class MTCNN_Crop_Face(object):
def __init__(self):
test_mode = "ONet"
thresh = [0.9, 0.7, 0.7]
min_face_size = 32
stride = 2
slide_window = False
shuffle = False
batch_size = [2048, 64, 16]
detectors = [None, None, None]
prefix = ['./data2/MTCNN_model/PNet_landmark/PNet', './data2/MTCNN_model/RNet_landmark/RNet', './data2/MTCNN_model/ONet_landmark/ONet']
epoch = [18, 14, 16]
model_path = ['%s-%s' % (x, y) for x, y in zip(prefix, epoch)]
PNet = FcnDetector(P_Net, model_path[0])
detectors[0] = PNet
RNet = Detector(R_Net, 24, 1, model_path[1])
detectors[1] = RNet
ONet = Detector(O_Net, 48, 1, model_path[2])
detectors[2] = ONet
self.mtcnn_detector = MtcnnDetector(detectors=detectors, min_face_size=min_face_size,
stride=stride, threshold=thresh, slide_window=slide_window)
def cropface(self, frame):
cropped_list = []
i = 0
image = np.array(frame)
all_boxes, landmarks = self.mtcnn_detector.detect(image)
# all_boxes,landmarks = self.mtcnn_detector.detect_face(test_data)
# image = cv2.imread(imagepath)
for bbox in all_boxes:
corpbbox = [int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])]
if abs(corpbbox[3] - corpbbox[1]) > abs(corpbbox[2] - corpbbox[0]):
n0 = int(0.5 * (corpbbox[2] + corpbbox[0]) - 0.5 * abs(corpbbox[3] - corpbbox[1]))
n1 = corpbbox[1]
n2 = int(0.5 * (corpbbox[2] + corpbbox[0]) + 0.5 * abs(corpbbox[3] - corpbbox[1]))
n3 = corpbbox[3]
else:
n0 = corpbbox[0]
n1 = int(0.5 * (corpbbox[3] + corpbbox[1]) - 0.5 * abs(corpbbox[2] - corpbbox[0]))
n2 = corpbbox[2]
n3 = int(0.5 * (corpbbox[3] + corpbbox[1]) + 0.5 * abs(corpbbox[2] - corpbbox[0]))
# cropped = image[n1:n3, n0:n2]
cropped = [int(n0), int(n1), int(n2), int(n3)]
cropped_list.append(cropped)
# cv2.rectangle(frame, (int(bbox[0]),int(bbox[1])),(int(bbox[2]),int(bbox[3])),(0,0,255))
# cv2.imwrite("../data/test_result/%d.png" %(count),image)
# cv2.imshow("image",frame)
# cv2.waitKey(0)
# return all_boxes
return cropped_list
slide_window=slide_window)
path = os.path.join(os.pardir, 'data/test/lfpw_testImage')
gt_imdb = [os.path.join(path, item) for item in os.listdir(path)]
test_data = TestLoader(gt_imdb)
# boxes, landmarks = mtcnn_detector.detect_face(test_data)
out_dir = os.path.dirname(gt_imdb[0]) + '_out'
if not os.path.exists(out_dir):
os.makedirs(out_dir)
for path in gt_imdb:
print(path)
image = cv2.imread(path)
boxes, landmarks = mtcnn_detector.detect(image)
# show rectangles
for bbox in boxes:
position = (int(bbox[0]), int(bbox[1]))
cv2.putText(image, str(np.round(bbox[4], 2)), position,
cv2.FONT_HERSHEY_TRIPLEX, 1, (255, 0, 255))
cv2.rectangle(image, position, (int(bbox[2]), int(bbox[3])),
(0, 0, 255))
# show landmarks
for landmark in landmarks:
for i, k in grouper(landmark, 2):
cv2.circle(image, (i, k), 3, (0, 0, 255))
path = os.path.join(out_dir, os.path.basename(path))
min_face_size=min_face_size,
stride=stride,
threshold=thresh,
slide_window=slide_window)
camera = cv2.VideoCapture(1)
name = 'test'
count = 0
path = os.getcwd() + '/data-face/' + name
if not os.path.exists(path):
os.makedirs(path)
while True:
_, frame = camera.read()
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
boxes_c, _ = mtcnn_detector.detect(frame)
for i in range(boxes_c.shape[0]):
bbox = boxes_c[i, :4]
score = boxes_c[i, 4]
corpbbox = [int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])]
cv2.rectangle(frame, (corpbbox[0], corpbbox[1]),
(corpbbox[2], corpbbox[3]), (255, 0, 0), 1)
cv2.putText(frame, '{:.3f}'.format(score),
(corpbbox[0], corpbbox[1] - 2), cv2.FONT_HERSHEY_SIMPLEX,
0.5, (0, 0, 255), 2)
try:
scipy.misc.imsave(
path + '/' + name + str(count) + '.png',
gray[int(bbox[1]):int(bbox[3]),
int(bbox[0]):int(bbox[2])])
except:
def build_camera(self, camera_id, path_name):
#print("Loading Deep Face Detector(MTCNN) recognition model ............")
#test_mode = "onet"
thresh = [0.9, 0.9, 0.8]
min_face_size = 81
stride = 2
slide_window = False
#shuffle = False
#vis = True
detectors = [None, None, None]
prefix = ['../execute_system/MTCNN_Tensorflow_fast/data/MTCNN_model/PNet_landmark/PNet',
'../execute_system/MTCNN_Tensorflow_fast/data/MTCNN_model/RNet_landmark/RNet',
'../execute_system/MTCNN_Tensorflow_fast/data/MTCNN_model/ONet_landmark/ONet']
epoch = [40, 36, 36]
model_path = ['%s-%s' % (x, y) for x, y in zip(prefix, epoch)]
PNet = FcnDetector(P_Net, model_path[0])
detectors[0] = PNet
RNet = Detector(R_Net, 24, 1, model_path[1])
detectors[1] = RNet
ONet = Detector(O_Net, 48, 1, model_path[2])
detectors[2] = ONet
mtcnn_detector = MtcnnDetector(detectors=detectors, min_face_size=min_face_size,
stride=stride, threshold=thresh, slide_window=slide_window)
# read names from dataset
name_list = read_name_list(path_name)
cap = cv2.VideoCapture(camera_id)
#fps1 = cap.get(cv2.CAP_PROP_FPS)
#size = (int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)),
# int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)))
#size = 640 x 480
cap.set(cv2.CAP_PROP_FRAME_WIDTH,528)
cap.set(cv2.CAP_PROP_FRAME_HEIGHT,384)
while True:
t1 = cv2.getTickCount()
success, frame = cap.read()
if success:
thickness = (frame.shape[0] + frame.shape[1]) // 350
image = np.array(frame)
boxes_c,landmarks = mtcnn_detector.detect(image)
#print(landmarks.shape)
t2 = cv2.getTickCount()
t = (t2 - t1) / cv2.getTickFrequency()
fps = 1.0 / t
print('fps:',fps)
for i in range(boxes_c.shape[0]):
bbox = boxes_c[i, :4]
#score = boxes_c[i, 4]
cropbbox = [int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])]
W = -int(cropbbox[0]) + int(cropbbox[2])
H = -int(cropbbox[1]) + int(cropbbox[3])
paddingH = 0.02 * H
paddingW = 0.01 * W
crop_img = frame[int(cropbbox[1]+paddingH):int(cropbbox[3]-paddingH), int(cropbbox[0]-paddingW):int(cropbbox[2]+paddingW)]
if crop_img is None:
continue
if crop_img.shape[0] < 0 or crop_img.shape[1] < 0:
continue
label,prob = self.model.face_predict(crop_img)
if prob > 0.7:
show_name = name_list[label]
else:
show_name = 'Stranger'
person_tag = "%s: %.2f" %(show_name, prob)
#text_end = (int(cropbbox[0]) + len(person_tag) * 10,int(cropbbox[1]) - 20 )
text_start = (max(int(cropbbox[0]), 10), max(int(cropbbox[1]), 10))
#cv2.rectangle(draw, text_end, text_start, (255, 255, 0), -1, cv2.LINE_AA )
cv2.putText(frame, person_tag, text_start, cv2.FONT_HERSHEY_SIMPLEX, 1, (255,0,255), 1)
# rectangle for face area
for i in range(thickness):
start = (int(cropbbox[0]) + i, int(cropbbox[1]) + i)
end = (int(cropbbox[2] - i), int(cropbbox[3]) - i)
frame = cv2.rectangle(frame, start, end, (0, 255, 0), 1)
# display the landmarks
for i in range(landmarks.shape[0]):
for j in range(len(landmarks[i])//2):
cv2.circle(frame, (int(landmarks[i][2*j]),int(int(landmarks[i][2*j+1]))), 2, (0,0,255))
cv2.imshow("Camera", frame)
k = cv2.waitKey(10)
if k & 0xFF == ord('q'):
break
else:
print ('device not find')
break
cap.release()
cv2.destroyAllWindows()
#imdb_['label'] = 5
path = "lala"
for item in os.listdir(path):
gt_imdb.append(os.path.join(path,item))
test_data = TestLoader(gt_imdb)
all_boxes,landmarks = mtcnn_detector.detect_face(test_data)
'''
image_path = 'lala/single_face.jpg'
img = cv2.imread(image_path)
img = cv2.resize(img, (640, 480))
t1 = time.time()
boxes_c,landmarks = mtcnn_detector.detect(img)
t = time.time() - t1
print('total time', t)
for i in range(boxes_c.shape[0]):
bbox = boxes_c[i, :4]
score = boxes_c[i, 4]
corpbbox = [int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])]
# if score > thresh:
cv2.rectangle(img, (corpbbox[0], corpbbox[1]), (corpbbox[2], corpbbox[3]), (255, 0, 0), 1)
cv2.putText(img, '{:.3f}'.format(score), (corpbbox[0], corpbbox[1] - 2), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255), 2)
cv2.imshow('show', img)
# exit press any key
cv2.waitKey(0)
'''
def RTrecognization(facenet_model_path, SVCpath, database_path):
#facenet_model_path为facenet模型路径
#SVCpath为SVM分类模型路径
#database_path为人脸库数据
with tf.Graph().as_default():
with tf.Session() as sess:
# Load the model
print('Loading feature extraction model')
facenet.load_model(facenet_model_path)
with open(SVCpath, 'rb') as infile:
(classifymodel, class_names) = pickle.load(infile)
print('Loaded classifier model from file "%s"' % SVCpath)
# Get input and output tensors
images_placeholder = tf.get_default_graph().get_tensor_by_name(
"input:0")
embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
phase_train_placeholder = tf.get_default_graph(
).get_tensor_by_name("phase_train:0")
embedding_size = embeddings.get_shape()[1]
Database = np.load(database_path)
test_mode = "onet"
thresh = [0.9, 0.6, 0.7]
min_face_size = 24
stride = 2
slide_window = False
shuffle = False
#vis = True
detectors = [None, None, None]
prefix = [
'../data/MTCNN_model/PNet_landmark/PNet',
'../data/MTCNN_model/RNet_landmark/RNet',
'../data/MTCNN_model/ONet_landmark/ONet'
]
epoch = [18, 14, 16]
model_path = ['%s-%s' % (x, y) for x, y in zip(prefix, epoch)]
PNet = FcnDetector(P_Net, model_path[0])
detectors[0] = PNet
RNet = Detector(R_Net, 24, 1, model_path[1])
detectors[1] = RNet
ONet = Detector(O_Net, 48, 1, model_path[2])
detectors[2] = ONet
mtcnn_detector = MtcnnDetector(detectors=detectors,
min_face_size=min_face_size,
stride=stride,
threshold=thresh,
slide_window=slide_window)
video_capture = cv2.VideoCapture(0)
# video_capture.set(3, 340)
# video_capture.set(4, 480)
video_capture.set(3, 800)
video_capture.set(4, 800)
corpbbox = None
while True:
t1 = cv2.getTickCount()
ret, frame = video_capture.read()
if ret:
image = np.array(frame)
img_size = np.array(image.shape)[0:2]
boxes_c, landmarks = mtcnn_detector.detect(image)
# print(boxes_c.shape)
# print(boxes_c)
# print(img_size)
t2 = cv2.getTickCount()
t = (t2 - t1) / cv2.getTickFrequency()
fps = 1.0 / t
for i in range(boxes_c.shape[0]):
bbox = boxes_c[i, :4] #检测出的人脸区域，左上x，左上y，右下x，右下y
score = boxes_c[i, 4] #检测出人脸区域的得分
corpbbox = [
int(bbox[0]),
int(bbox[1]),
int(bbox[2]),
int(bbox[3])
]
x1 = np.maximum(int(bbox[0]) - 16, 0)
y1 = np.maximum(int(bbox[1]) - 16, 0)
x2 = np.minimum(int(bbox[2]) + 16, img_size[1])
y2 = np.minimum(int(bbox[3]) + 16, img_size[0])
crop_img = image[y1:y2, x1:x2]
scaled = misc.imresize(crop_img, (160, 160),
interp='bilinear')
img = load_image(scaled, False, False, 160)
img = np.reshape(img, (-1, 160, 160, 3))
feed_dict = {
images_placeholder: img,
phase_train_placeholder: False
}
embvecor = sess.run(embeddings, feed_dict=feed_dict)
embvecor = np.array(embvecor)
#利用人脸特征与数据库中所有人脸进行一一比较的方法
# tmp=np.sqrt(np.sum(np.square(embvecor-Database['emb'][0])))
# tmp_lable=Database['lab'][0]
# for j in range(len(Database['emb'])):
# t=np.sqrt(np.sum(np.square(embvecor-Database['emb'][j])))
# if t<tmp:
# tmp=t
# tmp_lable=Database['lab'][j]
# print(tmp)
#利用SVM对人脸特征进行分类
predictions = classifymodel.predict_proba(embvecor)
best_class_indices = np.argmax(predictions, axis=1)
tmp_lable = class_names[best_class_indices]
best_class_probabilities = predictions[
np.arange(len(best_class_indices)),
best_class_indices]
print(best_class_probabilities)
if best_class_probabilities < 0.4:
tmp_lable = "others"
cv2.rectangle(frame, (corpbbox[0], corpbbox[1]),
(corpbbox[2], corpbbox[3]), (255, 0, 0),
1)
cv2.putText(frame, '{0}'.format(tmp_lable),
(corpbbox[0], corpbbox[1] - 2),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255),
2)
cv2.putText(
frame, '{:.4f}'.format(t) + " " + '{:.3f}'.format(fps),
(10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 255),
2)
for i in range(landmarks.shape[0]):
for j in range(len(landmarks[i]) // 2):
cv2.circle(frame,
(int(landmarks[i][2 * j]),
int(int(landmarks[i][2 * j + 1]))), 2,
(0, 0, 255))
# time end
cv2.imshow("", frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
else:
print('device not find')
break
video_capture.release()
cv2.destroyAllWindows()
# 如果从自己laptop 摄像头读入，cv2.VideoCapture(videopath) ===> cv2.VideoCapture(0)
# cv2.VideoCapture https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_gui/py_video_display/py_video_display.html
video_capture = cv2.VideoCapture(
videopath) # cv2.VideoCapture(0) for laptop camera
video_capture.set(3, 340)
video_capture.set(4, 480)
corpbbox = None
while True:
# fps = video_capture.get(cv2.CAP_PROP_FPS)
t1 = cv2.getTickCount()
ret, frame = video_capture.read(
) # capture frame-by-frame. read() returns a bool(True/False). If frame is read correctly, it's True.
if ret:
image = np.array(frame)
boxes_c, landmarks = mtcnn_detector.detect(
image) # use mtcnn detector to read image
t2 = cv2.getTickCount()
t = (t2 - t1) / cv2.getTickFrequency()
fps = 1.0 / t
for i in range(boxes_c.shape[0]):
bbox = boxes_c[i, :4]
score = boxes_c[i, 4]
corpbbox = [int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])]
# if score > thresh:
cv2.rectangle(frame, (corpbbox[0], corpbbox[1]),
(corpbbox[2], corpbbox[3]), (255, 0, 0), 1)
cv2.putText(frame, '{:.3f}'.format(score),
(corpbbox[0], corpbbox[1] - 2),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255), 2)
cv2.putText(frame, '{:.4f}'.format(t) + " " + '{:.3f}'.format(fps),
detectors[2] = ONet
videopath = "./video_test.avi"
mtcnn_detector = MtcnnDetector(detectors=detectors, min_face_size=min_face_size,
stride=stride, threshold=thresh, slide_window=slide_window)
video_capture = cv2.VideoCapture(videopath)
video_capture.set(3, 340)
video_capture.set(4, 480)
corpbbox = None
while True:
# fps = video_capture.get(cv2.CAP_PROP_FPS)
t1 = cv2.getTickCount()
ret, frame = video_capture.read()
if ret:
image = np.array(frame)
boxes_c,landmarks = mtcnn_detector.detect(image)
print landmarks.shape
t2 = cv2.getTickCount()
t = (t2 - t1) / cv2.getTickFrequency()
fps = 1.0 / t
for i in range(boxes_c.shape[0]):
bbox = boxes_c[i, :4]
score = boxes_c[i, 4]
corpbbox = [int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])]
# if score > thresh:
cv2.rectangle(frame, (corpbbox[0], corpbbox[1]),
(corpbbox[2], corpbbox[3]), (255, 0, 0), 1)
cv2.putText(frame, '{:.3f}'.format(score), (corpbbox[0], corpbbox[1] - 2), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(0, 0, 255), 2)
cv2.putText(frame, '{:.4f}'.format(t) + " " + '{:.3f}'.format(fps), (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
def RTrecognization(facenet_model_path,SVCpath,database_path):
#facenet_model_path为facenet模型路径
#SVCpath为SVM分类模型路径
#database_path为人脸库数据
with tf.Graph().as_default():
with tf.Session() as sess:
# Load the model
print('Loading feature extraction model')
facenet.load_model(facenet_model_path)
with open(SVCpath, 'rb') as infile:
(classifymodel, class_names) = pickle.load(infile)
print('Loaded classifier model from file "%s"' % SVCpath)
# Get input and output tensors
images_placeholder = tf.get_default_graph().get_tensor_by_name("input:0")
embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")
phase_train_placeholder = tf.get_default_graph().get_tensor_by_name("phase_train:0")
embedding_size = embeddings.get_shape()[1]
Database=np.load(database_path)
"""
model_folder : string
path for the models
训练好的模型路径
minsize : float number
minimal face to detect
threshold : float number
detect threshold for 3 stages
threshold参数是一个包含3个值的列表，这3个值在算法的3个stage中将分别用到，
可以看到这3个threshold值是递增的，是因为在3个stage中对一个bbox是否是人脸的置信度要求越来越高。
factor: float number
scale factor for image pyramid
factor表示和图像金字塔相关的一个参数，表示图像金字塔的每相邻两层之间的倍数关系是factor。
num_worker: int number
number of processes we use for first stage
accurate_landmark: bool
use accurate landmark localization or not
"""
test_mode = "onet"
thresh = [0.9, 0.6, 0.7]
min_face_size = 24
stride = 2
slide_window = False
shuffle = False
#vis = True
detectors = [None, None, None]
prefix = ['../data/MTCNN_model/PNet_landmark/PNet', '../data/MTCNN_model/RNet_landmark/RNet',
'../data/MTCNN_model/ONet_landmark/ONet']
epoch = [18, 14, 16]
model_path = ['%s-%s' % (x, y) for x, y in zip(prefix, epoch)]
PNet = FcnDetector(P_Net, model_path[0])
detectors[0] = PNet
RNet = Detector(R_Net, 24, 1, model_path[1])
detectors[1] = RNet
ONet = Detector(O_Net, 48, 1, model_path[2])
detectors[2] = ONet
mtcnn_detector = MtcnnDetector(detectors=detectors, min_face_size=min_face_size,
stride=stride, threshold=thresh, slide_window=slide_window)
video_capture = cv2.VideoCapture(0)
# video_capture.set(3, 340)
# video_capture.set(4, 480)
video_capture.set(3, 800)
video_capture.set(4, 800)
corpbbox = None
while True:
t1 = cv2.getTickCount()
ret, frame = video_capture.read()
if ret:
image = np.array(frame)
img_size=np.array(image.shape)[0:2]
boxes_c,landmarks = mtcnn_detector.detect(image)
# print(boxes_c.shape)
# print(boxes_c)
# print(img_size)
t2 = cv2.getTickCount()
t = (t2 - t1) / cv2.getTickFrequency()
fps = 1.0 / t
for i in range(boxes_c.shape[0]):
bbox = boxes_c[i, :4]#检测出的人脸区域，左上x，左上y，右下x，右下y
score = boxes_c[i, 4]#检测出人脸区域的得分
corpbbox = [int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])]
x1=np.maximum(int(bbox[0])-16,0)
y1=np.maximum(int(bbox[1])-16,0)
x2=np.minimum( int(bbox[2])+16,img_size[1])
y2=np.minimum( int(bbox[3])+16,img_size[0])
crop_img=image[y1:y2,x1:x2]
scaled=misc.imresize(crop_img,(160,160),interp='bilinear')
img=load_image(scaled,False, False,160)
img=np.reshape(img,(-1,160,160,3))
feed_dict = { images_placeholder:img, phase_train_placeholder:False }
embvecor=sess.run(embeddings, feed_dict=feed_dict)
embvecor=np.array(embvecor)
#利用人脸特征与数据库中所有人脸进行一一比较的方法
# tmp=np.sqrt(np.sum(np.square(embvecor-Database['emb'][0])))
# tmp_lable=Database['lab'][0]
# for j in range(len(Database['emb'])):
# t=np.sqrt(np.sum(np.square(embvecor-Database['emb'][j])))
# if t<tmp:
# tmp=t
# tmp_lable=Database['lab'][j]
# print(tmp)
# 利用SVM对人脸特征进行分类
# 通过输入的图片的编码，通过此分类器预测模型的名称
# 得到的结果是对于每张图片，都进行预测，得出可能是某个人物的可能性
predictions = classifymodel.predict_proba(embvecor)
# 返回每行最大值的索引,也就是返回这张图片对应的人物的可能性最大的姓名的索引
best_class_indices = np.argmax(predictions, axis=1)
tmp_lable=class_names[best_class_indices]
# 行索引，加列索引，就得到了所有图片预测的可能性
best_class_probabilities = predictions[np.arange(len(best_class_indices)), best_class_indices]
print(best_class_probabilities)
if best_class_probabilities<0.4:
tmp_lable="unknown"
cv2.rectangle(frame, (corpbbox[0], corpbbox[1]),
(corpbbox[2], corpbbox[3]), (255, 0, 0), 1)
cv2.putText(frame, '{0}'.format(tmp_lable), (corpbbox[0], corpbbox[1] - 2), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(0, 0, 255), 2)
cv2.putText(frame, '{:.4f}'.format(t) + " " + '{:.3f}'.format(fps), (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(255, 0, 255), 2)
for i in range(landmarks.shape[0]):
for j in range(len(landmarks[i])//2):
cv2.circle(frame, (int(landmarks[i][2*j]),int(int(landmarks[i][2*j+1]))), 2, (0,0,255))
# time end
cv2.imshow("", frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
else:
print('device not find')
break
video_capture.release()
cv2.destroyAllWindows()
if len(h264[n]) < 1000:
count += 1
if count == 3:
break
temp.append(h264[n])
break
if k > 2:
with open('temp.h264', 'wb') as fopen:
fopen.write(header + b''.join(temp[::-1]))
h264.clear()
cap = cv2.VideoCapture('temp.h264')
while True:
try:
last_time = time.time()
ret, img = cap.read()
boxes_c, _ = mtcnn_detector.detect(img)
for u in range(boxes_c.shape[0]):
bbox = boxes_c[u, :4]
# tl,tr,bl,br = [int(bbox[0]),int(bbox[1])],[int(bbox[2]),int(bbox[1])],[int(bbox[0]),int(bbox[3])],[int(bbox[2]),int(bbox[3])]
# (tltrX, tltrY) = midpoint(tl, tr)
# (blbrX, blbrY) = midpoint(bl, br)
# (tlblX, tlblY) = midpoint(tl, bl)
# (trbrX, trbrY) = midpoint(tr, br)
# print(land())
# # virtual width
# dA = dist.euclidean((tltrX, tltrY), (blbrX, blbrY))
# # virtual height
# dB = dist.euclidean((tlblX, tlblY), (trbrX, trbrY))
# distance = distance_to_camera(initial_flight_height, focal_length, dA)
# print(distance)
visualization_utils.draw_bounding_box_on_image_array(
detectors[1] = RNet
ONet = Detector(O_Net, 48, 1, model_path[2])
detectors[2] = ONet
image_path = "/home/tamvm/Pictures/test1/test_face_detect_2.jpeg"
mtcnn_detector = MtcnnDetector(detectors=detectors, min_face_size=min_face_size,
scale_factor=scale_factor,
stride=stride, threshold=thresh, slide_window=slide_window)
# fps = video_capture.get(cv2.CAP_PROP_FPS)
t1 = cv2.getTickCount()
image = cv2.imread(image_path)
image = cv2.resize(image, (input_img_size, input_img_size))
# frame = load_image_into_numpy_array(image)
frame = image
boxes_c, landmarks = mtcnn_detector.detect(frame)
print(landmarks.shape)
t2 = cv2.getTickCount()
t = (t2 - t1) / cv2.getTickFrequency()
fps = 1.0 / t
for i in range(boxes_c.shape[0]):
bbox = boxes_c[i, :4]
score = boxes_c[i, 4]
corpbbox = [int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])]
# if score > thresh:
cv2.rectangle(frame, (corpbbox[0], corpbbox[1]),
(corpbbox[2], corpbbox[3]), (255, 0, 0), 1)
cv2.putText(frame, '{:.3f}'.format(score), (corpbbox[0], corpbbox[1] - 2), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(0, 0, 255), 2)
cv2.putText(frame, '{:.4f}'.format(t) + " " + '{:.3f}'.format(fps), (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
def build_camera(self, camera_id, path):
count = 500
thresh = [0.9, 0.9, 0.8]
min_face_size = 100
stride = 2
slide_window = False
detectors = [None, None, None]
prefix = [
'E:/sign_system/execute_system/MTCNN_Tensorflow_fast/data/MTCNN_model_V2/PNet_landmark/PNet',
'E:/sign_system/execute_system/MTCNN_Tensorflow_fast/data/MTCNN_model_V2/RNet_landmark/RNet',
'E:/sign_system/execute_system/MTCNN_Tensorflow_fast/data/MTCNN_model_V2/ONet_landmark/ONet'
]
epoch = [40, 36, 36]
model_path = ['%s-%s' % (x, y) for x, y in zip(prefix, epoch)]
PNet = FcnDetector(P_Net, model_path[0])
detectors[0] = PNet
RNet = Detector(R_Net, 24, 1, model_path[1])
detectors[1] = RNet
ONet = Detector(O_Net, 48, 1, model_path[2])
detectors[2] = ONet
mtcnn_detector = MtcnnDetector(detectors=detectors,
min_face_size=min_face_size,
stride=stride,
threshold=thresh,
slide_window=slide_window)
cap = cv2.VideoCapture(camera_id)
num = 0
cur = self.read_feature(path)
while True:
success, frame = cap.read()
thickness = (frame.shape[0] + frame.shape[1]) // 350
if success:
t1 = time.time()
image = np.array(frame)
boxes_c, landmarks = mtcnn_detector.detect(image)
#print(boxes_c)
for i in range(boxes_c.shape[0]):
bbox = boxes_c[i, :4]
#score = boxes_c[i, 4]
cropbbox = [
int(bbox[0]),
int(bbox[1]),
int(bbox[2]),
int(bbox[3])
]
W = -int(cropbbox[0]) + int(cropbbox[2])
H = -int(cropbbox[1]) + int(cropbbox[3])
paddingH = 0.02 * H
paddingW = 0.01 * W
crop_img = frame[int(cropbbox[1] +
paddingH):int(cropbbox[3] - paddingH),
int(cropbbox[0] -
paddingW):int(cropbbox[2] + paddingW)]
image = cv2.resize(crop_img, (96, 96),
interpolation=cv2.INTER_CUBIC)
image = image.astype('float32')
image = image - 127.5
image = image * 0.0078125
f1_emb = self.get_feature(image)
f1 = f1_emb.reshape(256)
#计算距离
d1 = 100
show_name = ''
embed1 = sklearn.preprocessing.normalize(f1_emb)
for n, v in cur.items():
v = np.array(v)
v_emb = v.reshape(1, 256)
embed2 = sklearn.preprocessing.normalize(v_emb)
diff = np.subtract(embed1, embed2)
dist = np.sum(np.square(diff), 1)
d = np.dot(
v, f1) / (np.linalg.norm(v) * np.linalg.norm(f1))
print(
"name: %s total cosin distance %f and Euclidean distance %f"
% (n, d, dist))
if dist < d1:
d1 = dist
show_name = str(n)
else:
pass
#print(show_name)
t2 = time.time()
delta_t = t2 - t1
text_start = (max(int(cropbbox[0]),
10), max(int(cropbbox[1]), 10))
cv2.putText(frame, show_name, text_start,
cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 255), 1)
cv2.putText(frame, "time cost:" + '%.04f' % delta_t,
(10, 20), cv2.FONT_HERSHEY_SIMPLEX, 1,
(255, 0, 255), 2)
# rectangle for face area
for i in range(thickness):
start = (int(cropbbox[0]) + i, int(cropbbox[1]) + i)
end = (int(cropbbox[2] - i), int(cropbbox[3]) - i)
frame = cv2.rectangle(frame, start, end, (0, 255, 0),
1)
# display the landmarks
for i in range(landmarks.shape[0]):
for j in range(len(landmarks[i]) // 2):
cv2.circle(frame,
(int(landmarks[i][2 * j]),
int(int(landmarks[i][2 * j + 1]))), 2,
(0, 0, 255))
num = num + 1
cv2.imshow("Camera", frame)
k = cv2.waitKey(10)
# 如果输入q则退出循环
if (k & 0xFF == ord('q') or count == num):
break
else:
print('device not find')
break
cap.release()
cv2.destroyAllWindows()
