File size: 1,704 Bytes
39b4c8c |
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 |
import numpy as np
from scipy import signal
from .base import VHRMethod
class POS(VHRMethod):
"""
POS algorithm described in "Algorithmic Principles of Remote PPG"
(https://ieeexplore.ieee.org/document/7565547 )
Numbers in brackets refer to the line numbers in the "Algorithm 1" of the paper
"""
methodName = 'POS'
projection = np.array([[0, 1, -1], [-2, 1, 1]])
def __init__(self, **kwargs):
super(POS, self).__init__(**kwargs)
def apply(self, X):
# Run the pos algorithm on the RGB color signal c with sliding window length wlen
# Recommended value for wlen is 32 for a 20 fps camera (1.6 s)
wlen = int(1.6*self.video.frameRate)
# Initialize (1)
h = np.zeros(X.shape[1])
for n in range(X.shape[1]):
# Start index of sliding window (4)
m = n - wlen + 1
if m >= 0:
# Temporal normalization (5)
cn = X[:, m:(n+1)]
cn = np.dot(self.__get_normalization_matrix(cn), cn)
# Projection (6)
s = np.dot(self.projection, cn)
# Tuning (7)
hn = np.add(s[0, :], np.std(s[0, :])/np.std(s[1, :])*s[1, :])
# Overlap-adding (8)
h[m:(n+1)] = np.add(h[m:(n+1)], hn - np.mean(hn))
return h
def __get_normalization_matrix(self, x):
# Compute a diagonal matrix n such that the mean of n*x is a vector of ones
d = 0 if (len(x.shape) < 2) else 1
m = np.mean(x, d)
n = np.array([[1/m[i] if i == j and m[i] else 0 for i in range(len(m))] for j in range(len(m))])
return n
|