# 剔除掉多个人脸、大角度侧脸（鼻子不在两个眼之间）、部分人脸框在画面外、人脸像素低于80*80的

with mp_face_detection.FaceDetection(

model_selection=1, min_detection_confidence=0.5) as face_detection:

results = face_detection.process(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))

if not results.detections or len(results.detections) > 1:

return -1, None

rect = results.detections[0].location_data.relative_bounding_box

out_rect = [rect.xmin, rect.xmin + rect.width, rect.ymin, rect.ymin + rect.height]

nose_ = mp_face_detection.get_key_point(

results.detections[0], mp_face_detection.FaceKeyPoint.NOSE_TIP)

l_eye_ = mp_face_detection.get_key_point(

results.detections[0], mp_face_detection.FaceKeyPoint.LEFT_EYE)

r_eye_ = mp_face_detection.get_key_point(

results.detections[0], mp_face_detection.FaceKeyPoint.RIGHT_EYE)

# print(nose_, l_eye_, r_eye_)

if nose_.x > l_eye_.x or nose_.x < r_eye_.x:

return -2, out_rect

h, w = frame.shape[:2]

# print(frame.shape)

if rect.xmin < 0 or rect.ymin < 0 or rect.xmin + rect.width > w or rect.ymin + rect.height > h:

return -3, out_rect

if rect.width * w < 100 or rect.height * h < 100:

return -4, out_rect

x_min = min(face0[0], face1[0])

x_max = max(face0[1], face1[1])

y_min = min(face0[2], face1[2])

y_max = max(face0[3], face1[3])

tmp0 = ((face0[1] - face0[0]) * (face0[3] - face0[2])) / ((x_max - x_min) * (y_max - y_min))

tmp1 = ((face1[1] - face1[0]) * (face1[3] - face1[2])) / ((x_max - x_min) * (y_max - y_min))

with mp_face_mesh.FaceMesh(

static_image_mode=True,

max_num_faces=1,

refine_landmarks=True,

min_detection_confidence=0.5) as face_mesh:

results = face_mesh.process(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))

pts_3d = np.zeros([478, 3])

if not results.multi_face_landmarks:

print("****** WARNING! No face detected! ******")

else:

image_height, image_width = frame.shape[:2]

for face_landmarks in results.multi_face_landmarks:

for index_, i in enumerate(face_landmarks.landmark):

x_px = min(math.floor(i.x * image_width), image_width - 1)

y_px = min(math.floor(i.y * image_height), image_height - 1)

z_px = min(math.floor(i.z * image_width), image_width - 1)

pts_3d[index_] = np.array([x_px, y_px, z_px])

cap = cv2.VideoCapture(video_path)

if not cap.isOpened():

return 0

dir_path = os.path.dirname(video_path)

vid_width = cap.get(cv2.CAP_PROP_FRAME_WIDTH) # 宽度

vid_height = cap.get(cv2.CAP_PROP_FRAME_HEIGHT) # 高度

totalFrames = cap.get(cv2.CAP_PROP_FRAME_COUNT) # 总帧数

totalFrames = int(totalFrames)

pts_3d = np.zeros([totalFrames, 478, 3])

face_rect_list = []

# os.makedirs("../preparation/{}/image".format(model_name))

for frame_index in tqdm.tqdm(range(totalFrames)):

ret, frame = cap.read() # 按帧读取视频

# #到视频结尾时终止

if ret is False:

break

# cv2.imwrite("../preparation/{}/image/{:0>6d}.png".format(model_name, frame_index), frame)

tag_, rect = detect_face(frame)

if frame_index == 0 and tag_ != 1:

print("第一帧人脸检测异常，请剔除掉多个人脸、大角度侧脸（鼻子不在两个眼之间）、部分人脸框在画面外、人脸像素低于80*80")

pts_3d = -1

break

elif tag_ == -1: # 有时候人脸检测会失败，就用上一帧的结果替代这一帧的结果

rect = face_rect_list[-1]

elif tag_ != 1:

print("第{}帧人脸检测异常，请剔除掉多个人脸、大角度侧脸（鼻子不在两个眼之间）、部分人脸框在画面外、人脸像素低于80*80, tag: {}".format(frame_index, tag_))

# exit()

if len(face_rect_list) > 0:

face_area_inter = calc_face_interact(face_rect_list[-1], rect)

# print(frame_index, face_area_inter)

if face_area_inter < 0.6:

print("人脸区域变化幅度太大，请复查，超出值为{}, frame_num: {}".format(face_area_inter, frame_index))

pts_3d = -2

break

face_rect_list.append(rect)

x_min = rect[0] * vid_width

y_min = rect[2] * vid_height

x_max = rect[1] * vid_width

y_max = rect[3] * vid_height

seq_w, seq_h = x_max - x_min, y_max - y_min

x_mid, y_mid = (x_min + x_max) / 2, (y_min + y_max) / 2

# x_min = int(max(0, x_mid - seq_w * 0.65))

# y_min = int(max(0, y_mid - seq_h * 0.4))

# x_max = int(min(vid_width, x_mid + seq_w * 0.65))

# y_max = int(min(vid_height, y_mid + seq_h * 0.8))

crop_size = int(max(seq_w * 1.35, seq_h * 1.35))

x_min = int(max(0, x_mid - crop_size * 0.5))

y_min = int(max(0, y_mid - crop_size * 0.45))

x_max = int(min(vid_width, x_min + crop_size))

y_max = int(min(vid_height, y_min + crop_size))

frame_face = frame[y_min:y_max, x_min:x_max]

# cv2.imshow("s", frame_face)

# cv2.waitKey(20)

frame_kps = detect_face_mesh(frame_face)

pts_3d[frame_index] = frame_kps + np.array([x_min, y_min, 0])

cap.release() # 释放视频对象

cap = cv2.VideoCapture(video_path)

if not cap.isOpened():

return 0

vid_width = cap.get(cv2.CAP_PROP_FRAME_WIDTH) # 宽度

vid_height = cap.get(cv2.CAP_PROP_FRAME_HEIGHT) # 高度

totalFrames = cap.get(cv2.CAP_PROP_FRAME_COUNT) # 总帧数

totalFrames = int(totalFrames)

cap.release()

pts_3d = ExtractFromVideo(video_path)

if type(pts_3d) is np.ndarray and len(pts_3d) == totalFrames:

print("关键点已提取")

else:

print("error in video: {}!!!".format(video_path))

return

video_name = os.path.basename(video_path).split(".")[0]

video_data_path = os.path.join(os.path.dirname(video_path), video_name)

os.makedirs(video_data_path, exist_ok=True)

if export_imgs:

# 计算整个视频中人脸的范围

x_min, y_min, x_max, y_max = np.min(pts_3d[:, :, 0]), np.min(

pts_3d[:, :, 1]), np.max(

pts_3d[:, :, 0]), np.max(pts_3d[:, :, 1])

new_w = int((x_max - x_min) * 0.55) * 2

new_h = int((y_max - y_min) * 0.6) * 2

center_x = int((x_max + x_min) / 2.)

center_y = int(y_min + (y_max - y_min) * 0.6)

size = max(new_h, new_w)

x_min, y_min, x_max, y_max = int(center_x - size // 2), int(center_y - size // 2), int(

center_x + size // 2), int(center_y + size // 2)

# 确定裁剪区域上边top和左边left坐标

top = y_min

left = x_min

# 裁剪区域与原图的重合区域

top_coincidence = int(max(top, 0))

bottom_coincidence = int(min(y_max, vid_height))

left_coincidence = int(max(left, 0))

right_coincidence = int(min(x_max, vid_width))

print("人脸活动范围：{}:{}, {}:{}".format(top_coincidence, bottom_coincidence, left_coincidence, right_coincidence))

out_size = 512

scale = 512. / size

pts_3d = (pts_3d - np.array([left_coincidence, top_coincidence, 0])) * scale

Path_output_pkl = "{}/keypoint_rotate.pkl".format(video_data_path)

with open(Path_output_pkl, "wb") as f:

pickle.dump(pts_3d, f)

os.makedirs("{}/image".format(video_data_path), exist_ok=True)

ffmpeg_cmd = "ffmpeg -i {} -vf crop={}:{}:{}:{},scale=512:512:flags=neighbor -loglevel quiet -y {}/image/%06d.png".format(

video_path,

right_coincidence - left_coincidence,

bottom_coincidence - top_coincidence,

left_coincidence,

top_coincidence,

video_data_path

)

os.system(ffmpeg_cmd)

img_filelist = glob.glob("{}/image/*.png".format(video_data_path))

img_filelist.sort()

Path_output_pkl = "{}/keypoint_rotate.pkl".format(video_data_path)

with open(Path_output_pkl, "rb") as f:

images_info = pickle.load(f)[:, main_keypoints_index, :]

pts_driven = images_info.reshape(len(images_info), -1)

pts_driven = smooth_array(pts_driven).reshape(len(pts_driven), -1, 3)

face_pts_mean = np.loadtxt(os.path.join(current_dir, "../data/face_pts_mean_mainKps.txt"))

mat_list, pts_normalized_list, face_pts_mean_personal = calc_face_mat(pts_driven, face_pts_mean)

pts_normalized_list = np.array(pts_normalized_list)

# print(face_pts_mean_personal[INDEX_FACE_OVAL[:10], 1])

# print(np.max(pts_normalized_list[:,INDEX_FACE_OVAL[:10], 1], axis = 1))

face_pts_mean_personal[INDEX_FACE_OVAL[:10], 1] = np.max(pts_normalized_list[:, INDEX_FACE_OVAL[:10], 1],

axis=0) + np.arange(5, 25, 2)

face_pts_mean_personal[INDEX_FACE_OVAL[:10], 0] = np.max(pts_normalized_list[:, INDEX_FACE_OVAL[:10], 0],

axis=0) - (9 - np.arange(0, 10))

face_pts_mean_personal[INDEX_FACE_OVAL[-10:], 1] = np.max(pts_normalized_list[:, INDEX_FACE_OVAL[-10:], 1],

axis=0) - np.arange(5, 25, 2) + 28

face_pts_mean_personal[INDEX_FACE_OVAL[-10:], 0] = np.min(pts_normalized_list[:, INDEX_FACE_OVAL[-10:], 0],

axis=0) + np.arange(0, 10)

face_pts_mean_personal[INDEX_FACE_OVAL[10], 1] = np.max(pts_normalized_list[:, INDEX_FACE_OVAL[10], 1], axis=0) + 25

# for keypoints_normalized in pts_normalized_list:

# img = np.zeros([1000,1000,3], dtype=np.uint8)

# for coor in face_pts_mean_personal:

# # coor = (coor +1 )/2.

# cv2.circle(img, (int(coor[0]), int(coor[1])), point_size, (255, 0, 0), thickness)

# for coor in keypoints_normalized:

# # coor = (coor +1 )/2.

# cv2.circle(img, (int(coor[0]), int(coor[1])), point_size, point_color, thickness)

# cv2.imshow("a", img)

# cv2.waitKey(30)

with open("{}/face_mat_mask.pkl".format(video_data_path), "wb") as f:

# 检查命令行参数的数量

if len(sys.argv) != 2:

print("Usage: python data_preparation.py <data_dir>")

sys.exit(1) # 参数数量不正确时退出程序

# 获取video_name参数

data_dir = sys.argv[1]

print(f"Video dir is set to: {data_dir}")

# data_dir = r"F:\C\AI\CV\88"

video_files = glob.glob("{}/*.mp4".format(data_dir))

for video_path in tqdm.tqdm(video_files):