def rect_min_distance(self, box1, box2):
"""
Calculates the minimum distance between two given contours
Args:
box1:
box2:
Returns:
Returns minimum distance between box1 and box2
"""
x1, y1, x1b, y1b = box1
x2, y2, x2b, y2b = box2
left = x2b < x1
right = x1b < x2
bottom = y2b < y1
top = y1b < y2
if top and left:
return dist((x1, y1b), (x2b, y2))
elif left and bottom:
return dist((x1, y1), (x2b, y2b))
elif bottom and right:
return dist((x1b, y1), (x2, y2b))
elif right and top:
return dist((x1b, y1b), (x2, y2))
elif left:
return x1 - x2b
elif right:
return x2 - x1b
elif bottom:
x = y1 - y2b
return y1 - y2b
elif top:
return y2 - y1b
else: # rectangles intersect
return 0.
def kNN(gallery_set, probe_set, meanface, eigenfaces, gallery_labels,
probe_labels, numPCA):
'''
im_in : np.array((M,M))
input image
mean_faces : np.array
average face from training data
eigen_faces : np.array
eigenfaces from training data
numPCA : int
'''
Omega_in = data_projection(probe_set, numPCA, eigenfaces, meanface)
Omega_tr = data_projection(gallery_set, numPCA, eigenfaces, meanface)
# try to match a gallery image
correct = 0
start = timer()
for i in range(0, len(probe_labels)):
err_k = 100000
match = 99999
for j in range(0, len(gallery_labels)):
if dist(Omega_in[i], Omega_tr[j]) < err_k:
err_k = dist(Omega_in[i], Omega_tr[j])
# print dist(Omega_in[i],Omega_tr[j])
# print gallery_labels[j],probe_labels[i]
match = j
if gallery_labels[match] == probe_labels[i]:
correct += 1
end = timer()
print "Prinicipal Components = {}; ttl time = {:.5f}s; Percent correct = {:.2f}%".format(
numPCA, end - start,
float(correct) / len(probe_labels) * 100)
return float(correct) / len(probe_labels) * 100, end - start
def calculate_enthusiasm(l_eye,r_eye,l_eyebrow,r_eyebrow):
if len(l_eye)==6 and len(r_eye)==6 and len(l_eyebrow)==5 and len(r_eyebrow)==5:
r_enthu_dist = dist(r_eye[2],r_eyebrow[3])
l_enthu_dist = dist(l_eye[2],l_eyebrow[3])
enthusiasm_dist = (r_enthu_dist+l_enthu_dist)/2.0
return enthusiasm_dist
else:
print("Error in eye or eyebrow shape")
return -1
def closest(self, item):
best_dist = dist(self.X_train[0], item)
best_index = 0
for i in range(1, len(self.X_train)):
distance = dist(self.X_train[i], item)
if distance < best_dist:
best_dist = distance
best_index = i
return self.Y_train[best_index]
def calculate_EAR(eye):
if len(eye)==6:
width = dist(eye[0],eye[3])
A = dist(eye[1],eye[5])
B = dist(eye[2],eye[4])
EAR = (A+B)/width
return EAR
else:
print("Error in eye shape")
return -1
def _shuffle(self, items, ax):
items_count = len(items)
ra = items[0].a
rb = items[0].b
n_max = 20
motions_dist = 0
for _ in range(n_max):
for ip in range(items_count):
item = items[ip]
if not item.is_moved_enought():
way_to_update = random.randint(1, 3)
if way_to_update == 1:
new_p = item.try_to_move(ra, 0, 0, 0, ax, 0, ax)
if way_to_update == 2:
new_p = item.try_to_move(0, rb, 0, 0, ax, 0, ax)
if way_to_update == 3:
new_p = item.try_to_move(ra, rb, 0, 0, ax, 0, ax)
if self._is_valid_position(items, new_p):
d = dist([new_p.x, new_p.y], [item.x, item.y])
new_p.add_walked_dist(d)
motions_dist += d
items[ip] = new_p
if motions_dist > items_count * ra:
break
return items
def circle(self, location):
sqxs = range(max(0, location[0] - self.irad),
min(256, location[0] + self.irad + 1))
sqys = range(max(0, location[1] - self.irad),
min(256, location[1] + self.irad + 1))
square = [(x, y) for x in sqxs for y in sqys]
return [p for p in square if dist(p, location) <= self.rad]
def calculate_frown(l_eyebrow,r_eyebrow):
if len(l_eyebrow)==5 and len(r_eyebrow)==5:
frown_width = dist(l_eyebrow[0],r_eyebrow[4])
return frown_width
else:
print("Error in eyebrow shape")
return -1
def train_star_space(self):
train_positive_input_batches, train_positive_batch_targets, train_negative_batch_targets, train_dummy_outputs,\
test_positive_input_batches, test_positive_batch_targets, test_negative_batch_targets, test_dummy_outputs = self.prepare_features_targets()
self.test_positive_input_batches = test_positive_input_batches
self.test_positive_batch_targets = test_positive_batch_targets
self.test_negative_batch_targets = test_negative_batch_targets
self.build_model_architecture()
print(self.model.summary())
self.model.fit([train_positive_input_batches, train_positive_batch_targets, train_negative_batch_targets], train_dummy_outputs, epochs = 10,\
validation_data = ([test_positive_input_batches, test_positive_batch_targets, test_negative_batch_targets], test_dummy_outputs))
self.target_encodings = {
target_id:
self.target_encoder_model.predict(np.array([target_id]))[0, 0, :]
for target_id in range(10)
}
d = {}
for i in range(10):
for j in range(10):
if i != j and f'{j}_{i}' not in d:
d[f'{i}_{j}'] = 1 - dist(self.target_encodings[i],
self.target_encodings[j])
print({k: v for k, v in sorted(d.items(), key=lambda item: item[1])})
def calculate_mouth(mouth):
if len(mouth)==20:
mouth_dist = dist(mouth[14],mouth[18])
return mouth_dist
else:
print("Error in mouth shape")
return -1
def distance_matrix(self, gdx=None, metric='jaccard'):
from itertools import combinations
binary = ['jaccard', 'hamming']
if metric in binary:
if gdx:
M = self.E[:, gdx]
else:
M = self.E[:, :]
else:
if gdx:
M = self.M[:, gdx]
else:
M = self.M[:, :]
k = self.M.shape[0]
D = np.ones((k, k))
np.fill_diagonal(D, 1)
comb = combinations(range(k), 2)
if metric == 'jaccard':
dist = jaccard
elif metric == 'hamming':
from scipy.spatial.distance import hamming as dist
elif metric == 'pearsonr':
from scipy.stats import pearsonr
dist = correlation_dist
for (i, j) in comb:
r = dist(M[i, :], M[j, :])
D[i, j] = r
D[j, i] = r
return D
def minimum_distance(v, w, p, tolerance=1):
# Return minimum distance between line segment (v,w) and point p
l2 = dist(v, w) # i.e. |w-v|^2 - avoid a sqrt
if l2 == 0.0:
return dist(p, v)
# Consider the line extending the segment, parameterized as v + t (w - v).
# We find projection of point p onto the line.
# It falls where t = [(p-v) . (w-v)] / |w-v|^2
t = np.dot((p - v) / l2, (w - v) / l2)
if t < 0.0:
return dist(p, v) + tolerance # // Beyond the 'v' end of the segment
elif t > 1.0:
return dist(p, w) + tolerance # // Beyond the 'w' end of the segment
projection = v + t * (w - v) # // Projection falls on the segment
return dist(p, projection)
def onclick(event):
print('button=%d, x=%d, y=%d, xdata=%f, ydata=%f' %
(event.button, event.x, event.y, event.xdata, event.ydata))
node, d = min([(p, dist(p[:2], (event.xdata, event.ydata)))
for p in ts.g.nodes_iter()],
key=lambda x: x[1])
print node, d
assert ts.g.has_node(node)
with open(policy_filename, 'a') as fout:
print >> fout, ' {},'.format(node)
def face_detect(img, U, Omega_training, T_0, T_1):
'''
Returns the predicted subject if recognised
else returns 0 for a non-face, -1 for unknown face
'''
I = img.flatten()
I = I - m
Omega_I = np.dot(U.T, I)
I_R = np.dot(U, Omega_I)
d_0 = dist(I_R, I)
print('d_o is %d' % d_0)
dist_array = [dist(Omega_I, Omega) for Omega in Omega_training]
res = min(dist_array)
print('d_1 is %d' % res)
index = Img_dict[dist_array.index(res)]
if d_0 > T_0:
return 0
else:
if res > T_1:
return -1
return index
def onclick(event):
print('button=%d, x=%d, y=%d, xdata=%f, ydata=%f' %
(event.button, event.x, event.y, event.xdata, event.ydata))
node, d = min([(p, dist(p[:2], (event.xdata, event.ydata)))
for p in ts.g.nodes_iter()],
key=lambda x: x[1])
print node, d
assert ts.g.has_node(node)
with open(policy_filename, 'a') as fout:
print>>fout, ' {},'.format(node)
def astar(start, goal, grid):
openset = set()
closedset = set()
current = grid[start]
openset.add(current)
#While the open set is not empty
while openset:
current = min(openset, key=lambda o: o.G + o.H
) # current is item with lowest F score in openset
#If it is the item we want, retrace the path and return it
if current.point == goal:
path = []
while current.parent:
path.append(current)
current = current.parent
path.append(current)
return path[::-1]
openset.remove(current)
closedset.add(current)
for node in children(current.point, grid):
if node in closedset:
continue
if node in openset:
new_g = current.G + dist(current.point, node.point)
if node.G > new_g:
node.G = new_g
node.parent = current
else:
node.G = current.G + dist(current.point, node.point)
node.H = dist(node.point, goal)
node.parent = current
openset.add(node)
#Throw an exception if there is no path
raise ValueError('No Path Found')
def shuffle_extended_circles(self):
if self.verbose: print(f"circles shuffeling start.")
ranges = self.ranges
size = self.circle_radius
extended_circles_count = len(self.extended_circles)
shuffled_circles = deepcopy(self.extended_circles)
shuffle_step_count = self.shuffles_count
print_step = int(shuffle_step_count // 10)
for shuffle_step in range(shuffle_step_count):
if shuffle_step % print_step == 0:
if self.verbose:
print(
f"Percents done: {int(shuffle_step // print_step) * 10}%."
)
if self.verbose: print(f"Shuffles done: {shuffle_step}")
for i in range(extended_circles_count):
current_circle = deepcopy(shuffled_circles[i])
index = current_circle.pop('index', None)
for k, v in current_circle.items():
di = (random.random() - 1 / 2) * size
new_k = v + di
if new_k - size < 0:
current_circle[k] = size
elif new_k + size > ranges[k]:
current_circle[k] = ranges[k] - size
else:
current_circle[k] = new_k
current_circle['index'] = index
is_intersect = False
for j in [*range(i), *range(i + 1, extended_circles_count)]:
is_intersect = dist(
itemgetter('x', 'y')(current_circle),
itemgetter('x', 'y')(shuffled_circles[j])) <= 2 * size
if is_intersect:
break
if not is_intersect:
shuffled_circles[i] = current_circle
self.shuffled_extended_circles = shuffled_circles
if self.verbose: print(f"circles shuffeling end.\n")
return shuffled_circles
def update_routes_decmcts(netlogo,
cars,
GRID_SIZE,
network,
comm_radius,
initial=True):
#First Loop Updates Paths for the individual car
#only update this the first time to save on overhead
if initial:
for car in cars:
if car.stopped: # and car.direction == 'east':
route = UCTPlayGame(car, GRID_SIZE)
car.push_route_netlogo(netlogo, route, mode='both')
else:
#Second Loop Communicates route to neighbors and updates its own route
comm_rad = comm_radius #communications radius for cars ~1.5 road lengths
dist_array = np.zeros((len(cars), len(cars)))
for i in range(len(cars)):
for j in range(len(cars)):
if i != j:
if dist(cars[i].location, cars[j].location) <= comm_rad:
dist_array[i][j] = 1
neighbor_cars = []
for i, car in enumerate(cars):
if car.stopped: # or congested:# and
#if car is near the end, no need to communicate
if len(car.remaining_route) <= 1:
route = UCTPlayGame(car, GRID_SIZE)
car.push_route_netlogo(netlogo, route, mode='both')
continue
#cars only communicate if the sense congestions
congested = look_ahead(
car, network, GRID_SIZE
) #checks to see if the upcoming routes are conjested
if congested:
dist_cars = dist_array[i]
#used to find neighbors to communicate with, could also include degredation
neighbor_cars = [
cars[j] for j, distance in enumerate(dist_cars)
if distance != 0
]
if len(neighbor_cars) > 0:
route = UCTPlayGame(car, GRID_SIZE, neighbor_cars)
car.push_route_netlogo(netlogo, route, mode='both')
else:
continue
def cam_to_raft(self, camcoord):
'''
For a given set of coordinates camcoord = (cam_x, cam_y) in the camera coordinate system, returns:
- the ID of the raft whose center is the closest to camcoord
- the corresponding coordinates in the raft frame, where raftcoord = (raft_x, raft_y) = (0,0) at the raft centre
'''
self.camcoord = camcoord
self.dist2raft = {}
for raft in self.raft_list:
# print(self.camcoord, self.raftcentercoord[raft])
self.dist2raft[raft] = dist(camcoord, self.raftcentercoord[raft])
self.raft_id = min(self.dist2raft, key=self.dist2raft.get)
self.raftcoord = np.empty(2)
self.raftcoord[0] = self.camcoord[0] - self.raftcentercoord[
self.raft_id][0]
self.raftcoord[1] = self.camcoord[1] - self.raftcentercoord[
self.raft_id][1]
def get_MDS(data, n_components=2, metric="euclidean"):
"""
Calculates the Multi-Dimensional Scaling algorithm using different
distances
"""
# Compute the distance between every pair of points
dist_mat = dist(data, data, metric=metric) # N x N
# Calculate the matrix H
N = data.shape[0]
H = np.eye(N) - np.ones((N, N)) / N
# Apply double centering
S = -0.5 * np.dot(np.dot(H, np.power(dist_mat, 2)), H)
# Get PCA
return get_pca(S, n_components=n_components, center_data=False)
def solve(eqA, eqB):
points = []
for eqa in eqA:
for eqb in eqB:
a = [[1, -eqa[0]],[1, -eqb[0]]]
b = [eqa[1],eqb[1]]
x, y = np.linalg.solve(a, b)
points += [(x,y)]
filtered_points = []
for a in points:
nearest = []
for b in points:
if dist(a,b) < 20:
nearest += [b]
x, y = [ int(_) for _ in np.mean(nearest, axis=0) ]
filtered_points += [(x, y)]
return filtered_points
def evt_motion(self, evt):
if self._mode == 'dragging':
if evt.inaxes != self.ax:
self.reset()
return
x,y = evt.xdata,evt.ydata
if self.drag_patch is None:
self.drag_patch = pl.Circle((x,y), radius=self.r_init, edgecolor=(1,.6,.2,.8), facecolor=(1,1,1,.2), lw=1.5)
self.drag_pos0 = (x,y)
self.ax.add_patch(self.drag_patch)
else:
dx = x-self.drag_pos0[0]
dy = y-self.drag_pos0[1]
dp = np.sqrt(dx**2 + dy**2)
#dp = [dx,dy][np.argmax(np.abs([dx,dy]))]
new_r = max(0, self.r_init+dp*self.r_per_pix)
self.drag_patch.set_radius(new_r)
elif self._mode == 'remove':
if evt.inaxes != self.ax:
return
x,y = evt.xdata,evt.ydata
if self.rois is None or len(self.rois) == 0:
return
centers = [p.center for p in self.patches]
best = np.argmin([dist((x,y), c) for c in centers])
self.update_patches()
self.patches[best].set_color('red')
self.patches[best].set_alpha(1.)
self.fig.canvas.draw()
else:
return
self.fig.canvas.draw()
def predict_class(self, input_image=None):
actual_target = None
if input_image is None and self.test_positive_input_batches is not None:
random_idx = np.random.randint(
0, len(self.test_positive_input_batches))
input_image = self.test_positive_input_batches[
random_idx, :].reshape(1, 1, 784)
actual_target = self.test_positive_batch_targets[random_idx]
if self.input_encoder_model:
input_encodings = self.input_encoder_model.predict(input_image)[
0, 0, :]
distance_dict = {
i: 1.0 - dist(self.target_encodings[i], input_encodings)
for i in range(10)
}
return distance_dict, actual_target
def output_func(img, U, Omega_training, T_0, T_1):
sh = img.shape
I = img.flatten()
I = I - m
I_res = I.reshape(sh)
fig, (ax1, ax2) = plt.subplots(1, 2)
ax1.set_title('I - m face')
im1 = ax1.imshow(I_res, plt.cm.gray)
Omega_I = np.dot(U.T, I)
print('PCA coefficients are \n')
print(Omega_I)
I_R = np.dot(U, Omega_I)
Ir_res = I_R.reshape(sh)
ax2.set_title('Reconstructed face I_r')
im2 = ax2.imshow(Ir_res, plt.cm.gray)
dist_array = [dist(Omega_I, Omega) for Omega in Omega_training]
print('The distances are \n')
print(dist_array)
plt.show()
print('\n\n\n\n\n\n')
def raft_to_sensor(self, raftcoord):
'''
For a given set of coordinates raftcoord = (raft_x, raft_y) in the raft coordinate system, returns:
- the ID of the ccd whose center is the closest to raftcoord
- the corresponding coordinates in the ccd frame, where (slotcoord_x, slotcoord_y) = (0,0) at the CCD centre
'''
self.raftcoord = raftcoord
self.dist2sensor = {}
for sensor in self.sensor_list:
# print(self.camcoord, self.raftcentercoord[raft])
self.dist2sensor[sensor] = dist(raftcoord,
self.sensorcentercoord[sensor])
self.sensor_id = min(self.dist2sensor, key=self.dist2sensor.get)
self.sensorcoord = np.empty(2)
self.sensorcoord[0] = self.raftcoord[0] - self.sensorcentercoord[
self.sensor_id][0]
self.sensorcoord[1] = self.raftcoord[1] - self.sensorcentercoord[
self.sensor_id][1]
return self.sensor_id, self.sensorcoord
def compare_faces(self, encoding):
# If faces encoding match, then return fid, otherwise, create one
edists = []
if len(self.encodings) == 0:
return None
else:
for old_id, old_encoding in self.encodings.items():
edist = dist(encoding, old_encoding)
print(edist, old_id)
edists.append(edist)
mindist = min(edists)
if mindist > float(threshold):
return 'unknown'
else:
names = list(self.encodings.keys())
minID = names[edists.index(mindist)]
return minID
def converge_txt(s):
rms = []
val = []
headings = ["Iter"]
headings.extend(["Orb-" + str(i + 1) for i in range(10)])
headings.extend(["Total"])
for cnt in [[i, i + 1] for i in range(0, s.shape[0] - 1)]:
tmp = [
str(np.array(cnt) + 1).replace("[", "").replace("]",
"").replace(
" ", "->")
]
tmp.extend([
dist([s[cnt[0]][norb]], [s[cnt[1]][norb]])[0]
for norb in range(10)
])
tmp.extend([np.sum(tmp[1:])])
rms.append(tmp)
val.append(tmp[1:])
print(tabulate(rms, headers=headings))
fprint(tabulate(rms, headers=headings))
return (np.array(val))
def shape_context_shape(pc,
xr,
yr,
groups=None,
sampls=None,
r_inner=0.125,
r_outer=2,
nbins_r=5,
nbins_theta=12):
nbins = nbins_r * nbins_theta
def get_angle(p1, p2):
"""Return angle in radians"""
return math.atan2((p2[1] - p1[1]), (p2[0] - p1[0]))
def compute_one(pt, points, mean_dist):
distances = np.array([euclidean(pt, p) for p in points])
r_array_n = distances / mean_dist
r_bin_edges = np.logspace(np.log10(r_inner), np.log10(r_outer),
nbins_r)
r_array_q = np.zeros(len(points))
for m in xrange(nbins_r):
r_array_q += (r_array_n < r_bin_edges[m])
fz = r_array_q > 0
def _get_angles(self, x):
result = zeros((len(x), len(x)))
for i in xrange(len(x)):
for j in xrange(len(x)):
result[i, j] = get_angle(x[i], x[j])
return result
theta_array = np.array([get_angle(pt, p) for p in points])
# 2Pi shifted
theta_array_2 = theta_array + 2 * math.pi * (theta_array < 0)
theta_array_q = 1 + np.floor(theta_array_2 /
(2 * math.pi / nbins_theta))
sn = np.zeros((nbins_r, nbins_theta))
for j in xrange(len(points)):
if (fz[j]):
sn[r_array_q[j] - 1, theta_array_q[j] - 1] += 1
return sn.reshape(nbins)
rsi = binary_shape(pc, xr, yr, groups)
pixels = rsi.pixels.squeeze()
pts = np.argwhere(pixels > 0)
if sampls:
pts = pts[np.random.randint(0, pts.shape[0], sampls)]
mean_dist = dist([0, 0], [xr, yr]) / 2
sc = np.zeros((xr, yr, nbins))
for x in xrange(xr):
for y in xrange(yr):
sc[x, y, :] = compute_one(np.array([x, y]), pts, mean_dist)
return Image.init_from_rolled_channels(sc)
def find_target(self, frame):
'''returns yaw, centre zero, positive right. None if no target found'''
# Convert BGR to HSV
hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
# define range of blue color in HSV
# HUE IS 0 to 180
lower_orange = np.array([5, 150, 100])
upper_orange = np.array([20, 255, 255])
low_lower_red = np.array([2, 100, 70])
low_upper_red = np.array([5, 150, 150])
high_lower_red = np.array([160, 100, 30])
high_upper_red = np.array([200, 240, 240])
# Threshold the HSV image to get only blue colors
lmask = cv2.inRange(hsv, low_lower_red, low_upper_red)
hmask = cv2.inRange(hsv, high_lower_red, high_upper_red)
mask = cv2.bitwise_or(lmask, hmask)
# Bitwise-AND mask and original image
res = cv2.bitwise_and(frame, frame, mask=mask)
er_mask = cv2.erode(mask, None, iterations=self.erode_iterations)
dl_mask = cv2.dilate(er_mask, None, iterations=self.dilate_iterations)
# load the image, convert it to grayscale, blur it slightly,
# and threshold it
# find contours in the thresholded image
cnts = cv2.findContours(dl_mask.copy(), cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
cnts = cnts[1]
image = frame.copy()
# loop over the contours
target_list = []
for c in cnts:
# compute the center of the contour
M = cv2.moments(c)
cX = int(M["m10"] / M["m00"])
cY = int(M["m01"] / M["m00"])
# draw the contour and center of the shape on the image
cv2.drawContours(image, [c], -1, (0, 255, 0), 2)
cv2.circle(image, (cX, cY), 7, (255, 255, 255), -1)
if c.size < self.target_size:
cv2.putText(image, "small", (cX - 20, cY - 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 2)
else:
cv2.putText(image, "large", (cX - 20, cY - 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 2)
target_list.append([c.size, cX, cY])
targets = np.array(target_list)
if len(targets) == 0:
return None
target_idx = np.argmax(targets, axis=0)
if len(target_idx) > 1:
target_idx = target_idx[0]
largest_target = [
int(targets[target_idx][1]),
int(targets[target_idx][2])
]
#largest_target = targets.max(axis=0)
target_x = largest_target[1]
image_height, image_width, _ = [
frame.shape[0], frame.shape[1], frame.shape[2]
]
bottom_point = np.array([round(image_width / 2), round(image_height)])
cv2.line(image, (int(largest_target[0]), int(largest_target[1])),
(int(bottom_point[0]), int(bottom_point[1])), (255, 255, 255),
17, -1)
if self.display_windows:
cv2.imshow('im_with_keypoints',
cv2.resize(image, dsize=(0, 0), fx=0.5, fy=0.5))
cv2.imshow('mask', cv2.resize(mask, dsize=(0, 0), fx=0.5, fy=0.5))
pixel_x_location = target_x - image_width / 2
yaw_offset = pixel_x_location * self.heading_scale_factor
z = round(dist(largest_target, bottom_point))
x = image_height - largest_target[1]
self.angle = np.arcsin(x / z)
return
def openpose_face_detector(posePtr, threshold):
point_top_left = np.zeros(2)
face_size = 0.0
score = 0.0
points_used = set()
neckScoreAbove = posePtr[1, 2] > threshold
headNoseScoreAbove = posePtr[0, 2] > threshold
lEarScoreAbove = posePtr[16, 2] > threshold
rEarScoreAbove = posePtr[17, 2] > threshold
lEyeScoreAbove = posePtr[14, 2] > threshold
rEyeScoreAbove = posePtr[15, 2] > threshold
counter = 0.0
if neckScoreAbove and headNoseScoreAbove:
if (lEyeScoreAbove == lEarScoreAbove
and rEyeScoreAbove == rEarScoreAbove
and lEyeScoreAbove != rEyeScoreAbove):
if lEyeScoreAbove:
point_top_left += (posePtr[14, 0:2] + posePtr[16, 0:2] +
posePtr[0, 0:2]) / 3.0
face_size += 0.85 * (dist(posePtr[14, 0:2], posePtr[16, 0:2]) +
dist(posePtr[0, 0:2], posePtr[16, 0:2]) +
dist(posePtr[14, 0:2], posePtr[0, 0:2]))
points_used = points_used.union([0, 14, 16])
else:
point_top_left += (posePtr[15, 0:2] + posePtr[17, 0:2] +
posePtr[0, 0:2]) / 3.0
face_size += 0.85 * (dist(posePtr[15, 0:2], posePtr[17, 0:2]) +
dist(posePtr[0, 0:2], posePtr[17, 0:2]) +
dist(posePtr[15, 0:2], posePtr[0, 0:2]))
points_used = points_used.union([0, 15, 17])
else:
point_top_left += (posePtr[1, 0:2] + posePtr[0, 0:2]) / 2.0
face_size += 2.0 * dist(posePtr[1, 0:2], posePtr[0, 0:2])
points_used = points_used.union([0, 1])
counter += 1.0
if lEyeScoreAbove and rEyeScoreAbove:
point_top_left += (posePtr[14, 0:2] + posePtr[15, 0:2]) / 2.0
face_size += 3.0 * dist(posePtr[14, 0:2], posePtr[15, 0:2])
counter += 1.0
points_used = points_used.union([14, 15])
if lEarScoreAbove and rEarScoreAbove:
point_top_left += (posePtr[16, 0:2] + posePtr[17, 0:2]) / 2.0
face_size += 2.0 * dist(posePtr[16, 0:2], posePtr[17, 0:2])
counter += 1.0
points_used = points_used.union([16, 17])
if counter > 0:
point_top_left /= counter
face_size /= counter
score = np.mean(posePtr[list(points_used), 2])
return region.RectangleRegion(
point_top_left[0] - face_size / 2,
point_top_left[1] - face_size / 2,
face_size,
face_size,
score,
)
def is_intersect(self, other):
self_xy = [self.x, self.y]
other_xy = [other.x, other.y]
return dist(self_xy, other_xy) <= self.r + other.r