def recognize_gesture(self, source):
# calculating the optical flow of unknown gesture
# and reside to match with training data's
source_flow = measure(source, f=optical_flow_magnitude())
source_flow = cv2.resize(source_flow, (180, 320)).flatten()
# getting the prediction after passing the unknown data
# to SVM
i = int(self.model.predict(source_flow))
return GESTURES[i]
def recognize_gesture(self, source):
# Calculate the optical flow magnitute of unknown gesture
source_flow = measure(source, f=optical_flow_magnitude())
# resize to compare
source_flow = cv2.resize(source_flow, (180, 320))
avg_diffs = {}
for gesture in GESTURES:
# compare the unknown optical flow with data
diffs = [simple_difference(source_flow, flow)
for flow in self.known_gestures[gesture]]
avg_diffs[gesture] = np.mean(diffs)
print avg_diffs
# return the smallest difference -> the predicted gesture
return min(GESTURES, key=lambda x: avg_diffs[x])
def _process(gesture, idx, f):
source = 'videos/{0}{1}_blackbackground.mov'.format(gesture, idx)
print '\t', source
data = measure(source, f)
data = cv2.resize(data, (180, 320))
return data.flatten()