"""
Script used to prepare the URFD dataset:
1- It takes the downloaded RGB images of the camera 0, divided into two
folders: 'Falls' and 'ADLs', where all the images of a video (comprised in a
folder) are inside one of those folders.
2- It creates another folder for the new dataset with the 'Falls' and
'NotFalls' folders. All the ADL videos are moved to the 'NotFalls' folder.
The images within the original 'Falls' folder are divided in three stages:
(i) the pre-fall ADL images (they go to the new 'NotFalls' folder),
(ii) the fall itself (goes to the new 'Falls' folder) and
(iii) the post-fall ADL images (to 'NotFalls').
All the images are resized to size (W,H) - both W and H are variables of the
script.
The script should generate all the necessary folders.
"""
import cv2
import os
import csv
import sys
import glob
import numpy as np
import zipfile
# Path where the images are stored
downloads_folder = '/home/user/Downloads/'
data_folder = 'URFD_images_not_segmented/'
adl_folder = 'ADLs/'
fall_folder = 'Falls/'
# Path to save the images
output_path = 'URFD_images/'
# Label files, download them from the dataset's site
falls_labels = 'urfall-cam0-falls.csv'
notfalls_labels = 'urfall-cam0-adls.csv'
W, H = 224, 224 # shape of new images (resize is applied)
# =====================================================================
# UNZIP THE DATASET
# =====================================================================
if not os.path.exists(data_folder):
os.makedirs(data_folder + fall_folder)
os.makedirs(data_folder + adl_folder)
adl_zipped_files = glob.glob(downloads_folder + 'adl-*-cam0-rgb.zip')
fall_zipped_files = glob.glob(downloads_folder + 'fall-*-cam0-rgb.zip')
content = [
[adl_zipped_files, data_folder + adl_folder],
[fall_zipped_files, data_folder + fall_folder]
]
for zipped_files, dst_folder in content:
for zipped_file in zipped_files:
zfile = zipfile.ZipFile(zipped_file)
zfile.extractall(dst_folder)
# =====================================================================
# READ LABELS AND STORE THEM
# =====================================================================
labels = {'falls': dict(), 'notfalls': dict()}
# For falls videos: read the CSV where frame-level labels are given
with open(falls_labels, 'rb') as csvfile:
spamreader = csv.reader(csvfile, delimiter=' ', quotechar='|')
event_type = 'falls'
for row in spamreader:
elems = row[0].split(',') # read a line in the csv
if not elems[0] in labels[event_type]:
labels[event_type][elems[0]] = []
if int(elems[2]) == 1 or int(elems[2]) == -1:
labels[event_type][elems[0]].append(0)
elif int(elems[2]) == 0:
labels[event_type][elems[0]].append(1)
# For ADL videos: read the CSV where frame-level labels are given
with open(notfalls_labels, 'rb') as csvfile:
spamreader = csv.reader(csvfile, delimiter=' ', quotechar='|')
event_type ='notfalls'
for row in spamreader:
elems = row[0].split(',') # read a line in the csv
if not elems[0] in labels[event_type]:
labels[event_type][elems[0]] = []
if int(elems[2]) == 1 or int(elems[2]) == -1:
labels[event_type][elems[0]].append(0)
elif int(elems[2]) == 0:
labels[event_type][elems[0]].append(1)
print('Label files processed')
# =====================================================================
# PROCESS THE DATASET
# =====================================================================
if not os.path.exists(output_path):
os.makedirs(output_path + 'Falls')
os.makedirs(output_path + 'NotFalls')
# Get all folders: each one contains the set of images of the video
folders = [f for f in os.listdir(data_folder)
if os.path.isdir(os.path.join(data_folder, f))]
for folder in folders:
print('{} videos =============='.format(folder))
events = [f for f in os.listdir(data_folder + folder)
if os.path.isdir(os.path.join(data_folder + folder, f))]
events.sort()
for nb_event, event, in enumerate(events):
# Create the appropriate folder
if folder == 'ADLs':
event_id = event[:6]
#new_folder = output_path + 'NotFalls/notfall_{}'.format(event_id)
new_folder = output_path + 'NotFalls/{}'.format(event)
if not os.path.exists(new_folder):
os.makedirs(new_folder)
elif folder == 'Falls':
event_id = event[:7]
# "No falls" come before and after the fall, so the respective
# folders must be created
#new_folder = output_path + 'Falls/fall_{}'.format(event_id)
new_folder = output_path + 'Falls/{}'.format(event)
if not os.path.exists(new_folder):
os.makedirs(new_folder)
#folder_created = False
path_to_images = data_folder + folder + '/' + event + '/'
# Load all the images of the video
images = [f for f in os.listdir(path_to_images)
if os.path.isfile(os.path.join(path_to_images, f))]
images.sort()
fall_detected = False # whether a fall has been detected in the video
for nb_image, image in enumerate(images):
x = cv2.imread(path_to_images + image)
# If the image is part of an ADL video no fall need to be
# considered
if folder == 'ADLs':
# Save the image
save_path = (output_path +
#'NotFalls/notfall_{}'.format(event_id) +
'NotFalls/{}'.format(event) +
'/frame{:04}.jpg'.format(nb_image))
cv2.imwrite(save_path,
cv2.resize(x, (W,H)),
[int(cv2.IMWRITE_JPEG_QUALITY), 95])
elif folder == 'Falls':
event_type = 'falls'
if labels[event_type][event_id][nb_image] == 0: # ADL
if fall_detected:
# Create another folder for an ADL event,
# i.e. the post-fall ADL event
new_folder = (output_path +
#'NotFalls/notfall_{}_post'.format(
#event_id))
'NotFalls/{}_post'.format(event))
if not os.path.exists(new_folder):
os.makedirs(new_folder)
save_path = (output_path +
#'NotFalls/notfall_{}_post'.format(
#event_id) +
'NotFalls/{}_post'.format(event) +
'/frame{:04}.jpg'.format(nb_image))
else:
new_folder = (output_path +
#'NotFalls/notfall_{}_pre'.format(event_id))
'NotFalls/{}_pre'.format(event))
if not os.path.exists(new_folder):
os.makedirs(new_folder)
save_path = (output_path +
#'NotFalls/notfall_{}_pre'.format(
#event_id) +
'NotFalls/{}_pre'.format(event) +
'/frame{:04}.jpg'.format(nb_image))
cv2.imwrite(save_path,
cv2.resize(x, (W,H)),
[int(cv2.IMWRITE_JPEG_QUALITY), 95])
elif labels[event_type][event_id][nb_image] == 1: # actual fall
save_path = (output_path +
#'Falls/fall_{}'.format(event_id) +
'Falls/{}'.format(event) +
'/frame{:04}.jpg'.format(nb_image))
cv2.imwrite(save_path,
cv2.resize(x, (W,H)),
[int(cv2.IMWRITE_JPEG_QUALITY), 95])
# If fall is detected in a video set the variable to True
# used to discern between pre- and post-fall ADL events
fall_detected = True
print('End of the process, all the images stored within the {} folder'.format(
output_path))
Datasets
Models
