def main():
global colours, img_size
args = parse_args()
videos_dir = args.videos_dir
output_path = args.output_path
no_display = args.no_display
detect_interval = args.detect_interval # you need to keep a balance between performance and fluency
margin = args.margin # if the face is big in your video ,you can set it bigger for tracking easiler
scale_rate = args.scale_rate # if set it smaller will make input frames smaller
show_rate = args.show_rate # if set it smaller will dispaly smaller frames
face_score_threshold = args.face_score_threshold
mkdir(output_path)
# for display
if not no_display:
colours = np.random.rand(32, 3)
# init tracker
tracker = Sort() # create instance of the SORT tracker
logger.info('Start track and extract......')
with tf.Graph().as_default():
with tf.Session(
config=tf.ConfigProto(gpu_options=tf.GPUOptions(
allow_growth=True),
log_device_placement=False)) as sess:
pnet, rnet, onet = detect_face.create_mtcnn(
sess, os.path.join(project_dir, "align"))
minsize = 40 # minimum size of face for mtcnn to detect
threshold = [0.6, 0.7, 0.7] # three steps's threshold
factor = 0.709 # scale factor
for filename in os.listdir(videos_dir):
logger.info('All files:{}'.format(filename))
for filename in os.listdir(videos_dir):
suffix = filename.split('.')[1]
if suffix != 'mp4' and suffix != 'avi': # you can specify more video formats if you need
continue
video_name = os.path.join(videos_dir, filename)
directoryname = os.path.join(output_path,
filename.split('.')[0])
logger.info('Video_name:{}'.format(video_name))
#cam = cv2.VideoCapture(video_name)
cam = cv2.VideoCapture(0)
c = 0
while True:
final_faces = []
addtional_attribute_list = []
ret, frame = cam.read()
if not ret:
logger.warning("ret false")
break
if frame is None:
logger.warning("frame drop")
break
frame = cv2.resize(frame, (0, 0),
fx=scale_rate,
fy=scale_rate)
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
print('shape of gray')
print(gray.shape)
r_g_b_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
if c % detect_interval == 0:
img_size = np.asarray(frame.shape)[0:2]
mtcnn_starttime = time()
faces, points = detect_face.detect_face(
r_g_b_frame, minsize, pnet, rnet, onet, threshold,
factor)
logger.info(
"MTCNN detect face cost time : {} s".format(
round(time() - mtcnn_starttime,
3))) # mtcnn detect ,slow
face_sums = faces.shape[0]
if face_sums > 0:
face_list = []
for i, item in enumerate(faces):
score = round(faces[i, 4], 6)
if score > face_score_threshold:
det = np.squeeze(faces[i, 0:4])
# face rectangle
det[0] = np.maximum(det[0] - margin, 0)
det[1] = np.maximum(det[1] - margin, 0)
det[2] = np.minimum(
det[2] + margin, img_size[1])
det[3] = np.minimum(
det[3] + margin, img_size[0])
face_list.append(item)
# face cropped
bb = np.array(det, dtype=np.int32)
# use 5 face landmarks to judge the face is front or side
squeeze_points = np.squeeze(points[:, i])
tolist = squeeze_points.tolist()
facial_landmarks = []
for j in range(5):
item = [tolist[j], tolist[(j + 5)]]
facial_landmarks.append(item)
if args.face_landmarks:
for (x, y) in facial_landmarks:
cv2.circle(frame, (int(x), int(y)),
3, (0, 255, 0), -1)
cropped = frame[bb[1]:bb[3],
bb[0]:bb[2], :].copy()
dist_rate, high_ratio_variance, width_rate = judge_side_face(
np.array(facial_landmarks))
# face addtional attribute(index 0:face score; index 1:0 represents front face and 1 for side face )
item_list = [
cropped, score, dist_rate,
high_ratio_variance, width_rate
]
addtional_attribute_list.append(item_list)
final_faces = np.array(face_list)
emotion = 'Happy'
face_detection = cv2.CascadeClassifier(
'haarcascade_frontalface_default.xml')
emotion_classifier = load_model(
'models/_mini_XCEPTION.106-0.65.hdf5', compile=False)
EMOTIONS = [
"angry", "disgust", "scared", "happy", "sad",
"surprised", "neutral"
]
frontal_faces = face_detection.detectMultiScale(
gray,
scaleFactor=1.1,
minNeighbors=5,
minSize=(30, 30),
flags=cv2.CASCADE_SCALE_IMAGE)
if len(frontal_faces) > 0:
frontal_faces = sorted(frontal_faces,
reverse=True,
key=lambda x: (x[2] - x[0]) *
(x[3] - x[1]))[0]
(fX, fY, fW, fH) = frontal_faces
roi = gray[fY:fY + fH, fX:fX + fW]
roi = cv2.resize(roi, (48, 48))
roi = roi.astype("float") / 255.0
roi = img_to_array(roi)
roi = np.expand_dims(roi, axis=0)
np.reshape(roi, (48, 48, 1))
print(roi.shape)
preds = emotion_classifier.predict(roi)[0]
emotion_probability = np.max(preds)
label = EMOTIONS[preds.argmax()]
trackers = tracker.update(final_faces, img_size,
directoryname,
addtional_attribute_list,
detect_interval)
c += 1
emoTracker = ''
print(trackers)
for d in trackers:
if not no_display:
d = d.astype(np.int32)
cv2.rectangle(frame, (d[0], d[1]), (d[2], d[3]),
colours[d[4] % 32, :] * 255, 3)
if final_faces != []:
print('ID %d Detect' % (d[4]))
if label != emoTracker:
emoTracker = label
cv2.putText(
frame, 'ID : %d DETECT, EMOTION : %s' %
((d[4]), emoTracker),
(d[0] - 10, d[1] - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.4,
colours[d[4] % 32, :] * 255, 2)
else:
cv2.putText(frame, 'ID : %d' % (d[4]),
(d[0] - 10, d[1] - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.75,
colours[d[4] % 32, :] * 255, 2)
if not no_display:
frame = cv2.resize(frame, (0, 0),
fx=show_rate,
fy=show_rate)
cv2.imshow("Frame", frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
def detectExtract(a):
global colours, img_size
args = parse_args()
videos_dir = args.videos_dir
output_path = args.output_path
no_display = args.no_display
detect_interval = args.detect_interval # you need to keep a balance between performance and fluency
margin = args.margin # if the face is big in your video ,you can set it bigger for tracking easiler
scale_rate = args.scale_rate # if set it smaller will make input frames smaller
show_rate = args.show_rate # if set it smaller will dispaly smaller frames
face_score_threshold = args.face_score_threshold
mkdir(output_path)
# for display
if not no_display:
colours = np.random.rand(32, 3)
# init tracker
tracker = Sort() # create instance of the SORT tracker
logger.info('Start track and extract......')
with tf.Graph().as_default():
with tf.Session(
config=tf.ConfigProto(gpu_options=tf.GPUOptions(
allow_growth=True),
log_device_placement=True)) as sess:
pnet, rnet, onet = detect_face.create_mtcnn(
sess, os.path.join(project_dir, "align"))
minsize = 80 # minimum size of face for mtcnn to detect
threshold = [0.6, 0.7, 0.7] # three steps's threshold
factor = 0.709 # scale factor
for filename in os.listdir(videos_dir):
logger.info('All files:{}'.format(filename))
for filename in os.listdir(videos_dir):
suffix = filename.split('.')[1]
if suffix != 'mp4' and suffix != 'avi': # you can specify more video formats if you need
continue
video_name = os.path.join(videos_dir, filename)
directoryname = os.path.join(output_path,
filename.split('.')[0])
logger.info('Video_name:{}'.format(video_name))
cam = cv2.VideoCapture(video_name)
c = 0
while True:
final_faces = []
addtional_attribute_list = []
ret, frame = cam.read()
if not ret:
logger.warning("ret false")
break
if frame is None:
logger.warning("frame drop")
break
# frame = cv2.resize(frame, (0, 0), fx=scale_rate, fy=scale_rate)
r_g_b_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
if c % detect_interval == 0:
img_size = np.asarray(frame.shape)[0:2]
mtcnn_starttime = time()
faces, points = detect_face.detect_face(
r_g_b_frame, minsize, pnet, rnet, onet, threshold,
factor)
logger.info(
"MTCNN detect face cost time : {} s".format(
round(time() - mtcnn_starttime,
3))) # mtcnn detect ,slow
face_sums = faces.shape[0]
if face_sums > 0:
face_list = []
for i, item in enumerate(faces):
score = round(faces[i, 4], 6)
if score > face_score_threshold:
det = np.squeeze(faces[i, 0:4])
# face rectangle
det[0] = np.maximum(det[0] - margin, 0)
det[1] = np.maximum(det[1] - margin, 0)
det[2] = np.minimum(
det[2] + margin, img_size[1])
det[3] = np.minimum(
det[3] + margin, img_size[0])
face_list.append(item)
# face cropped
bb = np.array(det, dtype=np.int32)
# use 5 face landmarks to judge the face is front or side
squeeze_points = np.squeeze(points[:, i])
tolist = squeeze_points.tolist()
facial_landmarks = []
for j in range(5):
item = [tolist[j], tolist[(j + 5)]]
facial_landmarks.append(item)
if args.face_landmarks:
for (x, y) in facial_landmarks:
cv2.circle(frame, (int(x), int(y)),
3, (0, 255, 0), -1)
cropped = frame[bb[1]:bb[3],
bb[0]:bb[2], :].copy()
dist_rate, high_ratio_variance, width_rate = judge_side_face(
np.array(facial_landmarks))
# face addtional attribute(index 0:face score; index 1:0 represents front face and 1 for side face )
item_list = [
cropped, score, dist_rate,
high_ratio_variance, width_rate
]
addtional_attribute_list.append(item_list)
final_faces = np.array(face_list)
trackers = tracker.update(final_faces, img_size,
directoryname,
addtional_attribute_list,
detect_interval)
c += 1
for d in trackers:
if not no_display:
d = d.astype(np.int32)
cv2.rectangle(frame, (d[0], d[1]), (d[2], d[3]),
colours[d[4] % 32, :] * 255, 3)
if final_faces != []:
cv2.putText(frame, 'ID : %d DETECT' % (d[4]),
(d[0] - 10, d[1] - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.75,
colours[d[4] % 32, :] * 255, 2)
cv2.putText(frame, 'DETECTOR', (5, 45),
cv2.FONT_HERSHEY_SIMPLEX, 0.75,
(1, 1, 1), 2)
else:
cv2.putText(frame, 'ID : %d' % (d[4]),
(d[0] - 10, d[1] - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.75,
colours[d[4] % 32, :] * 255, 2)
if not no_display:
frame = cv2.resize(frame, (0, 0),
fx=show_rate,
fy=show_rate)
cv2.imshow("Frame", frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
def main():
global colours, img_size
args = parse_args()
# videos_dir = args.videos_dir
output_path = args.output_path
no_display = args.no_display
detect_interval = args.detect_interval # you need to keep a balance between performance and fluency
margin = args.margin # if the face is big in your video ,you can set it bigger for tracking easiler
scale_rate = args.scale_rate # if set it smaller will make input frames smaller
show_rate = args.show_rate # if set it smaller will dispaly smaller frames
face_score_threshold = args.face_score_threshold
mkdir(output_path)
# for display
if not no_display:
colours = np.random.rand(32, 3)
# init tracker
tracker = Sort() # create instance of the SORT tracker
logger.info('Start track and extract......')
with tf.Graph().as_default():
with tf.Session(config=tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True),
log_device_placement=False)) as sess:
pnet, rnet, onet = detect_face.create_mtcnn(sess, os.path.join(project_dir, "align"))
minsize = 40 # minimum size of face for mtcnn to detect
threshold = [0.6, 0.7, 0.7] # three steps's threshold
factor = 0.709 # scale factor
# for filename in os.listdir(videos_dir):
# logger.info('All files:{}'.format(filename))
# for filename in os.listdir(videos_dir):
# suffix = filename.split('.')[1]
# if suffix != 'mp4' and suffix != 'avi': # you can specify more video formats if you need
# continue
# video_name = os.path.join(videos_dir, filename)
# directoryname = os.path.join(output_path, filename.split('.')[0])
# logger.info('Video_name:{}'.format(video_name))
directoryname=output_path
cam = cv2.VideoCapture(0) #video_name --> 0 indicating feed from camera
c = 0
while True:
final_faces = []
addtional_attribute_list = []
ret, frame = cam.read()
if not ret:
logger.warning("ret false")
break
if frame is None:
logger.warning("frame drop")
break
frame = cv2.resize(frame, (0, 0), fx=scale_rate, fy=scale_rate)
r_g_b_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
if c % detect_interval == 0:
img_size = np.asarray(frame.shape)[0:2]
mtcnn_starttime = time()
faces, points = detect_face.detect_face(r_g_b_frame, minsize, pnet, rnet, onet, threshold,
factor)
logger.info("MTCNN detect face cost time : {} s".format(
round(time() - mtcnn_starttime, 3))) # mtcnn detect ,slow
face_sums = faces.shape[0]
# if len(face_boxes) > 0:
# for i in range(len(face_boxes)):
# box = face_boxes[i]
# cropped_face = frame[box[0]:box[2], box[1]:box[3], :]
# cropped_face = cv2.resize(cropped_face, (160, 160), interpolation=cv2.INTER_AREA)
# feature = face_recognition.recognize(cropped_face)
# (name , proba , conf) = face_classfier.classify(feature)
#
# cv2.rectangle(frame, (box[1], box[0]), (box[3], box[2]), (0, 255, 0), 2)
# name = name+":%"+ proba
# # plot result idx under box
# text_x = box[1]
# text_y = box[2] + 20
# if conf:
# cv2.putText(frame, name, (text_x, text_y), cv2.FONT_HERSHEY_COMPLEX_SMALL,
# 1, (0, 0, 255), thickness=1, lineType=2)
if face_sums > 0:
face_list = []
for i, item in enumerate(faces):
score = round(faces[i, 4], 2)
if score > face_score_threshold:
det = np.squeeze(faces[i, 0:4])
# face rectangle
det[0] = np.maximum(det[0] - margin, 0)
det[1] = np.maximum(det[1] - margin, 0)
det[2] = np.minimum(det[2] + margin, img_size[1])
det[3] = np.minimum(det[3] + margin, img_size[0])
face_list.append(item)
# face cropped
bb = np.array(det, dtype=np.int32)
# use 5 face landmarks to judge the face is front or side
squeeze_points = np.squeeze(points[:, i])
tolist = squeeze_points.tolist()
facial_landmarks = []
for j in range(5):
item = [tolist[j], tolist[(j + 5)]]
facial_landmarks.append(item)
if args.face_landmarks:
for (x, y) in facial_landmarks:
cv2.circle(frame, (int(x), int(y)), 3, (0, 255, 0), -1)
cropped = frame[bb[1]:bb[3], bb[0]:bb[2], :].copy()
cropped_face = cv2.resize(cropped, (160, 160), interpolation=cv2.INTER_AREA)
feature = face_recognition.recognize(cropped_face)
(name , proba , conf) = face_classfier.classify(feature)
#cv2.rectangle(frame, (bb[0], bb[1]), (bb[2], bb[3]), (0, 255, 0), 2)
name = name+":%"+ proba
# plot result idx under box
text_x = bb[0] - 10
text_y = bb[1] - 10
if conf:
cv2.putText(frame, name, (text_x, text_y), cv2.FONT_HERSHEY_COMPLEX_SMALL,
1, (0, 0, 255), thickness=1, lineType=2)
dist_rate, high_ratio_variance, width_rate = judge_side_face(
np.array(facial_landmarks))
# face addtional attribute(index 0:face score; index 1:0 represents front face and 1 for side face )
item_list = [cropped, score, dist_rate, high_ratio_variance, width_rate]
addtional_attribute_list.append(item_list)
final_faces = np.array(face_list)
trackers = tracker.update(final_faces, img_size, directoryname, addtional_attribute_list, detect_interval)
c += 1
for d in trackers:
# print("D values:",d)
if not no_display:
d = d.astype(np.int32)
cv2.rectangle(frame, (d[0], d[1]), (d[2], d[3]), colours[d[4] % 32, :] * 255, 3)
if final_faces != []:
# cv2.putText(frame, 'ID : %d DETECT' % (d[4]), (d[0] - 10, d[1] - 10),
# cv2.FONT_HERSHEY_SIMPLEX,
# 0.75,
# colours[d[4] % 32, :] * 255, 2)
cv2.putText(frame, 'DETECTOR', (5, 45), cv2.FONT_HERSHEY_SIMPLEX, 0.75,
(1, 1, 1), 2)
else:
cv2.putText(frame, 'ID : %d' % (d[4]), (d[0] - 10, d[1] - 10), cv2.FONT_HERSHEY_SIMPLEX,
0.75,
colours[d[4] % 32, :] * 255, 2)
if not no_display:
frame = cv2.resize(frame, (0, 0), fx=show_rate, fy=show_rate)
cv2.imshow("Frame", frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
# print("Debugging state:...")
# print("faces size",faces.shape)
# print("faces content",faces)
# print("RGB frame")
# print(r_g_b_frame)
# print("RGB frame size",r_g_b_frame.shape)
# print("face list shape",len(face_list))
# print(face_list)
# print("Det",det)
#
cam.release()
cv2.destroyAllWindows()
# modify the default parameters of np.load
np.load.__defaults__=(None, False, True, 'ASCII')
for box in decision_boxes:
a_bbox = []
for ii in range(0, 4):
a_bbox.append(box['bbox']['points'][ii])
a_bbox.append(0.99)
a_bbox.append(box['bbox']['score'])
if box['bbox']['label'] in vehicle:
a_bbox.append(
float(vehicle.index(box['bbox']['label'])))
else:
assert ('Vehicle TYPE ERROR')
sort_box.append(a_bbox)
sort_box = torch.tensor(sort_box).cuda()
tracked_objects = mot_tracker.update(sort_box.cpu())
for x1, y1, x2, y2, obj_id, cls_pred in tracked_objects:
cls = vehicle[int(cls_pred)]
#img_result = cv2.UMat(img_result)
#cv2.putText(img_result, cls + ":" + str(int(obj_id)),(int(x1+30), int(y1 - 10)), cv2.FONT_HERSHEY_SIMPLEX, 1, (225,225,0), 3)
############################ SORT ######################################
# save img_result
# 偵測違規
if violation_frame == -1:
for i in range(len(decision_boxes)):
if decision_boxes[i]['decision'] != 'pass':
#建立違規資料夾
directory = './violation/' + str(c)
createFolder(directory)
def main(called_from, stream_name, vid_strt_time, video_folder):
global colours, img_size
args = parse_args(video_folder)
videos_dir = args.videos_dir
output_path = args.output_path
no_display = args.no_display
detect_interval = args.detect_interval # you need to keep a balance between performance and fluency
margin = args.margin # if the face is big in your video ,you can set it bigger for tracking easiler
scale_rate = args.scale_rate # if set it smaller will make input frames smaller
show_rate = args.show_rate # if set it smaller will dispaly smaller frames
face_score_threshold = args.face_score_threshold
mkdir(output_path)
# for display
if not no_display:
colours = np.random.rand(100, 3)
# init tracker
tracker = Sort() # create instance of the SORT tracker
logger.info('Start track and extract......')
with tf.Graph().as_default():
with tf.Session(
config=tf.ConfigProto(gpu_options=tf.GPUOptions(
allow_growth=True),
log_device_placement=False)) as sess:
pnet, rnet, onet = detect_face.create_mtcnn(
sess, os.path.join(project_dir, "align"))
minsize = 70 # minimum size of face for mtcnn to detect
threshold = [0.6, 0.7, 0.7] # three steps's threshold
factor = 0.709 # scale factor
for filename in os.listdir(videos_dir):
logger.info('All files:{}'.format(filename))
for filename in os.listdir(videos_dir):
suffix = filename.split('.')[1]
if suffix != 'mp4' and suffix != 'avi' and suffix != 'flv': # you can specify more video formats if you need
continue
video_name = os.path.join(videos_dir, filename)
directoryname = os.path.join(output_path,
filename.split('.')[0])
logger.info('Video_name:{}'.format(video_name))
cam = cv2.VideoCapture(video_name)
c = 0
while True:
final_faces = []
addtional_attribute_list = []
ret, frame = cam.read()
start_time.append(time.time())
# print(time.time())
if not ret:
logger.warning("ret false")
break
if frame is None:
logger.warning("frame drop")
break
frame = cv2.resize(frame, (0, 0),
fx=scale_rate,
fy=scale_rate)
r_g_b_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
if c % detect_interval == 0:
img_size = np.asarray(frame.shape)[0:2]
mtcnn_starttime = time.time()
faces, points = detect_face.detect_face(
r_g_b_frame, minsize, pnet, rnet, onet, threshold,
factor)
logger.info(
"MTCNN detect face cost time : {} s".format(
round(time.time() - mtcnn_starttime,
3))) # mtcnn detect ,slow
face_sums = faces.shape[0]
if face_sums > 0:
face_list = []
for i, item in enumerate(faces):
score = round(faces[i, 4], 6)
if score > face_score_threshold:
det = np.squeeze(faces[i, 0:4])
# face rectangle
det[0] = np.maximum(det[0] - margin, 0)
det[1] = np.maximum(det[1] - margin, 0)
det[2] = np.minimum(
det[2] + margin, img_size[1])
det[3] = np.minimum(
det[3] + margin, img_size[0])
face_list.append(item)
# face cropped
bb = np.array(det, dtype=np.int32)
# use 5 face landmarks to judge the face is front or side
squeeze_points = np.squeeze(points[:, i])
tolist = squeeze_points.tolist()
facial_landmarks = []
for j in range(5):
item = [tolist[j], tolist[(j + 5)]]
facial_landmarks.append(item)
if args.face_landmarks:
for (x, y) in facial_landmarks:
cv2.circle(frame, (int(x), int(y)),
3, (0, 255, 0), -1)
cropped = frame[bb[1]:bb[3],
bb[0]:bb[2], :].copy()
dist_rate, high_ratio_variance, width_rate = judge_side_face(
np.array(facial_landmarks))
# face addtional attribute(index 0:face score; index 1:0 represents front face and 1 for side face )
item_list = [
cropped, score, dist_rate,
high_ratio_variance, width_rate
]
addtional_attribute_list.append(item_list)
final_faces = np.array(face_list)
trackers = tracker.update(final_faces, img_size,
directoryname,
addtional_attribute_list,
detect_interval)
c += 1
for d in trackers:
d = d.astype(np.int32)
g.append(str(start_time[4]))
if not no_display:
d = d.astype(np.int32)
if final_faces != []:
try:
os.mkdir('DB/' + called_from + str(d[4]))
first_time = round(start_time[0])
face_time = round(time.time())
entryTime = face_time - first_time
dr = parser.parse(vid_strt_time)
a = dr + timedelta(seconds=entryTime)
real_enter_time = a.strftime("%H:%M:%S")
f = open("DB/entryTime.txt", "a")
f.write(called_from + str(d[4]) + ',' +
real_enter_time + "\n")
f.close()
f = open("DB/stream_of_folder.txt", "a")
f.write(stream_name + ',' + called_from +
str(d[4]) + "\n")
f.close()
profile_id = called_from + str(d[4])
except Exception:
print("The folder already exists!!")
image_path = "DB/" + called_from + str(
d[4]) + '/' + str(
random.randint(32, 12141212)) + ".jpg"
d[0] = d[0]
d[1] = d[1]
save(frame, d, image_path)
original_height = (d[3] - d[1])
if not no_display:
frame = cv2.resize(frame, (0, 0),
fx=show_rate,
fy=show_rate)
def main():
global colours, img_size
args = parse_args()
#一个或多个视频存放路径
root_dir = args.root_dir
#采集并裁剪人脸保存路径
output_path = args.output_path
display = args.display
mkdir(output_path)
if display:
colours = np.random.rand(32, 3)
#初始化tracker
tracker = Sort()
logger.info('start track and extract......')
with tf.Graph().as_default():
with tf.Session(
config=tf.ConfigProto(gpu_options=tf.GPUOptions(
allow_growth=True),
log_device_placement=False)) as sess:
pnet, rnet, onet = detect_face.create_mtcnn(sess, "align")
margin = 50
minsize = 60
threshold = [0.6, 0.7, 0.7]
factor = 0.709
frame_interval = 1 # 每多少帧检测一次，默认3
scale_rate = 1 #对输入frame进行resize
show_rate = 1 #对输出frame进行resize
for filename in os.listdir(root_dir):
logger.info('all files:{}'.format(filename))
#遍历所有mp4格式的视频文件
for filename in os.listdir(root_dir):
if filename.split('.')[1] != 'mp4':
continue
video_name = os.path.join(root_dir, filename)
directoryname = os.path.join(output_path,
filename.split('.')[0])
logger.info('video_name:{}'.format(video_name))
cam = cv2.VideoCapture(video_name)
c = 0
while True:
final_faces = []
addtional_attribute_list = []
ret, frame = cam.read()
if not ret:
logger.warning("ret false")
break
if frame is None:
logger.warning("frame drop")
break
frame = cv2.resize(frame, (0, 0),
fx=scale_rate,
fy=scale_rate)
r_g_b_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
if c % frame_interval == 0:
img_size = np.asarray(frame.shape)[0:2]
faces, points = detect_face.detect_face(
r_g_b_frame, minsize, pnet, rnet, onet, threshold,
factor)
face_sums = faces.shape[0]
if face_sums > 0:
face_list = []
for i, item in enumerate(faces):
#检测人脸的置信度
f = round(faces[i, 4], 6)
if f > 0.95:
det = np.squeeze(faces[i, 0:4])
# face rectangle
det[0] = np.maximum(det[0] - margin, 0)
det[1] = np.maximum(det[1] - margin, 0)
det[2] = np.minimum(
det[2] + margin, img_size[1])
det[3] = np.minimum(
det[3] + margin, img_size[0])
face_list.append(item)
# face cropped
bb = np.array(det, dtype=np.int32)
frame_copy = frame.copy()
cropped = frame_copy[bb[1]:bb[3],
bb[0]:bb[2], :]
# use 5 face landmarks to judge the face is front or side
squeeze_points = np.squeeze(points[:, i])
tolist = squeeze_points.tolist()
facial_landmarks = []
for j in range(5):
item = [tolist[j], tolist[(j + 5)]]
facial_landmarks.append(item)
#可视化关键点
if args.face_landmarks:
for (x, y) in facial_landmarks:
cv2.circle(frame_copy,
(int(x), int(y)), 3,
(0, 0, 255), 2)
#计算三个重要值用于判断正脸
dist_rate, high_ratio_variance, width_rate = judge_side_face(
np.array(facial_landmarks))
# face addtional attribute(index 0:face score; index 1:0 represents front face and 1 for side face )
item_list = [
cropped, faces[i, 4], dist_rate,
high_ratio_variance, width_rate
]
addtional_attribute_list.append(item_list)
final_faces = np.array(face_list)
trackers = tracker.update(final_faces, img_size,
directoryname,
addtional_attribute_list,
r_g_b_frame)
c += 1
for d in trackers:
if display:
d = d.astype(np.int32)
cv2.rectangle(frame, (d[0], d[1]), (d[2], d[3]),
colours[d[4] % 32, :] * 255, 5)
cv2.putText(frame, 'ID : %d' % (d[4]),
(d[0] - 10, d[1] - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.75,
colours[d[4] % 32, :] * 255, 2)
if display:
frame = cv2.resize(frame, (0, 0),
fx=show_rate,
fy=show_rate)
cv2.imshow("Frame", frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
def main():
# 参数设置
global colours, img_size
args = parse_args()
videos_dir = args.videos_dir
output_path = args.output_path
no_display = args.no_display
detect_interval = args.detect_interval # you need to keep a balance between performance and fluency
margin = args.margin # if the face is big in your video ,you can set it bigger for tracking easiler
scale_rate = args.scale_rate # if set it smaller will make input frames smaller
show_rate = args.show_rate # if set it smaller will dispaly smaller frames
face_score_threshold = args.face_score_threshold
# 创建输出目录
mkdir(output_path)
# for display
if not no_display:
colours = np.random.rand(32, 3)
# 初始化SORT追踪器
tracker = Sort()
logger.info('开始追踪人脸并且提取......')
with tf.Graph().as_default():
with tf.Session(
config=tf.ConfigProto(gpu_options=tf.GPUOptions(
allow_growth=True),
log_device_placement=False)) as sess:
# 创建人脸检测MTCNN的三个网络
pnet, rnet, onet = detect_face.create_mtcnn(
sess, os.path.join(project_dir, "align"))
# 设置MTCNN能检测到的最小size的脸
minsize = 40 # 40
# 三个网络的人脸阈值
threshold = [0.6, 0.7, 0.7]
# 尺度因子
factor = 0.709 # scale factor
for filename in os.listdir(videos_dir):
logger.info('所需要检测并追踪人脸的文件:{}'.format(filename))
for filename in os.listdir(videos_dir):
suffix = filename.split('.')[1]
if suffix != 'mp4' and suffix != 'avi': # 可以设置过滤不同的文件格式
continue
video_name = os.path.join(videos_dir, filename)
directoryname = os.path.join(output_path,
filename.split('.')[0])
logger.info('当前检测的视频文件:{}'.format(video_name))
# 创建视频帧扑捉器，参数为文件路径或者0代表本地摄像头
cam = cv2.VideoCapture(video_name)
ret, frame = cam.read()
# 创建视频保存器
size = (frame.shape[1], frame.shape[0])
video_writer = cv2.VideoWriter("output/minisize_40_07.avi",
cv2.VideoWriter_fourcc(*'DIVX'),
24, size)
# 记录已经读取到的帧数
c = 0
while ret: # 循环读取整个视频文件的帧
final_faces = []
addtional_attribute_list = []
# ret表示有没有读取到帧true\false
ret, frame = cam.read()
if not ret:
logger.warning("ret false,没有读取到任何帧！")
break
if frame is None:
logger.warning("frame drop，读取到的帧为空！")
break
# cv2.resize(src, dsize, dst=None, fx=None, fy=None, interpolation=None)
# scr: 原图
# dsize：输出图像尺寸
# fx: 沿水平轴的比例因子
# fy: 沿垂直轴的比例因子
# interpolation：插值方法
frame = cv2.resize(frame, (0, 0),
fx=scale_rate,
fy=scale_rate)
# 将读取到的帧转换为rgb色彩空间
r_g_b_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
# 每隔detect_interval个帧检测跟踪一次
if c % detect_interval == 0:
img_size = np.asarray(frame.shape)[0:2]
# mtcnn开始检测的时间戳，用于计时
mtcnn_starttime = time()
# 使用MTCNN实施检测，返回检测到的所有可能为face的(x1,y1,x2,y2,score)和对应的landmark
faces, points = detect_face.detect_face(
r_g_b_frame, minsize, pnet, rnet, onet, threshold,
factor)
logger.info("MTCNN检测人脸花费时间: {} s".format(
round(time() - mtcnn_starttime, 3)))
face_sums = faces.shape[0]
# 如果检测到的脸大于0，将每张脸剪切出来，并进行侧脸评判
if face_sums > 0:
face_list = []
for i, item in enumerate(faces):
score = round(faces[i, 4], 6)
if score > face_score_threshold:
det = np.squeeze(faces[i, 0:4])
# face rectangle
det[0] = np.maximum(det[0] - margin, 0)
det[1] = np.maximum(det[1] - margin, 0)
det[2] = np.minimum(
det[2] + margin, img_size[1])
det[3] = np.minimum(
det[3] + margin, img_size[0])
face_list.append(item)
# face cropped
bb = np.array(det, dtype=np.int32)
# bounding box 距离fram左边的距离
bb_left = det[0]
# bounding box 距离fram上边的距离
bb_top = det[1]
# use 5 face landmarks to judge the face is front or side
squeeze_points = np.squeeze(points[:, i])
tolist = squeeze_points.tolist()
facial_landmarks = []
facial_landmarks_crap = []
for j in range(5):
item = [tolist[j], tolist[(j + 5)]]
facial_landmarks.append(item)
item_crap = [
tolist[j] - bb_left,
tolist[(j + 5)] - bb_top
]
facial_landmarks_crap.append(item_crap)
# 为脸画出标定点
if args.face_landmarks:
for (x, y) in facial_landmarks:
cv2.circle(frame, (int(x), int(y)),
3, (0, 255, 0), -1)
# 从帧中复制出人脸
cropped = frame[bb[1]:bb[3],
bb[0]:bb[2], :].copy()
# 计算五个标定点的高宽比例，高均方差，宽均方差
dist_rate, high_ratio_variance, width_rate = judge_side_face(
np.array(facial_landmarks))
# 整个视频中脸部附加属性addtional_attribute (index 0:face score; index 1:0代表正脸，1代表侧脸 )
item_list = [
cropped, score, dist_rate,
high_ratio_variance, width_rate,
facial_landmarks_crap
] ##### 添加参数facial_landmarks
addtional_attribute_list.append(item_list)
# 获取的每帧的face的score大于阈值的face
final_faces = np.array(face_list)
trackers = tracker.update(final_faces, img_size,
directoryname,
addtional_attribute_list,
detect_interval)
c += 1
for d in trackers:
if not no_display:
d = d.astype(np.int32)
cv2.rectangle(frame, (d[0], d[1]), (d[2], d[3]),
colours[d[4] % 32, :] * 255, 3)
if final_faces != []:
cv2.putText(frame, 'ID : %d DETECT' % (d[4]),
(d[0] - 10, d[1] - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.75,
colours[d[4] % 32, :] * 255, 2)
cv2.putText(frame, 'DETECTOR', (5, 45),
cv2.FONT_HERSHEY_SIMPLEX, 0.75,
(1, 1, 1), 2)
else:
cv2.putText(frame, 'ID : %d' % (d[4]),
(d[0] - 10, d[1] - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.75,
colours[d[4] % 32, :] * 255, 2)
# video_writer.write(frame)
data = {"type_code": type_code, "area_id": area_id}
file = {
"scene_img":
("scene_img.jpg", cv2.imencode(".jpg",
frame)[1].tobytes(),
"image/jpg")
}
res = requests.post(url=post_scene_url,
files=file,
data=data)
logger.info("检测场景发送到服务器，响应码：{}".format(res))
if not no_display:
frame = cv2.resize(frame, (0, 0),
fx=show_rate,
fy=show_rate)
cv2.imshow("Frame", frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break