-
Notifications
You must be signed in to change notification settings - Fork 0
/
postOpenPose.py
243 lines (213 loc) · 11.3 KB
/
postOpenPose.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
# This script is part of the bachelor thesis
# "Localization of medical measurement data from injured persons for use in rescue robotics"
# from Konstantin Wenig written for the university to luebeck's bachelor degree in robotics and autonomous systems
# Redistribution and change without consent it not allowed
import cv2
import numpy as np
import yaml
import os
from matplotlib import pyplot as plt
import wound_transformation
try:
import rospy
import wound_transformation_ros_node
except:
print('Could not import rospy.')
def extract_keypoints(path):
with open(path, 'r') as stream:
data = yaml.safe_load(stream)
try:
assert not len(data['people']) == 0
except:
return {}
keypoints_2d = data['people'][0]['pose_keypoints_2d']
# Convert the list to a dictionary that consists of entries following (x,y,confidence)
keypoints = {}
j = 0
for i in range(len(keypoints_2d)):
if i % 3 == 0:
keypoints[j] = {"x": keypoints_2d[i],
"y": keypoints_2d[i + 1],
"c": keypoints_2d[i + 2]}
j = j + 1
return keypoints
def handle_video(path, wound_transform, **kwargs):
_video_path = path + '/assets/videos/' + kwargs.get('video_name')
_video_name = kwargs.get('video_name').rsplit('.', 1)[0]
_wound_path = path + '/assets/videos/wounds_' + _video_name + '.yaml'
_keypoints_dir_path = path + '/assets/videos/keypoints_' + _video_name + '/'
try:
assert os.path.exists(_video_path)
assert os.path.exists(_wound_path)
except FileNotFoundError:
print('Could not find the files in the specified path.')
print('Video path: ', _video_path)
print('Wound path: ', _wound_path)
exit(-1)
# Load wound data
with open(_wound_path, 'r') as stream:
_video_wounds = yaml.safe_load(stream)
# Create VideoCapture object and load video
_video_cap = cv2.VideoCapture(_video_path)
# Define origin of frame counter on image
org = (10, 30)
# TEST!!!!!!
# Heatmap for wound location, only works with a single wound!
_standard_view = cv2.imread(path + '/assets/pictures/standard/male_100_rendered.png')
_heatmap = np.zeros(_standard_view.shape) # 348 x 612 is the shape of the standard view
_error = []
_out_path = path + '/assets/videos/' + _video_name + '_out.mp4'
_fps = _video_cap.get(cv2.CAP_PROP_FPS)
#_w = int(_video_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
#_h = int(_video_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
_h, _w, _ = wound_transform.standard_view_img.shape
_four_cc = cv2.VideoWriter_fourcc(*'mp4v')
#print(_out_path, _four_cc, _fps, (_w, _h))
_out = cv2.VideoWriter(_out_path, _four_cc, _fps, (_w, _h))
# TEST!!!!!!
# Read from _video_cap until video ends
while _video_cap.isOpened():
_ret, _frame = _video_cap.read()
if _ret:
_frame_num = int(_video_cap.get(cv2.CAP_PROP_POS_FRAMES)) - 1
_frame_name = _video_name + '_' + str(_frame_num).zfill(12)
# Extract keypoints, this only works with openpose data!
_keypoints = extract_keypoints(_keypoints_dir_path + _frame_name + '_keypoints.json')
_wounds = [_video_wounds[_frame_num]]
# print('_Wounds: ', _wounds)
if len(_wounds[0]) > 0:
try:
# Step 0: Display wound on original video-frame and write to video
#_image_before = cv2.circle(np.ndarray.copy(_frame), _wounds[0], 5, (0, 0, 255), 2)
#_out.write(_image_before)
# Step 1: Convert wounds from random pose on image to standard view reference frame
_conv_wounds = wound_transform.locate_wounds(_frame, _keypoints, _wounds)
if _conv_wounds[0][0] == -1 and _conv_wounds[0][1] == -1:
_heatmap[_conv_wounds[0][1], _conv_wounds[0][0]] += 0
else:
_heatmap[_conv_wounds[0][1], _conv_wounds[0][0]] += 1
# Step 2: Transfer converted wounds onto the standard view
_img_conv = wound_transform.transform_view(_conv_wounds)
# Step 3: Re-project converted wounds onto original image
_reprojected_wounds = wound_transform.reproject_wounds(_keypoints, _wounds, _conv_wounds)
# Step 4: Calculate reprojected error
_error.append(wound_transformation.calculate_reprojection_error(_wounds, _reprojected_wounds))
# Step 5: Write image of transformed view to video
_out.write(_img_conv)
#wound_transform.display_error(_frame, _wounds, _reprojected_wounds, video=_frame_name)
except:
_error.append([-1])
#_standard_view_no_detect = cv2.putText(np.ndarray.copy(_standard_view), 'Could not transfer wound.',
# org, cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 0), 2)
_out.write(_standard_view)
else:
_standard_view_no_wound = cv2.putText(np.ndarray.copy(_standard_view), 'No wound found.', org,
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), 2)
_out.write(_standard_view_no_wound)
_error.append([0])
# cv2.imshow('frame', _frame)
# cv2.waitKey(10)
else:
break
_error_per_wound = np.array(_error).flatten()
with open(path + '/assets/videos/error_per_wound_' + _video_name + '.yaml', 'w') as error_file:
yaml.dump(_error_per_wound.tolist(), error_file)
wound_transform.plot_error(_error_per_wound, video=_video_name)
_ksize = (3, 3) # Kernel size
_sigma = 10 # Standard deviation of the Gaussian distribution
_heatmap_extended = cv2.GaussianBlur(_heatmap, _ksize, _sigma)
_heatmap_normalized = _heatmap_extended / np.max(_heatmap_extended)
_heatmap_colored = cv2.applyColorMap(np.uint8(255 * _heatmap_extended), cv2.COLORMAP_JET)
_standard_view = cv2.imread(path + '/assets/pictures/standard/male_100_rendered.png')
#print(_heatmap_colored.shape)
#print(_standard_view.shape)
#cv2.imshow('heatmap', _heatmap_colored)
#cv2.waitKey(0)
_applied_heatmap = cv2.addWeighted(_standard_view, 0.7, _heatmap_colored, 0.3, 0)
cv2.imwrite(path + '/assets/videos/' + _video_name + '_converted_wound_heatmap.jpg', _applied_heatmap)
#wound_transform.plot_error(_video_wounds)
#cv2.imshow('Resultat', _applied_heatmap)
#cv2.waitKey(0)
def main(**kwargs):
if 'host' in kwargs and kwargs.get('host') == 'local':
_base_path = os.path.dirname(
os.path.abspath("postOpenPose.py")
)
if 'skeleton' in kwargs and kwargs.get('skeleton') == 'nuitrack':
wound_transform = wound_transformation.WoundTransformation(skeleton=kwargs.get('nuitrack'))
else:
wound_transform = wound_transformation.WoundTransformation(skeleton=kwargs.get('skeleton'))
# Analyze and transform wounds live or from saved images?
if 'mode' in kwargs and kwargs.get('mode') == 'video':
# Get video path
if 'video_name' in kwargs:
handle_video(_base_path, wound_transform, video_name=kwargs.get('video_name'))
elif 'mode' in kwargs and kwargs.get('mode') == 'image':
if 'image_num' in kwargs:
_num = kwargs['image_num']
else:
try:
# No number was given, choose a random number between 0 and the amount of pictures in the rs_color
# folder
_num_img = len(os.listdir(_base_path + '/assets/pictures/rs_color/'))
_num = np.random.randint(0, _num_img)
except:
print('Could not find path, please run the script preOpenPose.py '
'to calibrate the cameras and create pictures')
exit(1)
# Load image
_img = cv2.imread(_base_path + '/assets/pictures/rs_color/' + str(_num) + '.jpg')
try:
_img_d = cv2.imread(_base_path + '/assets/pictures/rs_depth/' + str(_num) + '.jpg')
except:
_img_d = None
# Load keypoints
if 'skeleton' in kwargs and kwargs.get('skeleton') == 'nui':
print('NuiTrack not yet implemented.')
exit(1)
else:
_keypoints = extract_keypoints(_base_path + '/assets/pictures/keypoints/' + str(_num) + '_keypoints.json')
# Load the wounds data
with open(_base_path + '/assets/pictures/wounds/' + str(_num) + '.yaml', 'r') as stream:
_wounds = yaml.safe_load(stream)
# _img, _wounds, _keypoints are loaded, proceed with wound transformation
#print(_wounds)
_img_copy = np.copy(_img)
for wound in _wounds:
_img_copy = cv2.circle(_img_copy, (np.round(wound[0]).astype('int'), np.round(wound[1]).astype('int')),
5, (0, 0, 255), 2)
# Step 1: Convert wounds from random pose on image to standard view reference frame
_conv_wounds = wound_transform.locate_wounds(_img, _keypoints, _wounds, _img_d, _num)
print('Converted Wounds: ', _conv_wounds)
# Step 2: Transfer converted wounds onto the standard view
_img_conv = wound_transform.transform_view(_conv_wounds)
# Step 2.5: Save image to output
#wound_transform.save_images(after=_img_conv, num=_num)
# Step 3: Re-project converted wounds onto original image
#_reprojected_wounds = wound_transform.reproject_wounds(_keypoints, _wounds, _conv_wounds)
# Step 4: Calculate and display reprojected error
#wound_transform.display_error(_img, _wounds, _reprojected_wounds, image_num=_num)
#error_per_wound = wound_transformation.calculate_reprojection_error(_wounds, _reprojected_wounds)
#wound_transform.plot_error(error_per_wound, image=_num)
else:
# No local data acquisition, use ROS
wound_transformation_ros_node.WoundTransformROSNode()
rospy.spin()
if __name__ == '__main__':
# Possible optional parameters:
# host: local If host='local' provided, non-ROS version will be started
# mode: image/video If any mode is provided and non-ROS version started, either an image or a video will be
# processed
# image_num: int If any image_num is provided and non-ROS, image version started, the image with the
# specified image_num will be used
# video_name: String If any video_name is provided and non-ROS, video version was started, the video with the
# specified name will be used
"""
Example for image mode using openpose:
main(mode='image', num=9, skeleton='openpose', host='local')
Example for video variant: (Not yet implemented)
main(mode='video', video_name=sampleVid.mp4, host='local')
Example for ROS:
main()
"""
main()