# -- TODO: use 3d model instead --
projectionImage = templateImage # The 3d model in obj format. For now it's just the template image
outVideo = cv2.VideoWriter(
'../results/outpy.avi', cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'), 10,
(backgroundVideo[0].shape[1], backgroundVideo[0].shape[0]))
templateImage = projectionImage
for i in range(len(backgroundVideo)):
backgroundImage = backgroundVideo[i]
projectionImageCropped = projectionImage # Currently no cropping is needed
# Extract features and match
# -- Note: if index out of range error occurs, lower numMatches variable in matchFeatures.py --
pts1, pts2 = matchFeatures(templateImage, backgroundImage)
# Compute homography using RANSAC
M, mask = cv2.findHomography(
pts1, pts2, cv2.RANSAC, 8.0
) # If performance sub-optimal, consider change the inlier threshold here
# Generate composite image
# TODO: -- replace the composite function with the 3D model version --
compositeImage = compositeWarp(M, projectionImageCropped, backgroundImage)
outVideo.write(compositeImage)
if i % 50 == 0:
print("Frame {} is recorded.".format(i))
# cv2.imshow("Composite image", compositeImage)
# cv2.waitKey(0)
num_tri = 30
print("Extracting logo image SIFT features... \n")
logopath = 'logos'
logoImages = extractFeaturesBulk.extractFeaturesBulk(logopath)
print("Done \n")
print("Extracting test image SIFT features... \n")
testpath = 'test'
testImages = extractFeaturesBulk.extractFeaturesBulk(testpath)
print("Done \n")
for i in range(len(testImages)):
for j in range(len(logoImages)):
print('\n')
print("********************************************************")
t = time.process_time()
pairs = matchFeatures.matchFeatures(logoImages[j], testImages[i], dthr)
elapsed_time = time.process_time() - t
print("Pairs created in ", elapsed_time)
t = time.process_time()
Dab, a, b = FGCT.extract_triangles(testImages[i], logoImages[j], pairs,
num_tri)
elapsed_time = time.process_time() - t
print("Triangles created in ", elapsed_time)
t = time.process_time()
correspodence1, correspodence2, consistensy, Dtemp = FGCT.FGCT(
Dab, alpha, sigma)
elapsed_time = time.process_time() - t
print("FGCT elapsed time is ", elapsed_time)
print("Correspodence 1 is", correspodence1)
print("Correspodence 2 is", correspodence2)
print("********************************************************")
import cv2
from matchFeatures import matchFeatures
from matchFeatures import plotMatches
from compositeWarp import compositeWarp
from skimage import color
from loadVid import loadVid
im1 = cv2.imread('../data/Squirtle_card.jpg', 0)
im1 = np.rot90(im1)
#im2 = cv2.imread('../data/cv_desk.png',0)
backgroundVideo = np.load("../readVideo/squirtle_parallel.npy")
im2 = backgroundVideo[0]
#cv2.imshow("Card image", im1)
#cv2.waitKey(0)
# Test for matchFeatures
pts1, pts2 = matchFeatures(im1, im2)
# Convert the images to grayscale if they are RGBs
im1Shape = im1.shape
im2Shape = im2.shape
im1Gray = None
im2Gray = None
numMatches = 10
if len(im1Shape) == 3:
im1Gray = color.rgb2gray(im1)
else:
im1Gray = im1
if len(im2Shape) == 3:
im2Gray = color.rgb2gray(im2)
else:
imageBPoints = getInterestPoints(imageB, feature_width)
'''
print('proj4.py: got interest points')
print(imageAPoints)
'''
# %% Create feature vectors at each interest point. Szeliski 4.1.2
# % !!! You will need to implement get_features. !!!
# intake a set of points. Output a set of features
imageA_features = getFeatures(imageA, imageAPoints, feature_width)
imageB_features = getFeatures(imageB, imageBPoints, feature_width)
matched = matchFeatures(imageA_features, imageB_features)
print('proj4.py: matched')
#print(matched)
showCorrespondence(imageA, imageB, matched)
#turn feature list into 2 lists of coords
#for feat in matched:
#coordAlist.append(feat[0])
#coordBlist.append(feat[2][1][0]) ??????
'''
# %% Match features. Szeliski 4.1.3