Lie detection method and system
View Patent ↗The present disclosure provides a heart rate variability (HRV) analysis-based method of lie detection, related computer program product, computer readable storage medium and system. Also provided is a method of HRV analysis using strange entropy.
1. A method of lie detection, comprising:
receiving an input associated with heart rate variability “HRV”; and
performing an HRV analysis based on the input associated with HRV to obtain a lie detection result; wherein said HRV analysis includes nonlinear HRV analysis, wherein the nonlinear HRV analysis includes the following:
taking a time series with N elements, u :{u(i): 1 ≦i≦N}, wherein u(i) represents signals carried in the input associated with HRV, and for each u(i), there is a corresponding vector with m elements:
X ( i )=[ u ( i ), u ( i+ 1), . . . , u ( i+m− 1)];
calculating a base scale BS(i) for each vector X(i);
transforming every X(i) to a m-dimensional symbol series S i ={s(i), . . . s(i+m−1)},sεA(A=0, 1, 2, 3) based on a scale of a×BS(i);
calculating the probability of S i , wherein the probability of each different combination among the entire N−m+1 m-dimensional vectors is:
p
(
π
)
=
#
{
t
|
(
u
i
,
…
,
u
t
+
m
-
1
)
has
type
π
}
N
-
m
+
1
,
wherein 1 ≦t ≦N−m+1; # is the number of states, and each possible combination π for S i represents a vibration mode for S i ;
obtaining nonlinear HRV parameters or m-word distribution graph according to the probability of S i ; and
obtaining a nonlinear HRV parameter PSS, which is defined as sum probabilities of all strange states calculated according to the following equation:
P
S
S
=
∑
P
s
∑
i
=
0
255
P
i
;
wherein P i is probability of combination i, P s is probability of strange state, and ΣP s is probability sum of all strange states.
2. A lie detection system, comprising:
a computing unit configured to
receive an input associated with HRV; and
perform an HRV analysis based on the input associated with HRV to obtain a lie detection result; wherein said HRV analysis includes nonlinear HRV analysis, wherein the nonlinear HRV analysis includes the following:
obtaining a nonlinear HRV parameter PSS, which is defined as sum probabilities of all strange states calculated according to the following equation:
P
S
S
=
∑
P
s
∑
i
=
0
255
P
i
;
wherein P i is probability of combination i, P s is probability of strange state, and ΣP s is probability sum of all strange states.
3. The lie detection system according to claim 2 , wherein said nonlinear HRV analysis is performed by using strange entropy method comprising:
taking a time series signal with N elements, u :{u(i): 1 ≦i≦N} ,wherein u(i) represents signals carried in the input associated with HRV, and for each u(i), there is a corresponding vector with m elements:
X ( i )=[ u ( i ), u ( i+ 1), . . . , u ( i+m− 1)];
calculating a base scale BS(i) for each vector X(i);
transforming every X(i) to a m-dimensional symbol series S i ={s(i), . . . s(i+m−1)},sεA(A=0, 1, 2, 3) based on a scale of a×BS(i);
calculating probability of S i such that the probability of each different combination among the entire N−m +1 m-dimensional vectors is:
p
(
π
)
=
#
{
t
|
(
u
i
,
…
,
u
t
+
m
-
1
)
has
type
π
}
N
-
m
+
1
,
wherein 1 ≦t ≦N −m+1 ,# is the number of states, and each possible combination π for S i represents a vibration mode for S i ; and
obtaining one or more of nonlinear HRV parameters or m-word distribution graph according to the probability of S i .
4. A method of lie detection, comprising:
receiving an input associated with heart rate variability “HRV”; and
performing an HRV analysis based on the input associated with HRV to obtain a lie detection result; wherein said HRV analysis includes one or more of nonlinear HRV analysis, or neural network-based linear HRV analysis, wherein said nonlinear HRV analysis is performed by using one or more methods selected from the group consisting of strange entropy (StEn), Chaos, correlation dimension, fractal theory, strange attractors, mode entropy (modEn), multifractal, multiscale multifractal, Lyapunov index, base-scale entropy, and approximate entropy (ApEn).
5. The method of lie detection according to claim 4 , wherein said input associated with HRV includes a time series of ECG recordings.
6. The method of lie detection according to claim 4 , wherein said HRV analysis is nonlinear HRV analysis.
7. The method of lie detection according to claim 6 , wherein said nonlinear HRV analysis is performed by using one or more methods selected from the group consisting of strange entropy (StEn) and base-scale entropy.
8. The method of lie detection according to claim 7 , wherein said nonlinear HRV analysis is performed by:
taking a time series signal with N elements, u:{u(i):1≦i≦N}, wherein u(i) represents signals carried in the input associated with HRV, and for each u(i), there is a corresponding vector with m elements:
X ( i )=[ u ( i ), u ( i+ 1), . . . , u ( i+m− 1)];
calculating a base scale BS(i) for each vector X(i);
transforming every X(i) to a m-dimensional symbol series S i ={s(i), . . . s(i+m−1)},sεA(A=0, 1, 2, 3) based on a scale of a×BS(i);
calculating probability of S i such that the probability of each different combination among the entire N−m+1m-dimensional vectors is:
p
(
π
)
=
#
{
t
|
(
u
i
,
…
,
u
t
+
m
-
1
)
has
type
π
}
N
-
m
+
1
,
wherein 1≦t≦N−m+1,# is the number of states, and each possible combination π for S i represents a vibration mode for S i ; and
obtaining one or more of nonlinear HRV parameters or m-word distribution graph according to the probability of S i .
9. The method of lie detection according to claim 8 , wherein u(i) represents R-R interval.
10. The method of lie detection according to claim 8 , wherein BS(i) is defined as square root average of difference between adjacent elements:
BS
(
i
)
=
∑
j
=
1
m
-
1
(
(
u
(
i
+
j
)
-
u
(
i
+
j
-
1
)
)
2
m
-
1
.
11. The method of lie detection according to claim 8 , wherein said every X(i) is transformed to a m-dimensional symbol series S i according to the transformation equation:
S
i
+
k
=
{
0
:
u
_
i
<
u
i
+
k
≤
u
_
i
+
a
×
BS
(
i
)
1
:
u
i
+
k
>
u
_
i
+
a
×
BS
(
i
)
2
:
u
_
i
-
a
×
BS
(
i
)
<
u
i
+
k
≤
u
_
i
3
:
u
i
+
k
≤
u
_
i
-
a
×
BS
(
i
)
;
wherein i =1, 2, . . . , N−m+1; k =0, 1, . . . m−1; S i+k is the k-th element of S i ; u i+k is the k-th element of X (i); ū i represents the average value of the m-dimensional vector X (i); and 0, 1, 2, 3 are notation for region partition.
12. The method of lie detection according to claim 8 , wherein said nonlinear HRV parameters include H(m) which is calculated according to the following equation:
H ( m )=−Σ P (π)log 2 P (π).
13. The method of lie detection according to claim 8 , wherein said nonlinear HRV parameters include PSS which is defined as sum of probabilities of all strange states calculated according to following equation:
P
S
S
=
∑
P
s
∑
i
=
0
255
P
i
;
wherein P i is probability of combination i, P s is probability of strange state, and ΣP s is probability sum of all strange states.
14. The method of lie detection according to claim 8 , wherein m is 4, a is 0.2 and N is larger than 4 m .
15. The method of lie detection according to claim 8 , wherein the lie detection result is obtained by real-time monitoring using a sliding window.
16. The method of lie detection according to claim 8 , wherein the lie detection result is obtained by comparing the analysis results with normal data.
17. The method of lie detection according to claim 4 , wherein said HRV analysis includes neural network-based linear HRV analysis comprising:
acquiring one or more linear HRV parameters from the input associated with HRV; and
performing a neural network analysis based on the linear HRV parameters.
18. The method of lie detection according to claim 17 , wherein the neural network analysis is performed based on at least three linear HRV parameters selected from the group consisting of Total Power, Low Frequency Power, High Frequency Power, LFnorm, HFnorm, Low Frequency Power/High Frequency Power, Heart Rate, and Standard Deviation of adjacent R-R Interval.
19. The method of lie detection according to claim 4 , wherein said HRV analysis includes neural network-based linear HRV analysis in combination with nonlinear HRV analysis, and wherein the neural network analysis is performed based on a combination of linear HRV parameters and nonlinear HRV parameters.
20. The method of lie detection according to claim 19 , wherein said combination of linear HRV parameters and nonlinear HRV parameters includes two nonlinear HRV parameters H(m) and PSS and one to six linear HRV parameters selected from the group consisting of Total Power, Low Frequency Power, High Frequency Power, LFnorm, HFnorm, Low Frequency Power/High Frequency Power, Heart Rate, and Standard Deviation of adjacent R-R Interval.
21. The method of lie detection according to claim 4 , further comprising outputting one or more of the analysis results or lie detection results.