CN110236523A - Gait-Cardiac RR interval the correlating method returned based on Gauss - Google Patents
Gait-Cardiac RR interval the correlating method returned based on Gauss Download PDFInfo
- Publication number
- CN110236523A CN110236523A CN201910521097.3A CN201910521097A CN110236523A CN 110236523 A CN110236523 A CN 110236523A CN 201910521097 A CN201910521097 A CN 201910521097A CN 110236523 A CN110236523 A CN 110236523A
- Authority
- CN
- China
- Prior art keywords
- gait
- interval
- cardiac
- follows
- gauss
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
- A61B5/366—Detecting abnormal QRS complex, e.g. widening
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Molecular Biology (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Signal Processing (AREA)
- Cardiology (AREA)
- Psychiatry (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Fuzzy Systems (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses a kind of gait-Cardiac RR interval correlating methods returned based on Gauss.The present invention is the following steps are included: step 1. is utilized respectively three-dimensional motion tracking system and electrocardiogram acquisition equipment, the three-dimensional coordinate data of synchronous acquisition subject motion and corresponding ECG signal；Step 2. carries out gait quantitative analysis, nonlinear analysis, time frequency analysis to three-dimensional coordinate data, extracts 17 kinds of features, and constitutive characteristic vector proposes fusion method；Step 3. carries out pretreatment to ECG signal and extracts the interval RR；Step 4. constructs the Gaussian process between gait feature and Cardiac RR interval and returns (GPR) interaction prediction model；Step 5. training and test Gaussian process regression model, are associated forecast analysis.The present invention adaptive strong, easy to accomplish, hyper parameter such as adaptively obtains at the advantages, has preferable robustness and Generalization Capability, can preferably disclose the related information between gait and electrocardio.
Description
Technical field
The present invention relates to a kind of gait-Cardiac RR interval correlating methods returned based on Gauss, are related to Digital Signal Processing
With the multiple technologies field such as machine learning, medical treatment & health, human motion analysis.
Background technique
In recent years, with the raising of human living standard, obesity, pressure etc. are the multiple factors for leading to cardiovascular disease.
Currently, cardiovascular disease is to threaten the one of the major reasons of human life and health, exist in particular in cardiovascular patient
When movement, the consequence of the especially severes such as uncomfortable or even sudden death is caused by excessive movement.Arrhythmia cordis is also a kind of extremely normal
The electrical activity abnormality seen even can cause to die suddenly, health is also increasingly appearing in movement environment when serious
On the person.
Currently, gait analysis includes qualitative analysis and quantitative analysis, qualitative analysis is studied by observation, but it has
There is certain subjectivity, it is difficult to be analyzed in multiple location, too many levels；The quantitative analysis of gait is by instrument and special to set
The standby method for obtaining objective data and being analyzed gait, i.e., can be measured using simple measuring tool, can also pass through complexity
Acquisition of the equipment and instrument to kinematics parameters, kinetic parameter, myoelectrical activity parameter and energy parameter is current using extensive
Method.Studies have shown that the movement of human body lower limbs is in periodically variation.In recent years, as the research of biomedicine signals is risen,
Nonlinear kinetics and the method for time frequency analysis are widely used, and achieve preferable achievement.Because nonlinear characteristic can embody
Its dynamic substantive characteristics, time frequency analysis have good locality in time domain and frequency domain, and analysis result is more stable, can be preferably
Embody the intrinsic propesties of signal.Electrocardiogram is the electrical activity state for recording heart, and the current potential that QRS complex represents ventricular muscles depolarization becomes
Change, is one narrow but that amplitude the is high wave group occurred after P wave.It is by Q wave, R wave, S wave component, be in electrocardiogram most
Part outstanding.The detection of QRS complex is the key that ECG signal, and current method mainly includes neural network and wavelet transformation,
But the training time of its initial stage is longer, and calculation amount is larger, is not suitable for real-time detection.At present about gait and cardiac electrical analysis
All it is the independent research of each subsystem, the very few of Journal of Sex Research is associated with cardiac electrical for gait.By from various gait informations
It accurately predicts Cardiac RR interval, has to human motion analysis, athletic rehabilitation treatment, the health supervision of patient and intelligent medical treatment
Highly important meaning.
Summary of the invention
For above-mentioned background and existing problems and shortcomings, in order between Accurate Prediction Cardiac RR in various gait informations
Every the invention proposes a kind of gait-Cardiac RR interval correlating methods returned based on Gauss.The present invention passes through Codamotion
Infrared three-dimensional real time motion capture system and TrignoTMWireless system, the movement of the body lower extremity of real-time capture is three-dimensional
Coordinate data and the ECG signal of synchronous acquisition are analyzed and processed.Firstly, by kinematics analysis, nonlinear dynamic analysis,
Time frequency analysis is extracted 17 kinds of gait features, including gait cycle, step-length, step height, step width, foot drift angle maximum value and minimum value,
Leg speed, leg speed acceleration, approximate entropy, Sample Entropy, fuzzy entropy, LZ complexity, C0 complexity and three kinds of small echos (Cmorlet,
Cgaus, mexh) transformation coefficient modulus value, constitutive characteristic vector, propose fusion method；Then, ECG signal is pre-processed,
QRS complex is detected by real time algorithm, extracts Cardiac RR corresponding with three-dimensional motion data interval；Establish gait feature vector
Gaussian process between Cardiac RR interval returns correlation model；Finally, carrying out Performance Evaluation.The experimental results showed that the present invention
The gait feature of extraction and the Gauss homing method proposed building gait and Cardiac RR interval prediction model, have compared with
High precision of prediction.
To achieve the above object, following technical scheme is used, the described method comprises the following steps:
Step 1. obtains the movement three-dimensional coordinate data of human body lower limbs foot mark point and corresponds to ECG signal；
Using three-dimensional motion tracking system and electrocardiogram acquisition equipment, when synchronization catch subject's lower limb foot mark point moves
Three-dimensional coordinate data and corresponding synchronous ECG signal.
Step 2. analyzes the movement three-dimensional coordinate data of real-time capture, extraction time-distance and angle parameter, non-
Totally 17 kinds of gait features, constitutive characteristic vector propose fusion method for linear character, time-frequency characteristics；
The three-dimensional motion data that noise is removed as obtained by step 1, select the mark point at human body lower limbs position to be analyzed.
We extract each gait cycle with the action data of heel direction of advance, and using every ten gait cycles as a subset,
Data are divided into multiple subsets.
By carrying out quantitative analysis to lower limb foot three-dimensional motion data, the when m- distance parameter and angle of gait is extracted
Parameter, including gait cycle, step-length, step height, step width, leg speed, leg speed acceleration, sufficient drift angle maximum value and minimum value are spent, is taken every
The average value of one subset；The three-dimensional motion data for extracting heel propulsion using the method for nonlinear dynamic analysis again are each
The nonlinear characteristic of subset, including approximate entropy, Sample Entropy, fuzzy entropy, LZ complexity, C0 complexity characteristics；Finally use time-frequency
The method of analysis calculates power spectral density maximum value by Fast Fourier Transform (FFT), passes through the C- of continuous wavelet transform
Tri- kinds of wavelet functions of morlet3-3, C-gaus1, mexh calculate the wavelet module value characteristics of mean in scale 2.It will be extracted
Feature constitute feature vector, propose fusion method.
The pretreatment of step 3.ECG signal and the interval RR are extracted；
3-1. is handled by the bandpass filter that cascade low pass and high-pass filter form, inhibit baseline drift and
It is interfered caused by T wave；The difference equation of its step low-pass and high-pass filter is respectively as follows:
Y (nT)=2y (nT-T)-y (nT-2T)+x (nT) -2x (nT-6T)+x (nT-12T)
Y (mT)=32x (mT-16T)-[y (mT-T)+x (mT)-x (mT-32T)]
Wherein, T is the sampling period, and x (nT) indicates the input of discrete-time system electrocardiosignal time series, and y (nT) is
Electrocardiosignal after low-pass filtering, cutoff frequency are set as 12Hz, gain 2, and processing delay is about 6 sampling periods.High pass
The cutoff frequency of filter is about 5Hz, the delay in 32,16 sampling periods of gain.
3-2. handles using derivative filter by bandpass filter treated electrocardiosignal, provides QRS negative slope
Information；Its difference equation are as follows:
Y (mT)=(1/8T) [- x (mT-2T) -2x (mT-T)+2x (mT+T)+x (mT+2T)]
Treated that output using non-linear squares function achievees the effect that nonlinear amplification through derivative filter by 3-3., by force
Adjust high frequency characteristics, the difference equation of non-linear squares function are as follows:
Y (mT)=[x (mT)]2
3-4. generates the signature waveform of envelope shape by moving window integrator (MVI) processing, to further enhance spy
The differentiation of sign and noise.Its difference equation are as follows:
Y (mT)=(1/N) [x (mT- (N-1) T)+x (mT- (N-2) T+ ...+x (mT)]
Wherein, the width of moving integration window should be roughly the same with QRS complex as wide as possible, believes with original electrocardiographicdigital
Number sampling period it is relevant.If acquiring the sample frequency 500Hz of electrocardiosignal, i.e. 500 sample points of acquisition per second, therefore one
As for N sample points be about 75 (the real-time width period of i.e. corresponding window is 0.15 second).
3-5. makes threshold value and other parameters carry out periodical adjustment, and detects the position of R wave or QRS wave.Finally to flat
The R wave output detected in sliding signal is analyzed, and is checked twice the output of bandpass signal, and detection essence is improved
Degree, eventually finds the original index of true R wave.
Step 4. constructs the Gaussian process between gait feature and Cardiac RR interval and returns (GPR) interaction prediction model；
By step 3 and the resulting Cardiac RR interval of step 4 and fused gait feature vector, by fused gait
Feature vector establishes height between the two as desired output sample set as input sample collection, corresponding Cardiac RR spaced features
This process regression model.
4.1 Gaussian processes (GP) principle
Gaussian process refers to that the random process of a normal state, the Joint Distribution of the limited variable of any dimension obey Gauss point
Cloth.For arbitrary finite x, i.e. x1,x2,…xn∈ N, corresponding limited stochastic variable F=(f (x1),f(x2),…f
(xn))T, remember F~GP (m (xi), K), i.e.,
Wherein, F is Joint Distribution；Mean function and covariance function (kernel function) are expressed as follows:
M (x)=E [f (x)]
K=E [(f (x)-m (x)) (f (x ')-m (x '))]
The foundation of 4.2GPR regression model
When establishing Gauss regression model, consider that target value y contains noise, i.e., is defined as:
Y=f (x)+ε.Wherein: f (x) is Gaussian process,Training set are as follows:
D:yi=f (xi)+εi(i=1,2 ..., n), in test set:In advance
Survey corresponding output valve f*Multivariate Gaussian distribution are as follows:
Wherein, K*=[k (x*, x1),k(x*,x2),…,k(x*,xn)], K**=k (x*, x*), σnFor the standard of noise
Difference.According to how far the condition of Gaussian Profile, Gaussian process forecast of regression model equation can be obtained:
(f* | X, y, X*)~GP (m (x*), cov (f*))
In formula: matrix X by training set input xiColumn vector groups at；Matrix X* by test set inputColumn vector
Composition.
Surveying the gait prediction cardiac electrical interval RR regression model indicates are as follows:
Wherein: ω ∈ RmIndicate weight vector, mean function m (x)=E [f (x)], kernel function k (x, x ')=E [(f (x)-
m(x))(f(x′)-m(x′))]。
The selection of 4.3 kernel functions
The selection of kernel function is most important for Gaussian process, covariance function (the i.e. core in Gaussian process regression model
Function) it must satisfy Mercer condition.Present invention employs 3 kinds of kernel functions:
Quadratic Rational kernel function (RQ):
Square exponential kernel functions (SE)
Matern kernel function (M)
Wherein, M=diag (l-2), l is variance measure,For signal variance.L is that relevance measures hyper parameter；δijFor symbol
Number function is 1 when i and j is identical, is not simultaneously 0；α is the form parameter of kernel function.
DefinitionFor hyper parameter vector, optimal solution is adaptively obtained by maximum-likelihood method.It initially sets up
The negative log-likelihood function of training sample conditional probability, and it is enabled to seek local derviation to hyper parameter, then using conjugate gradient method to this
Partial derivative is minimized, and hyper parameter optimal solution is obtained.
Step 5. training and test Gaussian process regression model, are associated forecast analysis；
Stochastic inputs training sample set and test sample collection accounting are 4:1, using three kinds of kernel functions: Rational Quadratic kernel function
(RQ), square exponential kernel functions (SE), Matern5/2 kernel function (MA), are associated forecast analysis, and test result ten times are taken
Its average value.
By using root-mean-square error (RMSE) and determine coefficient (R2) accuracy and validity of prediction model are carried out
Evaluation, RMSE is smaller, and prediction result is better.Its expression formula is as follows:
Wherein, n indicates the number of prediction data, yiIt is actual prediction data value, tiRepresent desired output.
The present invention has the beneficial effect that:
1, the present invention captures system and TrignoTM wireless system to human motion by Codamotion three-dimensional motion
Three-dimensional coordinate data is acquired with synchronous ECG signal, compared to the Portable acquiring equipment being commonly used, more precisely
Property；By proposing a kind of gait-Cardiac RR interval correlating method returned based on Gauss, this method has adaptive the present invention
By force, easy to accomplish, hyper parameter such as adaptively obtains at the advantages, has preferable robustness and Generalization Capability, can preferably disclose
Related information between gait and electrocardio.
2, the present invention by by gait when m- distance and angle parameter, nonlinear characteristic and time-frequency characteristics propose fusion
Method, the Gauss for establishing gait and Cardiac RR interval return interaction prediction model, and the interval the RR prediction compared to single feature has
Preferable prediction effect.Disclose people during exercise, speed is faster, the more accurate rule of the prediction at the interval RR.By non-linear
Dynamic analysis and Time-Frequency Analysis Method extract gait feature, and nonlinear characteristic can preferably embody body gait during exercise
Dynamic change, time-frequency characteristics have good locality in time domain and frequency domain, and analysis result is more stable, can be well reflected letter
Number itself.
Detailed description of the invention
Fig. 1 is body flow chart of the present invention；
Fig. 2 is the single gait feature of the embodiment of the present invention and the experiment prediction result table of fusion feature；
Fig. 3 is the experiment prediction result table of five kinds of motor patterns of the embodiment of the present invention；
Specific embodiment
In order to be more clear technical solution of the present invention, with reference to the accompanying drawings and examples, the present invention is made further
Detailed description.
Parameter declaration:
GC-gait cycle；ApEn-approximate entropy；SL-step-length SampEn-Sample Entropy；SH-step is high；FuzzyEn-mould
Paste entropy；SW-step width CLZ- LZ complexity；SV-leg speed；CC0- C0 complexity；SA-leg speed acceleration PSD-power spectrum
Degree；FAmax- sufficient drift angle maximum value；CWTCmorlet- answer morlet wavelet conversion coefficient modulus value；FAmin- sufficient drift angle minimum value；
CWTCgaus- answer Cgaus wavelet conversion coefficient modulus value；SV-leg speed；CWTmexhThe coefficient modulus value of-Mexican hat wavelet transform；
SA-leg speed acceleration；The warm set of eigenvectors of F-；GPR-RQ-Quadratic Rational kernel function Gaussian process returns；GPR-
The Gaussian process of SE-squares of exponential kernel functions returns；The Gaussian process of GPR-M-Matern5/2 kernel function returns；BP-is preceding
Present neural network；BT-boosted tree；RF-random forest.
As shown in Figure 1, gait-Cardiac RR interval correlating method overall flow of the Gauss recurrence for the embodiment of the present invention
Figure, includes the following steps:
Step 1. obtains the movement three-dimensional coordinate data of human body lower limbs foot mark point and corresponds to ECG signal；
The acquisition of movement three-dimensional coordinate data is tracked by Codamotion three-dimensional real time motion capture system, is passed through
WiFi is connect with computer, records body lower extremity movement position under the sample frequency using gait analysis software setting 100Hz.?
Measured's foot it is left and right with elbow lever, left and right followed by low-level, left elbow and right elbow, 8 marks are placed in the 5th left and right toe position
Note point.It is detected by the Coda CX1 unit being placed in laboratory work space to cover range of operation.The acquisition of ECG signal
By TrignoTM wireless system record electrocardiogram (ECG) activity, the electrocardiogram in 1 channel is acquired, sample frequency is
1000Hz, and it is synchronous with the physical feeling kinematics coordinate data of 100Hz.
Five kinds of motor pattern schemes include: that (1) comfortably walks: participant improves treadmill speed, until reaching comfortable row
Speed is walked, and with this speed walking 2min；(2) walking: participant increases or decreases treadmill speed, until 4km/h, and with
This speed walking 2min；(3) maximum speed is walked: participant improves speed, until its maximum walking speed, and with this
Speed walking 1min；(4) walk and run: after rest, the speed of treadmill is increased to maximum walking speed, walking by participant
Then 30s runs 1min at a same speed；(5) maximum speed is run: participant improves speed, until the limit that he runs, and
2min is carried out with the speed.17 healthy volunteers participate in data record, and all participants are not diagnosed any disease,
It and is normal type, and be all keen on sports.All subjects use same equipment, identical label sets and same experiment
Condition.
Step 2. analyzes the movement three-dimensional coordinate data of real-time capture, extracts gait feature；
By quantitative analysis, nonlinear dynamic analysis, Time-Frequency Analysis Method, extraction time-distance and angle parameter, non-
Totally 17 kinds of gait features, constitutive characteristic vector propose fusion method for linear character, time-frequency characteristics；Specifically include following methods:
(1) firstly, removing the three-dimensional motion data of noise as obtained by step 1, the mark point at human body lower limbs position is selected
It is analyzed.We extract each gait cycle with the gait data of heel direction of advance, and are with every ten gait cycles
Data are divided into multiple subsets by a subset.
(2) quantitative analysis often obtains one group of gait parameter using pertinent instruments, such as time-space distance parameter, including
With the measurement of gait and speed-related parameter, kinematics is related to the relevant parameter of kinematic geometry, such as joint angles, and then completes
Analysis to gait.We by lower limb three-dimensional motion data carry out quantitative analysis, extract gait cycle (GC), step-length (SL),
Walk height (SH), step width (SW), leg speed (SV), leg speed acceleration (SA), foot drift angle maximum value (FAmax) and minimum value (FAmin), then
The mean value of each subset is taken to be characterized.Character representation is specific as follows:
Gait cycle (GC) describes to follow from support leg contacts ground to same heel in the initial position on ground for the second time
Motion process.Longitudinal linear distance between a little is continuously contacted with when heel successively lands when step-length (SL) is walking or so.Often
The maximum height of heel lifts indicates stride height (SH) during walking period.When stride width then refers to people's walking, two
Lateral distance (SW) between heel.Sufficient angle refers to angle formed by center line and direction of advance through side vola.Leg speed
(SV) refer to that the average speed of human locomotion, leg speed acceleration (SA) are the mean changes of heel direction of advance speed in the unit time
Amount.
(3) each subset of three-dimensional motion data of heel propulsion is extracted using the method for nonlinear dynamic analysis again
Approximate entropy (ApEn), Sample Entropy (SampEn), fuzzy entropy (FuzzyEn), LZ complexity (CLZ), C0 complexity characteristics (CC0)；
Method is specific as follows:
Approximate entropy (ApEn) is the Nonlinear Dynamic of a kind of regularity for the fluctuation of quantization time sequence and unpredictability
Mechanics parameter reflects a possibility that new information occurs in time series, and the bigger entropy the more complicated.Then the value of approximate entropy can pass through
Following steps acquire:
1) by gait time sequence { xiForm m n dimensional vector n in order, i.e.,
X (i)=[x (i), x (i+1) ..., x (i+m-1)], wherein i=1,2 ..., N-m+1
2) difference maximum one in the two corresponding element, i.e., defining X (i) distance d [X (i), X (j)] between X (j) is
3) number and total vector by given threshold value r (r > 0), to each i Data-Statistics d [X (i), X (j)] < r
The ratio of number N-m+1, is denoted as
4) defined function Φm(i)
5) defining algorithm relevant parameter m=2, r=0.2*SD, SD is standard deviation.Calculate the approximate entropy of the sequence are as follows:
ApEn (m, r, N)=Φm(r)-Φm+1(r)
Sample Entropy (SampEn) is a kind of new time series Complexity Measurement method.Sample Entropy is algorithmically relative to close
Like the improvement of entropy algorithm, the value range of j is [1, N-m+1], but j ≠ i.
Sample estimates entropy are as follows: SampEn (m, r, N)=- ln [Bm+1(r)/Bm(r)]
We are chosen in the method in actual calculating, take m=2, r=0.2*SD
Fuzzy entropy (FuzzyEn) is a kind of method of measure time sequence regularity, is faintly defined in FuzzyEn
The similitude of vector.FuzzyEn possesses the good characteristic of SampEn entropy, and independent, distribution together random number and periodicity are just
String signal shows that FuzzyEn is capable of the regularity of more effectively time of measuring sequence.Its algorithm description is as follows:
1) for given N point sample time-series [x (1), x (2) ..., x (N)]
2) it defines phase space dimension m (m≤N-2) and similar tolerance r, phase space reconstruction is as follows: X (i)=[x (i), x (i
+1),…,x(i+m-1)]-x0(i), i=1,2 ..., N-m+1
Wherein,
3) ambiguity functionIt is exponential function, calculation formula is as follows:
For vector X (i), by the maximum absolute distance between X (i) and X (j)It is expressed as follows:
4) to each i'sAveraged obtains
5) defined function
6) therefore, FuzzyEn is indicated are as follows: FuzzyEn (m, r, N)=ln Φm(r)-lnΦm+1(r)
Lempel-Ziv complexity (CLZ) reflect a time series and the speed of new model occur with the growth of its length.
Complexity value is bigger, and the new change for illustrating that data occur at any time within length of window period is more, and the rate of new change occurs
It is faster, show that the data variation in this period is unordered and complicated；Conversely, complexity is smaller, then illustrate that new change occurs
Rate is slower, data variation be it is regular, periodically it is stronger.Specific algorithm is as follows:
1) using binarization method to sequence X={ X1,X2,…,XnCoarse processing is carried out, form " 0-1 " sequence P=
{S1,S2,…,Sn}.
2) to obtained " 0-1 " sequence above, new character strings therein are successively retrieved.New character strings need to meet uniqueness
And continuity, and will be separated with one " ", wherein the retrieving of new character strings is as follows in detail:
3) partial character string of sequence P, i.e. S={ S are indicated with S1,S2,…,Sr}(r≤n)；The substring of S, i.e. Q are indicated with Q
=| Q1,Q2,…,Qm|；The tandem compound of S, Q, i.e. SQ={ S, Q } are indicated with SQ
The resulting character string of the last character for leaving out SQ is indicated with SQ π；Substring obtained in SQ π is indicated with V (SQ π)
Set；The number of different substrings is indicated with d (n).For given sequence, P={ S1,S2,…,Sn, when beginning, take S=S1, Q=
S2, SQ π=S1, d (n)=1.Under normal circumstances,
Enable S={ S1,S2,…,Sr(r=2,3 ..., n-1)；So SQ π={ S1,S2,…,Sr(r=2,3 ..., n-
1),
Q=Sr+1, judge Q whether a substring for being S, if Q belongs to V (SQ π), Q is SQ π
A substring rather than new substring, S remain unchanged, enable Q=Sr+1Sr+2, only until Q is not belonging to V (SQ π), that is, have
Q={ Sr+1,Sr+2,…,Sr+iIt is not belonging to SQ π={ S1,S2,…,Sr+i-1, d (n) plus 1, once repeats the above process straight at this time
To a last character.The number for being divided into the character string of section with " " is calculated, defines complexity d (n).
4) complexity is normalized；In order to obtain the complexity independent of sequence length, need d (n) normalizing
Change.It was found that all symbol sebolic addressing complexity d (n) tend to a determining value, it may be assumed that
Wherein n is sequence length, and a is the number (" 0-1 " sequence a=2) of kinds of characters in character string
D (n) is normalized in b (n),
5) show that normalized LZ complexity is as follows:
C0 complexity (CC0) algorithm idea is mainly that the gait time argument sequence that will analyze is decomposed, it is divided into random
The part and sequence of rules part of sequence, when being defined as doing the gait of random motion part to the complexity prediction of sequence
Between argument sequence length and area ratio that all original gait time argument sequence length surrounds between time shaft respectively.It should
Method is specific as follows:
1) the gait time argument sequence x (t) for being N for a given length, wherein t ∈ 0,1,2 ..., N-1；To it
Carry out discrete Fourier transform:
I is imaginary unit in formula,
2) mean-square value of f (k) is found out:
3) a normal number δ greater than 1 is defined, the frequency spectrum in f (k) sequence being more than δ times of mean-square value is retained, it is on the contrary
Then its zero setting is handled, the sequence after being converted:
4) carrying out inverse Fourier transform to above formula can obtain:
WhereinIt is defined according to C0 product complexity theory, C0 complicated dynamic behaviour value can be obtained:
(4) method for finally using time frequency analysis, the power spectral density of each subset is calculated by Fast Fourier Transform (FFT)
Maximum value is calculated by three kinds of wavelet functions (Cmorlet3-3, Cgaus1, mexh) of continuous wavelet transform the three of scale 2
Kind wavelet module value characteristics of mean (CWTcmorlet、CWTcgaus、CWTmexh).Time-Frequency Analysis Method is specific as follows:
Fast Fourier Transform (FFT): the amplitude as obtained by calculating the Fourier transformation of each subset carries out power spectral density
(PSD) calculating forms feature vector using power spectral density maximum value as a feature.Fast Fourier Transform (FFT) is to calculate
The fast algorithm of discrete Fourier transform.The definition of DFT are as follows:
In all negative exponent ωN=e(-2πi)/NValue all calculated it is good in the case where, to calculate an X (k) and need n times multiple
Number multiplication and N-1 complex addition.Calculate whole N point X (k) needs N altogether2Secondary complex multiplication and N (N-1) secondary complex addition.
Power spectral density (PSD) calculation formula are as follows: P=y.*conj (y)/N.
Continuous wavelet transform: the multi-scale refinement point to signal is realized by the flexible peaceful in-migration to wavelet basis function
Analysis, its main feature is that statistical nature of the data in terms of part can be protruded sufficiently.Since wavelet transformation has good property, such as line
Property additive property, translation invariance, flexible co-variation, self-similarity, can more fine-characterization using these properties.F (t) indicates one
Continuous signal is tieed up, then the fundamental formular of continuous wavelet transform are as follows:
Wherein, CWT (f, a, b) is wavelet conversion coefficient of the signal function f (t) in scale a, position b, ψ*It is certain to meet
The conjugate function of the wavelet function ψ of condition.Become by using the small echo of tri- kinds of wavelet functions of morlet3-3, C-gaus1, mexh
It changes and calculates each subset in the wavelet module value (CWT of scale 2cmorlet、CWTcgaus、CWTmexh)。
(5) finally 17 kinds of Fusion Features of extraction are formed feature vector F by us.
The pretreatment of step 3.ECG signal and the interval RR are extracted；
(1) it is handled by the bandpass filter that cascade low pass and high-pass filter form, inhibits baseline drift and T
It is interfered caused by wave；The difference equation of its step low-pass and high-pass filter is respectively as follows:
Y (nT)=2y (nT-T)-y (nT-2T)+x (nT) -2x (nT-6T)+x (nT-12T)
Y (mT)=32x (mT-16T)-[y (mT-T)+x (mT)-x (mT-32T)]
Wherein, T is the sampling period, and x (nT) indicates the input of discrete-time system electrocardiosignal time series, and y (nT) is
Electrocardiosignal after low-pass filtering, cutoff frequency are set as 12Hz, gain 2, and processing delay is about 6 sampling periods.High pass
The cutoff frequency of filter is about 5Hz, the delay in 32,16 sampling periods of gain.
(2) it by bandpass filter treated electrocardiosignal, is handled using derivative filter, QRS negative slope is provided
Information；Its difference equation are as follows:
Y (mT)=(1/8T) [- x (mT-2T) -2x (mT-T)+2x (mT+T)+x (mT+2T)]
(3) through derivative filter, treated that output using non-linear squares function achievees the effect that nonlinear amplification, by force
Adjust high frequency characteristics, the difference equation of non-linear squares function are as follows:
Y (mT)=[x (mT)]2
(4) signature waveform of envelope shape is generated, by moving window integrator (MVI) processing to further enhance feature
With the differentiation of noise.Its difference equation are as follows:
Y (mT)=(1/N) [x (mT- (N-1) T)+x (mT- (N-2) T+ ...+x (mT)]
Wherein, the width of moving integration window should be roughly the same with QRS complex as wide as possible, believes with original electrocardiographicdigital
Number sampling period it is relevant.In our current research, the sample frequency 1000Hz of electrocardiosignal is acquired, i.e. 1000 samples of acquisition per second
This point, therefore the sample points of in general N are about 150 (the real-time width period of i.e. corresponding window is 0.15s).
(5) make threshold value and other parameters carry out periodical adjustment, and detect the position of R wave or QRS wave.Finally to smooth
The R wave output detected in signal is analyzed, and is checked twice the output of bandpass signal, and detection essence is improved
Degree, eventually finds the original index of true R wave.
If the ECG signal after Integral Processing is Y (n), ECG signal is trained by following formula, is obtained just
Beginning threshold values, it may be assumed that
Wherein, using 2s as the training period of initialization threshold values, then n=2fs, fs are the sample frequency of ECG signal, THRsig
Signal amplitude Initial Hurdle, THRnoiseTo interfere Initial Hurdle.In order to enable initial threshold has more reasonability, the number of winning the confidence X's (n)
The sample data of preceding 20s, and it is divided into 10 groups of carry out initial threshold value operations, 10 threshold values are calculated according to above-mentioned formula, then go
Except the maximum value and minimum value in 10 groups, in case situations such as causing the spike noise being likely to occur to interfere causes threshold value excessive or mistake
Error caused by small.Finally again remaining 8 values are averaged to obtain THRsigAnd THRnoiseIt is initial respectively as signal amplitude
Threshold value and interference initial threshold.If the signal peak detected is PEAK, if PEAK > THRsig, then position corresponding to the peak value
The QRS wave as pre-selection is set, and obtains an estimation signal level value LEVsig。
The estimation signal level LEV of updatesigElectricity LEV is interfered with estimationnoise, threshold values is adjusted, formula is as follows:
THRsig=α LEVsig+β·LEVnoise
Wherein, α, β are that the weighted factor of threshold values adjustment contribution takes α=0.25, β=0.75 by many experiments.In QRS
R wave is detected in wave group, and the time for calculating current two neighboring R peak is the interval RR, is averaged.
Step 4. constructs the Gaussian process between gait feature and Cardiac RR interval and returns (GPR) interaction prediction model；
By step 3 and the resulting Cardiac RR interval of step 4 and fused gait feature vector, by fused gait
Feature vector establishes height between the two as desired output sample set as input sample collection, corresponding Cardiac RR spaced features
This process regression model.
4.1 Gaussian processes (GP) principle
Gaussian process refers to that the random process of a normal state, the Joint Distribution of the limited variable of any dimension obey Gauss point
Cloth.For arbitrary finite x, i.e. x1,x2,…xn∈ N, corresponding limited stochastic variable F=(f (x1),f(x2),…f
(xn))T, remember F~GP (m (xi), K), i.e.,
Wherein, F is Joint Distribution；Mean function and covariance function (kernel function) are expressed as follows:
M (x)=E [f (x)]
K=E [(f (x)-m (x)) (f (x ')-m (x '))]
The foundation of 4.2GPR regression model
When establishing Gauss regression model, consider that target value y contains noise, i.e., is defined as:
Y=f (x)+ε.Wherein: f (x) is Gaussian process,Training set are as follows:
D:yi=f (xi)+εi(i=1,2 ..., n)
It needs in test set:Predict that corresponding output valve f*'s is polynary
Gaussian Profile are as follows:
Wherein, K*=[k (x*, x1),k(x*,x2),…,k(x*,xn)], K**=k (x*, x*), σnFor the standard of noise
Difference.
According to how far the condition of Gaussian Profile, Gaussian process forecast of regression model equation can be obtained:
(f* | X, y, X*)~GP (m (x*), cov (f*))
In formula: matrix X by training set input xiColumn vector groups at；Matrix X* by test set inputColumn vector
Composition.
Surveying the gait prediction cardiac electrical interval RR regression model indicates are as follows:
Wherein: ω ∈ RmIndicate weight vector, mean function m (x)=E [f (x)], kernel function k (x, x ')=E [(f (x)-
m(x))(f(x′)-m(x′))]。
The selection of 4.3 kernel functions
The selection of kernel function is most important for Gaussian process, covariance function (the i.e. core in Gaussian process regression model
Function) it must satisfy Mercer condition.Present invention employs 3 kinds of kernel functions:
Quadratic Rational kernel function (RQ):
Square exponential kernel functions (SE)
Matern kernel function (M)
Wherein, M=diag (l-2), l is variance measure,For signal variance.L is that relevance measures hyper parameter；δijFor symbol
Number function is 1 when i and j is identical, is not simultaneously 0；α is the form parameter of kernel function.
DefinitionFor hyper parameter vector, optimal solution is adaptively obtained by maximum-likelihood method.It initially sets up
The negative log-likelihood function of training sample conditional probability, and it is enabled to seek local derviation to hyper parameter, then using conjugate gradient method to this
Partial derivative is minimized, and hyper parameter optimal solution is obtained.
Step 5. training and test Gaussian process regression model, are associated forecast analysis；
Stochastic inputs training sample set and test sample collection accounting are 4:1, the three kinds of kernel functions returned using Gauss: reasonable
Secondary kernel function (GPR-RQ), square exponential kernel functions (GPR-SE), Matern5/2 kernel function (GPR-M), to five kinds of sports grounds
Scape is associated forecast analysis, takes its average value for test result ten times.
It is evaluated by using accuracy and validity of the root-mean-square error (RMSE) to prediction model, RMSE is got over
Small, prediction result is better.Its expression formula is as follows:
Wherein, n indicates the number of prediction data, yiIt is actual prediction data value, tiDesired output is represented,Represent the phase
Hope output mean value.
It is illustrated in figure 2 the single feature of the embodiment of the present invention and the experiment prediction result table of fusion feature, is tied by experiment
Known to fruit: in GPR-RQ, GPR-SE, GPR-M Gauss regression forecasting of three kernel functions, the Gaussian process of Matern kernel function
Regression model (GPR-M) is preferable to 17 kinds of single test effects of gait feature.Its fusion feature vector F is pre- compared to single feature
It is more preferable to survey effect, there is lower root-mean-square error.Wherein, the gait parameter that quantitative analysis is extracted has leg speed (SV), leg speed to add
Speed (SA), gait cycle (GC) have preferable prediction effect；The fuzzy entropy (FuzzyEn) and approximate entropy of nonlinear characteristic
(ApEn) and the Mexican hat wavelet transform coefficient modulus value (CWT of time-frequency characteristicsmexh) all there is preferable prediction effect, it can
Preferable reflection gait signal itself.And can be seen that by experimental result and be compared to traditional prediction technique, Gauss returns
The precision of prediction of method is higher.It is illustrated in figure 3 the experiment prediction result table of five kinds of motor patterns of the embodiment of the present invention.To five
In the prediction of kind moving scene, GPR-M method is also demonstrated by compared to feedforward BP neural network, boosted tree (BT) and random forest
(RF) three kinds of algorithms comfortable walking, walking, maximum speed are walked, walked and run, the prediction result for five kinds of motor patterns of running compared with
It is good.And with the increase of the speed of travel, the result for showing its prediction is more accurate, to disclose gait and Cardiac RR interval
High relevancy.It can be seen that the Generalization Capability of Gauss homing method is more preferable by above-mentioned experimental result, precision of prediction is higher, can be more preferable
Disclose the related information between gait and electrocardio in ground.
Claims (5)
1. gait-Cardiac RR interval the correlating method returned based on Gauss, it is characterised in that include the following steps:
Step 1. obtains the movement three-dimensional coordinate data of human body lower limbs foot mark point and corresponds to ECG signal；
Using three-dimensional motion tracking system and electrocardiogram acquisition equipment, when synchronization catch subject's lower limb foot mark point moves three
Tie up coordinate data and corresponding synchronous ECG signal；
Step 2. analyzes the movement three-dimensional coordinate data of real-time capture, extraction time-distance and angle parameter, non-linear
Totally 17 kinds of gait features, constitutive characteristic vector propose fusion method for feature, time-frequency characteristics；
The pretreatment of step 3.ECG signal and the interval RR are extracted；
Step 4. constructs the Gaussian process between gait feature and Cardiac RR interval and returns (GPR) interaction prediction model；
By step 3 and the resulting Cardiac RR interval of step 4 and fused gait feature vector, by fused gait feature
For vector as input sample collection, corresponding Cardiac RR spaced features establish Gauss mistake between the two as desired output sample set
Journey regression model；
Step 5. training and test Gaussian process regression model, are associated forecast analysis；
Stochastic inputs training sample set and test sample collection accounting are 4:1, using three kinds of kernel functions: Rational Quadratic kernel function
(RQ), square exponential kernel functions (SE), Matern5/2 kernel function (MA), are associated forecast analysis, and test result ten times are taken
Its average value.
2. gait-Cardiac RR interval the correlating method according to claim 1 returned based on Gauss, it is characterised in that will walk
The three-dimensional motion data of rapid 1 gained removal noise, select the mark point at human body lower limbs position to be analyzed；With heel direction of advance
Action data extract each gait cycle, and using every ten gait cycles as a subset, data are divided into multiple sons
Collection；
Extract the when m- distance parameter and angle parameter of gait, including gait cycle, step-length, step height, step width, leg speed, leg speed
Acceleration, sufficient drift angle maximum value and minimum value, take the average value of each subset；It is mentioned again using the method for nonlinear dynamic analysis
Take the nonlinear characteristic of each subset of three-dimensional motion data of heel propulsion, including approximate entropy, Sample Entropy, fuzzy entropy, LZ
Complexity, C0 complexity characteristics；The method for finally using time frequency analysis calculates power spectral density most by Fast Fourier Transform (FFT)
Big value calculates the small echo in scale 2 by tri- kinds of wavelet functions of C-morlet3-3, C-gaus1, mexh of continuous wavelet transform
Coefficient modulus value characteristics of mean；Extracted feature is constituted into feature vector, proposes fusion method.
3. gait-Cardiac RR interval the correlating method according to claim 2 returned based on Gauss, it is characterised in that step
3 are implemented as follows:
3-1. is handled by the bandpass filter that cascade low pass and high-pass filter form, and inhibits baseline drift and T wave
Caused by interfere；The difference equation of its step low-pass and high-pass filter is respectively as follows:
Y (nT)=2y (nT-T)-y (nT-2T)+x (nT) -2x (nT-6T)+x (nT-12T)
Y (mT)=32x (mT-16T)-[y (mT-T)+x (mT)-x (mT-32T)]
Wherein, T is the sampling period, and x (nT) indicates the input of discrete-time system electrocardiosignal time series, and y (nT) is low pass
Filtered electrocardiosignal, cutoff frequency are set as 12Hz, gain 2, and processing delay is about 6 sampling periods；High-pass filtering
The cutoff frequency of device is about 5Hz, the delay in 32,16 sampling periods of gain；
3-2. is handled by bandpass filter treated electrocardiosignal using derivative filter, provides QRS negative slope letter
Breath；Its difference equation are as follows:
Y (mT)=(1/8T) [- x (mT-2T) -2x (mT-T)+2x (mT+T)+x (mT+2T)]
Treated that output using non-linear squares function achievees the effect that nonlinear amplification through derivative filter by 3-3., emphasizes height
Frequency characteristic, the difference equation of non-linear squares function are as follows:
Y (mT)=[x (mT)]2
3-4. by moving window integrator (MVI) processing generate envelope shape signature waveform, come further enhance feature with
The differentiation of noise；Its difference equation are as follows:
Y (mT)=(1/N) [x (mT- (N-1) T)+x (mT- (N-2) T+ ...+x (mT)]
Wherein, if the sample frequency 500Hz of acquisition electrocardiosignal, i.e. 500 sample points of acquisition per second, the sample points of N are 75
It is a；
3-5. makes threshold value and other parameters carry out periodical adjustment, and detects the position of R wave or QRS wave；Finally to smooth letter
The R wave output detected in number is analyzed, and is checked twice the output of bandpass signal, and true R wave is eventually found
Original index.
4. gait-Cardiac RR interval the correlating method according to claim 3 returned based on Gauss, it is characterised in that step
4 are implemented as follows:
4.1 Gaussian processes (GP) principle
Gaussian process refers to the random process of a normal state, the Joint Distribution Gaussian distributed of the limited variable of any dimension；It is right
In arbitrary finite x, i.e. x1,x2,…xn∈ N, corresponding limited stochastic variable F=(f (x1),f(x2),…f(xn)
)T, remember F~GP (m (xi), K), i.e.,
Wherein, F is Joint Distribution；Mean function and covariance function (kernel function) are expressed as follows:
M (x)=E [f (x)]
K=E [(f (x)-m (x)) (f (x ')-m (x '))]
The foundation of 4.2 GPR regression models
When establishing Gauss regression model, consider that target value y contains noise, i.e., is defined as:
Y=f (x)+ε；Wherein: f (x) is Gaussian process,Training set are as follows:
D:yi=f (xi)+εi(i=1,2 ..., n), in test set:Prediction pair
The output valve f answered*Multivariate Gaussian distribution are as follows:
Wherein, K*=[k (x*,x1),k(x*,x2),…,k(x*,xn)], K**=k (x*,x*), σnFor the standard deviation of noise；According to
How far the condition of Gaussian Profile, Gaussian process forecast of regression model equation can be obtained:
(f*|X,y,X*)~GP (m (x*),cov(f*))
In formula: matrix X by training set input xiColumn vector groups at；Matrix X*By the input of test setColumn vector groups at；
Surveying the gait prediction cardiac electrical interval RR regression model indicates are as follows:
Wherein: ω ∈ RmIndicate weight vector, mean function m (x)=E [f (x)], kernel function k (x, x ')=E [(f (x)-m (x))
(f(x′)-m(x′))]；
The selection of 4.3 kernel functions
Using 3 kinds of kernel functions:
Quadratic Rational kernel function (RQ):
Square exponential kernel functions (SE)
Matern kernel function (M)
Wherein, M=diag (l-2), l is variance measure,For signal variance；L is that relevance measures hyper parameter；δijFor symbol letter
Number is 1 when i and j is identical, is not simultaneously 0；α is the form parameter of kernel function；
DefinitionFor hyper parameter vector, optimal solution is adaptively obtained by maximum-likelihood method；Initially set up training
The negative log-likelihood function of sample conditions probability, and it is enabled to seek local derviation to hyper parameter, then using conjugate gradient method to the local derviation
Number is minimized, and hyper parameter optimal solution is obtained.
5. gait-Cardiac RR interval the correlating method according to claim 4 returned based on Gauss, it is characterised in that step
5 are implemented as follows:
Stochastic inputs training sample set and test sample collection accounting are 4:1, using three kinds of kernel functions: Rational Quadratic kernel function
(RQ), square exponential kernel functions (SE), Matern5/2 kernel function (MA), are associated forecast analysis, and test result ten times are taken
Its average value；
By using root-mean-square error (RMSE) and determine coefficient (R2) accuracy and validity of prediction model are evaluated,
Its RMSE is smaller, and prediction result is better；Its expression formula is as follows:
Wherein, n indicates the number of prediction data, yiIt is actual prediction data value, tiRepresent desired output.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910521097.3A CN110236523A (en) | 2019-06-17 | 2019-06-17 | Gait-Cardiac RR interval the correlating method returned based on Gauss |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910521097.3A CN110236523A (en) | 2019-06-17 | 2019-06-17 | Gait-Cardiac RR interval the correlating method returned based on Gauss |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110236523A true CN110236523A (en) | 2019-09-17 |
Family
ID=67887499
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910521097.3A Withdrawn CN110236523A (en) | 2019-06-17 | 2019-06-17 | Gait-Cardiac RR interval the correlating method returned based on Gauss |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110236523A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110710976A (en) * | 2019-10-14 | 2020-01-21 | 深圳市迈步机器人科技有限公司 | Storage medium for personalized gait generation method, control system and exoskeleton robot |
| CN111797540A (en) * | 2020-07-17 | 2020-10-20 | 浙江工业大学 | Timely online Gaussian process regression model method for crystal size distribution in crystallization process |
| CN113009368A (en) * | 2021-02-24 | 2021-06-22 | 武汉科技大学 | MFF-based multi-core GPR algorithm-based lithium battery residual life prediction method |
| CN113094934A (en) * | 2021-05-10 | 2021-07-09 | 重庆大学 | Pipeline corrosion depth prediction method driven by physical model and data fusion |
| CN114397521A (en) * | 2021-12-24 | 2022-04-26 | 中国人民解放军海军航空大学 | Fault diagnosis method and system for electronic equipment |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101807245A (en) * | 2010-03-02 | 2010-08-18 | 天津大学 | Artificial neural network-based multi-source gait feature extraction and identification method |
| CN102270345A (en) * | 2011-06-02 | 2011-12-07 | 西安电子科技大学 | Image feature representing and human motion tracking method based on second-generation strip wave transform |
| CN103854268A (en) * | 2014-03-26 | 2014-06-11 | 西安电子科技大学 | Image super-resolution reconstruction method based on multi-core gaussian process regression |
| CN104299003A (en) * | 2014-11-02 | 2015-01-21 | 天津理工大学 | Gait recognition method based on similar rule Gaussian kernel function classifier |
| CN105411593A (en) * | 2014-09-12 | 2016-03-23 | 三星电子株式会社 | Method and apparatus for recognizing gait task |
| CN106529499A (en) * | 2016-11-24 | 2017-03-22 | 武汉理工大学 | Fourier descriptor and gait energy image fusion feature-based gait identification method |
| CN107403084A (en) * | 2017-07-21 | 2017-11-28 | 中国计量大学 | A kind of personal identification method based on gait data |
| CN107622260A (en) * | 2017-10-26 | 2018-01-23 | 杭州电子科技大学 | Lower limb gait phase identification method based on multi-source bio signal |
| CN109009143A (en) * | 2018-07-12 | 2018-12-18 | 杭州电子科技大学 | A method of ecg information is predicted by body gait |
| CN109446884A (en) * | 2018-09-06 | 2019-03-08 | 杭州电子科技大学 | A kind of gait recognition method that the speed based on nonlinear kinetics index is unrelated |
| CN109558834A (en) * | 2018-11-28 | 2019-04-02 | 福州大学 | A kind of multi-angle of view gait recognition method based on similarity study and kernel method |
| CN109635754A (en) * | 2018-12-17 | 2019-04-16 | 四川师范大学 | Gait feature fusion method based on Non-linear coupling metric learning |
| CN109717833A (en) * | 2018-11-26 | 2019-05-07 | 中国科学院软件研究所 | A kind of neurological disease assistant diagnosis system based on human motion posture |
-
2019
- 2019-06-17 CN CN201910521097.3A patent/CN110236523A/en not_active Withdrawn
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101807245A (en) * | 2010-03-02 | 2010-08-18 | 天津大学 | Artificial neural network-based multi-source gait feature extraction and identification method |
| CN102270345A (en) * | 2011-06-02 | 2011-12-07 | 西安电子科技大学 | Image feature representing and human motion tracking method based on second-generation strip wave transform |
| CN103854268A (en) * | 2014-03-26 | 2014-06-11 | 西安电子科技大学 | Image super-resolution reconstruction method based on multi-core gaussian process regression |
| CN105411593A (en) * | 2014-09-12 | 2016-03-23 | 三星电子株式会社 | Method and apparatus for recognizing gait task |
| CN104299003A (en) * | 2014-11-02 | 2015-01-21 | 天津理工大学 | Gait recognition method based on similar rule Gaussian kernel function classifier |
| CN106529499A (en) * | 2016-11-24 | 2017-03-22 | 武汉理工大学 | Fourier descriptor and gait energy image fusion feature-based gait identification method |
| CN107403084A (en) * | 2017-07-21 | 2017-11-28 | 中国计量大学 | A kind of personal identification method based on gait data |
| CN107622260A (en) * | 2017-10-26 | 2018-01-23 | 杭州电子科技大学 | Lower limb gait phase identification method based on multi-source bio signal |
| CN109009143A (en) * | 2018-07-12 | 2018-12-18 | 杭州电子科技大学 | A method of ecg information is predicted by body gait |
| CN109446884A (en) * | 2018-09-06 | 2019-03-08 | 杭州电子科技大学 | A kind of gait recognition method that the speed based on nonlinear kinetics index is unrelated |
| CN109717833A (en) * | 2018-11-26 | 2019-05-07 | 中国科学院软件研究所 | A kind of neurological disease assistant diagnosis system based on human motion posture |
| CN109558834A (en) * | 2018-11-28 | 2019-04-02 | 福州大学 | A kind of multi-angle of view gait recognition method based on similarity study and kernel method |
| CN109635754A (en) * | 2018-12-17 | 2019-04-16 | 四川师范大学 | Gait feature fusion method based on Non-linear coupling metric learning |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110710976A (en) * | 2019-10-14 | 2020-01-21 | 深圳市迈步机器人科技有限公司 | Storage medium for personalized gait generation method, control system and exoskeleton robot |
| CN111797540A (en) * | 2020-07-17 | 2020-10-20 | 浙江工业大学 | Timely online Gaussian process regression model method for crystal size distribution in crystallization process |
| CN113009368A (en) * | 2021-02-24 | 2021-06-22 | 武汉科技大学 | MFF-based multi-core GPR algorithm-based lithium battery residual life prediction method |
| CN113094934A (en) * | 2021-05-10 | 2021-07-09 | 重庆大学 | Pipeline corrosion depth prediction method driven by physical model and data fusion |
| CN114397521A (en) * | 2021-12-24 | 2022-04-26 | 中国人民解放军海军航空大学 | Fault diagnosis method and system for electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110236523A (en) | Gait-Cardiac RR interval the correlating method returned based on Gauss | |
| Yang et al. | Using PPG signals and wearable devices for atrial fibrillation screening | |
| Hou et al. | A real-time QRS detection method based on phase portraits and box-scoring calculation | |
| CN102961203B (en) | Method for identifying surface electromyography (sEMG) on basis of empirical mode decomposition (EMD) sample entropy | |
| Li et al. | Design of a continuous blood pressure measurement system based on pulse wave and ECG signals | |
| CN109446972B (en) | Gait recognition model establishing method, recognition method and device based on electromyographic signals | |
| Temko | Estimation of heart rate from photoplethysmography during physical exercise using Wiener filtering and the phase vocoder | |
| CN110464349A (en) | A kind of upper extremity exercise function score method based on hidden Semi-Markov Process | |
| CN108523881A (en) | A kind of accurate method for extracting abnormal potential in QRS | |
| CN114305364A (en) | Blood pressure detection method, system and equipment based on millimeter wave radar | |
| Jean Effil et al. | Wavelet scattering transform and long short-term memory network-based noninvasive blood pressure estimation from photoplethysmograph signals | |
| CN110215202A (en) | The pre- measuring/correlation method in Cardiac RR interval based on gait nonlinear characteristic | |
| CN113609975A (en) | Modeling method for tremor detection, hand tremor detection device and method | |
| CN117064410A (en) | Military physical training risk assessment method and system based on deep learning | |
| Veer | A flexible approach for segregating physiological signals | |
| Gordon et al. | PVC detection using a convolutional autoencoder and random forest classifier | |
| Hossain et al. | Analysis of Noise Sensitivity of Different ECG Detection Algorithms. | |
| CN114145725A (en) | PPG sampling rate estimation method based on noninvasive continuous blood pressure measurement | |
| Kocer | Classifying myopathy and neuropathy neuromuscular diseases using artificial neural networks | |
| CN114469041A (en) | Heart rate change data characteristic analysis method in exercise process | |
| Veer | Spectral and mathematical evaluation of electromyography signals for clinical use | |
| Zhang et al. | Computerized radial artery pulse signal classification for lung cancer detection | |
| CN114209281B (en) | Pulse data generation method based on generation countermeasure network | |
| Zhang et al. | Estimation model for blood pressure based on clustering and gradient boosting | |
| Apandi et al. | Noise Reduction Method based on Autocorrelation for Threshold-Based Heartbeat Detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190917 |
|
| WW01 | Invention patent application withdrawn after publication |