Pycharmでdlibを使って、webカメラを用いた顔検出を行なっています。(検出した顔の上に写真を貼り付けるプログラムになります。)
一人だけが顔検出されるところまではできたのですが、複数人の顔検出の仕方がわかりません。
コードは下記の通りです。どこを変えれば、複数人の顔が検出されるようになりますでしょうか?よろしくお願いします。
import cv2
import dlib
from datetime import datetime
IMAGE_PATH = "./niko.png"
CAPTURE_SCALE = 0.5
def main():
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor("shape_predictor_68_face_landmarks.dat")
cap = cv2.VideoCapture(0)
while True:
_, frame = cap.read()
frame = img_resize(frame, CAPTURE_SCALE)
h, w, _ = frame.shape
img = frame
dets = detector(frame[:, :, ::-1])
if len(dets) > 0:
parts = predictor(frame, dets[0]).parts()
distance, pos = calc_distance_and_pos(img, parts)
icon, icon_w, icon_h = load_icon(IMAGE_PATH, distance)
if (pos != None) and (distance > 0.0):
x = pos.x - int(icon_w / 2)
y = pos.y - int(icon_h / 2)
if (0 <= y) and (y <= (h - int(icon_h))) and (0 <= x) and (x <= (w - int(icon_w))):
img = merge_images(img, icon, x, y)
cv2.imshow("camera", img)
k = cv2.waitKey(1) & 0xff
if k == ord('q'):
break
elif k == ord('p'):
save_image(img)
cv2.imshow("saved", img)
cap.release()
cv2.destroyAllWindows()
def load_icon(path, distance):
icon = cv2.imread(path, -1)
icon_height, _ = icon.shape[:2]
icon = img_resize(icon, float(distance * 1.5 / icon_height))
icon_h, icon_w = icon.shape[:2]
return icon, icon_w, icon_h
def img_resize(img, scale):
h, w = img.shape[:2]
img = cv2.resize(img, (int(w * scale), int(h * scale)))
return img
def calc_distance_and_pos(img, parts):
cnt = 0
pos = None
p1 = None
distance = 0.0
for i in parts:
if (cnt == 0):
p1 = i
if (cnt == 16):
distance = ((p1.x - i.x) ** 2 + (p1.y - i.y) ** 2) ** 0.5
if (cnt == 33):
pos = i
cv2.putText(img, str(cnt), (i.x, i.y), cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 0, 255), thickness=1,
lineType=cv2.LINE_8)
cv2.circle(img, (i.x, i.y), 1, (255, 0, 0), -1)
cnt = cnt + 1
return distance, pos
def save_image(img):
date = datetime.now().strftime("%Y%m%d_%H%M%S")
path = "./" + date + ".png"
cv2.imwrite(path, img)
def merge_images(bg, fg_alpha, s_x, s_y):
alpha = fg_alpha[:, :, 3]
alpha = cv2.cvtColor(alpha, cv2.COLOR_GRAY2BGR)
alpha = alpha / 255.0
fg = fg_alpha[:, :, :3]
f_h, f_w, _ = fg.shape
b_h, b_w, _ = bg.shape
print("f_w:{} f_h:{} b_w:{} b_h:{} s({}, {})".format(f_w, f_h, b_w, b_h, s_x, s_y))
bg[s_y:f_h + s_y, s_x:f_w + s_x] = (bg[s_y:f_h + s_y, s_x:f_w + s_x] * (1.0 - alpha)).astype(
'uint8')
bg[s_y:f_h + s_y, s_x:f_w + s_x] = (bg[s_y:f_h + s_y, s_x:f_w + s_x] + (fg * alpha)).astype('uint8')
return bg
if name == 'main':
main()