audio = filename + '.wav'
fs, wav_data = wavfile.read(audio)
query_features = []
prev_frame = None
prev_colorhist = None
frame_nbr = int(args.s) * frame_rate
cap.set(cv2.CAP_PROP_POS_MSEC, int(args.s) * 1000)
while (cap.isOpened() and cap.get(cv2.CAP_PROP_POS_MSEC) <
(int(args.e) * 1000)):
ret, frame = cap.read()
if frame == None:
break
if args.f == features[0]:
h = ft.colorhist(frame)
elif args.f == features[1]:
h = temporal_diff(prev_frame, frame, 10)
elif args.f == features[2] or args.f == features[3]:
audio_frame = frame_to_audio(frame_nbr, frame_rate, fs, wav_data)
if args.f == features[2]:
h = np.mean(audio_frame**2)
elif args.f == features[3]:
h, mspec, spec = ft.extract_mfcc(audio_frame, fs)
elif args.f == features[4]:
colorhist = ft.colorhist(frame)
h = colorhist_diff(prev_colorhist, colorhist)
prev_colorhist = colorhist
if h != None:
query_features.append(h)
def process_videos(video_list, indx):
total = len(video_list)
progress_count = 0
for video in video_list:
progress_count += 1
print 'processing: ', video, ' (', progress_count, ' of ', total, ')'
cap = cv2.VideoCapture(video)
frame_rate = get_frame_rate(video)
total_frames = get_frame_count(video)
total_audio_frames = get_frame_count_audio(video)
# get corresponding audio file
filename, fileExtension = os.path.splitext(video)
audio = filename + '.wav'
fs, wav_data = read(audio)
colorhists = []
sum_of_differences = []
audio_powers = []
mfccs = []
colorhist_diffs = []
prev_colorhist = None
prev_frame = None
frame_nbr = 0
while (cap.isOpened()):
ret, frame = cap.read()
if frame == None:
break
audio_frame = frame_to_audio(frame_nbr, frame_rate, fs, wav_data)
# check if audio frame is long enough for mfcc transformation
if len(audio_frame) >= int(0.01 * fs):
power = np.mean(audio_frame**2)
audio_powers.append(power)
ceps, mspec, spec = ft.extract_mfcc(audio_frame, fs)
mfccs.append(ceps)
# calculate sum of differences
if not prev_frame == None:
tdiv = temporal_diff(prev_frame, frame, 10)
#diff = np.absolute(prev_frame - frame)
#sum = np.sum(diff.flatten()) / (diff.shape[0]*diff.shape[1]*diff.shape[2])
sum_of_differences.append(tdiv)
colorhist = ft.colorhist(frame)
colorhists.append(colorhist)
if not prev_colorhist == None:
ch_diff = colorhist_diff(prev_colorhist, colorhist)
colorhist_diffs.append(ch_diff)
prev_colorhist = colorhist
prev_frame = frame
frame_nbr += 1
print 'end:', frame_nbr
# prepare descriptor for database
# mfccs = descr['mfcc'] # Nx13 np array (or however many mfcc coefficients there are)
# audio = descr['audio'] # Nx1 np array
# colhist = descr['colhist'] # Nx3x256 np array
# tempdif = descr['tempdiff'] # Nx1 np array
descr = {}
descr['mfcc'] = np.array(mfccs)
descr['audio'] = np.array(audio_powers)
descr['colhist'] = np.array(colorhists)
descr['tempdiff'] = np.array(sum_of_differences)
descr['chdiff'] = np.array(colorhist_diffs)
indx.add_to_index(video, descr)
print 'added ' + video + ' to database'
indx.db_commit()
#Matrix for storing the frames
frames = []
#Reading the frames and calculating the color histograms and temporal differences
#Use the following line if you want to analyze the entire video, currently only the first 10% of the video is looked at
print("Start reading")
while (cap.isOpened() and cap.get(cv2.CAP_PROP_POS_MSEC) < q_total * 1000):
#while(cap.isOpened() and cap.get(cv2.CAP_PROP_POS_MSEC) < q_total * 100):
ret, frame = cap.read()
if frame == None:
break
h_ch = ft.colorhist(frame)
h_td = temporal_diff(prev_frame, frame, 10)
if (h_ch != None) and (h_td != None):
frames.append(frame)
ch_features.append(h_ch)
td_features.append(h_td)
prev_frame = frame
frame_nbr += 1
#Matrix for storing interesting frames
interesting_frames = []
#Integer that keeps track how much the frames have changed since the previous selected frame.
difference = 0
max_frame_skipped = 0
def process_videos(video_list, indx):
total = len(video_list)
progress_count = 0
for video in video_list:
progress_count += 1
print 'processing: ',video, ' (' ,progress_count, ' of ' ,total,')'
cap = cv2.VideoCapture(video)
frame_rate = get_frame_rate(video)
total_frames = get_frame_count(video)
total_audio_frames = get_frame_count_audio(video)
# get corresponding audio file
filename, fileExtension = os.path.splitext(video)
audio = filename + '.wav'
fs, wav_data = read(audio)
colorhists = []
sum_of_differences = []
audio_powers = []
mfccs = []
colorhist_diffs = []
prev_colorhist = None
prev_frame = None
frame_nbr = 0
while(cap.isOpened()):
ret, frame = cap.read()
if frame == None:
break
audio_frame = frame_to_audio(frame_nbr, frame_rate, fs, wav_data)
# check if audio frame is long enough for mfcc transformation
if len(audio_frame) >= 256:
power = np.mean(audio_frame**2)
audio_powers.append(power)
ceps, mspec, spec = ft.extract_mfcc(audio_frame, fs)
mfccs.append(ceps)
# calculate sum of differences
if not prev_frame == None:
tdiv = temporal_diff(prev_frame, frame, 10)
#diff = np.absolute(prev_frame - frame)
#sum = np.sum(diff.flatten()) / (diff.shape[0]*diff.shape[1]*diff.shape[2])
sum_of_differences.append(tdiv)
colorhist = ft.colorhist(frame)
colorhists.append(colorhist)
if not prev_colorhist == None:
ch_diff = colorhist_diff(prev_colorhist, colorhist)
colorhist_diffs.append(ch_diff)
prev_colorhist = colorhist
prev_frame = frame
frame_nbr += 1
print 'end:', frame_nbr
# prepare descriptor for database
# mfccs = descr['mfcc'] # Nx13 np array (or however many mfcc coefficients there are)
# audio = descr['audio'] # Nx1 np array
# colhist = descr['colhist'] # Nx3x256 np array
# tempdif = descr['tempdiff'] # Nx1 np array
descr = {}
descr['mfcc'] = np.array(mfccs)
descr['audio'] = np.array(audio_powers)
descr['colhist'] = np.array(colorhists)
descr['tempdiff'] = np.array(sum_of_differences)
descr['chdiff'] = np.array(colorhist_diffs)
indx.add_to_index(video,descr)
print 'added ' + video + ' to database'
indx.db_commit()
sod2 = []
cd = []
while(cap.isOpened()):
#
retVal, frame = cap.read()
#
if retVal == False:
break
#== Do your processing here ==#
# You might want to write separate functions and call those here
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
frame = cv2.blur(frame, (11,11))
if prev_frame is not None:
prev_hist = ft.colorhist(prev_frame)
hist = ft.colorhist(frame)
cd.append((np.abs(prev_hist - hist)))
diff = np.abs(prev_frame.astype('int8') - frame.astype('int8'))
normalized_diff = diff/(255.0)
t = 0.2
thresholded_diff = (normalized_diff > t) * normalized_diff
sod2.append(np.sum(normalized_diff/np.prod(diff.shape)))
sod.append(np.sum(thresholded_diff)/np.prod(diff.shape))
ax1.cla()
ax2.cla()
ax1.plot(sod)
ax1.plot(sod2)
ax2.plot(cd[-1])
fig1.canvas.draw()
filename, fileExtension = os.path.splitext(args.query)
audio = filename + '.wav'
fs, wav_data = wavfile.read(audio)
query_features = []
prev_frame = None
prev_colorhist = None
frame_nbr = int(args.s)*frame_rate
cap.set(cv2.CAP_PROP_POS_MSEC, int(args.s)*1000)
while(cap.isOpened() and cap.get(cv2.CAP_PROP_POS_MSEC) < (int(args.e)*1000)):
ret, frame = cap.read()
if frame == None:
break
if args.f == features[0]:
h = ft.colorhist(frame)
elif args.f == features[1]:
h = temporal_diff(prev_frame, frame, 10)
elif args.f == features[2] or args.f == features[3]:
audio_frame = frame_to_audio(frame_nbr, frame_rate, fs, wav_data)
if args.f == features[2]:
h = np.mean(audio_frame**2)
elif args.f == features[3]:
h, mspec, spec = ft.extract_mfcc(audio_frame, fs)
elif args.f == features[4]:
colorhist = ft.colorhist(frame)
h = colorhist_diff(prev_colorhist, colorhist)
prev_colorhist = colorhist
if h != None:
