def get_features(im):
im = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
detector = cv2.SURF(400, 5, 5)
t0 = time.time()
kps, descs = detector.detectAndCompute(im, None)
logger.debug('feature detection: %f sec' % (time.time() - t0))
return (kps, descs)
def __init__(self, visualwords,
index_filepath=None,
algorithm='kdtree',
loglevel='WARNING'):
"""Create or load index of visual words.
:param visualwords: 2D array of base vectors
:type visualwords: M x len(feature) `numpy.ndarray`,
where each element is `numpy.float32`
:param index_filepath: If not `None`, index is not created
but load from specified file.
The file must be created by :func:`BoFMaker.save`
:param algorithm: passed to `pyflann.FLANN().build_index`
"""
self._n_visualwords = visualwords.shape[0]
self._flann = pyflann.FLANN()
if index_filepath is not None:
# load index
logger.debug('Loading visual words index from %s ...' %
(index_filepath))
t0 = time.time()
with open(BoFMaker.meta_filepath(index_filepath)) as f:
self._index_param = json.load(f)
self._flann.load_index(index_filepath, visualwords)
logger.debug('%f sec to load index' % (time.time() - t0))
else:
# create index
logger.debug('Creating index of visual words...')
t0 = time.time()
self._index_param = self._flann.build_index(
visualwords, algorithm=algorithm)
logger.debug('%f sec to create index' % (time.time() - t0))
def __init__(self,
visualwords,
index_filepath=None,
algorithm='kdtree',
loglevel='WARNING'):
"""Create or load index of visual words.
:param visualwords: 2D array of base vectors
:type visualwords: M x len(feature) `numpy.ndarray`,
where each element is `numpy.float32`
:param index_filepath: If not `None`, index is not created
but load from specified file.
The file must be created by :func:`BoFMaker.save`
:param algorithm: passed to `pyflann.FLANN().build_index`
"""
self._n_visualwords = visualwords.shape[0]
self._flann = pyflann.FLANN()
if index_filepath is not None:
# load index
logger.debug('Loading visual words index from %s ...' %
(index_filepath))
t0 = time.time()
with open(BoFMaker.meta_filepath(index_filepath)) as f:
self._index_param = json.load(f)
self._flann.load_index(index_filepath, visualwords)
logger.debug('%f sec to load index' % (time.time() - t0))
else:
# create index
logger.debug('Creating index of visual words...')
t0 = time.time()
self._index_param = self._flann.build_index(visualwords,
algorithm=algorithm)
logger.debug('%f sec to create index' % (time.time() - t0))
def visualwords_union(features, loglevel='WARNING'):
"""Create simple visual words just by making union of feature vectors
Not suitable for large number of features.
:param features: feature vectors. Iterable of 2-D `numpy.ndarray`.
Example:
.. code-block: python
(
[[1.2, 3.5], [2.3, 4.1]], # image0's feature vectors
...
)
:rtype: 2-D `numpy.ndarray`
"""
t0 = time.time()
visualwords = np.array(list(itertools.chain.from_iterable(features)))
logger.debug('%f sec to union feature vectors' % (time.time() - t0))
return visualwords
def make(self, features, norm_order=None):
"""Create BoF representation of features.
:param features: 2D array of feature vectors
:type features: N x len(feature) `numpy.ndarray`,
where each element is `numpy.float32`
:param norm_order: `1` for L1-norm, `2` for L2-norm, ...
Histogram is not normalized when this is `None`.
"""
# nearest neighbor search
t0 = time.time()
nn_idx, dists = self._flann.nn_index(
features, 1, checks=self._index_param['checks'])
logger.debug('%f sec to search approx nearest neighbor' %
(time.time() - t0))
# make BoF histogram
bof_hist, bins = np.histogram(nn_idx, bins=self._n_visualwords)
if norm_order is None:
return bof_hist
return bof_hist / np.linalg.norm(bof_hist, ord=norm_order)
def make(self, features, norm_order=None):
"""Create BoF representation of features.
:param features: 2D array of feature vectors
:type features: N x len(feature) `numpy.ndarray`,
where each element is `numpy.float32`
:param norm_order: `1` for L1-norm, `2` for L2-norm, ...
Histogram is not normalized when this is `None`.
"""
# nearest neighbor search
t0 = time.time()
nn_idx, dists = self._flann.nn_index(
features, 1, checks=self._index_param['checks'])
logger.debug('%f sec to search approx nearest neighbor' %
(time.time() - t0))
# make BoF histogram
bof_hist, bins = np.histogram(nn_idx, bins=self._n_visualwords)
if norm_order is None:
return bof_hist
return bof_hist / np.linalg.norm(bof_hist, ord=norm_order)
def capdetector(imgpath, max_candidates, loglevel='WARNING'):
"""Detect circles from an image
:rtype: [{'x': <cap circle center x>, 'y': <cap circle center y>, 'r': <cap circle radius>}, ...]
"""
logger.setLevel(loglevel)
im = cv2.imread(imgpath, cv2.IMREAD_GRAYSCALE)
im_height, im_width = im.shape[0:2]
logger.debug('%s: size=(%d,%d)' % (basename(imgpath), im_width, im_height))
im = cv2.GaussianBlur(im, ksize=(5, 5), sigmaX=0)
# circles => [(x, y, r), ...] ; left one is most voted
t0 = time.time()
circles = cv2.HoughCircles(
im, cv2.cv.CV_HOUGH_GRADIENT,
dp=1,
minDist=1,
param1=85, param2=40,
minRadius=int(min(im_width, im_height) * 0.2),
maxRadius=int(min(im_width, im_height) * 0.6))
t1 = time.time()
logger.debug('cv2.HoughCircles(): %d circles detected in %f sec' %
(circles.size, t1 - t0))
# filter beer cap candidates from circles (at most `max_caididates`)
center_x, center_y = (im_width / 2, im_height / 2)
near_center_r = min(im_width, im_height) * 0.2
caps = []
t0 = time.time()
for x, y, r in circles[0, :]:
# only see near-center circle
if (x - center_x) ** 2 + (y - center_y) ** 2 > near_center_r ** 2:
continue
caps.append({'x': int(x), 'y': int(y), 'r': int(r)})
if len(caps) >= max_candidates:
break
t1 = time.time()
logger.debug('finding top-%d possible circles: %f sec' %
(max_candidates, t1 - t0))
return caps