def prepare_batch_frames(video_path):
fg_bg = cv2.createBackgroundSubtractorMOG2()
video = imageio.get_reader(video_path, 'ffmpeg')
frame_batch = np.zeros((240, 240))
frame_batch = frame_batch.reshape((1, 240, 240))
for i in range(len(video)):
frame = video.get_data(i)
edged_image = cv_utils.apply_canny(frame, 50, 150)
rect_pts = detect_person(frame)
fg_mask = fg_bg.apply(frame)
fg_mask = fg_mask[int(rect_pts[0]):int(rect_pts[2] - 120),
int(rect_pts[1]):int(rect_pts[3] - 50)]
edged_image = edged_image[int(rect_pts[0]):int(rect_pts[2] - 120),
int(rect_pts[1]):int(rect_pts[3] - 50)]
fg_mask[fg_mask > 0] = 255.0
print(fg_mask.shape)
fg_mask = cv2.addWeighted(fg_mask, 1, edged_image, 1, 0)
# fg_mask = cv2.bitwise_and(fg_mask, edged_image)
reshaped_img = cv_utils.resize(fg_mask, (240, 240))
reshaped_img = reshaped_img / 255.0
cv2.imshow("bg_subtraction", reshaped_img)
if cv2.waitKey(25) & 0xFF == ord('q'):
break
reshaped_img = reshaped_img.reshape((1, 240, 240))
frame_batch = np.vstack((frame_batch, reshaped_img))
frame_batch = frame_batch.reshape(frame_batch.shape[0], 240, 240, 1)
frame_batch = frame_batch[2:, :, :, :]
return frame_batch
def process_image(image):
edged_image = cv_utils.apply_canny(image, 50, 150)
rect_pts = detect_person(image)
fg_mask = fg_bg.apply(image)
fg_mask = fg_mask[int(rect_pts[0]):int(rect_pts[2] - 120),
int(rect_pts[1]):int(rect_pts[3] - 50)]
edged_image = edged_image[int(rect_pts[0]):int(rect_pts[2] - 120),
int(rect_pts[1]):int(rect_pts[3] - 50)]
fg_mask[fg_mask > 0] = 255.0
# print(fg_mask.shape)
fg_mask = cv2.addWeighted(fg_mask, 1, edged_image, 1, 0)
reshaped_img = cv_utils.resize(fg_mask, (500, 500))
reshaped_img = np.dstack((reshaped_img, np.zeros_like(reshaped_img),
np.zeros_like(reshaped_img)))
# cv2.imshow("bg_subtraction", reshaped_img)
return reshaped_img
def prepare_batch_frames_from_bg_data(video_path,
frame_limit=109,
resize=(240, 240)):
"""
This function prepares batches by reading the video and extracting
frames which is used as one mini-batch in training
:param video_path: string
path to video which is to be read
:param frame_limit: int
limiting the number frames which is to be returned
if the number of frames in the video is > frame_limit
then random sampling will be carried out to extract frames exactly of frame_limit
:param resize: tuple of shape 2 elements
resizing the frames
:return: frame_batch : numpy array of shape (batch_size, height, width, 1)
"""
sampling = False
video = imageio.get_reader(video_path, 'ffmpeg')
frame_batch = np.zeros(resize)
frame_batch = frame_batch.reshape((1, resize[0], resize[1]))
if frame_limit < len(video):
sampling = True
sampling_list = random.sample(range(0, len(video) - 1), frame_limit)
for i in range(len(video)):
if sampling and i not in sampling_list:
continue
frame = video.get_data(i)
red_channel = frame[:, :, 0]
red_channel = cv_utils.resize(red_channel, resize)
red_channel[red_channel > 0] == 255.0
red_channel = red_channel / 255.0
cv2.imshow("bg_subtraction", red_channel)
if cv2.waitKey(25) & 0xFF == ord('q'):
break
red_channel = red_channel.reshape((1, resize[0], resize[1]))
frame_batch = np.vstack((frame_batch, red_channel))
frame_batch = frame_batch.reshape(frame_batch.shape[0], resize[0],
resize[1], 1)
frame_batch = frame_batch[2:, :, :, :]
return frame_batch
def process_image(image):
edged_image = cv_utils.apply_canny(image, 50, 150)
rect_pts = detect_person(image)
print("**********************************")
print(rect_pts)
print("**********************************")
fg_bg = cv2.createBackgroundSubtractorMOG2()
IMAGE_SIZE = (12, 8)
pic_rand = random.randint(1, 100)
# plt.figure(figsize=IMAGE_SIZE)
# plt.imshow(image)
# plt.savefig("mask/before_mask_bg_temp{0}.png".format(pic_rand))
fg_mask = fg_bg.apply(image)
# plt.figure(figsize=IMAGE_SIZE)
# plt.imshow(fg_mask)
# plt.savefig("mask/after_mask_bg_temp{0}.png".format(pic_rand))
fg_mask = fg_mask[int(rect_pts[0]):int(rect_pts[2] - 120),
int(rect_pts[1]):int(rect_pts[3] - 50)]
edged_image = edged_image[int(rect_pts[0]):int(rect_pts[2] - 120),
int(rect_pts[1]):int(rect_pts[3] - 50)]
fg_mask[fg_mask > 0] = 255.0
# print(fg_mask.shape)
fg_mask = cv2.addWeighted(fg_mask, 0, edged_image, 1, 0)
reshaped_img = cv_utils.resize(fg_mask, (500, 500))
reshaped_img = np.dstack((reshaped_img, np.zeros_like(reshaped_img),
np.zeros_like(reshaped_img)))
# cv2_imshow(reshaped_img)
IMAGE_SIZE = (12, 8)
# plt.figure(figsize=IMAGE_SIZE)
# plt.imshow(reshaped_img)
# plt.savefig("bg_temp.png")
return reshaped_img