if videonummer < 10:
videopad = "{}MSK_0{}.mp4".format(cf.path_smartphone, videonummer)
undist_videopad = videopad
elif videonummer < 12:
videopad = "{}MSK_{}.mp4".format(cf.path_smartphone, videonummer)
undist_videopad = videopad
else:
videopad = "{}MSK_{}.mp4".format(cf.path_gopro, videonummer)
undist_videopad = videopad.replace(".mp4", "_undistorted.mp4")
print("Video:", videopad)
start_time = timeit.default_timer()
list_db_match_paths, list_db_match_scores, list_kps_and_matches = [],[],[]
#SAVING/LOADING DESCRIPTORS
database = m.makeListFromDatabase(cf.path_paintings + "*")
m.saveDescriptorFiles(database, cf.overwrite_descriptors)
all_descriptors = m.loadDescriptorFiles(cf.path_descriptor_files + "*")
#DETECTING SHARP FRAMES AND GETTING VIDEO PROPERTIES
total_frames, fps, height, width = pd.getVideoProperties(videopad)
_, frame_nrs, sharpness = pd.getSharpFrames(videopad, [0, ':'], 30)
#frame_nrs = np.arange(0,total_frames, 500)
blur_it = pd.calculateBlurIterations(height, width)
#STARTING DETECTION, MATCHING & LOCALIZATION
for i, frame_nr in enumerate(frame_nrs):
#print("\nFrame nr.:", frame_nr)
frame = m.getFrame(undist_videopad, frame_nr, False)
extracted_paintings, list_poly_corners = m.extractPaintings(
frame, frame_nr, blur_it)
def makeSmallerDatabase(
): # Downgrade database and store smaller resolution images
database = m.makeListFromDatabase(cf.path_paintings_high_res + "*")
m.downgradeDatabase(database)
def main_noSharpness(videopad, step=10):
start_time = timeit.default_timer()
all_scores = []
good_frame = False
success_frames = 0
correct_matches = 0
false_matches = 0
database = m.makeListFromDatabase(cf.path_paintings + "*")
m.saveDescriptorFiles(database, cf.overwrite_descriptors)
all_descriptors = m.loadDescriptorFiles(cf.path_descriptor_files + "*")
total_frames, fps, height, width = pd.getVideoProperties(videopad)
# Getting frame_nrs from video and returning useable sharp frames.
# _, frame_nrs, sharpness = pd.getSharpFrames(videopad, [0, ':'], 60) # step 60 for checking frames
# frame_nrs = np.arange(0,total_frames, 60)
# sharpness = np.zeros(len(frame_nrs))
# blur_it = pd.calculateBlurIterations(height, width)
list_db_match_paths = []
list_db_match_scores = []
list_kps_and_matches = []
i = 0
for frame_nr in range(0, total_frames, step):
print("\n{}".format('*' * 40), "\nFrame nr.:", frame_nr)
# voor gopro
undist_videopad = videopad.replace(".mp4", "_undistorted.mp4")
# voor smartphone
# undist_videopad = videopad
frame = m.getFrame(undist_videopad, frame_nr, False)
# if frame_nr in frame_nrs:
extracted_paintings, list_poly_corners = m.extractPaintings(
frame, frame_nr)
for j, painting in enumerate(extracted_paintings):
print("\nPolygon:", j + 1)
print("----------------------------------------")
db_match_path, db_match_score, kps_and_matches = m.findDatabaseMatch2(
painting, database, all_descriptors)
if db_match_path is not None:
list_db_match_paths.append(
db_match_path[db_match_path.find("zaal_"):db_match_path.
find("/Zaal")])
list_db_match_scores.append(db_match_score)
list_kps_and_matches.append(kps_and_matches)
inked_frame = m.inkFrame(frame, frame_nr, list_poly_corners[j],
0, db_match_score)
db_match_img = cv2.imread(db_match_path)
if m.toonCanvas(inked_frame, painting, db_match_img,
db_match_path, kps_and_matches, 1) == 'y':
correct_matches += 1
else:
false_matches += 1
all_scores.append(db_match_score)
if db_match_score is not None and db_match_score > 0:
good_frame = True
if good_frame:
success_frames += 1
good_frame = False
floorplan_img = main_localization(list_db_match_paths.copy(),
list_db_match_scores.copy(),
list_kps_and_matches.copy())
# print("\nall scores:")
# print(all_scores)
opslaan_pad = videopad.split('/')
naam_bestand = opslaan_pad[-1].replace(".mp4", "")
if not os.path.exists('./test_output/'):
os.makedirs('./test_output/')
ofile = open('./test_output/' + naam_bestand + "met_sharpness" + ".csv",
"w")
ofile.write('match_zaal;score\n')
for i, elem in enumerate(list_db_match_paths):
naam_zaal = elem.split('/')[-1]
ofile.write(naam_zaal + ';' + str(list_db_match_scores[i]) + '\n')
ofile.close()
# m.calcPerformance(success_frames, len(frame_nrs), correct_matches, false_matches)
print("Runtime:", str(round(timeit.default_timer() - start_time, 1)) + "s")
def main_realtime(videopad):
start_time = timeit.default_timer()
all_scores = []
good_frame = False
success_frames = 0
correct_matches = 0
false_matches = 0
database = m.makeListFromDatabase(cf.path_paintings + "*")
m.saveDescriptorFiles(database, cf.overwrite_descriptors)
all_descriptors = m.loadDescriptorFiles(cf.path_descriptor_files + "*")
total_frames, fps, height, width = pd.getVideoProperties(videopad)
# Getting frame_nrs from video and returning useable sharp frames.
_, frame_nrs, sharpness = pd.getSharpFrames(
videopad, [0, ':'], 60) # step 60 for checking frames
# frame_nrs = np.arange(0,total_frames, 60)
sharpness = np.zeros(len(frame_nrs))
# blur_it = pd.calculateBlurIterations(height, width)
list_db_match_paths = []
list_db_match_scores = []
list_kps_and_matches = []
i = 0
for frame_nr in range(0, total_frames, 30): # step 30 (real_time)
print("\n{}".format('*' * 40), "\nFrame nr.:", frame_nr)
# voor gopro
undist_videopad = videopad.replace(".mp4", "_undistorted.mp4")
# voor smartphone
# undist_videopad = videopad
frame = m.getFrame(undist_videopad, frame_nr, False)
cv2.imshow("video", frame)
cv2.waitKey(1)
if frame_nr in frame_nrs:
extracted_paintings, list_poly_corners = m.extractPaintings(
frame, frame_nr)
for j, painting in enumerate(extracted_paintings):
print("\nPolygon:", j + 1)
print("----------------------------------------")
db_match_path, db_match_score, kps_and_matches = m.findDatabaseMatch2(
painting, database, all_descriptors)
if db_match_path is not None:
list_db_match_paths.append(
db_match_path[db_match_path.find("zaal_"):db_match_path
.find("/Zaal")])
list_db_match_scores.append(db_match_score)
list_kps_and_matches.append(kps_and_matches)
inked_frame = m.inkFrame(frame, frame_nr,
list_poly_corners[j],
sharpness[i], db_match_score)
db_match_img = cv2.imread(db_match_path)
if m.toonCanvas(inked_frame, painting, db_match_img,
db_match_path, kps_and_matches, 1) == 'y':
correct_matches += 1
else:
false_matches += 1
all_scores.append(db_match_score)
if db_match_score is not None and db_match_score > 0:
good_frame = True
if good_frame:
success_frames += 1
good_frame = False
floorplan_img = main_localization(list_db_match_paths.copy(),
list_db_match_scores.copy(),
list_kps_and_matches.copy())
cv2.imshow("floorplan", floorplan_img)
cv2.waitKey(1)
# print("\nall scores:")
# print(all_scores)
return 0
m.calcPerformance(success_frames, len(frame_nrs), correct_matches,
false_matches)
print("Runtime:", str(round(timeit.default_timer() - start_time, 1)) + "s")