def __init__(self, capId, scaling_factor, win_name):
self.cap = cv2.VideoCapture(capId)
self.rect = None
self.win_name = win_name
self.scaling_factor = scaling_factor
self.tracker = PoseEstimator()
ret, frame = self.cap.read()
self.rect = None
self.frame = cv2.resize(frame,
None,
fx=scaling_factor,
fy=scaling_factor,
interpolation=cv2.INTER_AREA)
self.roi_selector = ROISelector(win_name, self.frame, self.set_rect)
self.overlay_vertices = np.float32([[0, 0, 0], [0, 1, 0], [1, 1, 0],
[1, 0, 0], [0.5, 0.5, 4]])
self.overlay_edges = [(0, 1), (1, 2), (2, 3), (3, 0), (0, 4), (1, 4),
(2, 4), (3, 4)]
self.color_base = (0, 255, 0)
self.color_lines = (0, 0, 0)
self.graphics_counter = 0
self.time_counter = 0
def __init__(self):
self.cap = cv2.VideoCapture(0)
self.frame = None
self.paused = False
self.tracker = PoseEstimator()
cv2.namedWindow('Augmented Reality')
self.roi_selector = ROISelector('Augmented Reality', self.on_rect)
self.overlay_vertices = np.float32([[0, 0, 0], [0, 1, 0], [1, 1, 0],
[1, 0, 0], [0.5, 0.5, 4]])
self.overlay_edges = [(0, 1), (1, 2), (2, 3), (3, 0), (0, 4), (1, 4),
(2, 4), (3, 4)]
self.color_base = (0, 255, 0)
self.color_lines = (0, 0, 0)
class Tracker(object):
def __init__(self):
self.cap = cv2.VideoCapture(0)
self.frame = None
self.paused = False
self.tracker = PoseEstimator()
cv2.namedWindow('Augmented Reality')
self.roi_selector = ROISelector('Augmented Reality', self.on_rect)
self.overlay_vertices = np.float32([[0, 0, 0], [0, 1, 0], [1, 1, 0],
[1, 0, 0], [0.5, 0.5, 4]])
self.overlay_edges = [(0, 1), (1, 2), (2, 3), (3, 0), (0, 4), (1, 4),
(2, 4), (3, 4)]
self.color_base = (0, 255, 0)
self.color_lines = (0, 0, 0)
self.graphics_counter = 0.5
self.time_counter = 0
self.sign = 1
self.frame_jump = 3
self.movement_size = 1
self.step_size = 0.1
def on_rect(self, rect):
self.tracker.add_target(self.frame, rect)
def start(self):
while True:
is_running = not self.paused and self.roi_selector.selected_rect is None
if is_running or self.frame is None:
ret, frame = self.cap.read()
scaling_factor = 0.5
frame = cv2.resize(frame,
None,
fx=scaling_factor,
fy=scaling_factor,
interpolation=cv2.INTER_AREA)
if not ret:
break
self.frame = frame.copy()
img = self.frame.copy()
if is_running:
tracked = self.tracker.track_target(self.frame)
for item in tracked:
cv2.polylines(img, [np.int32(item.quad)], True,
self.color_lines, 2)
for (x, y) in np.int32(item.points_cur):
cv2.circle(img, (x, y), 2, self.color_lines)
self.overlay_graphics(img, item)
self.roi_selector.draw_rect(img)
cv2.imshow('Augmented Reality', img)
ch = cv2.waitKey(1)
if ch == ord(' '):
self.paused = not self.paused
if ch == ord('c'):
self.tracker.clear_targets()
if ch == 27:
break
def overlay_graphics(self, img, tracked):
x_start, y_start, x_end, y_end = tracked.target.rect
quad_3d = np.float32([[x_start, y_start, 0], [x_end, y_start, 0],
[x_end, y_end, 0], [x_start, y_end, 0]])
h, w = img.shape[:2]
K = np.float64([[w, 0, 0.5 * (w - 1)], [0, w, 0.5 * (h - 1)],
[0, 0, 1.0]])
dist_coef = np.zeros(4)
ret, rvec, tvec = cv2.solvePnP(quad_3d, tracked.quad, K, dist_coef)
self.time_counter += 1
if not self.time_counter % self.frame_jump:
self.graphics_counter = self.graphics_counter + self.sign * self.step_size
if abs(self.graphics_counter) >= self.movement_size:
self.sign *= -1
self.overlay_vertices = np.float32([[0, 0, 0], [0, 1, 0], [1, 1, 0],
[1, 0, 0],
[self.graphics_counter, 0.5, 4]])
verts = self.overlay_vertices * [
(x_end - x_start), (y_end - y_start), -(x_end - x_start) * 0.3
] + (x_start, y_start, 0)
verts = cv2.projectPoints(verts, rvec, tvec, K,
dist_coef)[0].reshape(-1, 2)
verts_floor = np.int32(verts).reshape(-1, 2)
cv2.drawContours(img, [verts_floor[:4]], -1, self.color_base, -3)
cv2.drawContours(img, [np.vstack((verts_floor[:2], verts_floor[4:5]))],
-1, (0, 255, 0), -3)
cv2.drawContours(img,
[np.vstack((verts_floor[1:3], verts_floor[4:5]))], -1,
(255, 0, 0), -3)
cv2.drawContours(img,
[np.vstack((verts_floor[2:4], verts_floor[4:5]))], -1,
(0, 0, 150), -3)
cv2.drawContours(img, [
np.vstack((verts_floor[3:4], verts_floor[0:1], verts_floor[4:5]))
], -1, (255, 255, 0), -3)
for i, j in self.overlay_edges:
(x_start, y_start), (x_end, y_end) = verts[i], verts[j]
cv2.line(img, (int(x_start), int(y_start)),
(int(x_end), int(y_end)), self.color_lines, 2)
hours = time_left // 3600
minutes = time_left % 3600 // 60
seconds = time_left % 60
print("\rProgress: %.2f" % ((cur_idx - omit) * 100 / L) + "% "
+ str(hours) + " hours "
+ str(minutes) + " minutes "
+ str(seconds) + " seconds left",
end=" ")
if __name__ == '__main__':
dataset = pd.read_csv(os.path.join(DATASET_DIR, "annotation.csv"))
pose_estimator = PoseEstimator()
start_time = time.time()
omit = 0
for idx in range(dataset.shape[0]):
row = dataset.iloc[idx]
label = row["label"]
video_path = os.sep.join([DATASET_DIR, "videos", row["video_name"]])
data_path = os.sep.join([DATASET_DIR, "pose_data", row["video_name"] + ".pkl"])
if "shoot" in label or os.path.exists(data_path) or (not os.path.exists(video_path)):
omit += 1
continue
frame_n = 0
cap = cv2.VideoCapture(video_path)
video_data = []
class Tracker(object):
def __init__(self, capId, scaling_factor, win_name):
self.cap = cv2.VideoCapture(capId)
self.rect = None
self.win_name = win_name
self.scaling_factor = scaling_factor
self.tracker = PoseEstimator()
ret, frame = self.cap.read()
self.rect = None
self.frame = cv2.resize(frame,
None,
fx=scaling_factor,
fy=scaling_factor,
interpolation=cv2.INTER_AREA)
self.roi_selector = ROISelector(win_name, self.frame, self.set_rect)
self.overlay_vertices = np.float32([[0, 0, 0], [0, 1, 0], [1, 1, 0],
[1, 0, 0], [0.5, 0.5, 4]])
self.overlay_edges = [(0, 1), (1, 2), (2, 3), (3, 0), (0, 4), (1, 4),
(2, 4), (3, 4)]
self.color_base = (0, 255, 0)
self.color_lines = (0, 0, 0)
self.graphics_counter = 0
self.time_counter = 0
def set_rect(self, rect):
self.rect = rect
self.tracker.add_target(self.frame, rect)
def start(self):
paused = False
while True:
if not paused or self.frame is None:
ret, frame = self.cap.read()
scaling_factor = self.scaling_factor
frame = cv2.resize(frame, None, fx=scaling_factor, fy=scaling_factor,\
interpolation=cv2.INTER_AREA)
if not ret: break
self.frame = frame.copy()
img = self.frame.copy()
if not paused:
tracked = self.tracker.track_target(self.frame)
for item in tracked:
cv2.polylines(img, [np.int32(item.quad)], True,
self.color_lines, 2)
for (x, y) in np.int32(item.points_cur):
cv2.circle(img, (x, y), 2, self.color_lines)
self.overlay_graphics(img, item)
self.roi_selector.draw_rect(img, self.rect)
cv2.imshow(self.win_name, img)
ch = cv2.waitKey(1)
if ch == ord(' '): self.paused = not self.paused
if ch == ord('c'): self.tracker.clear_targets()
if ch == 27: break
def overlay_graphics(self, img, tracked):
x_start, y_start, x_end, y_end = tracked.target.rect
quad_3d = np.float32([[x_start, y_start, 0], [x_end, y_start, 0], \
[x_end, y_end, 0], [x_start, y_end, 0]])
h, w = img.shape[:2]
K = np.float64([[w, 0, 0.5 * (w - 1)], [0, w, 0.5 * (h - 1)],
[0, 0, 1.0]])
dist_coef = np.zeros(4)
ret, rvec, tvec = cv2.solvePnP(quad_3d, tracked.quad, K, dist_coef)
self.time_counter += 1
if not self.time_counter % 20:
self.graphics_counter = (self.graphics_counter + 1) % 8
self.overlay_vertices = np.float32([[0, 0, 0], [0, 1, 0],\
[1, 1, 0], [1, 0, 0], [0.5, 0.5, self.graphics_counter]])
verts = self.overlay_vertices * [(x_end-x_start), (y_end-y_start),\
-(x_end-x_start)*0.3] + (x_start, y_start, 0)
verts = cv2.projectPoints(verts, rvec, tvec, K,
dist_coef)[0].reshape(-1, 2)
verts_floor = np.int32(verts).reshape(-1, 2)
cv2.drawContours(img, [verts_floor[:4]], -1, self.color_base, -3)
cv2.drawContours(img, [np.vstack((verts_floor[:2], \
verts_floor[4:5]))], -1, (0,255,0), -3)
cv2.drawContours(img, [np.vstack((verts_floor[1:3], \
verts_floor[4:5]))], -1, (255,0,0), -3)
cv2.drawContours(img, [np.vstack((verts_floor[2:4], \
verts_floor[4:5]))], -1, (0,0,150), -3)
cv2.drawContours(img, [np.vstack((verts_floor[3:4], \
verts_floor[0:1], verts_floor[4:5]))], -1, (255,255,0), -3)
for i, j in self.overlay_edges:
(x_start, y_start), (x_end, y_end) = verts[i], verts[j]
cv2.line(img, (int(x_start), int(y_start)),
(int(x_end), int(y_end)), self.color_lines, 2)
omit += 1
continue
frame_n = 0
cap = cv2.VideoCapture(video_path)
with open(data_path, 'rb') as f:
video_data = pickle.load(f)
tracker_1 = Tracker()
tracker_2 = Tracker()
tracker_rand = Tracker()
num_samples = len(samples)
for frame_data in video_data:
frame_n = frame_data["frame_n"]
poses = PoseEstimator.filter(frame_data["poses"])
if frame_n % 2 == 0:
samples += tracker_1.update_tracks(None, frame_n, poses)
if (frame_n + 1) % 2 == 0:
samples += tracker_2.update_tracks(None, frame_n, poses)
if np.random.rand() > 0.6:
samples += tracker_rand.update_tracks(None, frame_n, poses)
num_samples = len(samples) - num_samples
labels += [row["label"]] * num_samples
splits += [row["split"]] * num_samples
if (idx) % 10 == 0: