File size: 4,420 Bytes
497f61b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import cv2
import numpy as np
import mediapipe as mp
mp_drawing = mp.solutions.drawing_utils
mp_drawing_styles = mp.solutions.drawing_styles
mp_hands = mp.solutions.hands

roi_bad_pixel_number = 500  # the number of pixel where 3 values are zero in RGB channel on ROI image
roi_aspect_ratio_threshold = 100  # the passable aspect ratio threshold of ROI image
roi_size_threshold = 0.23
padding_size = 300 # the extra marine size of the frame inputted into Google Mediapipe Graph (this value must be a multiple of two)

def img_padding(img):
    h, w, _ = img.shape
    image = np.zeros((h + padding_size, w + padding_size, 3), np.uint8)
    image[int(padding_size / 2):-int(padding_size / 2), int(padding_size / 2):-int(padding_size / 2), :] = img
    return image

def img_crop(img_original, x2, x1, y2, y1, label):
    
    h, w, _ = img_original.shape
    img = np.zeros((h + 20, w + 20, 3), np.uint8)
    img[10:-10, 10:-10, :] = img_original
    if label == "Right":
        v1 = np.array([x2 * w, y2 * h])
        v2 = np.array([x1 * w, y1 * h])
    else:
        v2 = np.array([x2 * w, y2 * h])
        v1 = np.array([x1 * w, y1 * h])
    
    theta = np.arctan2((v2 - v1)[1], (v2 - v1)[0]) * 180 / np.pi
    R = cv2.getRotationMatrix2D(tuple([int(v2[0]), int(v2[1])]), theta, 1)
    
    v1 = (R[:, :2] @ v1 + R[:, -1]).astype(int)
    v2 = (R[:, :2] @ v2 + R[:, -1]).astype(int)
    img_r = cv2.warpAffine(img, R, (w, h))
    
    if 1:
        ux = int(v1[0] - (v2 - v1)[0] * 0.05)
        uy = int(v1[1] + (v2 - v1)[0] * 0.05)
        lx = int(v2[0] + (v2 - v1)[0] * 0.05)
        ly = int(v2[1] + (v2 - v1)[0] * 1)
    else:
        ux = int(v1[0] - (v2 - v1)[0] * 0.1)
        uy = int(v1[1] + (v2 - v1)[0] * 0.1)
        lx = int(v2[0] + (v2 - v1)[0] * 0.1)
        ly = int(v2[1] + (v2 - v1)[0] * 1.2)

    # delta_y is movement value in y ward
    delta_y = (ly - uy) * 0.15

    ly = int(ly - delta_y)
    uy = int(uy - delta_y)

    delta_x = (lx - ux) * 0.01
    lx = int(lx + delta_x)
    ux = int(ux + delta_x)
    
    if label == "Right":
        delta_x = (lx - ux) * 0.05
        lx = int(lx + delta_x)
        ux = int(ux + delta_x)
    # roi = img_r
    roi = img_r[uy:ly, ux:lx]
    if roi.shape[0] == 0 or roi.shape[1] == 0:
        print("error 1")
        return None, 3
    
    if abs(roi.shape[0] - roi.shape[1]) > roi_aspect_ratio_threshold:
        print("error 2", abs(roi.shape[0] - roi.shape[1]))
        return None, 4
    if roi.shape[1] / w < roi_size_threshold:
        print("error 3", roi.shape[1] / w)
        return None, 7
    
    n_zeros = np.count_nonzero(roi == 0)
    if n_zeros > roi_bad_pixel_number:
        print("error 4", n_zeros)
        return None, 5
    return roi, 0

def cupped_hand_filter(hand_landmarks):
    return hand_landmarks.landmark[12].y - hand_landmarks.landmark[11].y
         
def get_roi(path, hand_type, x1, y1, x2, y2):
    img = cv2.imread(path)
    
    if hand_type != 0:
        label = "Left"
    else:
        label = "Right"
    roi, _ = img_crop(img, x1, x2, y1, y2, label)
    return roi

def get_roi_image(img):

    label = "" 
    with mp_hands.Hands(
            static_image_mode=True,
            max_num_hands=2,
            min_detection_confidence=0.5) as hands:
        # Read an image, flip it around y-axis for correct handedness output (see
        # above).
        if 1:            
            image = img_padding(img)
        else:
            image = cv2.flip(cv2.imread(file), 1)

        # Convert the BGR image to RGB before processing.
        results = hands.process(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))

        # Print handedness and draw hand landmarks on the image.
        if results.multi_handedness is not None:
            label = results.multi_handedness[0].classification[0].label
        
        if results.multi_hand_landmarks is None:
            return None, None

        image_height, image_width, _ = image.shape
        hand_landmarks = results.multi_hand_landmarks[0]
                    
        if cupped_hand_filter(hand_landmarks) > 0:
            return None, None
        else:                                
            roi, roi_msg_index = img_crop(image, hand_landmarks.landmark[5].x, hand_landmarks.landmark[17].x,
                                                hand_landmarks.landmark[5].y, hand_landmarks.landmark[17].y, label)
            
            return roi, label