File size: 8,057 Bytes
fa926f8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d15a19c
fa926f8
 
 
 
 
 
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
import numpy as np
import math
import cv2
import torch
from scipy import signal
from scipy.fft import fft
from scipy.signal import butter, filtfilt
# from facenet_pytorch import MTCNN
from face_detection import FaceDetection
import joblib


def butter_bandpass(sig, lowcut, highcut, fs, order=2):
    # butterworth bandpass filter

    sig = np.reshape(sig, -1)
    nyq = 0.5 * fs
    low = lowcut / nyq
    high = highcut / nyq
    b, a = butter(order, [low, high], btype='band')

    y = filtfilt(b, a, sig)
    return y


def face_detection(video_list):
    device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
    mtcnn = MTCNN(device=device)

    face_list = []
    for t, frame in enumerate(video_list):
        if t == 0:
            boxes, _, = mtcnn.detect(
                frame)  # we only detect face bbox in the first frame, keep it in the following frames.
        if t == 0:
            box_len = np.max([boxes[0, 2] - boxes[0, 0], boxes[0, 3] - boxes[0, 1]])
            box_half_len = np.round(box_len / 2 * 1.1).astype('int')
        box_mid_y = np.round((boxes[0, 3] + boxes[0, 1]) / 2).astype('int')
        box_mid_x = np.round((boxes[0, 2] + boxes[0, 0]) / 2).astype('int')
        cropped_face = frame[box_mid_y - box_half_len:box_mid_y + box_half_len,
                       box_mid_x - box_half_len:box_mid_x + box_half_len]
        cropped_face = cv2.resize(cropped_face, (128, 128))
        face_list.append(cropped_face)

        print('face detection %2d' % (100 * (t + 1) / len(video_list)), '%', end='\r', flush=True)

    face_list = np.array(face_list)  # (T, H, W, C)
    face_list = np.transpose(face_list, (3, 0, 1, 2))  # (C, T, H, W)
    face_list = np.array(face_list)[np.newaxis]

    return face_list


def face_detection_ROI(face_detection, frame_list):
    face_frame_list = []
    ROI1_list = []
    ROI2_list = []
    
    for i in range(0, frame_list.shape[0]):
        frame = frame_list[i]
        frame, face_frame, ROI1, ROI2, status, mask, face_region = FaceDetection.face_detect(face_detection, frame)
        face_frame_list.append(face_frame)
        ROI1_list.append(ROI1)
        ROI2_list.append(ROI2)
    return np.array(face_frame_list), np.array(ROI1_list), np.array(ROI2_list), status, face_region


def butter_bandpass(sig, lowcut, highcut, fs, order=2):
    # butterworth bandpass filter

    sig = np.reshape(sig, -1)
    nyq = 0.5 * fs
    low = lowcut / nyq
    high = highcut / nyq
    b, a = butter(order, [low, high], btype='band')

    y = filtfilt(b, a, sig)
    return y

def hr_fft(sig, fs, harmonics_removal=True):
    # get heart rate by FFT
    # return both heart rate and PSD

    sig = sig.reshape(-1)
    sig = sig * signal.windows.hann(sig.shape[0])
    sig_f = np.abs(fft(sig))
    low_idx = np.round(0.6 / fs * sig.shape[0]).astype('int')
    high_idx = np.round(4 / fs * sig.shape[0]).astype('int')
    sig_f_original = sig_f.copy()

    sig_f[:low_idx] = 0
    sig_f[high_idx:] = 0

    peak_idx, _ = signal.find_peaks(sig_f)
    sort_idx = np.argsort(sig_f[peak_idx])
    sort_idx = sort_idx[::-1]

    peak_idx1 = peak_idx[sort_idx[0]]
    peak_idx2 = peak_idx[sort_idx[1]]

    f_hr1 = peak_idx1 / sig_f.shape[0] * fs
    hr1 = f_hr1 * 60

    f_hr2 = peak_idx2 / sig_f.shape[0] * fs
    hr2 = f_hr2 * 60
    if harmonics_removal:
        if np.abs(hr1-2*hr2)<10:
            hr = hr2
        else:
            hr = hr1
    else:
        hr = hr1

    x_hr = np.arange(len(sig_f))/len(sig_f)*fs*60
    return hr, sig_f_original, x_hr


def hr_fft_2(processed, fps):
    L = len(processed)
    fps = 30#float(L) / (times[-1] - times[-L])  # calculate HR using a true fps of processor of the computer, not the fps the camera provide
    LEN = int(fps * 1.55)
    # even_times = np.linspace(times[-L], times[-1], LEN)

    processed = signal.detrend(processed)  # detrend the signal to avoid interference of light change
    # interpolated = np.interp(even_times, times[-L:], processed)  # interpolation by 1
    interpolated = processed
    # interpolated = np.hamming(LEN) * interpolated  # make the signal become more periodic (advoid spectral leakage)
    norm = (interpolated - np.mean(interpolated))/np.std(interpolated)#normalization

    # norm = interpolated / np.linalg.norm(interpolated)
    raw = np.fft.rfft(norm * 30)  # do real fft with the normalization multiplied by 10
    raw_r = raw.copy()

    freqs = float(fps) / LEN * np.arange(LEN / 2 + 1)
    freqs = 60. * freqs

    idx_remove = np.where((freqs < 50) & (freqs > 180))
    raw[idx_remove] = 0

    fft = np.abs(raw) ** 2  # get amplitude spectrum

    idx = np.where((freqs > 50) & (freqs < 180))  # the range of frequency that HR is supposed to be within
    pruned = fft[idx]
    pfreq = freqs[idx]

    # freqs = pfreq
    fft = pruned

    idx2 = np.argmax(pruned)  # max in the range can be HR
    bpm = pfreq[idx2]

    # calculate Respiratory Rate, 计算呼吸率
    idx_remove = np.where((freqs < 5) & (freqs > 60))
    raw_r[idx_remove] = 0
    fft_r = np.abs(raw_r) ** 2
    idx = np.where((freqs > 5) & (freqs < 60))
    pruned_r = fft_r[idx]
    pfreq = freqs[idx]
    idx3 = np.argmax(pruned_r)
    pruned_r[idx3] = 0
    idx3 = np.argmax(pruned_r)
    respiratory_rate = pfreq[idx3]

    return bpm, respiratory_rate


def calc_rr(peaklist, sample_rate, working_data={}):
    peaklist = np.array(peaklist) #cast numpy array to be sure or correct array type
    working_data['peaklist'] = peaklist  # Make sure, peaklist is always an np.array

    rr_list = (np.diff(peaklist) / sample_rate) * 1000.0
    rr_indices = [(peaklist[i], peaklist[i+1]) for i in range(len(peaklist) - 1)]
    rr_diff = np.abs(np.diff(rr_list))
    rr_sqdiff = np.power(rr_diff, 2)
    working_data['RR_list'] = rr_list
    working_data['RR_indices'] = rr_indices
    working_data['RR_diff'] = rr_diff
    working_data['RR_sqdiff'] = rr_sqdiff
    return working_data


def calculate_hrv(ippg, fps):
    peak_array, _ = signal.find_peaks(ippg[-200::2])  # down sample rate: 2
    peak_list = peak_array.tolist()
    result = calc_rr(peak_list, fps/2)
    # print(peak_list)
    # RR_list = result['RR_list'].tolist()
    # RR_diff = result['RR_diff'].tolist()
    # print(RR_list)
    # print(RR_diff)
    RR_std = 0
    if len(result['RR_list']) > 0:
        RR_std = np.std(result['RR_list'], ddof=1)  # calculate RR interval standard deviation
    if math.isnan(RR_std):
        RR_std = 0
    return RR_std

def RGB_SpO2(ROI_list):
    roi_avg = []
    roi_std = []
    for i in range(len(ROI_list)):
        roi_avg.append(np.average(ROI_list[i], axis=(0, 1)))
        roi_std.append(np.std(ROI_list[i], axis=(0, 1), ddof=1))
    roi_avg = np.array(roi_avg)
    roi_std = np.array(roi_std)
    mean_red = roi_avg[:, 0]
    std_red = roi_std[:, 0]
    mean_green = roi_avg[:, 1]
    std_green = roi_std[:, 1]
    mean_blue = roi_avg[:, 2]
    std_blue = roi_std[:, 2]
    A = -11.2
    B = 109.3
    R = (np.average(mean_red) / np.average(std_red)) / (np.average(mean_blue) / np.average(std_blue))
    SpO2 = A * R + B
    return SpO2


def RGB_HR(ROI_list):
    roi_avg = []
    roi_std = []
    for i in range(len(ROI_list)):
        roi_avg.append(np.average(ROI_list[i], axis=(0, 1)))
        roi_std.append(np.std(ROI_list[i], axis=(0, 1), ddof=1))
    roi_avg = np.array(roi_avg)
    roi_std = np.array(roi_std)
    mean_red = roi_avg[:, 0]
    std_red = roi_std[:, 0]
    mean_green = roi_avg[:, 1]
    std_green = roi_std[:, 1]
    mean_blue = roi_avg[:, 2]
    std_blue = roi_std[:, 2]

    

    ippg_chanel_data = np.array((mean_red,std_red,mean_green,std_green,mean_blue,std_blue)).T 
    print(ippg_chanel_data )
    # ippg_chanel_data = np.array(ippg_chanel_data).reshape(len(ippg_chanel_data),6)

    HR_pred = []
    model_list = joblib.load( './code/model_weight/lgb_model_threechanel2HR.pkl')
    for model in model_list:
        result = model.predict(ippg_chanel_data)
        HR_pred.append(result+10)
    HR = np.mean(HR_pred, axis=0)

    return np.mean(HR)