-
Notifications
You must be signed in to change notification settings - Fork 0
/
ampere_law.py
93 lines (72 loc) · 3.32 KB
/
ampere_law.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
import cv2
import mediapipe as mp
import numpy as np
THRESHOLD = 0.2 # 20%, 값이 클수록 손이 카메라와 가까워야 인식함
gesture = {
0:'fist', 1:'one', 2:'two', 3:'three', 4:'four', 5:'five',
6:'six', 7:'rock', 8:'spiderman', 9:'yeah', 10:'ok',
}
# MediaPipe hands model
mp_hands = mp.solutions.hands
mp_drawing = mp.solutions.drawing_utils
hands = mp_hands.Hands(
max_num_hands=1,
min_detection_confidence=0.5,
min_tracking_confidence=0.5)
# Gesture recognition model
file = np.genfromtxt('data/gesture_train.csv', delimiter=',')
angle = file[:,:-1].astype(np.float32)
label = file[:, -1].astype(np.float32)
knn = cv2.ml.KNearest_create()
knn.train(angle, cv2.ml.ROW_SAMPLE, label)
cap = cv2.VideoCapture(1)
while cap.isOpened():
ret, img = cap.read()
if not ret:
continue
img = cv2.flip(img, 1) # 이미지 좌우 반전
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
result = hands.process(img)
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
if result.multi_hand_landmarks is not None:
for res in result.multi_hand_landmarks:
joint = np.zeros((21, 3))
for j, lm in enumerate(res.landmark):
joint[j] = [lm.x, lm.y, lm.z]
# Compute angles between joints
v1 = joint[[0,1,2,3,0,5,6,7,0,9,10,11,0,13,14,15,0,17,18,19],:] # Parent joint
v2 = joint[[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20],:] # Child joint
v = v2 - v1 # [20,3]
# Normalize v
v = v / np.linalg.norm(v, axis=1)[:, np.newaxis]
# Get angle using arcos of dot product
angle = np.arccos(np.einsum('nt,nt->n',
v[[0,1,2,4,5,6,8,9,10,12,13,14,16,17,18],:],
v[[1,2,3,5,6,7,9,10,11,13,14,15,17,18,19],:])) # [15,]
angle = np.degrees(angle) # Convert radian to degree
# Inference gesture
data = np.array([angle], dtype=np.float32)
ret, results, neighbours, dist = knn.findNearest(data, 3)
idx = int(results[0][0])
if idx == 0 or idx == 6: # fist or six
thumb_end = res.landmark[4]
fist_end = res.landmark[17]
text = None
if thumb_end.x - fist_end.x > THRESHOLD:
text = 'RIGHT'
elif fist_end.x - thumb_end.x > THRESHOLD:
text = 'LEFT'
elif thumb_end.y - fist_end.y > THRESHOLD:
text = 'DOWN'
elif fist_end.y - thumb_end.y > THRESHOLD:
text = 'UP'
if text is not None:
cv2.putText(img, text=text, org=(int(res.landmark[0].x * img.shape[1]), int(res.landmark[0].y * img.shape[0] + 20)), fontFace=cv2.FONT_HERSHEY_SIMPLEX, fontScale=1, color=(255, 255, 255), thickness=2)
elif idx in [1, 2, 3, 4, 5, 9]: # 숫자 1,2,3,4,5 인식
if idx == 9:
idx = 2
cv2.putText(img, text=str(idx), org=(int(res.landmark[0].x * img.shape[1]), int(res.landmark[0].y * img.shape[0] + 20)), fontFace=cv2.FONT_HERSHEY_SIMPLEX, fontScale=1, color=(255, 255, 255), thickness=2)
mp_drawing.draw_landmarks(img, res, mp_hands.HAND_CONNECTIONS)
cv2.imshow('Ampere\'s Law', img)
if cv2.waitKey(1) == ord('q'):
break