CN108245172A - It is a kind of not by the human posture recognition method of position constraint - Google Patents
It is a kind of not by the human posture recognition method of position constraint Download PDFInfo
- Publication number
- CN108245172A CN108245172A CN201810022045.7A CN201810022045A CN108245172A CN 108245172 A CN108245172 A CN 108245172A CN 201810022045 A CN201810022045 A CN 201810022045A CN 108245172 A CN108245172 A CN 108245172A
- Authority
- CN
- China
- Prior art keywords
- feature
- data
- sensor
- posture
- model
- 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.)
- Granted
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
- A61B5/1123—Discriminating type of movement, e.g. walking or running
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Veterinary Medicine (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Physiology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Dentistry (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Signal Processing (AREA)
- Psychiatry (AREA)
- Navigation (AREA)
Abstract
The invention discloses a kind of not by the human posture recognition method of position constraint, attitude data is acquired using sensor, coordinate transform is carried out by Data Fusion of Sensor, sensing data is resolved to terrestrial coordinate system, extraction sensing station correlated characteristic and posture correlated characteristic are fastened in terrestrial coordinates;Sensing station identification model is established based on position feature, realizes the identification of sensing station;Gesture recognition model is established based on sensor position and posture correlated characteristic, realizes the identification of human body attitude.The present invention relieves constraint of the sensor to position using sensing data coordinate transform and sensing station identification, relieves dependence of the model to training sample using Increment Learning Algorithm, improves model generalization ability.
Description
Technical field
The present invention relates to gesture recognition technical field, more particularly to a kind of not by the human body attitude identification side of position constraint
Method.
Background technology
With the continuous development of medical treatment & health technology and Intelligent hardware, the health supervision technology of human body is more and more ripe.People
The daily posture of body is to evaluate an important indicator of its health status.The precisely daily posture of identification human body, to the health of human body
Monitoring is of great significance, and especially highlights in terms of the health supervision of the elderly.In daily life, the elderly is at one's side not necessarily
Constantly there is special caregiver, if identifying a series of postures of the elderly by intelligent terminal, and pass through and wirelessly pass
The human body attitude for embodying its animation, position and other health and fitness informations are sent to monitor center, community hospital by the modes such as defeated
Or the related personnel such as relatives, the timely processing for the accidents such as can both ensure to fall, wander away, can also be used modern technologies reduce it is old
The monitoring pressure of year people.On the other hand, also it can pass through digging by the new and high technologies such as medical data base and artificial intelligence and big data
Dig the relationship between the daily posture and its health status of human body, establish correlation model realize the detection of real-time health and
People and the disease forecasting of social groups.
Gesture recognition method mainly has human body attitude identification and the identification of sensor-based human body attitude based on camera
Two kinds.Human body attitude identification technology based on camera can recognize that fine human body attitude, but be fixed since there are privacy, places
Etc. restraining factors, can be only applied to the fixed-sites such as family, home for destitute, and the identification of sensor-based human body attitude can be effective
The deficiencies of camera place is avoided to fix, and have the characteristics that committed memory is small, equipment is simple, environmental suitability is strong.At present,
Mainly calculate corresponding threshold value by parameters such as height, weight judges people to sensor-based human body attitude identification technology again
The posture of body, this method can accurately identify less posture, relatively easy to environmental requirement, once environment is complicated or requirement is known
Other posture type increases, and this kind of method will be unable to meet demand.Meanwhile sensor-based conventional body's gesture recognition method
It is required that sensor is fixed on to the specific position of human body, and gesture recognition algorithms are established in the laying of the sensor based on fixed position,
Lead to gesture recognition algorithms dependent on sensor placement, algorithm does not have versatility and has to the wearing position of sensor stringent
It is required that.Based on this, research and develop under complex environment, do not constrained by sensing station and can identify the detection technique of a variety of human body attitudes
With important theory significance and application value.
Invention content
To solve the problems such as existing gesture recognition model is limited there are sensor by position constraint and gesture recognition classification,
The present invention gives a kind of not by the human posture recognition method of position constraint.This method is based on Fusion and seat
Mark transformation, the position identification model for initially setting up sensor obtain its position, then utilize sensor position and appearance
State correlated characteristic establishes human body attitude identification model, so as to efficiently solve sensor by position constraint and gesture recognition classification by
The problems such as limit, relieves dependence of the identification model to sensing station, enhances the robustness and universality of model.
It is a kind of not by the human posture recognition method of position constraint, including:
Step (1):Sensor is placed in the different location of human body, and it is different to acquire the human body under different location state
The sensing data of posture;
Step (2):The sensing data of acquisition is pre-processed, carries out coordinate transform later, data are transformed into ground
Spherical coordinate system;
Step (3):Feature extraction is carried out to the data after coordinate transform;By the feature of extraction respectively using position and posture as
Label carries out character subset screening, obtains the position feature subset of characterization sensing station and posture feature of characterization posture
Collection;
Step (4):The position feature subset obtained using step (3), establishes position identification model;Mould is identified using position
Position where type identification sensor;
Step (5):Gesture recognition mould is established with the human body attitude character subset that sensor position and step (3) obtain
Type is identified the posture of human body to be measured using gesture recognition model.
Further, the step of step (1) includes:Sensor is placed in the different location of human body, and is acquired
The sensing data of different postures under different location state;The sensor includes:Accelerometer, gyroscope and magnetometer;
For the sensor integration in mobile terminal, the data of the sensor acquisition are uploaded to controller processing;The different location,
Including:In trouser pocket, coat pocket, hand or in packet;The difference posture, including:Sit, lie, going upstairs, going downstairs, on elevator, walk
Road, running, cycling or jump.
Further, the pretreatment of the step (2), refers to:Low-pass filtering is carried out to data using low-pass filter, is disappeared
Except High-frequency Interference.
Further, the data to acquisition of the step (2) carry out coordinate transform, and data are transformed into terrestrial coordinates
System, including:
Step (201):Initialize quaternary number, acquisition acceleration, gyroscope and magnetometer data;
Step (202):Acceleration information and magnetometer data are normalized;
Step (203):Using the normalized acceleration information of step (202) and magnetometer data, calculate gravity direction and
Magnetic force direction;
Step (204):The gravity direction and magnetic force direction data obtained using step (203), is estimated corresponding error, obtained
Sum to overall error for the two error;
Step (205):The overall error obtained using step (204) is calculated using complementary filter method and corrects error;
Step (206):The amendment error obtained using step (205) is modified gyro data, after obtaining amendment
Gyro data;
Step (207):The revised gyro data obtained using step (206), using single order runge kutta method more
New quaternary number;
Step (208):The quaternary number obtained using step (207), calculates corresponding spin matrix;
Step (209):The spin matrix obtained using step (208) carries out acceleration, gyroscope and magnetometer data
Coordinate transform obtains terrestrial coordinates and fastens corresponding acceleration, gyroscope and magnetometer data;
Step (210):Acceleration, gyroscope and the magnetometer data obtained using step (209), uses Kalman filtering
Device filters, and eliminates noise jamming, obtains corresponding data.
Further, the feature extraction of the step (3), refers to:
Step (301):Setting time window length is 2s, and each time window is overlapped 50%;
Step (302):To each time series, temporally window extracts time domain, frequency domain and time and frequency domain characteristics;
The time series includes:
The terrestrial coordinate system x-axis acceleration signal ax, y-axis acceleration signal ay, z-axis acceleration signal azAccelerate with closing
Degrees of data a;
The terrestrial coordinate system x-axis gyro data ωx, y-axis gyro data ωyWith z-axis gyro data ωz;
The terrestrial coordinate system x-axis magnetometer data mx, y-axis magnetometer data myWith z-axis magnetometer data mz;
The feature of the extraction includes:
Temporal signatures, including:Mean value Mean, variance Std, related coefficient Corr, quartile spacing Iqr and intermediate value
Median;
Frequency domain character, including:Frequency domain energy Energy_Fre, frequency domain entropy Entropy_Fre, degree of bias SK and kurtosis KU;
Time and frequency domain characteristics, including:Low-frequency band energy LF and ultralow frequency energy band VLF.
Further, the step (3):Character subset screening is carried out to extraction feature,
Step (31):The weight of each feature is calculated using ReliefF algorithms, if feature weight is less than the first setting threshold
Value, then it is assumed that current signature is extraneous features, deletes extraneous features;
Step (32):Related coefficient between feature two-by-two is calculated by correlation coefficient process, if related coefficient is higher than the
Two given thresholds and feature weight are less than third given threshold, then it is assumed that current signature is redundancy feature, deletes redundancy feature;
Step (33):The coefficient of alienation between feature two-by-two is calculated using mutual information method, retains nonlinear factor and is higher than
4th given threshold and feature weight are less than the 5th given threshold, then it is assumed that current signature is redundancy feature, deletes redundancy feature;
Respectively using sensor position and posture as label, screened by the character subset of step (31)-step (33)
Step, obtain position feature subset relevant with sensing station and with the relevant human body attitude character subset of human body attitude.
Further, the step (4):Position identification model is established using sensing station character subset,
Sensing station character subset is divided into test set and training set;
By the use of training set as the input of decision tree, training decision tree;Obtain position identification model;
Further, the step (5) establishes gesture recognition model, refers to:
The position of the sensor identified by the use of step (4) and posture feature subset are built as the input of gesture recognition model
Standing position state identification model;
Posture feature subset and sensor position are divided into training set and several increment sample sets;
At the beginning of training set as support vector machines (Support Vector Machine, SVM) Incremental Learning Algorithm
Begin to input, training obtains initial attitude model;
By the use of increment sample set as input, initial attitude model is constantly updated, obtains gesture recognition model.
The beneficial effects of the invention are as follows:By accurate acquisition, Data Fusion of Sensor and the coordinate transform to attitude data,
Sensor position is identified, and realize the identification of human body attitude, had not by sensor using machine learning algorithm
The limitation and influence of position, accuracy is high and identifies the features such as posture type is more, can improve the applicable model of gesture recognition
It encloses and accuracy.
Description of the drawings
The accompanying drawings which form a part of this application are used for providing further understanding of the present application, and the application's shows
Meaning property embodiment and its explanation do not form the improper restriction to the application for explaining the application.
Fig. 1 is the realization flow of the present invention;
Fig. 2 is human body sensor different location schematic diagram;
Fig. 3 is illustrated for coordinate transform;
Fig. 4 is coordinate transformation algorithm flow chart;
Fig. 5 is characterized subset filtering algorithm flow chart;
Fig. 6 is sensing station-gesture recognition flow;
Fig. 7 is gesture recognition model incremental learning process figure;
Wherein, 1, handbag, 2, trouser pocket, 3, coat pocket, 4, hand.
Specific embodiment
It is noted that following detailed description is all illustrative, it is intended to provide further instruction to the application.It is unless another
It indicates, all technical and scientific terms used herein has usual with the application person of an ordinary skill in the technical field
The identical meanings of understanding.
The present invention provides a kind of not by the human posture recognition method of position constraint, and overall flow algorithm is as shown in Figure 1.It calculates
The main process of method is divided into the acquisition of sensing data and pretreatment, sensing data coordinate transform, feature extraction, character subset
Screening, the identification of sensing station and human body attitude identification etc., implementation step is as follows:
Step 1:The acquisition and pretreatment of sensing data;
(1) sensor includes accelerometer, gyroscope and magnetometer, and setting sample frequency is 32Hz, ensures the appearance of human body
State frequency is all in sample frequency;The placement of sensor is worn not by position constraint, can be placed on trouser pocket 2, coat pocket
3, in hand 4 or handbag 1 the inside, as shown in Figure 2;
(2) there are high-frequency noises in sensing data, it is filtered using low-pass filter, eliminates the influence of noise.
Step 2:Sensing data coordinate transform;
When sensing station is not fixed, the directions of the x-axis, y-axis and z-axis of sensor also disunity.Therefore quaternary is utilized
Number resolves spin matrix, then all resolves sensing data to terrestrial coordinate system, as shown in figure 3, coordinate X ', Y ' and Z ' tables
Show the coordinate system where sensor, E (X), N (Y), Z represent terrestrial coordinate system, respectively east, north, day direction.As shown in figure 4,
Sensing data coordinate transform is as follows:
(1) system initialization quaternary number obtains the acceleration, gyroscope and magnetometer data of pretreatment:
Quaternion Method is a kind of method for describing posture, is made of a real number and three imaginary numbers.If quaternary number is
q0、q1、q2And q3, can be expressed as:
WhereinIt is three directions perpendicular to each other;
(2) acceleration and magnetic flux normalization:
A in formulaiThe acceleration information of (i=x, y, z) for sensor x-axis, y-axis and z-axis, ai *(i=x, y, z) is normalization
The acceleration information of x-axis, y-axis and z-axis afterwards, miThe magnetometer data of (i=x, y, z) for sensor x-axis, y-axis and z-axis, mi *(i
=x, y, z) for normalization after x-axis, y-axis and z-axis magnetometer data;
(3) flux component is calculated:
bz=hz
H in formulax、hy、hzRepresent the flux component of sensor x-axis, y-axis and z-axis, bxAnd bzFor horizontal space and vertical sky
Between upper magnetic flux component;
(4) estimate gravity and magnetic flux direction:
υx=2 (q1q3-q0q2)
υy=2 (q0q1+q2q3)
μy=2bx(q1q2-q0q3)+2bz(q0q1+q2q3)
υ in formulax、υy、υzIt is sensor in the size of x-axis, y-axis and z-axis gravity direction, μx、μy、μzFor sensor x-axis,
Y-axis and the size in z-axis direction;
(5) accelerometer and magnetometer can generate corresponding error, and total error is the sum of the two error:
E in formulax、eyAnd ezRepresent the error that acceleration and magnetometer generate in x-axis, y-axis and z-axis direction;
(6) error and directly proportional to the integral error of gyroscope uses complementary filter method to correct gyroscope angle speed
Spend integral error:
E in formulaxint、eyintAnd ezintFor the integral term initial value of gyroscope correction amount,WithFor gyroscope
The integral term of correction amount, kiFor integral coefficient, t is the time interval of attitude data twice;
(7) correction of gyroscope angular speed:
ω in formulax、ωy、ωzFor initial angular velocity signal,For revised angular speed, kpFor ratio system
Number;
(8) quaternary number is updated:
Quaternary number is updated with single order runge kutta method:
In formulaWithRepresent updated quaternary number;
(9) spin matrix is resolved:
Quaternary number is the method for describing posture, and spin matrix can be represented with quaternary number:
It (10) will be in carrier coordinate system data calculation to terrestrial coordinates axis:
X in formulax、Xy、XzRepresent the data of acceleration, gyroscope and magnetometer sensor,It represents to accelerate
Degree, gyroscope and magnetometer data rotate to the data in terrestrial coordinate system east, north, day direction;
(11) Kalman filtering:
In actual test, although being corrected with acceleration and magnetometer to gyroscope, in the data of host computer
There are still noise jammings, are filtered using Kalman filter.
Step 3:Sensor characteristics are extracted and character subset screening;
Extraction feature is fastened, and the screening of character subset is carried out to it in the terrestrial coordinates, content includes:
(1) setting time window length is 2s, and each time window is overlapped 50%;
(2) to each time series, temporally window extracts a series of statistics/physical ones, when feature mainly includes
The feature in domain, frequency domain and time-frequency domain;
The time series includes:
The terrestrial coordinate system x-axis acceleration signal ax, y-axis acceleration signal ay, z-axis acceleration signal azAccelerate with closing
Degrees of data a:
The terrestrial coordinate system x-axis gyro data ωx, y-axis gyro data ωyWith z-axis gyro data ωz;
The terrestrial coordinate system x-axis magnetometer data mx, y-axis magnetometer data myWith z-axis magnetometer data mz;
The feature of the extraction includes:
Temporal signatures include:Mean value Mean, variance Std, related coefficient Corr, quartile spacing Iqr and intermediate value Median;
Frequency domain character includes:Frequency domain energy Energy_Fre, frequency domain entropy Entropy_Fre, degree of bias SK and kurtosis KU;
Time and frequency domain characteristics include:Low-frequency band energy LF and ultralow frequency energy band VLF
Step (4):Character subset screens;
There are unrelated and redundancy feature between the feature of the characterization sensing station and posture, to reduce calculation amount, section
About operation cost, the present invention remove unrelated and redundancy feature using assemblage characteristic subset screening technique.The assemblage characteristic subset
Screening technique includes ReliefF, correlation coefficient process and mutual information method, as shown in figure 5, idiographic flow is as follows:
(1) feature weight is calculated:The weight of each feature is calculated using ReliefF algorithms, if feature weight is less than
2%, then it is assumed that be characterized in extraneous features, delete this feature;During ReliefF algorithm process multi-class problems, every time from sample with
Machine takes out a sample R, and k neighbour's sample of R is then found out from sample similar with R samples, is looked for from each R differences sample
Go out k sample, then constantly update feature weight:
Diff (A, R in formula1,R2) represent sample R1And R2Difference on feature A, Mj(C) it representsIn j-th
Nearest samples;
(2) correlation coefficient process:Since ReliefF algorithms can only calculate the weight of each feature, redundancy feature can not be removed,
Therefore using linearly related method and mutual information method removal redundancy feature.Linearly related method is used for representing the linear correlation between feature
Property, the related coefficient of feature between any two is calculated using correlation coefficient process, is retained in feature of the linearly dependent coefficient higher than 70%
The larger feature of weight deletes redundancy feature.Using mean value linearly dependent coefficient method is gone, formula is the present invention:
(3) mutual information method:The coefficient of alienation of feature between any two is calculated using mutual information method, retains nonlinear correlation system
The larger feature of weight, deletes redundancy feature in feature of the number higher than 90%;Mutual information weighs the journey that two variables interdepend
Degree, represents jointly owned information content between two variables, and the mutual information between stochastic variable X and Y can be expressed as:
P (x), p (y) represent the edge distribution of X and Y in formula, and p (x, y) represents the joint probability distribution function of X and Y;
Step (5):Sensing station identifies;
Sensor is in body different location, having differences property of attitude data, in order to build not by the human body of position constraint
Gesture recognition model, the present invention devise a kind of Position-Attitude recognition methods, the identification sensor position first before gesture recognition
It puts, as shown in Figure 6;
The present invention chooses decision tree (Decision Tree, DT) as decision-making device, identification sensor position, model
Establishment step is as follows:
(1) attitude data of acquisition is divided into test set and training set;
(2) by the use of training set as the input of decision tree, training decision tree;
(3) beta pruning is carried out to the model of foundation, prevents its over-fitting;
(4) by the use of the decision tree established as grader, by the use of test set as inputting, test model generalization ability.
Step (6):Human body attitude identifies;
The present invention establishes human body attitude identification model using the Incremental Learning Algorithm based on SVM, i.e., by acquiring appearance offline
State data establish initial model, then model is constantly updated with incremental data, implementation model incremental learning mistake using SVM algorithm
Journey;The thought of SVM algorithm is to meet the optimal separating hyper plane of classificating requirement by finding so that in the case where precision is optimal
The white space of optimal hyperlane both sides is maximum;
SVM algorithm can only solve linear two classification problem, to nonlinear problem, by selecting radial basis function (Radial
BasisFunction, RBF) as kernel function, nonlinear problem is converted into higher dimensional space linear problem;To more classification problems
Multiple two classification problems can be converted into, and then the identification of many attitude is realized as assembled classifier;
Using grid optimizing algorithm, to SVM parameters, (penalty factor c, RBF kernel functional parameter g) carries out optimizing;Based on SVM's
There are two advantages for Incremental Learning Algorithm:It need not save historical data first, reduce memory space, next takes full advantage of history instruction
Practice as a result, save the subsequent training time, improve nicety of grading;
As shown in fig. 7, attitude mode establishment step is as follows:
(1) training set and increment sample are splitted data into;
(2) using RBF kernel functions
K(xi,xj)=exp (- γ | xi-xj| 2), and γ > 0
(3) grid optimizing obtains optimal c and g;
(4) it is trained using optimal c and g, obtains initial modelWith supporting vector collection SV1;
(5) model is utilizedTraining increment sample, obtains fault detection sample S;
(6) by SV1Training set re -training is done with S, obtains new graderWith supporting vector SV2;
(7) with SV1And SV2Union as new training set, divide mistake sample re -training, until there is no misclassification sample
Until;
Obtain final graderWith supporting vector collection SV.
The foregoing is merely the preferred embodiments of the application, are not limited to the application, for the skill of this field
For art personnel, the application can have various modifications and variations.It is all within spirit herein and principle, made any repair
Change, equivalent replacement, improvement etc., should be included within the protection domain of the application.
Claims (9)
1. it is a kind of not by the human posture recognition method of position constraint, it is characterized in that, including:
Step (1):Sensor is placed in the different location of human body, and acquires the human body difference posture under different location state
Sensing data;
Step (2):The sensing data of acquisition is pre-processed, carries out coordinate transform later, data are transformed into the earth sits
Mark system;
Step (3):Feature extraction is carried out to the data after coordinate transform;By the feature of extraction respectively using position and posture as label
Character subset screening is carried out, obtains the position feature subset of characterization sensing station and the posture feature subset of characterization posture;
Step (4):The position feature subset obtained using step (3), establishes position identification model;Known using position identification model
Position where individual sensor;
Step (5):Gesture recognition model is established with the human body attitude character subset that sensor position and step (3) obtain,
The posture of human body to be measured is identified using gesture recognition model.
2. it is as described in claim 1 a kind of not by the human posture recognition method of position constraint, it is characterized in that, the step
(1) the step of, includes:Sensor is placed in the different location of human body, and acquires the different postures under different location state
Sensing data;The sensor includes:Accelerometer, gyroscope and magnetometer;The sensor integration in mobile terminal,
The data of the sensor acquisition are uploaded to controller processing;The different location, including:In trouser pocket, coat pocket, hand or wrap
In;The difference posture, including:Sit, lie, going upstairs, going downstairs, on elevator, walk, run, cycle or jump.
3. it is as described in claim 1 a kind of not by the human posture recognition method of position constraint, it is characterized in that, the step
(2) pretreatment, refers to:Low-pass filtering is carried out to data using low-pass filter, eliminates High-frequency Interference.
4. it is as described in claim 1 a kind of not by the human posture recognition method of position constraint, it is characterized in that, the step
(3) feature extraction, refers to:
Step (301):Setting time window length is 2s, and each time window is overlapped 50%;
Step (302):To each time series, temporally window extracts time domain, frequency domain and time and frequency domain characteristics.
5. it is as claimed in claim 4 a kind of not by the human posture recognition method of position constraint, it is characterized in that, the time sequence
Row include:
The terrestrial coordinate system x-axis acceleration signal ax, y-axis acceleration signal ay, z-axis acceleration signal azWith resultant acceleration number
According to a;The terrestrial coordinate system x-axis gyro data ωx, y-axis gyro data ωyWith z-axis gyro data ωz;
The terrestrial coordinate system x-axis magnetometer data mx, y-axis magnetometer data myWith z-axis magnetometer data mz。
6. it is as claimed in claim 4 a kind of not by the human posture recognition method of position constraint, it is characterized in that, the extraction
Feature includes:
Temporal signatures, including:Mean value Mean, variance Std, related coefficient Corr, quartile spacing Iqr and intermediate value Median;
Frequency domain character, including:Frequency domain energy Energy_Fre, frequency domain entropy Entropy_Fre, degree of bias SK and kurtosis KU;
Time and frequency domain characteristics, including:Low-frequency band energy LF and ultralow frequency energy band VLF.
7. it is as described in claim 1 a kind of not by the human posture recognition method of position constraint, it is characterized in that, the step
(3):Character subset screening is carried out to extraction feature,
Step (31):The weight of each feature is calculated using ReliefF algorithms, if feature weight is less than the first given threshold,
Then think that current signature for extraneous features, deletes extraneous features;
Step (32):Related coefficient between feature two-by-two is calculated by correlation coefficient process, if related coefficient is set higher than second
Determine threshold value and feature weight is less than third given threshold, then it is assumed that current signature is redundancy feature, deletes redundancy feature;
Step (33):The coefficient of alienation between feature two-by-two is calculated using mutual information method, retains nonlinear factor and is higher than the 4th
Given threshold and feature weight are less than the 5th given threshold, then it is assumed that current signature is redundancy feature, deletes redundancy feature;
Respectively using sensor position and posture as label, by the character subset screening step of step (31)-step (33),
Obtain position feature subset relevant with sensing station and with the relevant human body attitude character subset of human body attitude.
8. it is as described in claim 1 a kind of not by the human posture recognition method of position constraint, it is characterized in that, the step
(4):Position identification model is established using sensing station character subset,
Sensing station character subset is divided into test set and training set;
By the use of training set as the input of decision tree, training decision tree;Obtain position identification model.
9. it is as described in claim 1 a kind of not by the human posture recognition method of position constraint, it is characterized in that, the step
(5) establish gesture recognition model, refer to:
The position of the sensor identified by the use of step (4) and posture feature subset establish appearance as the input of gesture recognition model
State identification model;
Posture feature subset and sensor position are divided into training set and several increment sample sets;
By the use of training set as the initial input of SVM Incremental Learning Algorithms, training obtains initial attitude model;
By the use of increment sample set as input, initial attitude model is constantly updated, obtains gesture recognition model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810022045.7A CN108245172B (en) | 2018-01-10 | 2018-01-10 | Human body posture recognition method free of position constraint |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810022045.7A CN108245172B (en) | 2018-01-10 | 2018-01-10 | Human body posture recognition method free of position constraint |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108245172A true CN108245172A (en) | 2018-07-06 |
CN108245172B CN108245172B (en) | 2020-04-07 |
Family
ID=62725670
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810022045.7A Active CN108245172B (en) | 2018-01-10 | 2018-01-10 | Human body posture recognition method free of position constraint |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108245172B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109635638A (en) * | 2018-10-31 | 2019-04-16 | 中国科学院计算技术研究所 | For the feature extracting method and system of human motion, recognition methods and system |
CN109883431A (en) * | 2019-02-21 | 2019-06-14 | 北京方位捷讯科技有限公司 | One kind walking appearance judgment method and system |
CN110299203A (en) * | 2019-08-12 | 2019-10-01 | 电子科技大学 | A kind of human body abnormal gait recognition methods based on smart phone |
CN110501008A (en) * | 2019-07-29 | 2019-11-26 | 西安电子科技大学 | The autonomous evolvement method of modelling of human body motion |
CN111345783A (en) * | 2020-03-26 | 2020-06-30 | 山东大学 | Vestibular dysfunction detection system based on inertial sensor |
CN111657918A (en) * | 2020-06-12 | 2020-09-15 | 电子科技大学 | Falling detection method and system combining electrocardio and inertial sensing data |
CN111839527A (en) * | 2020-07-31 | 2020-10-30 | 高新兴物联科技有限公司 | Fall detection method and device and computer readable storage medium |
CN112268521A (en) * | 2020-09-30 | 2021-01-26 | 西安理工大学 | Variable-angle synchronous phase shift interferometry method for gear tooth surface shape error |
CN115884120A (en) * | 2022-11-22 | 2023-03-31 | 南方医科大学珠江医院 | AD patient dynamic posture recognition system based on wireless signal |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101587546A (en) * | 2009-06-12 | 2009-11-25 | 浙江大学 | Identification method of lower limb action pattern based on supporting vector multivariate classification |
US20120218263A1 (en) * | 2009-10-12 | 2012-08-30 | Metaio Gmbh | Method for representing virtual information in a view of a real environment |
CN102779281A (en) * | 2012-06-25 | 2012-11-14 | 同济大学 | Vehicle type identification method based on support vector machine and used for earth inductor |
WO2015165260A1 (en) * | 2014-04-29 | 2015-11-05 | 华南理工大学 | Triaxial feature fusion method for human body movement identification |
CN105678222A (en) * | 2015-12-29 | 2016-06-15 | 浙江大学 | Human behavior identification method based on mobile equipment |
CN107092882A (en) * | 2017-04-19 | 2017-08-25 | 南京大学 | A kind of Activity recognition system and its method of work perceived based on sub- action |
-
2018
- 2018-01-10 CN CN201810022045.7A patent/CN108245172B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101587546A (en) * | 2009-06-12 | 2009-11-25 | 浙江大学 | Identification method of lower limb action pattern based on supporting vector multivariate classification |
US20120218263A1 (en) * | 2009-10-12 | 2012-08-30 | Metaio Gmbh | Method for representing virtual information in a view of a real environment |
CN102779281A (en) * | 2012-06-25 | 2012-11-14 | 同济大学 | Vehicle type identification method based on support vector machine and used for earth inductor |
WO2015165260A1 (en) * | 2014-04-29 | 2015-11-05 | 华南理工大学 | Triaxial feature fusion method for human body movement identification |
CN105678222A (en) * | 2015-12-29 | 2016-06-15 | 浙江大学 | Human behavior identification method based on mobile equipment |
CN107092882A (en) * | 2017-04-19 | 2017-08-25 | 南京大学 | A kind of Activity recognition system and its method of work perceived based on sub- action |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109635638A (en) * | 2018-10-31 | 2019-04-16 | 中国科学院计算技术研究所 | For the feature extracting method and system of human motion, recognition methods and system |
CN109635638B (en) * | 2018-10-31 | 2021-03-09 | 中国科学院计算技术研究所 | Feature extraction method and system and recognition method and system for human body motion |
CN109883431A (en) * | 2019-02-21 | 2019-06-14 | 北京方位捷讯科技有限公司 | One kind walking appearance judgment method and system |
CN110501008A (en) * | 2019-07-29 | 2019-11-26 | 西安电子科技大学 | The autonomous evolvement method of modelling of human body motion |
CN110501008B (en) * | 2019-07-29 | 2021-03-26 | 西安电子科技大学 | Autonomous evolution method of human motion model |
CN110299203A (en) * | 2019-08-12 | 2019-10-01 | 电子科技大学 | A kind of human body abnormal gait recognition methods based on smart phone |
CN111345783A (en) * | 2020-03-26 | 2020-06-30 | 山东大学 | Vestibular dysfunction detection system based on inertial sensor |
CN111345783B (en) * | 2020-03-26 | 2021-02-23 | 山东大学 | Vestibular dysfunction detection system based on inertial sensor |
CN111657918A (en) * | 2020-06-12 | 2020-09-15 | 电子科技大学 | Falling detection method and system combining electrocardio and inertial sensing data |
CN111839527A (en) * | 2020-07-31 | 2020-10-30 | 高新兴物联科技有限公司 | Fall detection method and device and computer readable storage medium |
CN112268521A (en) * | 2020-09-30 | 2021-01-26 | 西安理工大学 | Variable-angle synchronous phase shift interferometry method for gear tooth surface shape error |
CN112268521B (en) * | 2020-09-30 | 2022-06-14 | 西安理工大学 | Variable-angle synchronous phase shift interferometry method for gear tooth surface shape error |
CN115884120A (en) * | 2022-11-22 | 2023-03-31 | 南方医科大学珠江医院 | AD patient dynamic posture recognition system based on wireless signal |
Also Published As
Publication number | Publication date |
---|---|
CN108245172B (en) | 2020-04-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108245172A (en) | It is a kind of not by the human posture recognition method of position constraint | |
Gupta et al. | A survey on human activity recognition and classification | |
CN106384093B (en) | A kind of human motion recognition method based on noise reduction autocoder and particle filter | |
CN107219924B (en) | A kind of aerial gesture identification method based on inertial sensor | |
Hu et al. | A human body posture recognition algorithm based on BP neural network for wireless body area networks | |
US20130311134A1 (en) | Method for step detection and gait direction estimation | |
US10126108B2 (en) | Apparatus and method for classifying orientation of a body of a mammal | |
CN110236550B (en) | Human gait prediction device based on multi-mode deep learning | |
CN108509897A (en) | A kind of human posture recognition method and system | |
CN106096662A (en) | Human motion state identification based on acceleration transducer | |
Meghji et al. | An algorithm for the automatic detection and quantification of athletes’ change of direction incidents using IMU sensor data | |
CN108958482B (en) | Similarity action recognition device and method based on convolutional neural network | |
CN107588784A (en) | A kind of state recognition and the method, apparatus and terminal distinguished | |
Ahmad et al. | Human activity recognition using multi-head CNN followed by LSTM | |
Yang et al. | Human posture recognition in intelligent healthcare | |
Hnoohom et al. | Multi-sensor-based fall detection and activity daily living classification by using ensemble learning | |
CN107368790A (en) | Pedestrian detection method, system, computer-readable recording medium and electronic equipment | |
Wang et al. | A2dio: Attention-driven deep inertial odometry for pedestrian localization based on 6d imu | |
Lu et al. | MFE-HAR: multiscale feature engineering for human activity recognition using wearable sensors | |
Dhanraj et al. | Efficient smartphone-based human activity recognition using convolutional neural network | |
CN106971203A (en) | Personal identification method based on characteristic on foot | |
Ghobadi et al. | A robust automatic gait monitoring approach using a single IMU for home-based applications | |
Mekruksavanich et al. | Wearable-based activity recognition of construction workers using lstm neural networks | |
CN116465412A (en) | Improved PDR indoor positioning method based on LSTM and attention mechanism | |
Li et al. | [Retracted] Tracking of Gymnast’s Limb Movement Trajectory Based on MEMS Inertial Sensor |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |