def get_person_distance(person1: 'ClassPeopleReId',
person2: 'ClassPeopleReId'):
# Comparing person distance
# x factor penalty
distance = ClassUtils.get_euclidean_distance(
person1.global_pos[0] * 1.3, person1.global_pos[1],
person2.global_pos[0] * 1.3, person2.global_pos[1])
return distance
def main():
print('Warning: This code is deprecated')
print('Use testconvertmjpegxr.py instead')
FORMAT = "%(asctime)s [%(name)-16.16s] [%(levelname)-5.5s] %(message)s"
logging.basicConfig(level=logging.INFO, format=FORMAT)
print('Init re-id')
# Cleaning image tree
clean_base_folder()
list_cams = [419, 420, 421, 428, 429, 430]
date_init = datetime(2018, 2, 24, 14, 15, 0)
date_end = datetime(2018, 2, 24, 15, 15, 0)
date_video = date_init
list_readers = []
for i in range(len(list_cams)):
list_readers.append(ClassMjpegDate(list_cams[i]))
list_people = []
while date_video < date_end:
print(date_video)
for i in range(len(list_cams)):
frame_info = list_readers[i].load_frame(date_video)
dic_frame = frame_info[2]
image = frame_info[0]
frame_position = dic_frame['positions']
vectors = dic_frame['vectors']
guid_list = []
for candidate_position in frame_position:
guid = ''
score = candidate_position[2]
index = frame_position.index(candidate_position)
person_vectors = vectors[index]
# Only takes into account scores in one
if score == 1:
found = False
person_candidate = None
for person in list_people:
if not person['updated']:
person_position = person['position']
distance = ClassUtils.get_euclidean_distance(
candidate_position[0], candidate_position[1],
person_position[0], person_position[1])
if distance < min_distance:
found = True
person_candidate = person
break
if not found:
# Creates candidate into list
candidate = create_candidate(candidate_position,
date_video,
person_vectors, image)
list_people.append(candidate)
guid = candidate['guid']
print('Creating guid {0}'.format(guid))
else:
# Update only position
update_candidate(person_candidate, candidate_position,
date_video, person_vectors, image)
guid = person_candidate['guid']
print('Updating guid {0}'.format(guid))
guid_list.append(guid)
# Clean old records -> 5 seconds
people_to_remove = []
for person in list_people:
last_update = person['last_update']
delta_update = date_video - last_update
if delta_update.total_seconds() > clean_timeout_sec:
people_to_remove.append(person)
for person in people_to_remove:
# Remove from list and process action
list_people.remove(person)
get_cnn_images(person)
# Update person info again
for person in list_people:
person['updated'] = False
# Adding element to dict_frame
dic_frame['guids'] = guid_list
list_readers[i].update_frame(frame_info)
# Adding time to list
date_video = date_video + timedelta(milliseconds=game_period_ms)
# Done in all cycles -> Saving results
for reader in list_readers:
reader.save_frame_info()
def get_cnn_images(person):
# Getting cnn images
# Frame list must be ordered in list frames
# Custom frames between person
# Save vector, position
list_frames = person['list_frames']
new_person_vectors = []
for frame in list_frames:
person_vectors = frame['person_vectors']
torso_dis_1 = ClassUtils.get_euclidean_distance(
person_vectors[1][0], person_vectors[1][1], person_vectors[8][0],
person_vectors[8][1])
torso_dis_2 = ClassUtils.get_euclidean_distance(
person_vectors[1][0], person_vectors[1][1], person_vectors[11][0],
person_vectors[11][1])
logger.debug('Distance 1: {0} Distance 2: {1}'.format(
torso_dis_1, torso_dis_2))
torso_dis = (torso_dis_1 + torso_dis_2) / 2
print('Generating normalization')
neck_vector = person_vectors[1]
# Ignoring vectors from face
# Only consider vectors from 1 to 14
new_vectors = []
for i in range(14):
vector = person_vectors[i]
score = vector[2]
if score > ClassUtils.MIN_POSE_SCORE:
new_x = (vector[0] - neck_vector[0]) / torso_dis
new_y = (vector[1] - neck_vector[1]) / torso_dis
new_vectors.append([new_x, new_y])
else:
new_x = 0
new_y = 0
new_vectors.append([new_x, new_y])
new_person_vectors.append(new_vectors)
# Get min_x, max_x, min_y, max_y of the sequence
min_x = 0
min_y = 0
max_x = 0
max_y = 0
for new_vectors in new_person_vectors:
for vector in new_vectors:
if vector[0] < min_x:
min_x = vector[0]
if vector[0] > max_x:
max_x = vector[0]
if vector[1] < min_y:
min_y = vector[1]
if vector[1] > max_y:
max_y = vector[1]
# Normalizing vectors between min and max
factor_x = 255 / (max_x - min_x)
factor_y = 255 / (max_y - min_y)
for new_vectors in new_person_vectors:
for vector in new_vectors:
vector[0] = int((vector[0] - min_x) * factor_x)
vector[1] = int((vector[1] - min_y) * factor_y)
# Creating image based on opencv library
rows = len(new_person_vectors[0])
cols = len(new_person_vectors)
blank_image = np.zeros((rows, cols, 3), np.uint8)
for j in range(cols):
for i in range(rows):
# First item are frame, second item are joint position
# Image rows are joint position
# Image cols are frame
x_value = new_person_vectors[j][i][0]
y_value = new_person_vectors[j][i][1]
blank_image[i, j] = [x_value, y_value, 0]
# Creating folder from videos
if not os.path.isdir(CNN_BASE_FOLDER):
os.makedirs(CNN_BASE_FOLDER)
# Saving image as jpeg
path_image = os.path.join(CNN_BASE_FOLDER, person['guid'] + '.jpg')
cv2.imwrite(path_image, blank_image)
# Saving image list for debug purposes
for frame in list_frames:
index_frame = list_frames.index(frame)
image = frame['image']
image_folder = os.path.join(CNN_BASE_FOLDER, person['guid'])
if not os.path.isdir(image_folder):
os.makedirs(image_folder)
aux_image_path = os.path.join(image_folder, str(index_frame) + '.jpg')
ClassUtils.write_bin_to_file(aux_image_path, image)
print('Done!')