def testAerialSequence():
masks = []
for i in range(video_frames.shape[2] - 1):
It = video_frames[:, :, i]
It1 = video_frames[:, :, i + 1]
print(It.shape)
mask = SubtractDominantMotion(It, It1)
masks.append(mask)
# if i%50==0:
# plt.subplot(211)
# plt.imshow(mask)
# plt.show()
np.save('invercompose_masks.npy', masks)
return
def main():
frames = np.load('../data/aerialseq.npy')
fig, ax = plt.subplots(1)
rect_save = []
for i in range(frames.shape[-1] - 1):
p = SubtractDominantMotion.SubtractDominantMotion(
frames[:, :, i], frames[:, :, i + 1])
points = np.where(p)
plt.plot(points[1], points[0], 'g.')
plt.imshow(frames[:, :, i], cmap='gray')
plt.pause(0.0001)
if i in [30, 60, 90, 120]:
plt.savefig('airseq_' + str(i) + '.png')
plt.clf()
from SubtractDominantMotion import *
from LucasKanadeAffine import *
import time
frames = np.load('../data/aerialseq.npy')
time_total = 0
seq_len = frames.shape[2]
for i in range(seq_len):
if i == 0:
continue
print("Processing frame %d" % i)
start = time.time()
image1 = frames[:, :, i - 1]
image2 = frames[:, :, i]
mask = SubtractDominantMotion(image1, image2)
end = time.time()
time_total += end - start
if i == 31 or i == 61 or i == 91 or i == 121:
plt.figure()
plt.imshow(image1, cmap='gray')
for r in range(mask.shape[0] - 1):
for c in range(mask.shape[1] - 1):
if mask[r, c]:
plt.scatter(c, r, s=1, c='r', alpha=0.5)
plt.show()
print('Finished, the tracking frequency is %.4f' % (seq_len / time_total))
parser.add_argument('--tolerance', type=float, default=0.2, help='binary threshold of intensity difference when computing the mask')
args = parser.parse_args()
num_iters = args.num_iters
threshold = args.threshold
tolerance = args.tolerance
seq = np.load('../data/aerialseq.npy')
print(seq.shape)
num_frames = seq.shape[2]
fig = plt.figure()
for i in range(0, num_frames-1):
print('frame: ', i)
mask = SubtractDominantMotion(seq[:,:,i], seq[:,:,i+1], threshold,num_iters,tolerance)
img1 = seq[:,:,i-1]
img2 = seq[:,:,i]
#masks = mask[:,:,i-1]
highlight = np.where(mask,1.0,img1)
img_stack = np.dstack((img1,img1))
superimpose_img = np.dstack((img_stack,highlight))
plt.imshow(superimpose_img)
#plt.show()
if i in [30, 60, 90, 120]:
plt.savefig("aerialseq"+str(i)+".png")
plt.pause(1.0)
from SubtractDominantMotion import *
# write your script here, we recommend the above libraries for making your animation
video_path = '../data/aerialseq.npy'
frame_step = 1
data = np.load(video_path)
frame_num = data.shape[2]
masks = np.zeros(data.shape)
for i in np.arange(1, frame_num, frame_step):
print("Image:", i)
# pdb.set_trace()
img_pre = data[:, :, i - 1]
img_cur = data[:, :, i]
mask = SubtractDominantMotion(img_pre, img_cur)
masks[:, :, i] = mask
# plt.imshow(data[:,:,i], cmap='gray')
# plt.imshow(mask, cmap='Reds', alpha=0.4)
# plt.show()
# pdb.set_trace()
# visualization
frame = [30, 60, 90, 120]
fig, axs = plt.subplots(1, 4, figsize=(8, 2))
for i, k in enumerate(frame):
axs[i].imshow(data[:, :, k], cmap='gray')
axs[i].imshow(masks[:, :, k], cmap='Reds', alpha=0.4)
plt.show()
num_iters = args.num_iters
threshold = args.threshold
tolerance = args.tolerance
seq = np.load('../data/aerialseq.npy')
imH, imW, frames = np.shape(seq)
# print(imH,imW)
print(frames)
# rect = [101., 61., 155., 107.]
start = time.time()
for i in range(frames - 1):
#print(i)
image1 = seq[:, :, i]
image2 = seq[:, :, i + 1]
mask = SubtractDominantMotion.SubtractDominantMotion(
image1, image2, threshold, num_iters, tolerance)
if (i == 29) or (i == 59) or (i == 89) or (i == 119):
# since we are plotting framee 30, 60, 90, 120 are plotted
pic = plt.figure()
plt.imshow(image2, cmap='gray')
plt.axis('off')
plt.title("Frame %d " % (i + 1))
# print(mask)
for w in range(mask.shape[0] - 1):
for h in range(mask.shape[1] - 1):
if mask[w, h]:
plt.scatter(h, w, s=1, c='r', alpha=0.5)
plt.show()
stop = time.time()
from matplotlib import animation
import matplotlib.patches as patches
import os
import cv2
import SubtractDominantMotion
# write your script here, we recommend the above libraries for making your animation
if __name__ == '__main__':
result_dir = '../result/'
if not os.path.exists(result_dir):
os.makedirs(result_dir)
aerial_data = np.load('../data/aerialseq.npy')
frame = aerial_data[:, :, 0]
for i in range(1, aerial_data.shape[2]):
next_frame = aerial_data[:, :, i]
mask = SubtractDominantMotion.SubtractDominantMotion(frame, next_frame)
tmp_img = np.zeros((next_frame.shape[0], next_frame.shape[1], 3))
tmp_img[:, :, 0] = next_frame
tmp_img[:, :, 1] = next_frame
tmp_img[:, :, 2] = next_frame
tmp_img[:, :, 0][mask == 1] = 1
if i in [30, 60, 90, 120]:
cv2.imwrite(os.path.join(result_dir, 'q3-3_{}.jpg'.format(i)),
tmp_img * 255)
frame = next_frame
from matplotlib import animation
import matplotlib.patches as patches
import SubtractDominantMotion
carseq_data = np.load("../data/aerialseq.npy")
frame_num = carseq_data.shape[2]
for i in range(frame_num - 1):
# Load the template and the input images
It = carseq_data[:, :, i]
It1 = carseq_data[:, :, i + 1]
print(i)
# find out the mask that shows the main motion
mask = SubtractDominantMotion.SubtractDominantMotion(It, It1)
# plot the tracking
fig = plt.figure()
plt.imshow(It1)
locs = np.transpose(np.nonzero(mask))
plt.plot(locs[:, 1], locs[:, 0], 'b.')
plt.draw()
# save the tracking
file_name = 'md_%s.png' % (str(i + 1))
plt.savefig(file_name)
plt.close()