def __init__(self):
self.bpm_detector = findFaceGetPulse(
bpm_limits=[50, 160],
data_spike_limit=2500.,
face_detector_smoothness=10.,
cascade_path=self._face_cascade)
self.bpm_attempts = 0
self.last_bpm = 61.2
self.face_detector = self.make_face_detector()
self.smile_detector = Detector('smile')
self.heartbeat_deque = deque()
self.likeness = 0
class ImageScanner(object):
_face_cascade = os.path.join(fpath, '..', 'facerec/py/apps/videofacerec/haarcascade_frontalface_alt2.xml')
_dataset_location = os.path.join(fpath, '..', '..', 'data')
_smile_weight = 1.0
_heartbeat_weight = 0.1
def __init__(self):
self.bpm_detector = findFaceGetPulse(
bpm_limits=[50, 160],
data_spike_limit=2500.,
face_detector_smoothness=10.,
cascade_path=self._face_cascade)
self.bpm_attempts = 0
self.last_bpm = 61.2
self.face_detector = self.make_face_detector()
self.smile_detector = Detector('smile')
self.heartbeat_deque = deque()
self.likeness = 0
def make_face_detector(self):
image_size = (100, 100) # TODO: Fiddle with this
images, labels, subject_names = fr_tools.read_images(self._dataset_location, image_size)
list_of_labels = list(xrange(max(labels) + 1))
subject_dictionary = dict(zip(list_of_labels, subject_names))
self.face_model = fr_tools.get_model(image_size=image_size, subject_names=subject_dictionary)
self.face_model.compute(images, labels)
face_detector = CascadedDetector(cascade_fn=self._face_cascade, minNeighbors=5, scaleFactor=1.1)
return face_detector
def downsample(self, image):
img = cv2.resize(image, (image.shape[1] / 2, image.shape[0] / 2), interpolation=cv2.INTER_CUBIC)
return img
def detect_face(self, img):
detections = self.face_detector.detect(img)
for i, bounding_box in enumerate(detections):
x0, y0, x1, y1 = bounding_box
return x0, y0, x1, y1
def recognize_face(self, img):
# TODO: Accumulate/make probabilistic
detections = self.face_detector.detect(img)
for i, bounding_box in enumerate(detections):
if i >= 1:
print "Ignoring face at index > 1"
break
x0, y0, x1, y1 = bounding_box
face = img[y0:y1, x0:x1]
face = cv2.cvtColor(face, cv2.COLOR_BGR2GRAY)
# print len(self.smile_detector.classify(face))
face = cv2.resize(face, self.face_model.image_size, interpolation=cv2.INTER_CUBIC)
smiling = self.smile_detector.classify(face)
if smiling != []:
self.likeness += self._smile_weight
else:
self.likeness -= self._smile_weight * 0.3
prediction = self.face_model.predict(face)[0]
person_name = self.face_model.subject_names[prediction]
return person_name, bounding_box, smiling
else:
return None
def get_bpm(self, image):
self.bpm_attempts += 1
# TODO: Check if we've detected a face
# Try a few times first
self.bpm_detector.frame_in = image
self.bpm_detector.find_faces = bool(self.bpm_attempts < 5)
self.bpm_detector.run(1)
if self.bpm_attempts > 5:
bpm = self.bpm_detector.bpm
if bpm == 0:
bpm = self.last_bpm
elif bpm > 105:
bpm = bpm / 2.0
else:
self.last_bpm = bpm
else:
bpm = 61.2 + np.random.normal(3.0)
if len(self.heartbeat_deque) > 15:
self.heartbeat_deque.popleft()
self.heartbeat_deque.append(bpm)
average = np.average(self.heartbeat_deque)
self.likeness += (average - 60.0) * self._heartbeat_weight
return average
