def tryload():
# -- Carga de cascadas e info OPENCV/CPU
print("-"*100)
print("Hilos:", cv2.getNumThreads())
print("Nucleos:", cv2.getNumberOfCPUs())
print("OpenCV optimizado:", cv2.useOptimized())
print("-"*100)
file = frontalface_cascades.read()
if not frontalface_cascade.load(cv2.samples.findFile(file)):
print(' -- (!)Error cargando "face cascade"')
exit(0)
file = profileface_cascades.read()
if not profileface_cascade.load(cv2.samples.findFile(file)):
print(' -- (!)Error cargando "face cascade"')
exit(0)
file = eyes_cascades.read()
if not eyes_cascade.load(cv2.samples.findFile(file)):
print(' -- (!)Error cargando "eyes cascade"')
exit(0)
file = smile_cascades.read()
if not smile_cascade.load(cv2.samples.findFile(file)):
print(' -- (!)EError cargando "smile cascade"')
exit(0)
playvideo()
def process_video(bgsubImpl,
vidreader,
vidwriter,
num_blocks=6,
threaded=False):
vidwriter.build()
threadn = cv2.getNumberOfCPUs()
pool = ThreadPool(processes=threadn)
pending = deque()
N = vidreader.frames
frameIdx = num_blocks
while True:
while len(pending) > 0 and pending[0].ready():
task = pending.popleft()
frame, idx = task.get()
vidwriter.write(frame)
print 'Proceesing ... {0}%\r'.format((idx * 100 / N)),
if len(pending) < threadn and frameIdx < N:
(cnt, frames) = vidreader.read(frameIdx - num_blocks, num_blocks)
if cnt == num_blocks:
if threaded:
task = pool.apply_async(bgsub_process,
[bgsubImpl, frames, frameIdx])
else:
task = DummyTask(bgsub_process(bgsubImpl, frames,
frameIdx))
pending.append(task)
frameIdx += 1
if len(pending) == 0:
break
def run(self):
threadn = cv2.getNumberOfCPUs() - 1
pool = ThreadPool(processes=threadn)
process_frame = Joint_View()
pending = deque()
cv2.namedWindow('Threaded Video', cv2.WINDOW_NORMAL)
for _ in range(self.duration):
frame = next(self)
task = pool.apply_async(process_frame.transform, (frame, ))
pending.append(task)
print("processing done")
for _ in range(self.duration):
if len(pending) > 0 and pending[0].ready():
result = pending.popleft().get()
if self.isLabeling:
data, label = next(self.label)
cv2.polylines(result, [data], False, (0, 0, 255), 2)
cv2.polylines(result, [label], False, (180, 105, 255), 2)
self.combine_writer.write(result)
if (self.isViz):
cv2.imshow('Threaded Video', result)
cv2.waitKey(100)
def detect_video(self, inputFname, outputFname=None, skip_frames=1):
"""Detects FEX from a video file.
Args:
inputFname (str): Path to video file
outputFname (str, optional): Path to output file. Defaults to None.
skip_frames (int, optional): Number of every other frames to skip for speed or if not all frames need to be processed. Defaults to 1.
Returns:
dataframe: Prediction results dataframe if outputFname is None. Returns True if outputFname is specified.
"""
self.info['inputFname'] = inputFname
self.info['outputFname'] = outputFname
init_df = pd.DataFrame(columns=self["output_columns"])
if outputFname:
init_df.to_csv(outputFname, index=False, header=True)
cap = cv.VideoCapture(inputFname)
# Determine whether to use multiprocessing.
n_jobs = self['n_jobs']
if n_jobs==-1:
thread_num = cv.getNumberOfCPUs() # get available cpus
else:
thread_num = n_jobs
pool = ThreadPool(processes=thread_num)
pending_task = deque()
counter = 0
processed_frames = 0
frame_got = True
detected_faces = []
print("Processing video.")
while True:
# Consume the queue.
while len(pending_task) > 0 and pending_task[0].ready():
df = pending_task.popleft().get()
# Save to output file.
if outputFname:
df.to_csv(outputFname, index=True, header=False, mode='a')
else:
init_df = pd.concat([init_df, df], axis=0)
processed_frames = processed_frames + 1
if not frame_got:
break
# Populate the queue.
if len(pending_task) < thread_num:
frame_got, frame = cap.read()
# Process at every seconds.
if counter%skip_frames == 0:
if frame_got:
task = pool.apply_async(self.process_frame, (frame.copy(), counter))
pending_task.append(task)
counter = counter + 1
cap.release()
if outputFname:
return True
else:
return init_df
def run(self):
self.results = np.array([])
# Let subscriber know the total
self.startProgress.emit(config.NISSL_COUNT)
# For updating the progress in the UI
self.total = 1
# Compute SIFT for region before starting
self.kp1, self.des1 = feature.extract_sift(self.im)
# Set multithreading if capable
if config.MULTITHREAD:
pool = ThreadPool(processes = cv2.getNumberOfCPUs())
else:
pool = ThreadPool(processes = 1)
# Total number of images to compare against
nissl = range(1, config.NISSL_COUNT + 1)
# Begin mapping process
pool.map(self.process_level, nissl)
pool.close()
pool.join()
# Tell UI the results
self.endProgress.emit(self.results)
def __init__(self, templates, ratio=0.75):
self.templates = templates
self.ratio = ratio
flann_params = dict(algorithm=FLANN_INDEX_KDTREE, trees=5)
self.matcher = cv2.FlannBasedMatcher(flann_params, {})
self.pool = ThreadPool(processes=cv2.getNumberOfCPUs())
def detect_video(self,
inputFname,
outputFname=None,
skip_frames=1,
verbose=False):
"""Detects FEX from a video file.
Args:
inputFname (str): Path to video file
outputFname (str, optional): Path to output file. Defaults to None.
skip_frames (int, optional): Number of every other frames to skip for speed or if not all frames need to be processed. Defaults to 1.
Returns:
dataframe: Prediction results dataframe if outputFname is None. Returns True if outputFname is specified.
"""
self.info["inputFname"] = inputFname
self.info["outputFname"] = outputFname
init_df = pd.DataFrame(columns=self["output_columns"])
if outputFname:
init_df.to_csv(outputFname, index=False, header=True)
cap = cv2.VideoCapture(inputFname)
length = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
frames_to_process = int(np.ceil(length / skip_frames))
# Determine whether to use multiprocessing.
n_jobs = self["n_jobs"]
if n_jobs == -1:
thread_num = cv2.getNumberOfCPUs() # get available cpus
else:
thread_num = n_jobs
if verbose:
print(f"Using {thread_num} cpus")
pool = ThreadPool(processes=thread_num)
pending_task = deque()
counter = 0
processed_frames = 0
frame_got = True
detected_faces = []
if verbose:
print("Processing video.")
# single core
while True:
frame_got, frame = cap.read()
if counter % skip_frames == 0:
df = self.process_frame(frame, counter=counter)
df["input"] = inputFname
if outputFname:
df[init_df.columns].to_csv(outputFname,
index=False,
header=False,
mode="a")
else:
init_df = pd.concat([init_df, df[init_df.columns]], axis=0)
counter = counter + 1
if not frame_got:
break
cap.release()
if outputFname:
return True
else:
return init_df
def init__threads(self):
if self.threadn == 0 :
from multiprocessing.pool import ThreadPool
self.threadn = cv2.getNumberOfCPUs()
self.pool = ThreadPool(processes = self.threadn)
from collections import deque
self.pending = deque()
def dump_matching_result(fn2, testset_full_path):
#外部ファイルに出力する
#個々のファイル毎にデータを出力する
fn, ext = os.path.splitext(fn2)
testcase_full_path = os.path.join(testset_full_path, fn2)
imgT = cv2.imread(testcase_full_path, 0)
if imgT is None:
logger.info('Failed to load fn2:', testcase_full_path)
raise ValueError('Not found the file')
logger.info("Using Training: {}".format(fn2))
pool = ThreadPool(processes=cv2.getNumberOfCPUs())
with Timer('Detection'):
kpT, descT = spla.affine_detect(detector,
imgT,
pool=pool,
simu_param='test')
logger.info('imgQ - %d features, imgT - %d features' %
(spla.count_keypoints(splt_kpQ), len(kpT)))
with Timer('matching'):
mesh_pQ, mesh_pT, mesh_pairs = spla.match_with_cross(
matcher, splt_descQ, splt_kpQ, descT, kpT)
index_mesh_pairs = format4pickle_pairs(mesh_pairs)
import joblib
dump_match_testcase_dir = myfsys.setup_output_directory(dump_match_dir, fn)
joblib.dump(mesh_pQ,
os.path.join(dump_match_testcase_dir, 'mesH_pQ.pikle'),
compress=True)
joblib.dump(mesh_pT,
os.path.join(dump_match_testcase_dir, 'mesH_pT.pikle'),
compress=True)
import pickle
with open(os.path.join(dump_match_testcase_dir, 'mesh_pairs.pickle'),
'wb') as f:
pickle.dump(index_mesh_pairs, f)
f.close()
# for i, mesh_pair in enumerate(index_mesh_pairs):
# joblib.dump(mesh_pair, os.path.join(dump_detected_testcase_dir, "mesh_pairs_{0:02d}.pikle".format(i)),
# compress=True)
with Timer('estimation'):
Hs, statuses, pairs = spla.calclate_Homography4splitmesh(mesh_pQ,
mesh_pT,
mesh_pairs,
median=median)
joblib.dump(Hs,
os.path.join(dump_match_testcase_dir, 'Hs.pikle'),
compress=True)
joblib.dump(statuses,
os.path.join(dump_match_testcase_dir, 'statuses.pikle'),
compress=True)
index_pairs = tuple(
tuple((p.pt, p.size, p.angle, p.response, p.octave, p.class_id)
for p in pair) for pair in pairs)
with open(os.path.join(dump_match_testcase_dir, 'pairs.pickle'),
'wb') as f:
pickle.dump(index_pairs, f)
def compute_fgbg_masks(vidreader,
sal,
bg,
num_prev_frames=1,
threaded=False,
num_blocks=20):
start = time.time()
def compute_mask(frameIdx, frames):
num_frames = len(frames)
_idx = num_prev_frames
fgMasks = []
bgMasks = []
while _idx < num_frames:
prev_frames = frames[_idx - num_prev_frames:_idx]
bg_variation = bg.process(frames[_idx], prev_frames)
saliency = sal.process(frames[_idx])
_idx += 1
saliency_prob = normalize(6 * bg_variation + 4 * saliency)
_, fgmask = cv2.threshold(saliency_prob, 0.6, 1, cv2.THRESH_BINARY)
_, bgmask = cv2.threshold(bg_variation, 0.1, 1,
cv2.THRESH_BINARY_INV)
fgMasks.extend([fgmask])
bgMasks.extend([bgmask])
return (frameIdx, fgMasks, bgMasks)
frameFgMasks = []
frameBgMasks = []
frameIdx = num_prev_frames + num_blocks
N = vidreader.frames
threadn = cv2.getNumberOfCPUs()
pool = ThreadPool(processes=threadn / 2)
pending = deque()
N = vidreader.frames
while True:
while len(pending) > 0 and pending[0].ready():
task = pending.popleft()
idx, fgmask, bgmask = task.get()
frameFgMasks.extend(fgmask)
frameBgMasks.extend(bgmask)
print 'Computing Mask ... {0}%\r'.format((idx * 100 / N)),
if len(pending
) < threadn and frameIdx - num_prev_frames - num_blocks < N:
(cnt,
frames) = vidreader.read(frameIdx - num_prev_frames - num_blocks,
num_prev_frames + num_blocks)
if cnt >= num_prev_frames:
if threaded:
task = pool.apply_async(compute_mask,
[min(frameIdx, N), frames])
else:
task = DummyTask(compute_mask(min(frameIdx, N), frames))
pending.append(task)
frameIdx += num_blocks
if len(pending) == 0:
break
time_taken = time.time() - start
print "Computing Mask ... [DONE] in ", time_taken, " seconds"
return (frameFgMasks, frameBgMasks)
class ThreadedReader:
thread_num = cv2.getNumberOfCPUs()
pool = ThreadPool(processes=thread_num)
pending = deque()
latency = StatValue()
frame_interval = StatValue()
last_frame_time = clock()
def run_jobs(self, f, jobs):
if self.usecloud:
jids = cloud.map(f, jobs, _env=self.cloud_env, _profile=True, _depends_on=self.preprocess_job)
ires = cloud.iresult(jids)
else:
pool = ThreadPool(processes=cv2.getNumberOfCPUs())
ires = pool.imap_unordered(f, jobs)
return ires
def main():
# 背景分割识别器序号
algo_index = 0
subtractor = ALGORITHMS_TO_EVALUATE[algo_index][0]
videoPath = "./video/vtest.avi"
show_fgmask = False
# 获得运行环境CPU的核心数
nthreads = cv2.getNumberOfCPUs()
# 设置线程数
cv2.setNumThreads(nthreads)
# 读取视频
capture = cv2.VideoCapture(videoPath)
# 当前帧数
frame_num = 0
# 总执行时间
sum_Time = 0.0
while True:
ret, frame = capture.read()
if not ret:
return
begin_time = time()
fgmask = subtractor.apply(frame)
end_time = time()
run_time = end_time - begin_time
sum_Time = sum_Time + run_time
# 平均执行时间
average_Time = sum_Time / (frame_num + 1)
if show_fgmask:
segm = fgmask
else:
segm = (frame * 0.5).astype('uint8')
cv2.add(frame, (100, 100, 0, 0), segm, fgmask)
# 显示当前方法
cv2.putText(segm, ALGORITHMS_TO_EVALUATE[algo_index][1], (10, 30), cv2.FONT_HERSHEY_PLAIN, 2.0, (255, 0, 255),
2,
cv2.LINE_AA)
# 显示当前线程数
cv2.putText(segm, str(nthreads) + " threads", (10, 60), cv2.FONT_HERSHEY_PLAIN, 2.0, (255, 0, 255), 2,
cv2.LINE_AA)
# 显示当前每帧执行时间
cv2.putText(segm, "averageTime {} s".format(average_Time), (10, 90), cv2.FONT_HERSHEY_PLAIN, 2.0,
(255, 0, 255), 2, cv2.LINE_AA);
cv2.imshow('some', segm)
key = cv2.waitKey(1) & 0xFF
frame_num = frame_num + 1
# 按'q'健退出循环
if key == ord('q'):
break
cv2.destroyAllWindows()
def __init__(self, camera, cameraName, cameraDevice, initCmd=None):
threading.Thread.__init__(self)
self._stop_event = threading.Event()
self.camera = camera
self.cameraName = cameraName
self.cameraDevice = cameraDevice
self.threadn = cv2.getNumberOfCPUs()
self.pools = {}
self.initCmd = initCmd
def asift_detect(detector, fn):
img = read_image(fn)
pool = ThreadPool(processes=cv2.getNumberOfCPUs())
with Timer('Detection with [ ASIFT ]'):
splt_kp, splt_desc = affine_detect(detector,
img,
pool=pool,
simu_param='asift')
return img, splt_kp, splt_desc
def split_asift_detect(detector, fn, split_num):
img = emod.read_image(fn)
pool = ThreadPool(processes=cv2.getNumberOfCPUs())
with Timer('Detection with [ ASIFT ]'):
splt_kp, splt_desc = saf.affine_detect_into_mesh(detector,
split_num,
img,
simu_param='asift')
return img, splt_kp, splt_desc
def affine_func(folder_number, base_image, image_sequence, rand_img):
starttime = timeit.default_timer()
FLANN_INDEX_KDTREE = 1 # bug: flann enums are missing
FLANN_INDEX_LSH = 6
feature_name = 'sift'
detector = cv2.SIFT()
norm = cv2.NORM_L2
flann_params = dict(algorithm = FLANN_INDEX_KDTREE, trees = 5)
matcher = cv2.FlannBasedMatcher(flann_params, {})
img1 = cv2.imread(base_image, 0)
features = []
inliers = []
matches = []
base_features = []
img_features = []
print 'Processing image sequence using ASIFT'
for i in range(0,len(image_sequence)):
print 'Processing image %d of %d images' % (i, len(image_sequence)-1)
fn2 = image_sequence[i]
img2 = cv2.imread(fn2, 0)
pool=ThreadPool(processes = cv2.getNumberOfCPUs())
kp1, desc1 = affine_detect(detector, img1, pool=pool)
kp2, desc2 = affine_detect(detector, img2, pool=pool)
base_features.append(len(kp1))
img_features.append(len(kp2))
raw_matches = matcher.knnMatch(desc1, trainDescriptors = desc2, k = 2)
p1, p2, kp_pairs = rect.filter_matches(kp1, kp2, raw_matches)
if len(p1) >= 4:
H, status = cv2.findHomography(p1, p2, cv2.RANSAC, 5.0)
a = np.sum(status)
matches.append(len(status))
inliers.append(a)
print '%d / %d inliers/matched' % (np.sum(status), len(status))
# Save randomly selected feature mapping
if i == rand_img:
print 'Random image selected...'
img_draw = img2
kp_pairs_draw = kp_pairs
status_draw = status
H_draw = H
filename = "Image"+str(i)+"_asift.png"
vis = rect.explore_match(folder_number, filename, img1, img_draw, kp_pairs_draw, status_draw, H_draw)
else:
H, status = None, None
print '%d matches found, not enough for homography estimation' % len(p1)
matches.append(len(p1))
inliers.append(0)
stoptime = timeit.default_timer()
run_time = stoptime - starttime
return (base_features, img_features, matches, inliers, run_time)
def __init__(self):
# Configuracion del pool
self.threadn = cv2.getNumberOfCPUs()
self.pool = ThreadPool(processes = self.threadn)
self.pending = deque()
# Defino los clock para tomar tiempos
self.latency = StatValue()
self.frame_interval = StatValue()
self.last_frame_time = clock()
def detect_and_match(detector, matcher, set_fn, splt_num=64, simu_type="default"):
"""
SplitA実験
set_fn:
"""
fnQ, testcase, fnT = set_fn
def get_expt_names():
tmpf, tmpext = os.path.splitext(fnT)
return (os.path.basename(__file__), testcase, tmpf)
expt_names = get_expt_names()
logger = setup(expt_names)
logger.info(__doc__)
full_fnQ = myfsys.getf_template((fnQ,))
full_fnT = myfsys.getf_input(testcase, fnT)
imgQ, imgT = read_images(full_fnQ, full_fnT, logger)
pool = ThreadPool(processes=cv2.getNumberOfCPUs())
with Timer('Detection with SPLIT-ASIFT', logger):
splt_kpQ, splt_descQ = spltA.affine_detect_into_mesh(detector, splt_num, imgQ, simu_param=simu_type)
with Timer('Detection with SFIT', logger):
kpT, descT = affine_detect(detector, imgT, pool=pool, simu_param='test')
logger.info('imgQ - {0} features, imgT - {1} features'.format(spltA.count_keypoints(splt_kpQ), len(kpT)))
with Timer('matching', logger):
mesh_pQ, mesh_pT, mesh_pairs = spltA.match_with_cross(matcher, splt_descQ, splt_kpQ, descT, kpT)
Hs = []
statuses = []
kp_pairs_long = []
Hs_stable = []
kp_pairs_long_stable = []
for pQ, pT, pairs in zip(mesh_pQ, mesh_pT, mesh_pairs):
pairs, H, status = calclate_Homography(pQ, pT, pairs)
Hs.append(H)
statuses.append(status)
if status is not None and not len(status) == 0 and np.sum(status)/len(status) >= 0.4:
Hs_stable.append(H)
else:
Hs_stable.append(None)
for p in pairs:
kp_pairs_long.append(p)
if status is not None and not len(status) == 0 and np.sum(status)/len(status) >= 0.4:
kp_pairs_long_stable.append(p)
vis = draw_matches_for_meshes(imgQ, imgT, Hs=Hs)
cv2.imwrite(myfsys.getf_output(expt_names, 'meshes.png'), vis)
visS = draw_matches_for_meshes(imgQ, imgT, Hs=Hs_stable)
cv2.imwrite(myfsys.getf_output(expt_names, 'meshes_stable.png'), visS)
viw = explore_match_for_meshes('affine find_obj', imgQ, imgT, kp_pairs_long_stable, Hs=Hs_stable)
cv2.imwrite(myfsys.getf_output(expt_names, 'meshes_and_keypoints_stable.png'), viw)
return vis, visS, viw
def video_parse(input_video = None, output_video = None, display = False):
#Get capture object and read frame
if input_video is not None:
cap = cv2.VideoCapture(input_video)
else:
cap = cv2.VideoCapture(0)
# Space for opencv setup
#Generate output objects if is required
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
out_v = None
if output_video is not None:
out_v = cv2.VideoWriter(output_video,fourcc, 20.0, (int(previous_frame.shape[1]),int(previous_frame.shape[0])))
## Generate a queue for the processes
n_threads = cv2.getNumberOfCPUs()
pool = ThreadPool(processes = n_threads)
pending_frames = deque()
## While the video is running
while(cap.isOpened()):
# Unload the ready frames
while len(pending_frames) > 0 and pending_frames[0].ready():
#Get the completed job
res = pending_frames.popleft().get()
#Write out the file to video if requested
if out_va is not None:
out_va.write(res)
#Display ready frame
if display:
cv2.imshow(' Frame ', res)
#If there are approx threads free, get more jobs
if len(pending_frames) < n_threads:
## Generate the new job
ret, frame = cap.read()
if ret == False:
break
# drawing componenet to the fraem - only the compoenents not culled
task = pool.apply_async(process_frame, (frame.copy()))
pending_frames.append(task)
# Clean up the used elements
if out_v is not None:
out_v.release()
cap.release()
cv2.destroyAllWindows()
def main():
import sys, getopt
opts, args = getopt.getopt(sys.argv[1:], '', ['feature='])
opts = dict(opts)
feature_name = opts.get('--feature', 'brisk-flann')
try:
fn1, fn2 = args
except:
fn1 = 'data/aero1.jpg'
fn2 = 'data/aero3.jpg'
img1 = cv.imread(cv.samples.findFile(fn1), cv.IMREAD_GRAYSCALE)
img2 = cv.imread(cv.samples.findFile(fn2), cv.IMREAD_GRAYSCALE)
detector, matcher = init_feature(feature_name)
if img1 is None:
print('Failed to load fn1:', fn1)
sys.exit(1)
if img2 is None:
print('Failed to load fn2:', fn2)
sys.exit(1)
if detector is None:
print('unknown feature:', feature_name)
sys.exit(1)
print('using', feature_name)
pool = ThreadPool(processes=cv.getNumberOfCPUs())
kp1, desc1 = affine_detect(detector, img1, pool=pool)
kp2, desc2 = affine_detect(detector, img2, pool=pool)
print('img1 - %d features, img2 - %d features' % (len(kp1), len(kp2)))
def match_and_draw(win):
with Timer('matching'):
raw_matches = matcher.knnMatch(desc1, trainDescriptors=desc2,
k=2) #2
p1, p2, kp_pairs = filter_matches(kp1, kp2, raw_matches)
if len(p1) >= 4:
H, status = cv.findHomography(p1, p2, cv.RANSAC, 5.0)
print('%d / %d inliers/matched' % (np.sum(status), len(status)))
# do not draw outliers (there will be a lot of them)
kp_pairs = [kpp for kpp, flag in zip(kp_pairs, status) if flag]
else:
H, status = None, None
print('%d matches found, not enough for homography estimation' %
len(p1))
explore_match(win, img1, img2, kp_pairs, None, H)
match_and_draw('affine find_obj')
cv.waitKey()
print('Done')
def main():
import sys, getopt
opts, args = getopt.getopt(sys.argv[1:], '', ['feature='])
opts = dict(opts)
feature_name = opts.get('--feature', 'brisk-flann')
try:
fn1, fn2 = args
except:
fn1 = 'aero1.jpg'
fn2 = 'aero3.jpg'
img1 = cv.imread(cv.samples.findFile(fn1), cv.IMREAD_GRAYSCALE)
img2 = cv.imread(cv.samples.findFile(fn2), cv.IMREAD_GRAYSCALE)
detector, matcher = init_feature(feature_name)
if img1 is None:
print('Failed to load fn1:', fn1)
sys.exit(1)
if img2 is None:
print('Failed to load fn2:', fn2)
sys.exit(1)
if detector is None:
print('unknown feature:', feature_name)
sys.exit(1)
print('using', feature_name)
pool=ThreadPool(processes = cv.getNumberOfCPUs())
kp1, desc1 = affine_detect(detector, img1, pool=pool)
kp2, desc2 = affine_detect(detector, img2, pool=pool)
print('img1 - %d features, img2 - %d features' % (len(kp1), len(kp2)))
def match_and_draw(win):
with Timer('matching'):
raw_matches = matcher.knnMatch(desc1, trainDescriptors = desc2, k = 2) #2
p1, p2, kp_pairs = filter_matches(kp1, kp2, raw_matches)
if len(p1) >= 4:
H, status = cv.findHomography(p1, p2, cv.RANSAC, 5.0)
print('%d / %d inliers/matched' % (np.sum(status), len(status)))
# do not draw outliers (there will be a lot of them)
kp_pairs = [kpp for kpp, flag in zip(kp_pairs, status) if flag]
else:
H, status = None, None
print('%d matches found, not enough for homography estimation' % len(p1))
explore_match(win, img1, img2, kp_pairs, None, H)
match_and_draw('affine find_obj')
cv.waitKey()
print('Done')
def matching_image(fn1='aero1.jpg', fn2='aero3.jpg', feature='sift'):
import sys
feature_name = feature
detector, matcher = init_feature(feature_name)
try:
img1 = cv.imread(cv.samples.findFile(fn1), cv.IMREAD_GRAYSCALE)
img2 = cv.imread(cv.samples.findFile(fn2), cv.IMREAD_GRAYSCALE)
except:
img1 = fn1
img2 = fn2
if img1 is None:
print('Failed to load fn1:', fn1)
sys.exit(1)
if img2 is None:
print('Failed to load fn2:', fn2)
sys.exit(1)
if detector is None:
print('unknown feature:', feature_name)
sys.exit(1)
print('using', feature_name)
pool=ThreadPool(processes = cv.getNumberOfCPUs())
kp1, desc1 = affine_detect(detector, img1, pool=pool)
kp2, desc2 = affine_detect(detector, img2, pool=pool)
print('img1 - %d features, img2 - %d features' % (len(kp1), len(kp2)))
def match_and_draw(win):
with Timer('matching'):
raw_matches = matcher.knnMatch(desc1, trainDescriptors = desc2, k = 2) #2
p1, p2, kp_pairs = filter_matches(kp1, kp2, raw_matches)
if len(p1) >= 4:
H, status = cv.findHomography(p1, p2, cv.RANSAC, 5.0)
print('%d / %d inliers/matched' % (np.sum(status), len(status)))
# do not draw outliers (there will be a lot of them)
kp_pairs = [kpp for kpp, flag in zip(kp_pairs, status) if flag]
else:
H, status = None, None
print('%d matches found, not enough for homography estimation' % len(p1))
explore_match(win, img1, img2, kp_pairs, None, H)
return blending(img1, img2, H)
final = match_and_draw('affine find_obj')
cv.imwrite('{}_panorama.jpg'.format(feature), final)
cv.waitKey()
print('Done')
def affine_detect_into_mesh(detector,
split_num,
img1,
mask=None,
simu_param='default'):
pool = ThreadPool(processes=cv2.getNumberOfCPUs())
kp, desc = affine_detect(detector,
img1,
mask,
pool=pool,
simu_param=simu_param)
return split_kd(kp, desc, split_num)
def run_jobs(self, f, jobs):
if self.usecloud:
jids = cloud.map(f,
jobs,
_env=self.cloud_env,
_profile=True,
_depends_on=self.preprocess_job)
ires = cloud.iresult(jids)
else:
pool = ThreadPool(processes=cv2.getNumberOfCPUs())
ires = pool.imap_unordered(f, jobs)
return ires
def main():
import sys
try:
fn = sys.argv[1]
except:
fn = 0
cap = video.create_capture(fn)
def process_frame(frame, t0):
# some intensive computation...
frame = cv.medianBlur(frame, 19)
frame = cv.medianBlur(frame, 19)
return frame, t0
threadn = cv.getNumberOfCPUs()
pool = ThreadPool(processes=threadn)
pending = deque()
threaded_mode = True
latency = StatValue()
frame_interval = StatValue()
last_frame_time = clock()
while True:
while len(pending) > 0 and pending[0].ready():
res, t0 = pending.popleft().get()
latency.update(clock() - t0)
draw_str(res, (20, 20), "threaded : " + str(threaded_mode))
draw_str(res, (20, 40),
"latency : %.1f ms" % (latency.value * 1000))
draw_str(
res, (20, 60),
"frame interval : %.1f ms" % (frame_interval.value * 1000))
cv.imshow('threaded video', res)
if len(pending) < threadn:
_ret, frame = cap.read()
t = clock()
frame_interval.update(t - last_frame_time)
last_frame_time = t
if threaded_mode:
task = pool.apply_async(process_frame, (frame.copy(), t))
else:
task = DummyTask(process_frame(frame, t))
pending.append(task)
ch = cv.waitKey(1)
if ch == ord(' '):
threaded_mode = not threaded_mode
if ch == 27:
break
print('Done')
def compute_fgbg_masks(vidreader,gtreader,sal,bg,num_prev_frames=1,skip_frames=0,
last_frame=1000,threaded=False,num_blocks=10,):
start = time.time();
def compute_mask(frameIdx,frames,gtframes):
num_frames = len(frames); _idx = num_prev_frames
fgMasks = []; bgMasks = []; gtMasks = [];
while _idx<num_frames:
prev_frames = frames[_idx-num_prev_frames:_idx]
gtmask = getMask(gtframes[_idx-num_prev_frames]);
bg_variation = bg.process(frames[_idx],prev_frames);
saliency = sal.process(frames[_idx]);
_idx += 1; saliency_prob = normalize(7*bg_variation + 3*saliency);
_,fgmask = cv2.threshold(saliency_prob ,0.5,1,cv2.THRESH_BINARY)
fgmask = cv2.morphologyEx(fgmask, cv2.MORPH_OPEN, KERNEL)
EL_KERNEL = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(2,2))
fgmask = cv2.dilate(np.uint8(fgmask),EL_KERNEL,iterations = 2)
fgmask = cv2.medianBlur(fgmask,3)
fgmask = binary_fill_holes(fgmask)
fgmask = cv2.dilate(np.uint8(fgmask),EL_KERNEL,iterations = 2)
fgmask = cv2.morphologyEx(np.uint8(fgmask), cv2.MORPH_CLOSE,KERNEL)
_,bgmask = cv2.threshold(bg_variation,0.1,1,cv2.THRESH_BINARY_INV)
zero_frame= np.zeros(frames[0].shape,dtype = np.uint8);
zero_frame[:,:,2]=fgmask*255;
zero_frame[:,:,1]=bgmask*255;
_prob = cv2.cvtColor(saliency_prob*255,cv2.COLOR_GRAY2RGB);
fgMasks.extend([fgmask]); bgMasks.extend([bgmask]); gtMasks.extend([gtmask])
return (frameIdx,fgMasks,bgMasks,gtMasks);
frameFgMasks = []; frameBgMasks = []; frameGTMasks = [];
frameIdx = skip_frames+num_prev_frames+num_blocks; N = vidreader.frames;
threadn = cv2.getNumberOfCPUs(); pool = ThreadPool(processes = threadn/2);
pending = deque(); N = min(vidreader.frames,last_frame);
while True:
while len(pending) > 0 and pending[0].ready():
task = pending.popleft()
idx,fgmask,bgmask,gtmask = task.get()
frameFgMasks.extend(fgmask); frameBgMasks.extend(bgmask); frameGTMasks.extend(gtmask);
print 'Computing Mask ... {0}%\r'.format((idx*100/N)),
if len(pending) < threadn and frameIdx-num_prev_frames-num_blocks < N:
(cnt,frames) = vidreader.read(frameIdx-num_prev_frames-num_blocks,num_prev_frames+num_blocks);
(_,gtframes) = gtreader.read(frameIdx-num_blocks,num_blocks);
if cnt >= num_prev_frames:
if threaded:
task = pool.apply_async(compute_mask,[min(frameIdx,N),frames,gtframes]);
else:
task = DummyTask(compute_mask(min(frameIdx,N),frames,gtframes));
pending.append(task)
frameIdx += num_blocks;
if len(pending) == 0:
break;
time_taken = time.time()-start;
print "Computing Mask ... [DONE] in ",time_taken," seconds"
return (frameFgMasks,frameBgMasks,frameGTMasks)
def __init__(self, stream):
self.stream = stream
self.numThread = cv2.getNumberOfCPUs()
#self.numThread = 1
self.workerPool = ThreadPool(processes = self.numThread)
self.pendingWorker = deque()
self.latency = StatValue()
self.frameInterval = StatValue()
self.lastFrameTime = clock()
self.outFrames = deque(maxlen = self.numThread)
self.faces = []
def __init__(self, stream):
self.stream = stream
self.numThread = cv2.getNumberOfCPUs()
#self.numThread = 5
self.workerPool = ThreadPool(processes=self.numThread)
self.pendingWorker = deque()
self.latency = StatValue()
self.frameInterval = StatValue()
self.lastFrameTime = clock()
self.outFrames = deque(maxlen=self.numThread)
self.faces = []
def check_build_info():
success = True
print("OpenCV Version: {}".format(cv2.__version__))
if (cv2.getVersionMajor() != CURRENT_OPENCV_BUILD_VERSION[0]) and (
cv2.getVersionMinor() != CURRENT_OPENCV_BUILD_VERSION[1]) and (
cv2.getVersionRevision() != CURRENT_OPENCV_BUILD_VERSION[2]):
print("ERROR: OpenCV version is different than the expected.")
success = False
print("Available CPUs: ", cv2.getNumberOfCPUs())
print("Available threads: ", cv2.getNumThreads())
if cv2.getNumThreads() < cv2.getNumberOfCPUs():
print("ERROR: TBB is not enabled.")
success = False
cv2.CPU_NEON = 100 # Value taken from OpenCV doc. CPU labels don't work correctly in Python
print("Cpu NEON support: ", cv2.checkHardwareSupport(cv2.CPU_NEON))
if not cv2.checkHardwareSupport(cv2.CPU_NEON):
print("ERROR: NEON is not enabled.")
success = False
return success
def find_asift_homography(fn1, fn2, scale=1.0):
feature_name = 'orb-flann'
img1 = cv2.imread(fn1, 0)
img2 = cv2.imread(fn2, 0)
detector, matcher = init_feature(feature_name, n=400)
if img1 is None:
print('Failed to load fn1:', fn1)
return
if img2 is None:
print('Failed to load fn2:', fn2)
return
if detector is None:
print('unknown feature:', feature_name)
return
print('using', feature_name)
# cv2.INTER_LINEAR
# img1 = cv2.resize(img1, (0, 0), fx=fx, fy=fy, interpolation=cv2.INTER_AREA)
w, h = get_size(img1)
img1 = cv2.resize(img1, (int(w * scale), int(h * scale)))
w, h = get_size(img2)
img2 = cv2.resize(img2, (int(w * scale), int(h * scale)))
pool = ThreadPool(processes=cv2.getNumberOfCPUs())
kp1, desc1 = affine_detect(detector, img1, pool=pool)
kp2, desc2 = affine_detect(detector, img2, pool=pool)
print('img1 - %d features, img2 - %d features' % (len(kp1), len(kp2)))
with Timer('matching'):
raw_matches = matcher.knnMatch(desc1, trainDescriptors=desc2, k=2) #2
p1, p2, kp_pairs = filter_matches(kp1, kp2, raw_matches)
if len(p1) >= 4:
H, status = cv2.findHomography(p1, p2, cv2.RANSAC, 5.0)
print('%d / %d inliers/matched' % (np.sum(status), len(status)))
# do not draw outliers (there will be a lot of them)
kp_pairs = [kpp for kpp, flag in zip(kp_pairs, status) if flag]
else:
H, status = None, None
print('%d matches found, not enough for homography estimation' %
len(p1))
return H, kp_pairs
def __init__(self, queue, cmd, user_queue, cct):
multiprocessing.Process.__init__(self)
CameraServer.queue = queue
self.trainor_queue = multiprocessing.Queue(64)
self.svm_queue = multiprocessing.Queue(64)
self.cmd = cmd
from bvps.camera.camera import Camera
self.user_queue = user_queue
self.cct = cct
self.detector = detector()
self.recognizer = recognizer(None)
self.threadn = cv2.getNumberOfCPUs()
self.pools = {}
def detect_image(img1, img2):
import sys, getopt
opts, args = getopt.getopt(sys.argv[1:], '', ['feature='])
opts = dict(opts)
feature_name = opts.get('--feature', 'brisk-flann')
detector, matcher = init_feature(feature_name)
if img1 is None:
print('Failed to load fn1:', fn1)
sys.exit(1)
if img2 is None:
print('Failed to load fn2:', fn2)
sys.exit(1)
if detector is None:
print('unknown feature:', feature_name)
sys.exit(1)
print('using', feature_name)
img1 = cv.GaussianBlur(img1, (5, 5), sigmaX=5)
pool = ThreadPool(processes=cv.getNumberOfCPUs())
kp1, desc1 = affine_detect(detector, img1, pool=pool)
kp2, desc2 = affine_detect(detector, img2, pool=pool)
print('img1 - %d features, img2 - %d features' % (len(kp1), len(kp2)))
def match_and_draw(win):
with Timer('matching'):
raw_matches = matcher.knnMatch(desc1, trainDescriptors=desc2,
k=2) #2
p1, p2, kp_pairs = filter_matches(kp1, kp2, raw_matches)
if len(p1) >= 4:
H, status = cv.findHomography(p1, p2, cv.RANSAC, 5.0)
print('%d / %d inliers/matched' % (np.sum(status), len(status)))
# do not draw outliers (there will be a lot of them)
kp_pairs = [kpp for kpp, flag in zip(kp_pairs, status) if flag]
else:
H, status = None, None
print('%d matches found, not enough for homography estimation' %
len(p1))
img, corners = explore_match(win, img1, img2, kp_pairs, None, H)
return img, corners, kp_pairs
return match_and_draw('affine find_obj')
def run(self):
"""Run the main loop."""
threadn = cv2.getNumberOfCPUs()
pool = ThreadPool(processes=threadn)
pending = deque()
# latency = StatValue()
# frame_interval = StatValue()
# last_frame_time = clock()
# TODO: Camera Calibration, Video Stabilization
self._windowManager.create_window()
while self._windowManager.is_window_created:
self._captureManager.enter_frame()
original = self._captureManager.original
self._captureManager.frame = original
# if original is not None:
# output = self.process_and_detect(original)
# self._captureManager.frame = output§
while len(pending) > 0 and pending[0].ready():
output = pending.popleft().get()
# latency.update(clock() - t0)
cv2.putText(output, "threaded : " + str(self._thread_mode),
(15, 80), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2)
# draw_str(res, (20, 40), "latency : %.1f ms" % (latency.value * 1000))
# draw_str(res, (20, 60), "frame interval : %.1f ms" % (frame_interval.value * 1000))
self._captureManager.frame = output
self._captureManager.exit_frame()
if len(pending) < threadn:
# ret, frame = cap.read()
# t = clock()
# frame_interval.update(t - last_frame_time)
# last_frame_time = t
if self._thread_mode:
task = pool.apply_async(self.process_and_detect, (original.copy(),))
else:
task = DummyTask(self.process_and_detect(original))
pending.append(task)
self._captureManager.exit_frame()
self._windowManager.process_events()
def extract_sift(im):
if im.dtype != np.uint8:
im = (im * 255).astype(np.uint8)
if len(im.shape) == 3:
im = cv2.cvtColor(im, cv2.COLOR_RGB2GRAY)
# Use multithreading to split the affine detection work
pool = ThreadPool(processes=cv2.getNumberOfCPUs())
#kp, des = affine_detect(SIFT, im, pool=pool)
kp, des = SIFT.detectAndCompute(im, None)
pool.close()
pool.join()
return kp, des
def compute_fgbg_masks(vidreader,sal,bg,num_prev_frames=1,threaded=False,num_blocks=20):
start = time.time();
def compute_mask(frameIdx,frames):
num_frames = len(frames); _idx = num_prev_frames
fgMasks = []; bgMasks = [];
while _idx<num_frames:
prev_frames = frames[_idx-num_prev_frames:_idx]
bg_variation = bg.process(frames[_idx],prev_frames);
saliency = sal.process(frames[_idx]);
_idx += 1; saliency_prob = normalize(7*bg_variation + 3*saliency);
_,fgmask = cv2.threshold(saliency_prob ,0.5,1,cv2.THRESH_BINARY)
_,bgmask = cv2.threshold(bg_variation,0.1,1,cv2.THRESH_BINARY_INV)
name = "test_results/smoothening/{0}_sal.png".format(frameIdx-num_blocks-num_prev_frames+_idx);
zero_frame= np.zeros(frames[0].shape,dtype = np.uint8);
zero_frame[:,:,2]=fgmask*255;
zero_frame[:,:,1]=bgmask*255;
_prob = cv2.cvtColor(saliency_prob*255,cv2.COLOR_GRAY2RGB);
out = cv2.addWeighted(np.uint8(_prob),0.6,zero_frame,0.4,0);
cv2.imwrite(name,out)
fgMasks.extend([fgmask]); bgMasks.extend([bgmask])
return (frameIdx,fgMasks,bgMasks);
frameFgMasks = []; frameBgMasks = [];
frameIdx = num_prev_frames+num_blocks; N = vidreader.frames;
threadn = cv2.getNumberOfCPUs(); pool = ThreadPool(processes = threadn/2);
pending = deque(); N = vidreader.frames;
while True:
while len(pending) > 0 and pending[0].ready():
task = pending.popleft()
idx,fgmask,bgmask = task.get()
frameFgMasks.extend(fgmask); frameBgMasks.extend(bgmask);
print 'Computing Mask ... {0}%\r'.format((idx*100/N)),
if len(pending) < threadn and frameIdx-num_prev_frames-num_blocks < N:
(cnt,frames) = vidreader.read(frameIdx-num_prev_frames-num_blocks,num_prev_frames+num_blocks);
if cnt >= num_prev_frames:
if threaded:
task = pool.apply_async(compute_mask,[min(frameIdx,N),frames]);
else:
task = DummyTask(compute_mask(min(frameIdx,N),frames));
pending.append(task)
frameIdx += num_blocks;
if len(pending) == 0:
break;
time_taken = time.time()-start;
print "Computing Mask ... [DONE] in ",time_taken," seconds"
return (frameFgMasks,frameBgMasks)
def test_result(self):
pool = ThreadPool(processes=cv2.getNumberOfCPUs())
meshList_kpT, meshList_descT = splta2.affine_detect_into_mesh(
self.detector, self.splt_num, self.img1, pool=pool)
kpQ, descQ = affine_detect(self.detector, self.img2, pool=pool)
def count_keypoints():
c = 0
for s_kpT in meshList_kpT:
for kpT in s_kpT:
c += len(kpT)
return c
print('imgQ - %d features, imgT - %d features' %
(count_keypoints(), len(kpQ)))
mesh_pT, mesh_pQ, mesh_pairs = splta2.match_with_cross(
self.matcher, meshList_descT, meshList_kpT, descQ, kpQ)
self.assertTrue(True)
def process_video(bgsubImpl,vidreader,vidwriter,num_blocks=6, threaded = False):
vidwriter.build();
threadn = cv2.getNumberOfCPUs(); pool = ThreadPool(processes = threadn);
pending = deque(); N = vidreader.frames; frameIdx = num_blocks;
while True:
while len(pending) > 0 and pending[0].ready():
task = pending.popleft()
frame, idx = task.get()
vidwriter.write(frame);
print 'Proceesing ... {0}%\r'.format((idx*100/N)),
if len(pending) < threadn and frameIdx < N:
(cnt,frames) = vidreader.read(frameIdx-num_blocks,num_blocks);
if cnt == num_blocks:
if threaded:
task = pool.apply_async(bgsub_process,[bgsubImpl,frames,frameIdx]);
else:
task = DummyTask(bgsub_process(bgsubImpl,frames,frameIdx));
pending.append(task)
frameIdx += 1;
if len(pending) == 0:
break;
def compute_fgbg_masks(vidreader,sal,bg,num_prev_frames=1,threaded=False,num_blocks=20):
start = time.time();
def compute_mask(frameIdx,frames):
num_frames = len(frames); _idx = num_prev_frames
fgMasks = []; bgMasks = [];
while _idx<num_frames:
prev_frames = frames[_idx-num_prev_frames:_idx]
bg_variation = bg.process(frames[_idx],prev_frames);
saliency = sal.process(frames[_idx]);
_idx += 1; saliency_prob = normalize(6*bg_variation + 4*saliency);
_,fgmask = cv2.threshold(saliency_prob ,0.6,1,cv2.THRESH_BINARY)
_,bgmask = cv2.threshold(bg_variation,0.1,1,cv2.THRESH_BINARY_INV)
fgMasks.extend([fgmask]); bgMasks.extend([bgmask])
return (frameIdx,fgMasks,bgMasks);
frameFgMasks = []; frameBgMasks = [];
frameIdx = num_prev_frames+num_blocks; N = vidreader.frames;
threadn = cv2.getNumberOfCPUs(); pool = ThreadPool(processes = threadn/2);
pending = deque(); N = vidreader.frames;
while True:
while len(pending) > 0 and pending[0].ready():
task = pending.popleft()
idx,fgmask,bgmask = task.get()
frameFgMasks.extend(fgmask); frameBgMasks.extend(bgmask);
print 'Computing Mask ... {0}%\r'.format((idx*100/N)),
if len(pending) < threadn and frameIdx-num_prev_frames-num_blocks < N:
(cnt,frames) = vidreader.read(frameIdx-num_prev_frames-num_blocks,num_prev_frames+num_blocks);
if cnt >= num_prev_frames:
if threaded:
task = pool.apply_async(compute_mask,[min(frameIdx,N),frames]);
else:
task = DummyTask(compute_mask(min(frameIdx,N),frames));
pending.append(task)
frameIdx += num_blocks;
if len(pending) == 0:
break;
time_taken = time.time()-start;
print "Computing Mask ... [DONE] in ",time_taken," seconds"
return (frameFgMasks,frameBgMasks)
feature_name = opts.get('--feature', 'sift-flann')
try: fn1, fn2 = args
except:
fn1 = 'C:/opencv-2.4.9/sources/samples/python2/data/aero1.jpg'#入力画像
fn2 = 'C:/opencv-2.4.9/sources/samples/python2/data/aero3.jpg'
img1 = cv2.imread(fn1, 0)
img2 = cv2.imread(fn2, 0)
detector, matcher = init_feature(feature_name)
if detector != None:
print 'using', feature_name
else:
print 'unknown feature:', feature_name
sys.exit(1)
pool=ThreadPool(processes = cv2.getNumberOfCPUs())
kp1, desc1 = affine_detect(detector, img1, pool=pool)
kp2, desc2 = affine_detect(detector, img2, pool=pool)
print 'img1 - %d features, img2 - %d features' % (len(kp1), len(kp2))
def match_and_draw(win):
with Timer('matching'):
raw_matches = matcher.knnMatch(desc1, trainDescriptors = desc2, k = 2) #2
p1, p2, kp_pairs = filter_matches(kp1, kp2, raw_matches)
if len(p1) >= 4:
H, status = cv2.findHomography(p1, p2, cv2.RANSAC, 5.0)
print '%d / %d inliers/matched' % (np.sum(status), len(status))
# do not draw outliers (there will be a lot of them)
kp_pairs = [kpp for kpp, flag in zip(kp_pairs, status) if flag]
else:
H, status = None, None
def imrestore(images, **opts):
"""
Restore images by merging and stitching techniques.
:param images: list of images or string to path which uses glob filter in path.
Loads image array from path, url, server, string
or directly from numpy array (supports databases)
:param debug: (0) flag to print debug messages
0 -> do not print messages.
1 -> print messages.
2 -> print messages and show results
(consumes significantly more memory).
3 -> print messages, show results and additional data
(consumes significantly more memory).
:param feature: (None) feature instance. It contains the configured
detector and matcher.
:param pool: (None) use pool Ex: 4 to use 4 CPUs.
:param cachePath: (None) saves memoization to specified path and
downloaded images.
:param clearCache: (0) clear cache flag.
* 0 do not clear.
* 1 All CachePath is cleared before use.
* 2 re-compute data but other cache data is left intact.
Notes: using cache can result in unspected behaviour
if some configurations does not match to the cached data.
:param loader: (None) custom loader function used to load images
to merge. If None it loads the original images in color.
:param pshape: (400,400) process shape, used to load pseudo images
to process features and then results are converted to the
original images. If None it loads the original images to
process the features but it can incur to performance penalties
if images are too big and RAM memory is scarce.
:param baseImage: (None) First image to merge to.
* None -> takes first image from raw list.
* True -> selects image with most features.
* Image Name.
:param selectMethod: (None) Method to sort images when matching. This
way the merging order can be controlled.
* (None) Best matches.
* Histogram Comparison: Correlation, Chi-squared,
Intersection, Hellinger or any method found in hist_map
* Entropy.
* custom function of the form: rating,fn <-- selectMethod(fns)
:param distanceThresh: (0.75) filter matches by distance ratio.
:param inlineThresh: (0.2) filter homography by inlineratio.
:param rectangularityThresh: (0.5) filter homography by rectangularity.
:param ransacReprojThreshold: (5.0) maximum allowed reprojection error
to treat a point pair as an inlier.
:param centric: (False) tries to attach as many images as possible to
each matching. It is quicker since it does not have to process
too many match computations.
:param hist_match: (False) apply histogram matching to foreground
image with merge image as template
:param mergefunc: (None) function used to merge foreground with
background image using the given transformation matrix.
The structure is as follows:
merged, H_back, H_fore= mergefunc(back,fore,H)
..where::
back: background image
fore: foreground image
H: calculated Transformation Matrix
merged: new image of fore in back image
H_back: transformation matrix that modifies
background key-points
H_fore: transformation matrix that modifies
foreground key-points
:param postfunc: (None) function used for post processing
the merging result. The function is called with the merging
image and must return the processed image.
:param save: (False)
* True, saves in path with name restored_{base_image}
* False, does not save
* Image name used to save the restored image.
:return: restored image
"""
# for debug
FLAG_DEBUG = opts.get("debug",1)
# for multiprocessing
pool = opts.get("pool",None)
if pool is not None: # convert pool count to pool class
NO_CPU = cv2.getNumberOfCPUs()
if pool <= NO_CPU:
pool = Pool(processes = pool)
else:
raise Exception("pool of {} exceeds the number of processors {}".format(pool,NO_CPU))
# for features
feature = opts.get("feature",None)
if feature is None:
feature = Feature(pool=pool,debug=FLAG_DEBUG)
feature.config(name='a-sift-flann') # init detector and matcher to compute descriptors
else:
feature.pool = pool
feature.debug = FLAG_DEBUG
# select method to order images to feed in superposition
selectMethod = opts.get("selectMethod",None)
best_match_list = ("bestmatches", "best matches")
entropy_list = ("entropy",)
if callable(selectMethod):
orderValue = 3
elif selectMethod in hist_map:
orderValue = 2
elif selectMethod in entropy_list:
orderValue = 1
elif selectMethod in best_match_list or selectMethod is None:
orderValue = 0
else:
raise Exception("selectMethod {} not recognized".format(selectMethod))
# distance threshold to filter best matches
distanceThresh = opts.get("distanceThresh",0.75) # filter ratio
# threshold for inlineRatio
inlineThresh = opts.get("inlineThresh",0.2) # filter ratio
assert inlineThresh<=1 and inlineThresh>=0 # ensures adequate value [0,1]
# threshold for rectangularity
rectangularityThresh = opts.get("rectangularityThresh",0.5) # filter ratio
assert rectangularityThresh<=1 and rectangularityThresh>=0 # ensures adequate value [0,1]
# threshold to for RANSAC reprojection
ransacReprojThreshold = opts.get("ransacReprojThreshold",5.0)
centric = opts.get("centric",False) # tries to attach as many images as possible
pshape = opts.get("process_shape",(400,400))# it is not safe to compute descriptors from big images
usepshape = False # output is as process_shape if True, else process with process_shape but output is as loader
minKps = 3 # minimum len of key-points to find Homography
histMatch = opts.get("hist_match",False)
mergefunc = opts.get("mergefunc",None)
if mergefunc is None:
mergefunc = superpose
assert callable(mergefunc)
postfunc = opts.get("postfunc",None)
assert postfunc is None or callable(postfunc)
############################## OPTIMIZATION MEMOIZEDIC #############################
cachePath = opts.get("cachePath",None)
if cachePath is not None:
feature_dic = MemoizedDict(cachePath + "descriptors")
if FLAG_DEBUG: print("Cache path is in {}".format(feature_dic._path))
clearCache = opts.get("clearCache",0)
if clearCache==1:
feature_dic.clear()
if FLAG_DEBUG: print("Cache path cleared")
else:
feature_dic = {}
expert = opts.get("expert",None)
if expert is not None:
expert = MemoizedDict(expert) # convert path
################################## LOADING IMAGES ###################################
if images is None or len(images)==0: # if images is empty use demonstration
test = MANAGER.TESTPATH
if FLAG_DEBUG: print("Looking in DEMO path {}".format(test))
fns = glob(test + "*")
elif isinstance(images,basestring):
# if string assume it is a path
if FLAG_DEBUG:print("Looking as {}".format(images))
fns = glob(images)
elif not isinstance(images,basestring) and len(images) == 1 and "*" in images[0]:
images = images[0] # get string
if FLAG_DEBUG:print("Looking as {}".format(images))
fns = glob(images)
else: # iterator containing data
fns = images # list file names
# check images
if not len(fns)>1:
raise Exception("list of images must be greater than 1, got {}".format(len(fns)))
# to select base image ahead of any process
baseImage = opts.get("baseImage",None)
if isinstance(baseImage,basestring):
base_old = baseImage
try: # tries user input
if baseImage not in fns:
base, path, name, ext = getData(baseImage)
if not path: # if name is incomplete look for it
base, path, _,_ = getData(fns[0])
baseImage = lookinglob(baseImage, "".join((base, path)))
# selected image must be in fns
if baseImage is None:
raise IndexError
except IndexError: # tries to find image based in user input
# generate informative error for the user
raise Exception("{} or {} is not in image list"
"\n A pattern is {}".format(
base_old,baseImage,fns[0]))
if FLAG_DEBUG: print("No. images {}...".format(len(fns)))
# make loader
loader = opts.get("loader",None) # BGR loader
if loader is None: loader = loadFunc(1)
######################## Local features: Key-points and descriptors ####################
with TimeCode("Computing features...\n",
endmsg="Computed feature time was {time}\n", enableMsg=FLAG_DEBUG):
feature_list = [] # list of key points and descriptors
index = 0
tries = 0 # prevents infinite loop
while index < len(fns):
try:
path = fns[index]
try:
if cachePath is None or clearCache==2 and path in feature_dic:
raise KeyError # clears entry from cache
kps, desc, shape = feature_dic[path] # thread safe
# check memoized is the same
if pshape is None:
sh = loader(path).shape # and checks that image exists
else:
sh = pshape
# solves possible difference in object id instead of if sh != shape
for ii,jj in zip(sh,shape):
if ii!=jj: raise KeyError
except (KeyError, ValueError) as e: # not memorized
if FLAG_DEBUG: print("Processing features for {}...".format(path))
img = loader(path) #cv2.imread(path)
img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
if pshape is None: # get features directly from original
kps, desc = feature.detectAndCompute(img) # get features
else: # optimize by getting features from scaled image and the rescaling
oshape = img.shape # cache original shape
img = cv2.resize(img, pshape) # pseudo image
kps, desc = feature.detectAndCompute(img) # get features
# re-scale keypoints to original image
if not usepshape:
# this necessarily does not produce the same result when not used
"""
# METHOD 1: using Transformation Matrix
H = getSOpointRelation(process_shape, oshape, True)
for kp in kps:
kp["pt"]=tuple(cv2.perspectiveTransform(
np.array([[kp["pt"]]]), H).reshape(-1, 2)[0])
"""
# METHOD 2:
rx,ry = getSOpointRelation(pshape, oshape)
for kp in kps:
x,y = kp["pt"]
kp["pt"] = x*rx,y*ry
shape = img.shape
feature_dic[path] = kps, desc, shape # to memoize
# add paths to key-points
for kp in kps:
kp["path"] = path
# number of key-points, index, path, key-points, descriptors
feature_list.append((len(kps),index,path,kps,desc))
if FLAG_DEBUG: print("\rFeatures {}/{}...".format(index+1,len(fns)), end=' ')
index += 1
tries = 0
except Exception as e:
tries += 1
warnings.warn("caught error 139")
if tries > 2:
raise e
############################## Pre-selection from a set ###############################
# initialization and base image selection
if baseImage is None:
_,_,path,kps_base,desc_base = feature_list[0] # select first for most probable
elif isinstance(baseImage,basestring):
kps_base,desc_base = feature_dic[baseImage]
path = baseImage
elif baseImage is True: # sort images
feature_list.sort(reverse=True) # descendant: from bigger to least
_,_,path,kps_base,desc_base = feature_list[0] # select first for most probable
else:
raise Exception("baseImage must be None, True or String")
if FLAG_DEBUG: print("baseImage is", path)
used = [path] # select first image path
restored = loader(path) # load first image for merged image
if usepshape:
restored = cv2.resize(restored,pshape)
failed = [] # registry for failed images
############################# Order set initialization ################################
if orderValue: # obtain comparison with structure (value, path)
if orderValue == 1: # entropy
comparison = list(zip(*entropy(fns,loadfunc=loadFunc(1,pshape),invert=False)[:2]))
if FLAG_DEBUG: print("Configured to sort by entropy...")
elif orderValue == 2: # histogram comparison
comparison = hist_comp(fns,loadfunc=loadFunc(1,pshape),method=selectMethod)
if FLAG_DEBUG: print("Configured to sort by {}...".format(selectMethod))
elif orderValue == 3:
comparison = selectMethod(fns)
if FLAG_DEBUG: print("Configured to sort by Custom Function...")
else:
raise Exception("DEBUG: orderValue {} does "
+"not correspond to {}".format(orderValue,selectMethod))
elif FLAG_DEBUG: print("Configured to sort by best matches")
with TimeCode("Restoring ...\n",
endmsg= "Restoring overall time was {time}\n", enableMsg= FLAG_DEBUG):
while True:
with TimeCode("Matching ...\n",
endmsg= "Matching overall time was {time}\n", enableMsg= FLAG_DEBUG):
####################### remaining keypoints to match ##########################
kps_remain,desc_remain = [],[] # initialize key-point and descriptor base list
for _,_,path,kps,desc in feature_list:
if path not in used: # append only those which are not in the base image
kps_remain.extend(kps)
desc_remain.extend(desc)
if not kps_remain: # if there is not image remaining to stitch break
if FLAG_DEBUG: print("All images used")
break
desc_remain = np.array(desc_remain) # convert descriptors to array
################################## Matching ###################################
# select only those with good distance (hamming, L1, L2)
raw_matches = feature.matcher.knnMatch(desc_remain,
trainDescriptors = desc_base, k = 2) #2
# If path=2, it will draw two match-lines for each key-point.
classified = {}
for m in raw_matches:
# filter by Hamming, L1 or L2 distance
if m[0].distance < m[1].distance * distanceThresh:
m = m[0]
kp1 = kps_remain[m.queryIdx] # keypoint in query image
kp2 = kps_base[m.trainIdx] # keypoint in train image
key = kp1["path"] # ensured that key is not in used
if key in classified:
classified[key].append((kp1,kp2))
else:
classified[key] = [(kp1,kp2)]
############################ Order set ########################################
if orderValue: # use only those in classified of histogram or entropy comparison
ordered = [(val,path) for val,path in comparison if path in classified]
else: # order with best matches
ordered = sorted([(len(kps),path) for path,kps in list(classified.items())],reverse=True)
for rank, path in ordered: # feed key-points in order according to order set
########################### Calculate Homography ##########################
mkp1,mkp2 = list(zip(*classified[path])) # probably good matches
if len(mkp1)>minKps and len(mkp2)>minKps:
p1 = np.float32([kp["pt"] for kp in mkp1])
p2 = np.float32([kp["pt"] for kp in mkp2])
if FLAG_DEBUG > 1: print('Calculating Homography for {}...'.format(path))
H, status = cv2.findHomography(p1, p2, cv2.RANSAC, ransacReprojThreshold)
else:
if FLAG_DEBUG > 1: print('Not enough key-points for {}...'.format(path))
H = None
if H is not None: #first test
fore = loader(path) # load fore image
if usepshape:
fore = cv2.resize(fore,pshape)
"""
else:
# METHOD 3 for rescaling points. FIXME
#shapes = fore.shape,process_shape,fore.shape,process_shape
#H = sh2oh(H, *shapes) #### sTM to oTM
H = getSOpointRelation(process_shape,fore.shape, True)*H
for kp in feature_dic[path][0]:
kp["pt"]=tuple(cv2.perspectiveTransform(
np.array([[kp["pt"]]]), H).reshape(-1, 2)[0])"""
if histMatch: # apply histogram matching
fore = hist_match(fore, restored)
h,w = fore.shape[:2] # image shape
# get corners of fore projection over back
projection = getTransformedCorners((h,w),H)
c = ImCoors(projection) # class to calculate statistical data
lines, inlines = len(status), np.sum(status)
# ratio to determine how good fore is in back
inlineratio = inlineRatio(inlines,lines)
text = "inlines/lines: {}/{}={} and rectangularity {}".format(
inlines, lines, inlineratio, c.rotatedRectangularity)
if FLAG_DEBUG>1: print(text)
if FLAG_DEBUG > 2: # show matches
MatchExplorer("Match " + text, fore,
restored, classified[path], status, H)
######################### probability test ############################
if inlineratio>inlineThresh \
and c.rotatedRectangularity>rectangularityThresh: # second test
if FLAG_DEBUG>1: print("Test succeeded...")
while path in failed: # clean path in fail registry
try: # race-conditions safe
failed.remove(path)
except ValueError:
pass
###################### merging and stitching #######################
if FLAG_DEBUG > 1: print("Merging...")
restored, H_back, H_fore= mergefunc(restored,fore,H)
if H_fore is None: # H is not modified use itself
H_fore = H
if FLAG_DEBUG > 1: # show merging result
Plotim("Last added with " + text, restored).show()
####################### update base features #######################
# make projection to test key-points inside it
if FLAG_DEBUG > 1: print("Updating key-points...")
projection = getTransformedCorners((h,w),H_fore)
newkps, newdesc = [], []
for _,_,p,kps,desc in feature_list:
# append all points in the base image and update their position
if p in used: # transform points in back
for kp,dsc in zip(kps,desc): # kps,desc
pt = kp["pt"] # get point
if H_back is not None: # update point
pt = tuple(transformPoint(pt,H_back))
kp["pt"] = pt
# include only those points outside foreground
if cv2.pointPolygonTest(projection, pt, False) == -1:
newkps.append(kp)
newdesc.append(dsc)
elif p == path: # transform points in fore
# include only those points inside foreground
for kp,dsc in zip(kps,desc): # kps,desc
kp["pt"] = tuple(transformPoint(kp["pt"],H_fore))
newkps.append(kp)
newdesc.append(dsc)
# update kps_base and desc_base
kps_base = newkps
desc_base = np.array(newdesc)
if FLAG_DEBUG > 2: # show keypints in merging
Plotim("merged Key-points", # draw key-points in image
cv2.drawKeypoints(
im2shapeFormat(restored,restored.shape[:2]+(3,)),
[dict2keyPoint(index) for index in kps_base],
flags=4, color=(0,0,255))).show()
if FLAG_DEBUG: print("This image has been merged: {}...".format(path))
used.append(path) # update used
if not centric:
break
else:
failed.append(path)
else:
failed.append(path)
# if all classified have failed then end
if set(classified.keys()) == set(failed):
if FLAG_DEBUG:
print("Ended, these images do not fit: ")
for index in list(classified.keys()):
print(index)
break
if postfunc is not None:
if FLAG_DEBUG: print("Applying post function...")
restored = postfunc(restored)
####################################### Save image ####################################
save = opts.get("save",False)
if save:
base, path, name, ext = getData(used[0])
if isinstance(save,basestring):
fn = save.format(path="".join((base, path)), ext=ext,)
else:
fn = "".join((base,path,"restored_"+ name,ext))
mkPath(getPath(fn))
r = cv2.imwrite(fn,restored)
if FLAG_DEBUG and r:
print("Saved: {}".format(fn))
else:
print("{} could not be saved".format(fn))
return restored # return merged image
def _setup_multithreaded(self):
self.thread_num = cv2.getNumberOfCPUs()
self.thread_pool = ThreadPool(self.thread_num)
def extract_ball_from_capture_threaded(cap, max_frames_count=-1, skip_count=0,
carpet_mask=None, get_mask=None,
carpet_lowerb=None,
carpet_upperb=None,
ball_lowerb=None,
ball_upperb=None):
"""
Read frames from capture until we detect motion of the ball
Return tuple of (original frames, ball mask frames)
:param carpet_lowerb:
:rtype : (list(np.ndarray), list(np.ndarray))
"""
frames = []
mask_frames = []
# mog = cv2.BackgroundSubtractorMOG(history=5, nmixtures=4,backgroundRatio=0.7)
mog = cv2.BackgroundSubtractorMOG()
motion_started = False
for _ in xrange(skip_count):
if cap.isOpened():
cap.read()
prev_mask = None
threadn = cv2.getNumberOfCPUs()
pool = ThreadPool(processes=threadn)
pending = deque()
def process_frame(frame, prev_mask, carpet_mask):
if carpet_mask is None:
carpet_mask = img_util.green_carpet_mask(frame,
carpet_lowerb,
carpet_upperb)
mask = get_mask(frame, lowerb=ball_lowerb, upperb=ball_upperb,
carpet_mask=carpet_mask)
if prev_mask is None:
move_mask = np.zeros_like(mask)
else:
move_mask = cv2.bitwise_not(prev_mask, mask=mask)
prev_mask = mask
is_ball, move_mask = img_util.detect_ball(move_mask)
return is_ball, move_mask, prev_mask, frame
while cap.isOpened():
while len(pending) > 0 and pending[0].ready():
is_ball, move_mask, prev_mask, frame = pending.popleft().get()
if is_ball:
if not motion_started:
logger.debug('ball appeared')
motion_started = True
frames.append(frame)
mask_frames.append(move_mask)
else:
if motion_started:
logger.debug(
'ball disappeared. Frames count: {}'.format(
len(frames)))
break
else:
continue
if 0 < max_frames_count < len(frames):
logger.debug('max frames count reached')
break
if True:
ret, frame = cap.read()
task = pool.apply_async(process_frame,
(frame, prev_mask, carpet_mask))
pending.append(task)
else:
logger.debug('Cannot read more frames')
break
return frames, mask_frames
def __init__(self, context, poller):
super(NewClient, self).__init__(context, poller)
self._threadn = cv2.getNumberOfCPUs()
self._pool = ThreadPool(processes = self._threadn)
self._pending = deque()
def multi_thread(f, jobs):
pool = ThreadPool(processes=cv2.getNumberOfCPUs())
ires = pool.imap_unordered(f, jobs)
return ires
cv2.CPU_NEON: "NEON",
cv2.CPU_POPCNT: "POPCNT",
cv2.CPU_SSE: "SSE",
cv2.CPU_SSE2: "SSE2",
cv2.CPU_SSE3: "SSE3",
cv2.CPU_SSE4_1: "SSE4.1",
cv2.CPU_SSE4_2: "SSE4.2",
cv2.CPU_SSSE3: "SSSE3"};
for feat in featDict:
res = cv2.checkHardwareSupport(feat);
print("%s = %d" % (featDict[feat], res));
#cv2.setUseOptimized(onoff)!!!!
# "Returns the number of logical CPUs available for the process."
common.DebugPrint("cv2.getNumberOfCPUs() (#logical CPUs) is %s" % str(cv2.getNumberOfCPUs()));
common.DebugPrint("cv2.getTickFrequency() is %s" % str(cv2.getTickFrequency()));
"""
Available only in C++:
# "getNumThreads - Returns the number of threads used by OpenCV for parallel regions."
common.DebugPrint("cv2.getNumThreads() (#logical CPUs) is %s" % str(cv2.getNumThreads()));
"""
videoPathFileNameQ = sys.argv[1]; # input/current video
videoPathFileNameR = sys.argv[2]; # reference video
#!!!!TODO: use getopt() to run Evangelidis' or "Alex's" algorithm, etc
def init__threads(self):
if self.threadn == 0:
self.threadn = cv2.getNumberOfCPUs()
self.pool = ThreadPool(processes=self.threadn)
self.pending = deque()
def main():
opts, args = getopt.getopt(sys.argv[1:], "", ["video="])
opts = dict(opts)
video = int(opts.get("--video", "0"))
vsrc = None
index = 0
lastFn = None
cmode = "rgb"
mainImg = None
value = 0
algostate = {}
print(
"keys:\n"
" ESC: exit\n\n"
" c: rgb\n"
" r: red\n"
" g: green\n"
" b: blue,\n"
" h: hue\n"
" s: sat\n"
" v: val\n"
" 0: adaptive threshold\n"
" 1: threshold\n"
" 2: huerange*valuerange\n"
" 3: canny edges\n"
" 4: simple blobs\n"
"\n"
" <home>: reset algorithm input\n"
" <up arrow>: increase algorithm input\n"
" <down arrow>: decrease algorithm input\n"
"\n"
" <right arrow>: increase img seq frame\n"
" <left arrow>: decrease img seq frame\n"
)
if video:
vsrc = cv2.VideoCapture(0)
if not vsrc or not vsrc.isOpened():
print("Problem opening video source")
vsrc = None
video = 0
if 0:
vsrc.set(cv2.cv.CV_CAP_PROP_FRAME_WIDTH, w)
vsrc.set(cv2.cv.CV_CAP_PROP_FRAME_HEIGHT, h)
else:
vsrc = None
def processFrame(frame, t0):
keypoints = None
if cmode != "rgb":
if cmode in ["h", "s", "v"]:
# mode = (cv2.COLOR_BGR2HLS, cv2.COLOR_HLS2BGR)
mode = (cv2.COLOR_BGR2HSV, cv2.COLOR_HSV2BGR)
hsv = cv2.cvtColor(frame, mode[0])
if cmode == "h":
if mode[0] == cv2.COLOR_BGR2HSV:
hsv[:, :, 1] = 255 # s = 1
hsv[:, :, 2] = 128 # v = .5
else:
hsv[:, :, 1] = 128 # l = .5
hsv[:, :, 2] = 255 # s = 1
frame = cv2.cvtColor(hsv, mode[1])
elif cmode == "s":
# extract the s as grayscale
if mode[0] == cv2.COLOR_BGR2HSV:
h, frame, v = cv2.split(hsv)
else:
h, l, frame = cv2.split(hsv)
elif cmode == "v":
if mode[0] == cv2.COLOR_BGR2HSV:
h, s, frame = cv2.split(hsv)
else:
h, frame, s = cv2.split(hsv)
elif cmode in ["r", "g", "b"]:
if cmode == "r":
b, g, frame = cv2.split(frame)
elif cmode == "g":
b, frame, r = cv2.split(frame)
elif cmode == "b":
frame, g, r = cv2.split(frame)
elif cmode == "0":
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
frame = cv2.adaptiveThreshold(
gray, 200, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 5, value # value to draw
)
elif cmode == "1":
# simple thresholding
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
(ret, frame) = cv2.threshold(gray, 75 + value, 200, cv2.THRESH_BINARY) # value to draw
elif cmode == "2":
# threshold value (want bright areas)
# multiply result by hue in range
mode = cv2.COLOR_BGR2HSV # cv2.COLOR_HSV2BGR)
hsv = cv2.cvtColor(frame, mode)
h, s, v = cv2.split(hsv)
vimg = cv2.inRange(v, 65, 255)
himg = cv2.inRange(h, 50, 90)
frame = cv2.multiply(vimg, himg, scale=1.0 / 255)
elif cmode == "3":
t1 = 2000 + value * 20
t2 = t1 + 2000
frame = cv2.Canny(frame, t1, t2, apertureSize=5)
elif cmode == "4":
if not "blobdetector" in algostate:
bp = cv2.SimpleBlobDetector_Params()
bp.filterByColor = True
bp.blobColor = 0
# Change thresholds
bp.minThreshold = 50
bp.maxThreshold = 150
bp.thresholdStep = 5
# Filter by Area.
bp.filterByArea = True
bp.minArea = 500
bp.maxArea = 10000
# Filter by Circularity
bp.filterByCircularity = False
bp.minCircularity = 0.1
# Filter by Convexity
bp.filterByConvexity = False
bp.minConvexity = 0.87
# Filter by Inertia
bp.filterByInertia = False
bp.minInertiaRatio = 0.01
algostate["blobdetector"] = cv2.SimpleBlobDetector(bp)
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
keypoints = algostate["blobdetector"].detect(frame)
else:
print("unknown cmode: " + cmode)
return frame, t0, keypoints
def showImg(frame, keypoints):
if keypoints:
frame = cv2.drawKeypoints(
frame, keypoints, np.array([]), (0, 0, 255), cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS
)
cv2.imshow("img", frame)
# cv2.namedWindow("img")
# cv2.createTrackbar("value", "img", 0, 100, trackbarChange)
threadn = cv2.getNumberOfCPUs()
pool = ThreadPool(processes=threadn)
pending = deque()
threadedMode = True
latency = common.StatValue()
frameT = common.StatValue()
lastFrameTime = common.clock()
while True:
while len(pending) > 0 and pending[0].ready():
frame, t0, keypoints = pending.popleft().get()
latency.update(common.clock() - t0)
putStr(frame, (20, 20), "latency : %.1f ms" % (latency.value * 1000))
putStr(frame, (20, 40), "frame interval: %.1f ms" % (frameT.value * 1000))
showImg(frame, keypoints)
if video:
if len(pending) < threadn:
ret, mainImg = vsrc.read()
t = common.clock()
frameT.update(t - lastFrameTime)
lastFrameTime = t
task = pool.apply_async(processFrame, (mainImg.copy(), t))
pending.append(task)
done, update, cmode, index, value = checkKey(1, cmode, index, value)
if done:
break
else:
if index < 0:
# index < 0 signals user's intent to go backward in framelist
goback = True
index = abs(index)
else:
goback = False
fn = fnpat % index
base = os.path.basename(fn)
if not os.path.isfile(fn) and index >= 0:
sys.stdout.write(base + " not found\r")
sys.stdout.flush()
if goback:
index = -(index - 1)
else:
index = index + 1
continue
if fn != lastFn:
mainImg = cv2.imread(fn, cv2.IMREAD_ANYCOLOR)
(img, t0, keypts) = processFrame(mainImg.copy(), common.clock())
frameT.update(common.clock() - t0)
if cmode == "rgb":
str = "%s (%.2f ms)" % (base, frameT.value * 1000)
else:
str = "%s[%s] (%.2f ms)" % (base, cmode, frameT.value * 1000)
putStr(img, (20, 20), str)
showImg(img, keypts)
lastFn = fn
done, update, cmode, index, value = checkKey(10, cmode, index, value)
if done:
break
elif update:
lastFn = None
if __name__ == '__main__': source = 'V:\\Video AG\\archiv\\14ws-infin-141014\\14ws-infin-141014-dozent.mp4' # 0:12:40 source = 0 if sys.argv[1:]: source = sys.argv[1] framescale = 0.25 #0.5 prevframe = None curframe = None tracks = [] tracklen = 10 flowmethod = 0 numthreads = cv2.getNumberOfCPUs() scalerpool = ThreadPool(processes=1) qscaling = deque() flowpool = ThreadPool(processes=1) # must be serial! qflow = deque() vid = cv2.VideoCapture(source) #vid.set(cv2.CAP_PROP_FPS, 15) #if isinstance(source, str): # vid.set(cv2.CAP_PROP_POS_MSEC, 1000 * (12*60 + 40)) vid.set(cv2.CAP_PROP_POS_MSEC, 1000 * (10*60 + 0)) if source == 0:
def run(self):
# init big image fro stitching
ret, frame = self.cam.read()
frame=cv2.resize(frame,(320,240))
h,w,d=frame.shape
big_image = np.zeros((h*12,w*3,3), np.uint8)
starty=h*11
startx=w
total_transl_x=0
total_transl_y=0
frame_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
self.prev_gray=frame_gray
self.prev_frame=frame
detector, matcher = init_feature('sift-flann')
pool=ThreadPool(processes = cv2.getNumberOfCPUs())
while True:
for i in range(skip_frames):
ret, frame = self.cam.read()
ret, frame = self.cam.read()
frame=cv2.resize(frame,(320,240))
frame_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
vis = frame.copy()
img0, img1 = self.prev_gray, frame_gray
kp1, desc1 = affine_detect(detector, img0[10:h-50,10:w-10], pool=pool)
kp2, desc2 = affine_detect(detector, img1[10:h-50,10:w-10], pool=pool)
print 'img1 - %d features, img2 - %d features' % (len(kp1), len(kp2))
with Timer('matching'):
raw_matches = matcher.knnMatch(desc1, trainDescriptors = desc2, k = 2) #2
p1, p2, kp_pairs = filter_matches(kp1, kp2, raw_matches)
if len(p1) >= 4:
H, status = cv2.findHomography(p1, p2, cv2.RANSAC, 5.0)
print '%d / %d inliers/matched' % (np.sum(status), len(status))
# do not draw outliers (there will be a lot of them)
kp_pairs = [kpp for kpp, flag in zip(kp_pairs, status) if flag]
warp = cv2.warpPerspective(img0[10:h-50,10:w-10], H, (w, h*2))
cv2.imshow("warped",warp)
else:
H, status = None, None
print '%d matches found, not enough for homography estimation' % len(p1)
vis = explore_match('affine find_obj', img0, img1, kp_pairs, None, H)
# stitching-----------------------
translation = np.zeros((3,1)) #3,1
if len(p1)>4 and len(p2)>4:
# temp1=[]
# # temp2=[]
# for i in range(len(kp1)):
# print kp1[i].pt+ (0,)
# temp1.append(kp1[i].pt+ (0,))
# # for i in range(len(kp2)):
# # temp2.append(kp2[i].pt)
# points1.astype(np.uint8)
# points1 = np.array(temp1)
# print points1
# # points2 = np.array(temp2)
# Hr=cv2.estimateRigidTransform(points1, points1,False)
translation[:,0] = H[:,2] #Hr[:,2]
# rotation = np.zeros((3,3))
# rotation[:,0] = H[:,0]
# rotation[:,1] = H[:,1]
# rotation[:,2] = np.cross(H[0:3,0],H[0:3,1])
# print "x translation:",translation[0]
# print "y translation:",translation[1]
draw_str(vis, (20, 40), 'x-axis translation: %.1f' % translation[0])
draw_str(vis, (20, 60), 'y-axis translation: %.1f' % translation[1])
if translation[0]<60 and translation[1]<60: #check for bad H
total_transl_x+=int(translation[0])
total_transl_y+=int(translation[1])
draw_str(vis, (20, 80), 'tot x-axis translation: %.1f' % total_transl_x)
draw_str(vis, (20, 100), 'tot y-axis translation: %.1f' % total_transl_y)
#h,w,d=frame.shape
frame_over=self.prev_frame[10:h-50,10:w-10].copy()
overlay = cv2.warpPerspective(frame_over, H, (w, h))
frame_h,frame_w,d=frame_over.shape
cv2.imshow('overlay',overlay)
#vis = cv2.addWeighted(vis, 0.5, overlay, 0.5, 0.0)
big_image[starty-int(total_transl_y):starty-int(total_transl_y)+frame_h,startx-int(total_transl_x):startx-int(total_transl_x)+frame_w]=overlay[0:frame_h,0:frame_w].copy()
#small_image=big_image.copy()
big_h,big_w,d=big_image.shape
small_image=cv2.resize(big_image,(big_w/4,big_h/4))
cv2.imshow('stitching', small_image)
#cv2.imwrite("result.jpg",big_image);
self.frame_idx += 1
self.prev_gray = frame_gray
self.prev_frame=frame
ch = 0xFF & cv2.waitKey(5)
if ch == 27:
break
from builtins import zip
from builtins import range
from builtins import object
from past.utils import old_div
import itertools as it
from multiprocessing.pool import ThreadPool as Pool
# three-party
import cv2
import numpy as np
# custom
from RRtool.RRtoolbox.lib import plotter, cache, config, image
from RRtool.RRtoolbox import filter,basic as ar
from RRtool.RRtoolbox.lib.arrayops import convert
# ---------------------------- GLOBALS ---------------------------- #
cpc = cv2.getNumberOfCPUs()
print("configured to use {} cpus".format(cpc))
pool = Pool(processes = cpc) # DO NOT USE IT when module is imported and this runs with it. It creates a deadlock"
feature_name = 'sift-flann'
paths = config.ConfigFile()
# ---------------------------- DECORATORS ---------------------------- #
def getalfa(foregray,backgray,window = None):
""" get alfa transparency for merging to retinal images
:param foregray: image on top
:param backgray: image at bottom
:param window: window used to customizing alfa, values go from 0 for transparency to any value
where the maximum is visible i.e a window with all the same values does nothing.
a binary image can be used, where 0 is transparent and 1 is visible.
If not window is given alfa is left as intended.
:return: float window modified by alfa
def run_jobs(self, f, jobs):
pool = ThreadPool(processes=cv.getNumberOfCPUs())
ires = pool.imap_unordered(f, jobs)
return ires
from sklearn.cluster import KMeans
from sklearn.cluster import MiniBatchKMeans
from sklearn.linear_model.logistic import LogisticRegression
from utils import *
import warnings
warnings.filterwarnings("ignore")
TEST = False
SAVE = False
LOWE = False
train_images, train_labels, test_images, test_labels = get_train_test(TEST)
pool = Pool(cv2.getNumberOfCPUs()-2)
if LOWE:
print " [!] Lowe's SIFT"
train_sift_with_null = pool.map(get_sift_lowe, train_images)
test_sift_with_null = pool.map(get_sift_lowe, test_images)
else:
print " [!] OpenCV2's SIFT"
train_sift_with_null = pool.map(get_sift, train_images)
test_sift_with_null = pool.map(get_sift, test_images)
pool.terminate()
train_sift = removing_null(train_sift_with_null, train_labels)
reduced_train_sift = np.concatenate(train_sift, axis = 0)
descriptor = "SIFT"
descriptor = "spSIFT"
if TEST:
prefix = "%s_%s_" % (descriptor, "test")
else:
prefix = "%s_%s_" % (descriptor, "full")
train_images, train_labels, test_images, test_labels = get_train_test(TEST)
if descriptor == "SIFT":
if os.path.isfile(prefix % "kmeans"):
kmeans = load_pickle(prefix + "kmeans.pkl")
else:
pool = Pool(getNumberOfCPUs() - 2)
if LOWE:
print " [!] Lowe's SIFT"
train_sift_with_null = pool.map(get_sift_lowe, train_images)
test_sift_with_null = pool.map(get_sift_lowe, test_images)
else:
print " [!] OpenCV2's SIFT"
train_sift_with_null = pool.map(get_sift, train_images)
test_sift_with_null = pool.map(get_sift, test_images)
pool.close()
pool.join()
pool.terminate()
del pool
blank += 1
else:
if mx and mx.number.value() > 9 and blank >= _wait(5):
puts(mx)
wait = _wait(5)
blank += 1
if __name__ == '__main__':
from optparse import OptionParser
parser = OptionParser()
parser.add_option("-s", "--start", type="int", default=0, help="set the start time offset")
parser.add_option("-l", "--length", type="int", default=99999999, help="set the length")
parser.add_option("-X", "--offset_x", type="int", help="horizontal offset of the crop")
parser.add_option("-Y", "--offset_y", type="int", help="vertical offset of the crop")
parser.add_option("--crop_size", help="width and height of the crop")
parser.add_option("-p", "--processes", type="int", default=cv2.getNumberOfCPUs(), help="num of processes")
parser.add_option("--gui", action="store_true", help="use the GUI")
parser.add_option("--profile", action="store_true", help="enable profiling")
(op, args) = parser.parse_args()
crop_size = [None, None]
if op.crop_size:
crop_size = [int(e) for e in op.crop_size.split('x')]
if op.profile:
import cProfile
cProfile.run('Scout(args[0], op.start, op.length, op.offset_x, op.offset_y, *crop_size, op.processes, op.gui)')
else:
Scout(args[0], op.start, op.length, op.offset_x, op.offset_y, *crop_size, op.processes, op.gui)
except:
fn = 0
'''camera = PiCamera()
camera.resolution = (640, 480)
camera.framerate = 32
rawCap= PiRGBArray(camera, size = (640, 480))'''
cap = PiVideoStream().start()
def process_frame(frame, t0):
# some intensive computation...
frame = cv2.medianBlur(frame, 19)
# frame = cv2.medianBlur(frame, 19)
return frame, t0
threadn = cv2.getNumberOfCPUs()
pool = ThreadPool(processes = threadn)
pending = deque()
threaded_mode = True
latency = StatValue()
frame_interval = StatValue()
last_frame_time = clock()
while True:
while len(pending) > 0 and pending[0].ready():
''' '''
res, t0 = pending.popleft().get()
latency.update(clock() - t0)
draw_str(res, (20, 20), "Latency: %.1f ms" % (latency.value*1000))
draw_str(res, (100, 20), "Frame interval: %.1f ms" % (frame_interval.value*1000))
def init(self):
# args is argparser.Namespace: an object whose members are
# accessed as: self.args.cannedimages
self.robotCnx = robotCnx.RobotCnx(self.args.fakerobot)
self.vsrc = None
if self.args.cannedimages:
self.fnpat = '../../pictures/RealFullField/%d.jpg'
else:
self.fnpat = None
self.index = 0
self.lastFn = None
self.mainImg = None
self.algostate = {} # stores persistant objects depending on cmode
self.threadn = cv2.getNumberOfCPUs()
self.pool = ThreadPool(processes = self.threadn)
self.pending = deque()
self.threadedMode = True
self.update = False
self.latency = common.StatValue()
self.frameT = common.StatValue()
self.lastFrameTime = common.clock()
self.lastStashTime = 0
self.stashFilename = "/var/tmp/imgServer.home/currentImage.jpg"
self.stashParams = [int(cv2.IMWRITE_JPEG_QUALITY), 50]
self.zeros = (0,0,0,0,0,0)
self.LUT = np.array(range(0,256)).astype('uint8')
self.indent = ' ' * 50
self.lastStatus = ""
# cmodelist maps a number [0, len] to an cmode/algorithm
self.cmodelist = [
'rgb',
'adaptiveThreshold',
'threshold',
'huerange*valrange',
'canny',
'simpleblob',
'houghlines',
'contours',
'ORB',
'dance1',
'gamma',
'b',
'g',
'h',
'r',
's',
'v',
]
# cmodeValueCache stores the edited values associated with each cmode.
# It starts empty, then builds up state as cmode changes.
self.cmodeValueCache = {
'adaptiveThreshold': [-4, 0, 0, 0, 0, 0],
'threshold': [75, 0, 0, 0, 0, 0],
'huerange*valrange': [55, 100, 255, 255, 0, 0], # works for g LED
'canny': [10, 200, 0, 0, 0, 0],
'simpleblob': [75, 150, 20, 40**2, 0, 0],
# minThresh
# maxThresh
# thresStep
# minSize (maxsize is 'large')
'houghlines': [2, 5, 10, 30, 2, 0],
# rho is distance resolution in pixels
# theta is angle in degress (larger -> more lines)
# threshold is measured in 'votes', higher -> fewer
# minLineLength
# maxLineGap
'contours': [1, 0, 0, -1, 1, 0],
# mode: [0-3]
# method: [CHAIN_APPROX_NONE,_SIMPLE,_TC89_L1, KOS]
# offset shift for contour
# unused
# which contour to draw, -1 is all
# which level of the contours to draw
'ORB': [40, 10, 8, 31, 0, 0],
# nfeatures
# scaleFactor [1 -> 2 (0->255)]
# nlevels
# patchSizea == edgeThreshold
'dance1': [20,20,0,0,0,0],
#X(degrees)
#Y(degrees)
#Not rotating correctly (not interpreting correctly)
'gamma': [34,0,0,0,0,0],
#alpha
#beta
'r': self.zeros,
'g': self.zeros,
'b': self.zeros,
'h': [0, 61, 0, 0, 0, 0],
's': self.zeros,
'v': self.zeros,
'rgb': self.zeros
}
#keyToCmode maps a key to a cmod
self.keyToCmode = {
49: self.cmodelist[1], # adaptiveThreshold on '1' key
50: self.cmodelist[2],
51: self.cmodelist[3],
52: self.cmodelist[4],
53: self.cmodelist[5],
54: self.cmodelist[6], # houghlines
55: self.cmodelist[7], # contours
56: self.cmodelist[8], # ORB
57: self.cmodelist[9], # dance1
48: self.cmodelist[10],# 0 key: gamma
98: 'b',
99: 'rgb',
103: 'g',
104: 'h',
114: 'r',
115: 's',
118: 'v',
}
self.keyToFrameChange = {
81: 'b',
83: 'n',
}
self.algoValueChange = {
127: ("reset", self.zeros), # keypad clear
190: ("v1Down", (-1, 0, 0, 0, 0, 0)), # f1
191: ("v1Up ", (1, 0, 0, 0, 0, 0)), # f2
84: ("v1Down", (-1, 0, 0, 0, 0, 0)), # downarrow
82: ("v1Up ", (1, 0, 0, 0, 0, 0)), # uparrow
191: ("v1Up ", (1, 0, 0, 0, 0, 0)), # downarrow
192: ("v2Down", (0, -1, 0, 0, 0, 0)), # f3
193: ("v2Up ", (0, 1, 0, 0, 0, 0)), # f4
194: ("v3Down", (0, 0, -1, 0, 0, 0)), # f5
195: ("v3Up ", (0, 0, 1, 0, 0, 0)), # f6
196: ("v4Down", (0, 0, 0, -1, 0, 0)), # f7
197: ("v4Up ", (0, 0, 0, 1, 0, 0)), # f8
198: ("v5Down", (0, 0, 0, 0, -1, 0)), # f9
199: ("v5Up ", (0, 0, 0, 0, 1, 0)), # f10
200: ("v6Down", (0, 0, 0, 0, 0, -1)), # f11
201: ("v6Up ", (0, 0, 0, 0, 0, 1)), # f12
}
self.cmode = self.getCmode(self.args.algorithm)
self.values = self.getCmodeValues(self.cmode)