Attempted implementation of the following paper: " GOLFPOSE:GOLFSWINGANALYSESWITHAMONOCULARCAMERABASEDHUMAN POSEESTIMATION Zhongyu Jiang1⋆, Haorui Ji2⋆, Samuel Menaker2 and Jenq-Neng Hwang1 1Dept. Electrical & Computer Engineering , University of Washington 2SPORTSBOX.AI INC. zyjiang@uw.edu, haoruij@sportsbox.ai, samm@sportsbox.ai, hwang@uw.edu ABSTRACT With the rapid developments of computer vision and deep learning technologies, artificial intelligence takes a more and more important role in sports analyses. In this paper, to at tain the objective of automated golf swing analyses, we pro pose a lightweight temporal-based 2D human pose estimation (HPE) method, called GolfPose, which achieves improved performance than the state-of-the-art image-based HPE meth ods. Unlike traditional image-based methods, our temporal based method, designed for efficient and effective golf swing analyses, takes advantage of the temporal information to im prove the estimation accuracy of fast-moving and partially self-occluded keypoints. Furthermore, in order to make sure the golf swing analyses can run on mobile devices, we op timize the model architecture to achieve real-time inference. With around 10% of the parameters and half of the GFLOPs used in the state-of-the-art HRNet, our proposed GolfPose model can achieve 9.16 mean pixel error (MPE) in our golf swing dataset, compared with 9.20 MPE for HRNet. Further more, the proposed temporal-based method, facilitated with golf club detection(GCD), significantly improves the accu racy of keypoints on the golf club from 13.98 to 9.21 MPE. Index Terms— SportsAnalysis, HumanPoseEstimation, Golf Swing, Line Segment Detection"

1. 1. import math
2. import cv2
3. import numpy as np
4. from time import time
5. import mediapipe as mp
6. import matplotlib.pyplot as plt
7. from mpl_toolkits.mplot3d import Axes3D
8. import depthai as dai
9.
10. import keyboard #load keyboard package
11. # Initializing mediapipe pose class.
12. mp_pose = mp.solutions.pose
13. mp_face_detection = mp.solutions.face_detection
14. mp_drawing_styles = mp.solutions.drawing_styles
15. mp_face_mesh = mp.solutions.face_mesh
16. mp_hands = mp.solutions.hands
17. # Initializing mediapipe drawing class, useful for annotation.
18. mp_drawing = mp.solutions.drawing_utils
19. mp_holistic = mp.solutions.holistic
20.
21. ## Initializing mp_pose for Pose capture
22. ######################################################################
#################
23. drawing_spec = mp_drawing.DrawingSpec(thickness=1, circle_radius=1)
24. # Setup Pose function for video.
25. pose_video = mp_pose.Pose(static_image_mode=False,
min_detection_confidence=0.5,min_tracking_confidence = 0.5 ,model_complexity=2,
smooth_landmarks =True)
26. holistic = mp_holistic.Holistic(min_detection_confidence=0.6,
min_tracking_confidence=0.5)
27. hands = mp_hands.Hands(static_image_mode=True, max_num_hands=2,
min_detection_confidence=0.6, min_tracking_confidence = 0.7)
28.
29. # # Setting up the Pose function.
30. # pose = mp_pose.Pose(static_image_mode=True, min_detection_confidence=0.7,
model_complexity=2)
31. ##################################################################
32. ##################################################################
33. ##################################################################
34. def detectPose(image, pose, display=True):
35. '''
36. This function performs pose detection on an image.
37. Args:
38. image: The input image with a prominent person whose pose landmarks needs to
be detected.
39. pose: The pose setup function required to perform the pose detection.
40. display: A boolean value that is if set to true the function displays the original input
image, the resultant image,
41. and the pose landmarks in 3D plot and returns nothing.
42. Returns:
43. output_image: The input image with the detected pose landmarks drawn.

2. 44. landmarks: A list of detected landmarks converted into their original scale.
45. '''
46.
47. # Create a copy of the input image.
48. output_image = image.copy()
49.
50. # Convert the image from BGR into RGB format.
51. # imageRGB = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
52.
53. # Perform the Pose Detection.
54. results = pose.process(image)
55.
56. # Retrieve the height and width of the input image.
57. height, width, _ = image.shape
58. # Initialize a list to store the detected landmarks.
59. landmarks = []
60. landmarks_world = []
61. # print(height, width)
62. # Check if any landmarks are detected.
63. if results.pose_landmarks:
64. # Draw Pose landmarks on the output image.
65. mp_drawing.draw_landmarks(image=output_image,
landmark_list=results.pose_landmarks,
66. connections=mp_pose.POSE_CONNECTIONS)
67.
68. mp_drawing.draw_landmarks(image=output_image,
landmark_list=results.pose_landmarks,
69. connections=mp_pose.POSE_CONNECTIONS)
70. # Iterate over the detected landmarks.
71. for landmark in results.pose_landmarks.landmark:
72. # # Append the landmark into the list.
73. landmarks.append((int(landmark.x * width), int(landmark.y * height),
74. (landmark.z * width)))
75.
76. for landmark in results.pose_world_landmarks.landmark:
77. # # Append the landmark into the list.
78. landmarks_world.append((int(landmark.x * width), int(landmark.y * height),
79. (landmark.z * width)))
80.
81. # Check if the original input image and the resultant image are specified to be
displayed.
82. if display:
83. # # Display the original input image and the resultant image.
84. # plt.figure(figsize=[22,22])
85. # plt.subplot(121);plt.imshow(image[:,:,::-1]);plt.title("Original Image");plt.axis('off');
86. # plt.subplot(122);plt.imshow(output_image[:,:,::-1]);plt.title("Output
Image");plt.axis('off');
87. # Also Plot the Pose landmarks in 3D.
88. mp_drawing.plot_landmarks(results.pose_landmarks,
mp_pose.POSE_CONNECTIONS)
89. return output_image, landmarks, landmarks_world

3. 90.
91. # Otherwise
92. else:
93.
94. # Return the output image and the found landmarks.
95. return output_image, landmarks, landmarks_world
96. def calculateAngle(landmark1, landmark2, landmark3):
97.
98. '''
99. This function calculates angle between three different landmarks.
100. Args:
101. landmark1: The first landmark containing the x,y and z coordinates.
102. landmark2: The second landmark containing the x,y and z coordinates.
103. landmark3: The third landmark containing the x,y and z coordinates.
104. Returns:
105. angle: The calculated angle between the three landmarks.
106.
107. '''
108.
109. # Get the required landmarks coordinates.
110. x1, y1, _ = landmark1
111. x2, y2, _ = landmark2
112. x3, y3, _ = landmark3
113.
114. # Calculate the angle between the three points
115. angle = math.degrees(math.atan2(y3 - y2, x3 - x2) - math.atan2(y1 - y2, x1 - x2))
116.
117. # Check if the angle is less than zero.
118. if angle < 0:
119. # Add 360 to the found angle.
120. angle += 360
121.
122. # Return the calculated angle.
123. return angle
124. def classifyPose_Golfswing_RIGHT_SIDE_view(landmarks_0, output_image_0,
landmarks_1, output_image_1, display=False):
125. '''
126. This function classifies poses depending upon the angles of various body joints.
127. Args:
128. landmarks: A list of detected landmarks of the person whose pose needs to be
classified.
129. output_image: A image of the person with the detected pose landmarks drawn.
130. display: A boolean value that is if set to true the function displays the resultant
image with the pose label
131. written on it and returns nothing.
132. Returns:
133. output_image: The image with the detected pose landmarks drawn and pose
label written.
134. label: The classified pose label of the person in the output_image.
135.
136. '''

4. 137. # Initialize the label of the pose. It is not known at this stage.
138. label = '.'
139. label_1 = '.'
140.
141.
142. # Specify the color (Red) with which the label will be written on the image.
143. color = (0, 0, 255)
144.
145. # Calculate the required angles for the ROMS we are interested in.
146.
147.
#----------------------------------------------------------------------------------------------------------------
148.
149. # Get the angle between the left shoulder, elbow and wrist points.
150. left_elbow_angle =
calculateAngle(landmarks_0[mp_pose.PoseLandmark.LEFT_SHOULDER.value],
151. landmarks_0[mp_pose.PoseLandmark.LEFT_ELBOW.value],
152. landmarks_0[mp_pose.PoseLandmark.LEFT_WRIST.value])
153.
154. # Get the angle between the right shoulder, elbow and wrist points.
155. right_elbow_angle =
calculateAngle(landmarks_0[mp_pose.PoseLandmark.RIGHT_SHOULDER.value],
156.
landmarks_0[mp_pose.PoseLandmark.RIGHT_ELBOW.value],
157. landmarks_0[mp_pose.PoseLandmark.RIGHT_WRIST.value])
158.
159. # Get the angle between the left elbow, shoulder and hip points.
160. left_shoulder_angle =
calculateAngle(landmarks_0[mp_pose.PoseLandmark.LEFT_ELBOW.value],
161.
landmarks_0[mp_pose.PoseLandmark.LEFT_SHOULDER.value],
162. landmarks_0[mp_pose.PoseLandmark.LEFT_HIP.value])
163.
164. # Get the angle between the right hip, shoulder and elbow points.
165. right_shoulder_angle =
calculateAngle(landmarks_0[mp_pose.PoseLandmark.RIGHT_HIP.value],
166.
landmarks_0[mp_pose.PoseLandmark.RIGHT_SHOULDER.value],
167.
landmarks_0[mp_pose.PoseLandmark.RIGHT_ELBOW.value])
168.
169. # Get the angle between the left hip, knee and ankle points.
170. left_knee_angle =
calculateAngle(landmarks_0[mp_pose.PoseLandmark.LEFT_HIP.value],
171. landmarks_0[mp_pose.PoseLandmark.LEFT_KNEE.value],
172. landmarks_0[mp_pose.PoseLandmark.LEFT_ANKLE.value])
173.
174. # Get the angle between the right hip, knee and ankle points
175. right_knee_angle =
calculateAngle(landmarks_0[mp_pose.PoseLandmark.RIGHT_HIP.value],
176. landmarks_0[mp_pose.PoseLandmark.RIGHT_KNEE.value],

5. 177. landmarks_0[mp_pose.PoseLandmark.RIGHT_ANKLE.value])
178.
179. ##
180. ##
181. ##
182. # Get the angle between the right hip, knee and ankle points
183. right_bending_angle =
calculateAngle(landmarks_0[mp_pose.PoseLandmark.RIGHT_SHOULDER.value],
184. landmarks_0[mp_pose.PoseLandmark.RIGHT_HIP.value],
185. landmarks_0[mp_pose.PoseLandmark.RIGHT_KNEE.value])
186. # Get the angle between the right hip, knee and ankle points
187. left_bending_angle =
calculateAngle(landmarks_0[mp_pose.PoseLandmark.LEFT_SHOULDER.value],
188. landmarks_0[mp_pose.PoseLandmark.LEFT_HIP.value],
189. landmarks_0[mp_pose.PoseLandmark.LEFT_KNEE.value])
190.
191.
192.
193. ##
194. x1,y1,z1 = landmarks_0[mp_pose.PoseLandmark.LEFT_ANKLE.value]
195. x2,y2,z2 = landmarks_0[mp_pose.PoseLandmark.RIGHT_ANKLE.value]
196. mid_GROUND_x,mid_GROUND_y,mid_GROUND_z = (x1+x2)/2 , (y1+y2)/2,
(z1+z2)/2
197.
198. x3,y3,z3 = landmarks_0[mp_pose.PoseLandmark.LEFT_HIP.value]
199. x4,y4,z4 = landmarks_0[mp_pose.PoseLandmark.RIGHT_HIP.value]
200. mid_HIP_x,mid_HIP_y,mid_HIP_z = (x3+x4)/2 , (y3+y4)/2, (z3+z4)/2
201.
202. GROUND_HIP_NOSE_angle =
calculateAngle((mid_GROUND_x,mid_GROUND_y,mid_GROUND_z),
203. (mid_HIP_x,mid_HIP_y,mid_HIP_z),
204. landmarks_0[mp_pose.PoseLandmark.NOSE.value])
205.
206. x5,y5,z5 = landmarks_0[mp_pose.PoseLandmark.LEFT_SHOULDER.value]
207. x6,y6,z6 = landmarks_0[mp_pose.PoseLandmark.RIGHT_SHOULDER.value]
208. mid_SHOULDER_x,mid_SHOULDER_y,mid_SHOULDER_z = (x5+x6)/2 ,
(y5+y6)/2, (z5+z6)/2
209.
210. dist_between_shoulders = round(math.sqrt((int(x5)-int(x6))**2 +
(int(y5)-int(y6))**2))
211.
212. x7,y7,z7 = landmarks_0[mp_pose.PoseLandmark.NOSE.value]
213. lenght_of_body = round(math.sqrt((int(x7)-int(mid_GROUND_x))**2 +
(int(y7)-int(mid_GROUND_y))**2))
214.
215.
216. x8,y8,z8 = landmarks_0[mp_pose.PoseLandmark.LEFT_PINKY.value]
217. x9,y9,z9 = landmarks_0[mp_pose.PoseLandmark.RIGHT_PINKY.value]
218.
219. x10,y10,z10 = landmarks_0[mp_pose.PoseLandmark.NOSE.value]
220.

6. 221. x11,y11,z11 = landmarks_0[mp_pose.PoseLandmark.LEFT_KNEE.value]
222. x12,y12,z12 = landmarks_0[mp_pose.PoseLandmark.RIGHT_KNEE.value]
223.
224. ## Drawing shoulder vertical lines
225. cv2.line(output_image_0, (x1,y1), (x1,y1-300), [0,0,255], thickness = 2, lineType =
cv2.LINE_8, shift = 0)
226. cv2.line(output_image_0, (x2,y2), (x2,y2-300), [0,0,255], thickness = 2, lineType =
cv2.LINE_8, shift = 0)
227.
228. cv2.circle(output_image_0,
(round(mid_SHOULDER_x),round(mid_SHOULDER_y)), round(lenght_of_body/2), color
= [128,0,0], thickness =2)
229.
230. cv2.line(output_image_0, (x3,y3), (x4,y4), [0,255,255], thickness = 4, lineType =
cv2.LINE_8, shift = 0)
231. cv2.line(output_image_0, (x5,y5), (x6,y6), [0,255,255], thickness = 4, lineType =
cv2.LINE_8, shift = 0)
232.
233. cv2.line(output_image_0,
(round(mid_SHOULDER_x),round(mid_SHOULDER_y)),
(round(mid_HIP_x),round(mid_HIP_y)), [0,255,255], thickness = 4, lineType =
cv2.LINE_8, shift = 0)
234. cv2.line(output_image_0, (x11,y11), (x12,y12), [0,255,255], thickness = 4, lineType
= cv2.LINE_8, shift = 0)
235.
236. cv2.line(output_image_0, (round(mid_HIP_x),round(mid_HIP_y)),
(round(mid_HIP_x),round(mid_HIP_y-300)), [0,0,255], thickness = 2, lineType =
cv2.LINE_8, shift = 0)
237.
238. midHip_midShoulder_angle_fromVertical =
calculateAngle((mid_HIP_x,mid_HIP_y-300,mid_HIP_z),
239. (mid_HIP_x,mid_HIP_y,mid_HIP_z),
240.
(mid_SHOULDER_x,mid_SHOULDER_y,mid_SHOULDER_z))
241.
242. if midHip_midShoulder_angle_fromVertical > 180:
243. midHip_midShoulder_angle_fromVertical = 360
-midHip_midShoulder_angle_fromVertical
244. cv2.putText(output_image_0, "Core Angle:" +
str(midHip_midShoulder_angle_fromVertical),
(round(mid_HIP_x)+20,round(mid_HIP_y)), fontFace =
cv2.FONT_HERSHEY_SIMPLEX, fontScale = 1,color = [0,0,255], thickness = 2)
245.
246.
247. try:
248. # Check if person is in ADDRESS stage
249. if x8/x9 >0/8 and x8/x9 < 1.2 and y8 > mid_SHOULDER_y and y9 >
mid_SHOULDER_y: #Checking if both hands are on club grip
250. if x8 > x2 and x8 < x1 and x9 >x2 and x9 < x1:
251. label = "ADDRESS pose established"
252. # Check if person HEAD has left the boundary of ankle vertical lines

7. 253. if x10 < x2 or x10 > x1:
254. cv2.circle(output_image_0, (x10,y10), 20, color = [0,0,255], thickness
=2)
255. # print("Keep Head posture within red line boundary")
256. cv2.putText(output_image_0, "Keep Head posture within red line
boundary", (x10+20,y10), fontFace = cv2.FONT_HERSHEY_SIMPLEX, fontScale =
1,color = [0,0,255], thickness = 2)
257.
258. ## While in ADDRESS pose, we want to check the pose correction from
back view as well
259. output_image_1, label_1 =
classifyPose_Golfswing_BACK_SIDE_view(landmarks_1, output_image_1,
display=False)
260.
261. except:
262. pass
263.
264. # Check if person is in TAKE BACK stage
265. try:
266. if x8/x9 >0/8 and x8/x9 < 1.2: #Checking if both hands are on club grip
267. if x8 < x2 and x9 < x2:
268. label = "TAKE BACK pose in process"
269. if x10 < x2 or x10 > x1:
270. cv2.circle(output_image_0, (x10,y10), 20, color = [0,0,255], thickness
=2)
271. print("Keep Head posture within red line boundary")
272. cv2.putText(output_image_0, "Keep Head posture within red line
boundary", (x10+20,y10), fontFace = cv2.FONT_HERSHEY_SIMPLEX, fontScale =
1,color = [0,0,255], thickness = 2)
273. except:
274. pass
275.
276. # Check if person has reached BACKSWING TOP
277. try:
278. if x8/x9 >0/8 and x8/x9 < 1.2 and y8 < mid_SHOULDER_y and y9 <
mid_SHOULDER_y: #Checking if both hands are on club grip
279. if x8 < x1 and x9 < x1: #and x8 > x2 and x9 >x2 #It is not neccasary that the
right and left hands be in the Red Lines defined by Left and Right ankle. It is important
the Left and Right hand be left of the Left ankle and hgiher than the shoulder mid point
280. label = "BACKSWING TOP reached. Ready to launch swing"
281. if x10 < x2 or x10 > x1:
282. cv2.circle(output_image_0, (x10,y10), 20, color = [0,0,255], thickness
=2)
283. print("Keep Head posture within red line boundary")
284. cv2.putText(output_image_0, "Keep Head posture within red line
boundary", (x10+20,y10), fontFace = cv2.FONT_HERSHEY_SIMPLEX, fontScale =
1,color = [0,0,255], thickness = 2)
285.
286. except:
287. pass
288.

8. 289. # Check if person is in FOLLOW THROUGH stage
290. try:
291. if x8/x9 >0/8 and x8/x9 < 1.2 and y8 > mid_SHOULDER_y and y9 >
mid_SHOULDER_y: #Checking if both hands are on club grip
292. if x8 > x1 and x9 > x1:
293. label = "FOLLOW THROUGH stage"
294. except:
295. pass
296.
297. try:
298. if x8/x9 >0/8 and x8/x9 < 1.2 and y8 < mid_SHOULDER_y and y9 <
mid_SHOULDER_y: #Checking if both hands are on club grip
299. if x8 > x1 and x9 > x1:
300. label = "FINISH SWING"
301. except:
302. pass
303.
304. #
#----------------------------------------------------------------------------------------------------------------
305.
306. # Check if the pose is classified successfully
307. if label != 'Unknown Pose':
308. # Update the color (to green) with which the label will be written on the image.
309. color = (0, 255, 0)
310. # Write the label on the output image.
311. cv2.putText(output_image_0, label, (10, 30),cv2.FONT_HERSHEY_PLAIN, 2,
color, 2)
312.
313.
314. ###########################################
315. # Check if the resultant image is specified to be displayed.
316. if display:
317. # Display the resultant image.
318. plt.figure(figsize=[10,10])
319. plt.imshow(output_image_0[:,:,::-1]);plt.title("Output Image");plt.axis('off');
320.
321. else:
322. ######################################333
323. # Return the output image and the classified label.
324. return output_image_0, label, output_image_1, label_1
325.
326. def classifyPose_Golfswing_BACK_SIDE_view(landmarks, output_image,
display=False):
327. '''
328. This function classifies poses depending upon the angles of various body joints.
329. Args:
330. landmarks: A list of detected landmarks of the person whose pose needs to be
classified.
331. output_image: A image of the person with the detected pose landmarks drawn.
332. display: A boolean value that is if set to true the function displays the resultant
image with the pose label

9. 333. written on it and returns nothing.
334. Returns:
335. output_image: The image with the detected pose landmarks drawn and pose
label written.
336. label: The classified pose label of the person in the output_image.
337.
338. '''
339. # Initialize the label of the pose. It is not known at this stage.
340. label = '.'
341.
342. # Specify the color (Red) with which the label will be written on the image.
343. color = (0, 0, 255)
344.
345. # Calculate the required angles for the ROMS we are interested in.
346.
347.
#----------------------------------------------------------------------------------------------------------------
348.
349. # Get the angle between the left shoulder, elbow and wrist points.
350. left_elbow_angle =
calculateAngle(landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value],
351. landmarks[mp_pose.PoseLandmark.LEFT_ELBOW.value],
352. landmarks[mp_pose.PoseLandmark.LEFT_WRIST.value])
353.
354. # Get the angle between the right shoulder, elbow and wrist points.
355. right_elbow_angle =
calculateAngle(landmarks[mp_pose.PoseLandmark.RIGHT_SHOULDER.value],
356. landmarks[mp_pose.PoseLandmark.RIGHT_ELBOW.value],
357. landmarks[mp_pose.PoseLandmark.RIGHT_WRIST.value])
358.
359. # Get the angle between the left elbow, shoulder and hip points.
360. left_shoulder_angle =
calculateAngle(landmarks[mp_pose.PoseLandmark.LEFT_ELBOW.value],
361.
landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value],
362. landmarks[mp_pose.PoseLandmark.LEFT_HIP.value])
363.
364. # Get the angle between the right hip, shoulder and elbow points.
365. right_shoulder_angle =
calculateAngle(landmarks[mp_pose.PoseLandmark.RIGHT_HIP.value],
366.
landmarks[mp_pose.PoseLandmark.RIGHT_SHOULDER.value],
367.
landmarks[mp_pose.PoseLandmark.RIGHT_ELBOW.value])
368.
369. # Get the angle between the left hip, knee and ankle points.
370. left_knee_angle =
calculateAngle(landmarks[mp_pose.PoseLandmark.LEFT_HIP.value],
371. landmarks[mp_pose.PoseLandmark.LEFT_KNEE.value],
372. landmarks[mp_pose.PoseLandmark.LEFT_ANKLE.value])
373.

10. 374. # Get the angle between the right hip, knee and ankle points
375. right_knee_angle =
calculateAngle(landmarks[mp_pose.PoseLandmark.RIGHT_HIP.value],
376. landmarks[mp_pose.PoseLandmark.RIGHT_KNEE.value],
377. landmarks[mp_pose.PoseLandmark.RIGHT_ANKLE.value])
378.
379. ##
380. ##
381. ##
382. # Get the angle between the right hip, knee and ankle points
383. right_bending_angle =
calculateAngle(landmarks[mp_pose.PoseLandmark.RIGHT_SHOULDER.value],
384. landmarks[mp_pose.PoseLandmark.RIGHT_HIP.value],
385. landmarks[mp_pose.PoseLandmark.RIGHT_KNEE.value])
386.
387. # Get the angle between the right hip, knee and ankle points
388. left_bending_angle =
calculateAngle(landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value],
389. landmarks[mp_pose.PoseLandmark.LEFT_HIP.value],
390. landmarks[mp_pose.PoseLandmark.LEFT_KNEE.value])
391. #########################################
392. ## Bringing angles to within 0 - 180 degree range
393.
394. if right_knee_angle > 180:
395. right_knee_angle = 360 -right_knee_angle
396.
397. if right_bending_angle > 180:
398. right_bending_angle = 360 -right_bending_angle
399. ###############################3
400. ##
401. x1,y1,z1 = landmarks[mp_pose.PoseLandmark.LEFT_ANKLE.value]
402. x2,y2,z2 = landmarks[mp_pose.PoseLandmark.RIGHT_ANKLE.value]
403. mid_GROUND_x,mid_GROUND_y,mid_GROUND_z = (x1+x2)/2 , (y1+y2)/2,
(z1+z2)/2
404.
405. x3,y3,z3 = landmarks[mp_pose.PoseLandmark.LEFT_HIP.value]
406. x4,y4,z4 = landmarks[mp_pose.PoseLandmark.RIGHT_HIP.value]
407. mid_HIP_x,mid_HIP_y,mid_HIP_z = (x3+x4)/2 , (y3+y4)/2, (z3+z4)/2
408.
409. GROUND_HIP_NOSE_angle =
calculateAngle((mid_GROUND_x,mid_GROUND_y,mid_GROUND_z),
410. (mid_HIP_x,mid_HIP_y,mid_HIP_z),
411. landmarks[mp_pose.PoseLandmark.NOSE.value])
412.
413. x5,y5,z5 = landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value]
414. x6,y6,z6 = landmarks[mp_pose.PoseLandmark.RIGHT_SHOULDER.value]
415. mid_SHOULDER_x,mid_SHOULDER_y,mid_SHOULDER_z = (x5+x6)/2 ,
(y5+y6)/2, (z5+z6)/2
416.
417. dist_between_shoulders = round(math.sqrt((int(x5)-int(x6))**2 +
(int(y5)-int(y6))**2))

11. 418.
419. x7,y7,z7 = landmarks[mp_pose.PoseLandmark.NOSE.value]
420. lenght_of_body = round(math.sqrt((int(x7)-int(mid_GROUND_x))**2 +
(int(y7)-int(mid_GROUND_y))**2))
421.
422.
423. x8,y8,z8 = landmarks[mp_pose.PoseLandmark.LEFT_PINKY.value]
424. x9,y9,z9 = landmarks[mp_pose.PoseLandmark.RIGHT_PINKY.value]
425.
426. x10,y10,z10 = landmarks[mp_pose.PoseLandmark.NOSE.value]
427.
428. x11,y11,z11 = landmarks[mp_pose.PoseLandmark.LEFT_KNEE.value]
429. x12,y12,z12 = landmarks[mp_pose.PoseLandmark.RIGHT_KNEE.value]
430.
431.
432. # # Check if leading (RIGHT) leg is straight
433.
434.
#----------------------------------------------------------------------------------------------------------------
435.
436. cv2.line(output_image, (x12,y12), (x12,y12-300), [0,255,255], thickness = 4,
lineType = cv2.LINE_8, shift = 0)
437.
438. # Check if one leg is straight
439. if right_knee_angle > 165 and right_knee_angle < 179:
440. # Specify the label of the pose that is tree pose.
441. cv2.putText(output_image, "Leadside Knee Flexion angle" + str(180
-right_knee_angle), (x12+20,y12), fontFace = cv2.FONT_HERSHEY_SIMPLEX,
fontScale = 1,color = [0,0,255], thickness = 2)
442. label = '1. Bend your knees more'
443. else:
444. cv2.putText(output_image, "Leadside Knee Flexion angle" + str(180
-right_knee_angle), (x12+20,y12), fontFace = cv2.FONT_HERSHEY_SIMPLEX,
fontScale = 1,color = [0,0,255], thickness = 2)
445. label = 'Knees flexion posture CORRECT!'
446.
447. # Check if one leg is straight
448. if right_bending_angle > 165 and right_bending_angle < 179:
449. # Specify the label of the pose that is tree pose.
450. cv2.putText(output_image, "Leadside Spine Flexion angle" + str(180
-right_knee_angle), (x4+20,y4), fontFace = cv2.FONT_HERSHEY_SIMPLEX, fontScale
= 1,color = [0,0,255], thickness = 2)
451. label = '2. Bend your Spine- more'
452. else:
453. cv2.putText(output_image, "Leadside Spine Flexion angle" + str(180
-right_knee_angle), (x4+20,y4), fontFace = cv2.FONT_HERSHEY_SIMPLEX, fontScale
= 1,color = [0,0,255], thickness = 2)
454. label = 'Spine Flexion angle posture CORRECT!'
455.
456.

12. 457. #
#----------------------------------------------------------------------------------------------------------------
458.
459. # Check if the pose is classified successfully
460. if label != 'Unknown Pose':
461. # Update the color (to green) with which the label will be written on the image.
462. color = (0, 255, 0)
463. # Write the label on the output image.
464. cv2.putText(output_image, label, (10, 30),cv2.FONT_HERSHEY_PLAIN, 2, color,
2)
465.
466. # Check if the resultant image is specified to be displayed.
467. if display:
468. # Display the resultant image.
469. plt.figure(figsize=[10,10])
470. plt.imshow(output_image[:,:,::-1]);plt.title("Output Image");plt.axis('off');
471.
472. else:
473. # Return the output image and the classified label.
474. return output_image, label
475.
476. ###################################################################
####################
477. ###################################################################
####################
478. # Initialize the VideoCapture object to read video recived from PT session
479. #############
480. ##############################
481. camera_video_0 = cv2.VideoCapture(1)
482. camera_video_1 = cv2.VideoCapture(0)
483. #
484. ##
485. # cap = camera_video
486. # camera_video_0.set(3,1280)
487. # camera_video_0.set(4,960)
488. # camera_video_1.set(3,1280)
489. # camera_video_1.set(4,960)
490.
491. print( "Frame capture Initialized from RIGHT side and BACK side video camera")
492. # print("Select the camera footage you are interested in applying CV Models on:
'1'for RIGHT SIDE VIEW, '2' for LEFT SIDE VIEW")
493. # #listening to input
494. # cam_input = keyboard.read_key()
495. ##
496. # cam_input = 2
497. ##
498.
499.
500. with
mp_face_mesh.FaceMesh(max_num_faces=1,refine_landmarks=True,min_detection_co
nfidence=0.5,min_tracking_confidence=0.5) as face_mesh:

13. 501. # Iterate until the webcam is accessed successfrully.
502. while camera_video_0.isOpened() and camera_video_1.isOpened():
503. # Read a frame.
504. ok, frame_0 = camera_video_0.read()
505. ok, frame_1 = camera_video_1.read()
506. # Check if frame is not read properly.
507. if not ok:
508. continue
509. frame_height, frame_width, _ = frame_0.shape
510. # Resize the frame while keeping the aspect ratio.
511. frame_0 = cv2.resize(frame_0, (int(frame_width * (640 / frame_height)), 640))
512. frame_1 = cv2.resize(frame_1, (int(frame_width * (640 / frame_height)), 640))
513. frame_final_0 = frame_0
514. frame_final_1 = frame_1
515.
516. # # Perform Pose landmark detection.
517. # Check if frame is not read properly.
518. if not ok:
519. # Continue to the next iteration to read the next frame and ignore the empty
camera frame.
520. continue
521. #################################################
522. #################################################
523.
524. # if cam_input=='1':
525. frame_0, landmarks_0, landmarks_world = detectPose(frame_0, pose_video,
display=False)
526. frame_1, landmarks_1, landmarks_world = detectPose(frame_1, pose_video,
display=False)
527.
528. if landmarks_0 and landmarks_1:
529. frame_final_0, label_0, frame_final_1, label_1 =
classifyPose_Golfswing_RIGHT_SIDE_view(landmarks_0, frame_0,landmarks_1,
frame_1, display=False)
530. else:
531. continue
532.
533.
534.
535. if cv2.waitKey(1) & 0xFF==ord('q'): ## EXTRACT THE LABEL OF THE ANGLE
MEASUREMENT AT A PARTICULAR FRAME
536. # breakw
537. print(label_0)
538. print(label_1)
539. #returns the value of the LABEL when q is pressed
540. ###################################################################
######################################
541.
542.
543. stream_final_img = cv2.hconcat([frame_final_0, frame_final_1])
544. cv2.imshow('Combined Video', stream_final_img)

14. 545.
546.
547. k = cv2.waitKey(1) & 0xFF
548. # Check if 'ESC' is pressed.
549. if(k == 27):
550. # Break the loop.
551. break
552. camera_video_0.release()
553. camera_video_1.release()
554.
555.