def calculate_poses(option: Option, nn_classifier: ClassNN, svm_classifier: ClassSVM):
print('Calculating poses using nn')
# Recalculate all poses and get confidence
for classInfo in list_classes:
folder = classInfo['folderPath']
for root, _, files in os.walk(folder):
for file in files:
full_path = os.path.join(root, file)
ext = ClassUtils.get_filename_extension(full_path)
if '_rawdata' in file and ext == '.json':
print('Processing file: {0}'.format(full_path))
with open(full_path, 'r') as f:
file_txt = f.read()
file_json = json.loads(file_txt)
list_poses = file_json['listPoses']
for pose in list_poses:
angles = pose['angles']
transformed_points = pose['transformedPoints']
# Using transformed points and angles
list_desc = list()
list_desc += ClassUtils.get_flat_list(transformed_points)
# Convert to numpy
list_desc_np = np.asanyarray(list_desc, np.float)
if option == Option.NN:
result = nn_classifier.predict_model_fast(list_desc_np)
else:
result = svm_classifier.predict_model(list_desc_np)
pose['class'] = int(result['classes'])
pose['probabilities'] = result['probabilities'].tolist()
# Writing again into file
file_txt = json.dumps(file_json, indent=4)
new_full_path = ClassUtils.change_ext_training(full_path,
'{0}_posedata'.format(option.value))
with open(new_full_path, 'w') as f:
f.write(file_txt)
# Done
print('Done processing elements')
def apply_classifier(option: Option,
base_data_1,
base_data_2,
instance_nn=None,
hmm_models=None,
accuracy=0,
real_accuracy=0):
# Apply classification for each item!
for item in list_classes:
folder = item['folderPath']
for root, _, files in os.walk(folder):
for file in files:
full_path = os.path.join(root, file)
if '{0}_{1}_actiondata'.format(base_data_1,
base_data_2) in full_path:
with open(full_path, 'r') as f:
json_txt = f.read()
json_data = json.loads(json_txt)
list_actions = json_data['listActions']
if option == Option.HMM:
res = predict_data_hmm(list_actions, hmm_models)
else:
words = get_bow_descriptors(list_actions)
res = instance_nn.predict_model_fast(words)
obj_to_write = {
'class': int(res['classes']),
'probabilities': res['probabilities'].tolist(),
'modelAccuracy': accuracy,
'realAccuracy': real_accuracy
}
obj_txt = json.dumps(obj_to_write, indent=4)
new_filename = ClassUtils.change_ext_training(
full_path, '{0}_ensembledata'.format(option.value))
with open(new_filename, 'w') as f:
f.write(obj_txt)
# Done Writing
print('Done!')
def process_list_partial(filename, zone_data, idx_cls):
# Loading json data from filename
with open(filename, 'r') as f:
person_txt = f.read()
person = json.loads(person_txt)
list_poses = person['listPoses']
# Create partial list of ele
moving = True
index = 0
num_poses_future = 5
min_distance_x = 100
min_distance_y = 100
list_poses_action = list()
list_poses_partial = list()
while index < len(list_poses):
pose = list_poses[index]
remaining = len(list_poses) - index - 1
pos = pose['globalPosition']
if moving:
valid = True
if remaining >= num_poses_future:
count = 0
count_min = 0
for i in range(index + 1, index + num_poses_future):
count += 1
pos_future = list_poses[index +
num_poses_future]['globalPosition']
distance_x = math.fabs(pos[0] - pos_future[0])
distance_y = math.fabs(pos[1] - pos_future[1])
if distance_x < min_distance_x and distance_y < min_distance_y:
count_min += 1
# All points must be in a range
if count == count_min:
valid = False
if not valid:
if len(list_poses_partial) != 0:
# Check total moving changes
# Using movement variations
_, list_action_mov = ClassDescriptors.get_moving_action_poses(
list_poses_partial)
print('Total action moving: {0}'.format(
len(list_action_mov)))
for idx_act, action_mov in enumerate(list_action_mov):
if idx_act == len(action_mov) - 1:
# Take last action as num_pose_future position
final_pos = list_poses[
index + num_poses_future]['globalPosition']
else:
# Take final pos as last item
final_pos = action_mov[-1]['globalPosition']
initial_pos = action_mov[0]['globalPosition']
# Get if action ends in zone
# Except for walk, loitering actions
if idx_cls == 2 or idx_cls == 6:
in_zone = False
in_zone_before = False
else:
in_zone = is_point_in_zone(final_pos, zone_data)
# Get if action init in zone
in_zone_before = is_point_in_zone(
initial_pos, zone_data)
# Saving
list_poses_action.append({
'moving':
moving,
'listPoses':
copy.deepcopy(action_mov),
'finalPos':
final_pos,
'inZone':
in_zone,
'inZoneBefore':
in_zone_before
})
list_poses_partial.clear()
moving = False
list_poses_partial.append(pose)
else:
# Generating elements into list
valid = True
if remaining >= num_poses_future:
for i in range(index + 1, index + num_poses_future):
pos_future = list_poses[index +
num_poses_future]['globalPosition']
distance_x = math.fabs(pos[0] - pos_future[0])
distance_y = math.fabs(pos[1] - pos_future[1])
# Inverse
# If there is some point with greater distance!
if distance_x >= min_distance_x or distance_y >= min_distance_y:
valid = False
break
list_poses_partial.append(pose)
if not valid:
final_pos = list_poses[index]['globalPosition']
in_zone = is_point_in_zone(final_pos, zone_data)
# New pose list
for i in range(num_poses_future):
list_poses_partial.append(list_poses[index + i])
list_poses_action.append({
'moving':
moving,
'listPoses':
copy.deepcopy(list_poses_partial),
'finalPos':
final_pos,
'inZone':
in_zone,
'inZoneBefore':
False # In zone before does not apply
})
list_poses_partial.clear()
index += num_poses_future
moving = True
# Iterator!
index += 1
if len(list_poses_partial) != 0:
final_pos = list_poses_partial[-1]['globalPosition']
initial_pos = list_poses_partial[0]['globalPosition']
if moving:
if idx_cls == 2 or idx_cls == 6:
in_zone = False
in_zone_before = False
else:
in_zone = is_point_in_zone(final_pos, zone_data)
in_zone_before = is_point_in_zone(initial_pos, zone_data)
_, list_action_mov = ClassDescriptors.get_moving_action_poses(
list_poses_partial)
for action_mov in list_action_mov:
list_poses_action.append({
'moving': moving,
'listPoses': copy.deepcopy(action_mov),
'finalPos': final_pos,
'inZone': in_zone,
'inZoneBefore': in_zone_before
})
else:
in_zone = is_point_in_zone(final_pos, zone_data)
# Saving elements in partial format
list_poses_action.append({
'moving':
moving,
'listPoses':
copy.deepcopy(list_poses_partial),
'finalPos':
final_pos,
'inZone':
in_zone,
'inZoneBefore':
False # In zone before does not apply
})
list_poses_partial.clear()
# Generating elements into list
person_action = {
'personGuid': person['personGuid'],
'listPosesAction': list_poses_action
}
new_filename = ClassUtils.change_ext_training(filename, 'partialdata')
action_txt = json.dumps(person_action, indent=4)
with open(new_filename, 'w') as f:
f.write(action_txt)
# Saving list poses action for debugging
save_list_poses_action(filename, list_poses_action, zone_data)
print('Done!')
def process_file_action(full_path,
option: Option,
instance_model=None,
hmm_models=None,
accuracy=0):
with open(full_path, 'r') as f:
json_txt = f.read()
json_data = json.loads(json_txt)
list_actions = list()
list_list_poses = json_data['listPosesAction']
for list_poses_data in list_list_poses:
list_poses = list_poses_data['listPoses']
moving = list_poses_data['moving']
in_zone = list_poses_data['inZone']
in_zone_before = list_poses_data['inZoneBefore']
# Descriptor not moving
if not moving:
if option == Option.NN:
descriptor = get_nn_descriptor(list_poses)
descriptor_nn = np.asanyarray(descriptor, dtype=np.float)
res = instance_model.predict_model_fast(descriptor_nn)
elif option == Option.CNN:
descriptor = get_cnn_descriptor_pos(list_poses)
descriptor_nn = np.asanyarray(descriptor, dtype=np.float)
res = instance_model.predict_model_fast(descriptor_nn)
else:
res = predict_data(hmm_models, list_poses)
cls = int(res['classes'])
if in_zone:
cls += 4
probabilities = res['probabilities'].tolist()
list_prob = list()
if in_zone:
for i in range(4):
list_prob.append(0)
list_prob += probabilities
else:
list_prob += probabilities
for i in range(4):
list_prob.append(0)
for i in range(4):
list_prob.append(0)
data_action = {
'class': cls,
'probabilities': list_prob,
'count': len(list_poses)
}
else:
if not in_zone and not in_zone_before:
cls = 8
elif in_zone and not in_zone_before:
cls = 9
elif not in_zone and in_zone_before:
cls = 10
else:
cls = 11
probabilities = list()
for i in range(12):
if i == cls:
probabilities.append(1)
else:
probabilities.append(0)
data_action = {
'class': cls,
'probabilities': probabilities,
'count': len(list_poses)
}
list_actions.append(data_action)
json_data = {'listActions': list_actions, 'accuracy': accuracy}
# Saving full action info into list
json_txt = json.dumps(json_data, indent=4)
new_file_path = ClassUtils.change_ext_training(
full_path, '{0}_actiondata'.format(option.value))
with open(new_file_path, 'w') as f:
f.write(json_txt)
