+100)
return int((pitch_compliance+roll_compliance+yaw_compliance)/3)
def topoint(dp):
return dlib.point(dp.x, dp.y)
def pointcloud_to_dlib_parts(pcloud):
return [dlib.point(int(p[1]), int(p[0])) for p in pcloud.points]
def pointcloud_to_dlib_parts(pcloud):
return [dlib.point(int(p[1]), int(p[0])) for p in pcloud.points]
def _create_full_object_detection(css, landmark_points):
points = [dlib.point(*point) for point in landmark_points]
return dlib.full_object_detection(_css_to_rect(css), points)
def cv_point_2_dlib(self, point):
"""Helper function to convert an OpenCV point to Dlib format (dlib.point)."""
return dlib.point(point[0], point[1])
