def stabilize(self, new_frame: np.ndarray) -> np.ndarray:
self.frame_n += 1
frame_gray = cv2.cvtColor(new_frame, cv2.COLOR_BGR2GRAY)
flow = np.zeros((self.frame_height, self.frame_width, 2))
flow = self.retval.calc(self.old_gray, frame_gray, flow)
good_new = np.copy(self.good_old)
# good_new_matrix=np.copy(good_old_matrix)
index = 0
for i in range(self.frame_height):
index = i * self.frame_width
good_new[index:index + self.frame_width, 0] += flow[i, :, 0]
good_new[index:index + self.frame_width, 1] += flow[i, :, 1]
x_motion_mesh, y_motion_mesh = motion_propagate(
self.good_old, good_new, new_frame, self.good_old_matrix)
self.x_paths, self.y_paths = generate_vertex_profiles(
self.x_paths, self.y_paths, x_motion_mesh, y_motion_mesh)
if (self.frame_n > self.BUFFER_SZ):
self.x_paths = self.x_paths[:, :, 1:]
self.y_paths = self.y_paths[:, :, 1:]
if self.frame_n == 2:
tmp = 1
while (tmp <= 2):
self.sx_paths, self.sy_paths, self.x_paths_pre, self.y_paths_pre = self.__stabilize(
self.x_paths, self.y_paths, self.x_paths_pre,
self.y_paths_pre, self.sx_paths, self.sy_paths, tmp)
new_x_motion_meshes, new_y_motion_meshes = self.__get_frame_warp(
self.x_paths, self.y_paths, self.sx_paths, self.sy_paths)
if tmp == 1:
new_frame = self.__generate_stabilized_video(
self.old_frame, tmp, new_x_motion_meshes,
new_y_motion_meshes)
else:
new_frame = self.__generate_stabilized_video(
new_frame, tmp, new_x_motion_meshes,
new_y_motion_meshes)
tmp += 1
else:
self.sx_paths, self.sy_paths, self.x_paths_pre, self.y_paths_pre = self.__stabilize(
self.x_paths, self.y_paths, self.x_paths_pre, self.y_paths_pre,
self.sx_paths, self.sy_paths, self.frame_n)
new_x_motion_meshes, new_y_motion_meshes = self.__get_frame_warp(
self.x_paths, self.y_paths, self.sx_paths, self.sy_paths)
new_frame = self.__generate_stabilized_video(
new_frame, self.frame_n, new_x_motion_meshes,
new_y_motion_meshes)
self.old_frame = new_frame.copy()
self.old_gray = frame_gray.copy()
# self.good_old = temp_good_old.copy()
return new_frame
def read_video(cap):
"""
@param: cap is the cv2.VideoCapture object that is
instantiated with given video
Returns:
returns mesh vertex motion vectors &
mesh vertex profiles
"""
# params for ShiTomasi corner detection
feature_params = dict(maxCorners=1000,
qualityLevel=0.3,
minDistance=7,
blockSize=7)
# Parameters for lucas kanade optical flow
lk_params = dict(winSize=(15, 15),
maxLevel=2,
criteria=(cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT,
20, 0.03))
# Take first frame
cap.set(cv2.CAP_PROP_POS_FRAMES, 0)
ret, old_frame = cap.read()
old_gray = cv2.cvtColor(old_frame, cv2.COLOR_BGR2GRAY)
frame_count = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
# preserve aspect ratio
global HORIZONTAL_BORDER
HORIZONTAL_BORDER = 30
global VERTICAL_BORDER
VERTICAL_BORDER = (HORIZONTAL_BORDER *
old_gray.shape[1]) / old_gray.shape[0]
# motion meshes in x-direction and y-direction
x_motion_meshes = []
y_motion_meshes = []
# path parameters
x_paths = np.zeros(
(old_frame.shape[0] / PIXELS, old_frame.shape[1] / PIXELS, 1))
y_paths = np.zeros(
(old_frame.shape[0] / PIXELS, old_frame.shape[1] / PIXELS, 1))
frame_num = 1
bar = tqdm(total=frame_count)
while frame_num < frame_count:
# processing frames
ret, frame = cap.read()
frame_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# find corners in it
p0 = cv2.goodFeaturesToTrack(old_gray, mask=None, **feature_params)
# calculate optical flow
p1, st, err = cv2.calcOpticalFlowPyrLK(old_gray, frame_gray, p0, None,
**lk_params)
# Select good points
good_new = p1[st == 1]
good_old = p0[st == 1]
# estimate motion mesh for old_frame
x_motion_mesh, y_motion_mesh = motion_propagate(
good_old, good_new, frame)
try:
x_motion_meshes = np.concatenate(
(x_motion_meshes, np.expand_dims(x_motion_mesh, axis=2)),
axis=2)
y_motion_meshes = np.concatenate(
(y_motion_meshes, np.expand_dims(y_motion_mesh, axis=2)),
axis=2)
except:
x_motion_meshes = np.expand_dims(x_motion_mesh, axis=2)
y_motion_meshes = np.expand_dims(y_motion_mesh, axis=2)
# generate vertex profiles
x_paths, y_paths = generate_vertex_profiles(x_paths, y_paths,
x_motion_mesh,
y_motion_mesh)
# updates frames
bar.update(1)
frame_num += 1
old_frame = frame.copy()
old_gray = frame_gray.copy()
bar.close()
return [x_motion_meshes, y_motion_meshes, x_paths, y_paths]
def read_video(cap):
"""
@param: cap - это объект cv2.VideoCapture, который
инициализруется с подаваемым видео
Returns:
возвращает векторы движения вершины сетки и
профили вершин сетки
"""
# Параметры для определиетля угла ShiTomasi
feature_params = dict( maxCorners = 1000,
qualityLevel = 0.3,
minDistance = 7,
blockSize = 7 )
# Параметры для оптического потока Лукаса Канаде
lk_params = dict( winSize = (15, 15),
maxLevel = 2,
criteria = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 20, 0.03))
cap.set(cv2.CAP_PROP_POS_FRAMES, 0)
ret, old_frame = cap.read()
old_gray = cv2.cvtColor(old_frame, cv2.COLOR_BGR2GRAY)
frame_count = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
# Сохранить соотношение сторон
global HORIZONTAL_BORDER
HORIZONTAL_BORDER = 30
global VERTICAL_BORDER
VERTICAL_BORDER = (HORIZONTAL_BORDER*old_gray.shape[1])//old_gray.shape[0]
# движение сетки в направлении х и у
x_motion_meshes = []; y_motion_meshes = []
x_paths = np.zeros((old_frame.shape[0]//PIXELS, old_frame.shape[1]//PIXELS, 1))
y_paths = np.zeros((old_frame.shape[0]//PIXELS, old_frame.shape[1]//PIXELS, 1))
frame_num = 1
bar = tqdm(total=frame_count)
while frame_num < frame_count:
ret, frame = cap.read()
frame_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
p0 = cv2.goodFeaturesToTrack(old_gray, mask=None, **feature_params)
p1, st, err = cv2.calcOpticalFlowPyrLK(old_gray, frame_gray, p0, None, **lk_params)
good_new = p1[st==1]
good_old = p0[st==1]
# Оценка движения сетки для старого кадра
x_motion_mesh, y_motion_mesh = motion_propagate(good_old, good_new, frame)
try:
x_motion_meshes = np.concatenate((x_motion_meshes, np.expand_dims(x_motion_mesh, axis=2)), axis=2)
y_motion_meshes = np.concatenate((y_motion_meshes, np.expand_dims(y_motion_mesh, axis=2)), axis=2)
except:
x_motion_meshes = np.expand_dims(x_motion_mesh, axis=2)
y_motion_meshes = np.expand_dims(y_motion_mesh, axis=2)
# Генерация профилей вершин
x_paths, y_paths = generate_vertex_profiles(x_paths, y_paths, x_motion_mesh, y_motion_mesh)
bar.update(1)
frame_num += 1
old_gray = frame_gray.copy()
bar.close()
return [x_motion_meshes, y_motion_meshes, x_paths, y_paths]