class TrainPose():
def __init__(self, id, nsamples, dst):
self._id = id
self._nsamples = nsamples
self._dst = dst
self._kinect = Kinect()
self._body = BodyDetector()
self._hand = HandDetector()
self._contour = HandContourDetector()
self._fdesc = FourierDescriptors()
self._train = []
def run(self):
warmup = True
for (depth8, depth, rgb) in self._kinect.get_data():
contour = self._get_hand_contour(depth8, depth, rgb)
if not contour:
continue
self._contour.draw()
if warmup:
key = cv2.waitKey(5)
if key == GO:
warmup = False
continue
fd = self._fdesc.run(contour)
self._train.append(fd)
if len(self._train) == self._nsamples:
self._save()
break
cv2.waitKey(5)
def _get_hand_contour(self, depth8, depth, rgb):
body = self._body.run(depth8)
(hand, _) = self._hand.run(body)
(cont, box, hc) = self._contour.run(hand)
if self._contour.not_valid():
return []
(cont, _, _) = self._contour.run(rgb, True, box, hc, depth)
return cont
def _save(self):
data = np.array(self._train)
model = EmpiricalCovariance().fit(np.array(self._train))
output = {'id': self._id, 'data': data, 'model': model}
pickle.dump(output, open(self._dst, 'wb'))
class TrainPose():
def __init__(self, id, nsamples, dst):
self._id = id
self._nsamples = nsamples
self._dst = dst
self._kinect = Kinect()
self._body = BodyDetector()
self._hand = HandDetector()
self._contour = HandContourDetector()
self._fdesc = FourierDescriptors()
self._train = []
def run(self):
warmup = True
for (depth8, depth, rgb) in self._kinect.get_data():
contour = self._get_hand_contour(depth8, depth, rgb)
if not contour:
continue
self._contour.draw()
if warmup:
key = cv2.waitKey(5)
if key == GO:
warmup = False
continue
fd = self._fdesc.run(contour)
self._train.append(fd)
if len(self._train) == self._nsamples:
self._save()
break
cv2.waitKey(5)
def _get_hand_contour(self, depth8, depth, rgb):
body = self._body.run(depth8)
(hand, _) = self._hand.run(body)
(cont, box, hc) = self._contour.run(hand)
if self._contour.not_valid():
return []
(cont, _, _) = self._contour.run(rgb, True, box, hc, depth)
return cont
def _save(self):
data = np.array(self._train)
model = EmpiricalCovariance().fit(np.array(self._train))
output = {'id': self._id, 'data': data, 'model': model}
pickle.dump(output, open(self._dst, 'wb'))
class MultiPoseClassifier():
def __init__(self, src):
self._fourier_desc = FourierDescriptors()
self._models = self._load_models(src)
def run(self, contour):
desc = self._fourier_desc.run(contour)
dist = np.zeros(len(self._models))
for i, model in enumerate(self._models):
dist[i] = model['model'].mahalanobis([desc])
pose = self._models[np.argsort(dist)[0]]['id']
return pose
def _load_models(self, src):
models = []
for s in src:
data = pickle.load(open(s, 'rb'))
models.append({'id': data['id'], 'model': data['model']})
return models
class MultiPoseClassifier():
def __init__(self, src):
self._fourier_desc = FourierDescriptors()
self._models = self._load_models(src)
def run(self, contour):
desc = self._fourier_desc.run(contour)
dist = np.zeros(len(self._models))
for i, model in enumerate(self._models):
dist[i] = model['model'].mahalanobis([desc])
pose = self._models[np.argsort(dist)[0]]['id']
return pose
def _load_models(self, src):
models = []
for s in src:
data = pickle.load(open(s, 'rb'))
models.append({'id': data['id'], 'model': data['model']})
return models