-
Notifications
You must be signed in to change notification settings - Fork 1
/
test9.py
284 lines (232 loc) · 11.6 KB
/
test9.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
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
import cv2
import time
import random
import numpy as np
from playsound import playsound
def f_dist(p1, p2):
return (p1[0] - p2[0]) * (p1[0] - p2[0]) + (p1[1] - p2[1]) * (p1[1] - p2[1])
def output_keypoints(frame, net, threshold, BODY_PARTS, now_frame, total_frame):
global points
# 입력 이미지의 사이즈 정의
image_height = 368
image_width = 368
# 네트워크에 넣기 위한 전처리
input_blob = cv2.dnn.blobFromImage(
frame, 1.0 / 255, (image_width, image_height), (0, 0, 0), swapRB=False, crop=False)
# 전처리된 blob 네트워크에 입력
net.setInput(input_blob)
# 결과 받아오기
out = net.forward()
# The output is a 4D matrix :
# The first dimension being the image ID ( in case you pass more than one image to the network ).
# The second dimension indicates the index of a keypoint.
# The model produces Confidence Maps and Part Affinity maps which are all concatenated.
# For COCO model it consists of 57 parts – 18 keypoint confidence Maps + 1 background + 19*2 Part Affinity Maps. Similarly, for MPI, it produces 44 points.
# We will be using only the first few points which correspond to Keypoints.
# The third dimension is the height of the output map.
out_height = out.shape[2]
# The fourth dimension is the width of the output map.
out_width = out.shape[3]
# 원본 이미지의 높이, 너비를 받아오기
frame_height, frame_width = frame.shape[:2]
# 포인트 리스트 초기화
points = []
for i in range(len(BODY_PARTS)):
# 신체 부위의 confidence map
prob_map = out[0, i, :, :]
# 최소값, 최대값, 최소값 위치, 최대값 위치
min_val, prob, min_loc, point = cv2.minMaxLoc(prob_map)
# 원본 이미지에 맞게 포인트 위치 조정
x = (frame_width * point[0]) / out_width
x = int(x)
y = (frame_height * point[1]) / out_height
y = int(y)
if prob > threshold: # [pointed]
cv2.circle(frame, (x, y), 5, (0, 255, 255),
thickness=-1, lineType=cv2.FILLED)
cv2.putText(frame, str(i), (x, y), cv2.FONT_HERSHEY_SIMPLEX,
0.6, (0, 0, 255), 1, lineType=cv2.LINE_AA)
points.append((x, y))
else: # [not pointed]
cv2.circle(frame, (x, y), 5, (0, 255, 255),
thickness=-1, lineType=cv2.FILLED)
cv2.putText(frame, str(i), (x, y), cv2.FONT_HERSHEY_SIMPLEX,
0.6, (255, 0, 0), 1, lineType=cv2.LINE_AA)
points.append(None)
return frame
def output_keypoints_with_lines(frame, POSE_PAIRS):
for pairs in POSE_PAIRS :
cv2.line(frame, points[pairs[0]], points[pairs[1]], (random.randrange(0, 256), random.randrange(0, 256), random.randrange(0, 256)), thickness=4, lineType=cv2.LINE_AA)
return frame
def output_keypoints_with_lines_video(proto_file, weights_file, threshold, BODY_PARTS, POSE_PAIRS):
# 네트워크 불러오기
net = cv2.dnn.readNetFromCaffe(proto_file, weights_file)
# GPU 사용
# net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)
# net.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA)
capture = cv2.VideoCapture(0)
#capture = cv2.VideoCapture('http://192.168.200.166:4747/mjpegfeed?640x480')
#capture = cv2.VideoCapture('/home/kimdoyoung/Downloads/test3.mp4')
stime = time.time()
while True :
wait_img = np.zeros((80, 240, 3), np.uint8)
cv2.putText(wait_img, str(round(time.time() - stime, 1)), (100, 50), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2)
wait_img = np.reshape(wait_img, (80, 240, 3))
cv2.imshow("wait_img", wait_img)
if cv2.waitKey(10) == 27:
break
if time.time() - stime > 3.0:
cv2.destroyWindow('wait_img')
break
while(True):
points.clear()
ret, frame_boy = capture.read()
#cv2.imshow("test", frame_boy)
if cv2.waitKey(1) == ord('q'):
break
#frame_boy = cv2.resize(frame_boy,dsize = (0,0), fx = 0.5, fy = 0.5)
frame_boy = cv2.rotate(frame_boy, cv2.ROTATE_90_COUNTERCLOCKWISE)
frame_boy = cv2.resize(frame_boy, (480, 640))
template = frame_boy.copy()
now_frame_boy = capture.get(cv2.CAP_PROP_POS_FRAMES)
total_frame_boy = capture.get(cv2.CAP_PROP_FRAME_COUNT)
frame_boy = output_keypoints(frame=frame_boy, net=net, threshold=threshold,
BODY_PARTS=BODY_PARTS, now_frame=now_frame_boy, total_frame=total_frame_boy)
frame_boy = output_keypoints_with_lines(frame=frame_boy, POSE_PAIRS=POSE_PAIRS)
if points[5] and points[2] and points[11] and points[8] and points[4] and points[7] is not None :
print(points[5])
print(points[2])
print(points[11])
print(points[8])
print(points[4])
print(points[7])
standard = int((points[4][1] + points[7][1]) / 2)
if(points[5][0] > points[2][0]):
points[5], points[2] = points[2], points[5]
if(points[11][0] > points[8][0]):
points[11], points[8] = points[8], points[11]
#template = template[min(points[5][1], points[2][1]): max(points[11][1], points[8][1]), points[5][0]: points[2][0]].copy()
template = template[points[1][1]: max(points[11][1], points[8][1]), points[5][0]: points[2][0]].copy()
cv2.destroyAllWindows()
break
else :
wait_img = np.zeros((80, 240, 3), np.uint8)
cv2.putText(wait_img, "please wait", (30, 50), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2)
wait_img = np.reshape(wait_img, (80, 240, 3))
cv2.imshow("wait_img", wait_img)
if cv2.waitKey(10) == 27:
break
print("None")
#y_gradient = int((max(points[11][1], points[8][1]) - min(points[5][1], points[2][1])) / 2)
y_gradient = int((max(points[11][1], points[8][1]) - points[1][1])/2)
x_gradient = int((points[2][0] - points[5][0]) / 2)
BGR_template = list(range(0, 3))
BGR_frame_boy = list(range(0, 3))
res = list(range(0, 3))
BGR_template = cv2.split(template)
'''
cv2.imshow("test", frame_boy)
cv2.waitKey()
cv2.imshow("test2", template)
cv2.waitKey()
'''
maxloc = points[14]
maxloc_t = list(range(0, 3))
cnt = 0
flag = False
end_flag = True
stime = time.time()
playsound("./Github/HCI/Wistle_long.mp3", block=False)
while True :
if time.time() - stime > 1.5 :
break
stime = time.time()
while True:
ret, frame_boy = capture.read()
#frame_boy = cv2.resize(frame_boy,dsize = (0,0), fx = 0.5, fy = 0.5)
frame_boy = cv2.rotate(frame_boy, cv2.ROTATE_90_COUNTERCLOCKWISE)
frame_boy = cv2.resize(frame_boy, (480, 640))
test = frame_boy.copy()
#cv2.TM_CCORR_NORMED, cv2.TM_CCOEFF, cv2.TM_CCOEFF_NORMED
BGR_frame_boy = cv2.split(frame_boy)
for i in range(0, 3):
res[i] = cv2.matchTemplate(
BGR_frame_boy[i], BGR_template[i], cv2.TM_CCOEFF_NORMED)
_, _, _, maxloc_t[0] = cv2.minMaxLoc(res[0])
_, _, _, maxloc_t[1] = cv2.minMaxLoc(res[1])
_, _, _, maxloc_t[2] = cv2.minMaxLoc(res[2])
tmax = -3.0
n = maxloc[0]
m = maxloc[1]
'''
cv2.rectangle(frame_boy, (n-5, m - 5),
(n + 5, m + 5), (255, 255, 255), 2)
'''
range_gradient = 40
res_y_max = len(res[0])
res_x_max = len(res[0][0])
for i in range(max(n - x_gradient - int(range_gradient/2), 0), min(n - x_gradient + int(range_gradient/2), res_x_max)):
for j in range(max(0, m - y_gradient - range_gradient, 0), min(m - y_gradient + range_gradient, res_y_max)):
temp = res[0][j][i] + res[1][j][i] + res[2][j][i]
if tmax < temp:
tmax = temp
maxloc = (i + x_gradient, j + y_gradient)
cv2.rectangle(frame_boy, (maxloc[0] - x_gradient, maxloc[1] - y_gradient),
(maxloc[0] + x_gradient, maxloc[1] + y_gradient), (255, 255, 255), 2)
print(str(maxloc[1] - y_gradient) + " " + str(standard))
if flag == False and maxloc[1] - y_gradient < (standard + 30) - 15 and time.time() - stime < 30.0:
cnt = cnt + 1
playsound("./Github/HCI/beef.mp3", block=False)
print("cnt : " + str(cnt))
flag = True
elif flag == True and maxloc[1] - y_gradient > (standard + 30) + 15 and time.time() - stime < 30.0:
flag = False
'''
cv2.rectangle(frame_boy, maxloc_t[0], (maxloc_t[0][0] + x_gradient *
2, maxloc_t[0][1] + y_gradient * 2), (255, 0, 0), 2)
cv2.rectangle(frame_boy, maxloc_t[1], (maxloc_t[1][0] + x_gradient *
2, maxloc_t[1][1] + y_gradient * 2), (0, 255, 0), 2)
cv2.rectangle(frame_boy, maxloc_t[2], (maxloc_t[2][0] + x_gradient *
2, maxloc_t[2][1] + y_gradient * 2), (0, 0, 255), 2)
'''
#cv2.imshow("frame_boy", frame_boy)
cnt_img = np.zeros((80, 240, 3), np.uint8)
time_img = np.zeros((80, 240, 3), np.uint8)
if time.time() - stime > 30.0:
frame_boy = test.copy()
cv2.putText(cnt_img, str(cnt), (100, 50),
cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2)
cv2.putText(time_img, str(round(30.0, 1
)), (100, 50), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2)
else :
cv2.putText(cnt_img, str(cnt), (100, 50),
cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2)
cv2.putText(time_img, str(round(time.time() - stime, 1
)), (100, 50), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2)
cnt_img = np.reshape(cnt_img, (80, 240, 3))
time_img = np.reshape(time_img, (80, 240, 3))
cnt_time_img = cv2.hconcat([cnt_img, time_img])
frame_cnt_time_img = cv2.vconcat([frame_boy, cnt_time_img])
cv2.imshow("frame_cnt_time_img", frame_cnt_time_img)
if cv2.waitKey(10) == 27:
break
if time.time() - stime > 30.0 and end_flag:
end_flag = False
playsound("./Github/HCI/Wistle_long.mp3", block=False)
capture.release()
cv2.destroyAllWindows()
if __name__ == "__main__" :
BODY_PARTS_MPI = {0: "Head", 1: "Neck", 2: "RShoulder", 3: "RElbow", 4: "RWrist",
5: "LShoulder", 6: "LElbow", 7: "LWrist", 8: "RHip", 9: "RKnee",
10: "RAnkle", 11: "LHip", 12: "LKnee", 13: "LAnkle", 14: "Chest",
15: "Background"}
POSE_PAIRS_MPI = [[0, 1], [1, 2], [1, 5], [1, 14], [2, 3], [3, 4], [5, 6],
[6, 7], [8, 9], [9, 10], [11, 12], [12, 13], [14, 8], [14, 11]]
# 신경 네트워크의 구조를 지정하는 prototxt 파일 (다양한 계층이 배열되는 방법 등)
protoFile_mpi = "./Github/HCI/models/pose/mpi/pose_deploy_linevec.prototxt"
# 훈련된 모델의 weight 를 저장하는 caffemodel 파일
weightsFile_mpi = "./Github/HCI/models/pose/mpi/pose_iter_160000.caffemodel"
# 키포인트를 저장할 빈 리스트
points = []
output_keypoints_with_lines_video(proto_file=protoFile_mpi, weights_file=weightsFile_mpi,
threshold=0.1, BODY_PARTS=BODY_PARTS_MPI, POSE_PAIRS=POSE_PAIRS_MPI)