CN109934179A - Human motion recognition method based on automated characterization selection and Ensemble Learning Algorithms - Google Patents
Human motion recognition method based on automated characterization selection and Ensemble Learning Algorithms Download PDFInfo
- Publication number
- CN109934179A CN109934179A CN201910203015.0A CN201910203015A CN109934179A CN 109934179 A CN109934179 A CN 109934179A CN 201910203015 A CN201910203015 A CN 201910203015A CN 109934179 A CN109934179 A CN 109934179A
- Authority
- CN
- China
- Prior art keywords
- data
- feature
- human
- learning algorithms
- human action
- 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
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000012512 characterization method Methods 0.000 title claims abstract description 21
- 238000012952 Resampling Methods 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 15
- 230000002123 temporal effect Effects 0.000 claims abstract description 12
- 238000011156 evaluation Methods 0.000 claims description 16
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000003595 spectral effect Effects 0.000 claims description 4
- 230000001133 acceleration Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 7
- 238000010801 machine learning Methods 0.000 description 6
- 238000011160 research Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 210000003423 ankle Anatomy 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000002354 daily effect Effects 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000003203 everyday effect Effects 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 210000000707 wrist Anatomy 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000006806 disease prevention Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001953 sensory effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of human motion recognition methods based on automated characterization selection and Ensemble Learning Algorithms, comprising the following steps: A obtains the data information collection of human action;B carries out resampling to the data information collection of human action, constructs sample characteristics space using resampling point data, the corresponding temporal signatures of window data, the corresponding frequency domain character of window data;C, using based on Ka Te tree characteristic evaluating algorithm and Ensemble Learning Algorithms the sample data in sample characteristics space is trained, obtain trained bilayer model, wherein the bilayer model includes feature selecting layer and action recognition layer;D carries out Classification and Identification to human action with trained bilayer model, realizes automated characterization selection and human action identification.The present invention can automatically select out to the advantageous feature of model, reduce additional manual operation, improve engineering efficiency;Human action identification is carried out using Ensemble Learning Algorithms, the training time is few, and discrimination is high.
Description
Technical field
It is the present invention relates to sports science and artificial intelligence field, in particular to a kind of to select and integrate based on automated characterization to learn
Practise the human motion recognition method of algorithm.
Background technique
With the development of technology of Internet of things, wearable body-building and medical sensory device are being developed rapidly and vigorously, it can be used for
The acquisition and data sharing of human body physiological data, can be carried out data processing and analysis beyond the clouds, finally send its result
To doctor to provide diagnosis and rehabilitation suggestion.This " closed loop " can continue to carry out daily, by the everyday actions to human body
It is identified and is recorded, grasp its behavioural habits, preferably analyze daily work and rest and exercise intensity, to be established for everyone strong
The activity program of health.The technology of a most critical is exactly human action identification technology among these, it be applied to control and prevention of disease,
All various aspects such as rehabilitation training, energy consumption prediction and health guidance.
The method that traditional human action identification technology uses " Cut Point (cut-point) ", but the identification of this method
Precision is lower, and especially partially static movement, the selection accuracy of cut-point are affected to accuracy of identification.With artificial intelligence
The development of technology, machine learning method are introduced in human action identification.It is used in human action identification technology at present
More machine learning algorithm mainly has decision tree, artificial neural network, Hidden Markov Model, support vector machines etc..But
In these researchs, feature selection process and action recognition process are completely separable independent, the even basic no spies of some
Levy the process of selection.It is understood that the upper limit of data and feature meeting decision model recognition effect, and algorithm only adjusts on this
Limit, so feature selecting is most important relative to human action identification technology.In addition, in terms of action recognition model, it is existing
Research demonstrates similar decision tree or this single learner of support vector machines in machine learning, it cannot be guaranteed that any given
Human action identification problem on performance it is good, as long as and have enough data, weak learning algorithm can pass through integrated side
Formula generates any high-precision strong learning method, and research at present has been proven that the identification effect of the algorithm model of integrated study
Fruit is better than single learner.However independent feature selection process is manual operation, is all often according to personal experience and
Some results of study carry out feature selecting, may result in and filter out the redundancy feature useless to identification model, or eliminate
To the important feature that identification model is affected, this phenomenon can all influence the discrimination of identification model.If model can be certainly
It is dynamic that feature selecting is carried out to initial data, the higher feature of different degree is selected according to certain evaluation mechanism, it can not only Lifting Modules
The discrimination of type can also reduce manual operation bring inconvenience, so as to improve engineering efficiency.There is presently no use certainly
Dynamic feature selecting and the combined model of action recognition algorithm carry out the research of human action identification.
The object of the invention is to the human action datas acquired using wearable device to train one to be based on Ka Te tree
The human action identification model of characteristic evaluating algorithm and Ensemble Learning Algorithms, the model can carry out feature selecting and human body automatically
Action recognition.
Summary of the invention
In the prior art in the human motion recognition method based on machine learning algorithm, feature selection process and action recognition
Process is two separated independent processes, is all independently and artificially to carry out feature selecting, selects by subjective and experience
Feature will affect the recognition effect of model.It is an object of the present invention to aiming at the shortcomings in the prior art, provide one kind and be based on
The human motion recognition method of automated characterization selection and Ensemble Learning Algorithms, can be realized automated characterization selection and human action is known
Not, it can automatically select out to the advantageous feature of model, higher discrimination can be obtained, reduce additional manual operation, mention
High engineering efficiency.
In order to solve the above technical problems, the technical scheme adopted by the invention is that:
A kind of human motion recognition method based on automated characterization selection and Ensemble Learning Algorithms, its main feature is that including following
Step:
Step A obtains the data information collection of human action;
Step B, it is dynamic to human body in the sliding window in one sliding window of data information concentrated setting of human action
The data information collection of work carries out resampling, corresponding using resampling point data, the corresponding temporal signatures of window data, window data
Frequency domain character construct sample characteristics space;
Step C, using based on Ka Te tree characteristic evaluating algorithm and Ensemble Learning Algorithms to the sample in sample characteristics space
Notebook data is trained, and obtains trained bilayer model --- and RAFs model, wherein the bilayer model includes feature selecting layer
With action recognition layer;
Step D carries out Classification and Identification to human action with trained double-deck (RAFs) model, realizes automated characterization selection
It is identified with human action.
As a preferred method, in the step A, the data information collection process for obtaining human action includes: firstly, obtaining
The raw data set for obtaining human action (such as passes through and installs the initial data that inertial sensor acquires human action at human synovial
Collection);Then, initial data is pre-processed, to obtain the data information collection of human action.
The step B includes: as a preferred method,
In one sliding window of data information concentrated setting of human action, sliding distance is set and between the resampling time
Every to the progress resampling of data information collection in the sliding window;
Temporal signatures and frequency domain character are calculated in the sliding window;
Sample is constructed using resampling point data, the corresponding temporal signatures of window data, the corresponding frequency domain character of window data
Eigen space.
The step C includes: as a preferred method,
Step C1, if the data information of human action integrates as T, sample characteristics space is F, Characteristic Number K, with loom sky
Between factor alpha, Ka Te tree number be M, feature selection index β;
Step C2 randomly selects α K feature-modeling subcharacter space F in feature space Fsub={ f1,f2,...,fj},
(j=1,2 ..., α K);
Step C3 chooses subcharacter space FsubIn feature data information collection T is mapped, obtain new data set
Close Ti, i=1,2 ..., M, it may be assumed that
Wherein, Π represents map operation,
Step C4 uses new data acquisition system TiTraining card spy tree CTi;
Step C5, for i from 1 to M, repetition step C2~step C4 is M times total, obtains M Ka Te tree CTi;
Step C6, if feature fjNode is in Ka Te tree CTiAmong, then j-th of characteristic node is in i-th Ka Te tree
Depth is dij, then different degree evaluation of estimate E of j-th of feature in i-th Ka Te treeijAre as follows:If feature fj
Node is not in Ka Te tree CTiAmong, then Eij=0;
Step C7, for i from 1 to M, for j from 1 to K, repetition step C6 is M*K times total, obtains each feature in every Ka Te tree
In different degree evaluation of estimate Eij;
Step C8 calculates general comment value E of each characteristic node in all Ka Te trees for j from 1 to Kj:
Step C9, according to EjEach feature in the F of sample characteristics space is ranked up, it is high to select different degree evaluation of estimate
Preceding β K feature forms new feature space Fselected, Fselected={ f1,f2,…,fβK}
Step C10 chooses new feature space FselectedIn feature data information collection T is mapped, obtain new collection
Close Tselected, that is,
Wherein, Π represents map operation,
Step C11, by new data set TselectedIt is input to the second layer of model, i.e. XGBoost algorithm action recognition layer,
For the training of human action identification layer, to obtain final trained bilayer model.
As a preferred method, in the step A, preprocessing process includes: to eliminate initial data using mean filter
The noise of concentration.
As a preferred method, in the step A, preprocessing process includes: to remove most starting for initial data concentration
Preceding 10 second data and most end rear 10 second data, to remove movement preparation stage and ending phase hash.
The temporal signatures include average value, minimum value, maximum value, median, variance, mark as a preferred method,
One of the quasi- poor, degree of bias, kurtosis, zero-crossing values number, the interrelated index of acceleration reference axis are a variety of.
The frequency domain character includes one of spectral energy, dominant frequency, the corresponding amplitude of dominant frequency as a preferred method,
Or it is a variety of.
Compared with prior art, the present invention carries out feature choosing using the characteristic importance evaluation algorithms based on Ka Te tree automatically
It selects, can automatically select out to the advantageous feature of model, to avoid artificially carrying out feature selecting bring inconvenience, reduce volume
Outer manual operation improves engineering efficiency;Meanwhile human action identification, time complexity are carried out using Ensemble Learning Algorithms
Small, the training time greatly reduces, and improves the discrimination of human action.
Detailed description of the invention
Fig. 1 is the model framework figure in the present invention.
Fig. 2 is characterized selection front and back recognition effect comparison diagram.
Specific embodiment
The present invention establishes model using the Ensemble Learning Algorithms in machine learning, and model structure is as shown in Figure 1, we are referred to as
For RAFs (Human motion recognition model based on automatic feature selection) mould
Type.The model includes two layers altogether, and first layer carries out feature selecting using the characteristic evaluating algorithm based on Ka Te tree, and the second layer uses
XGBoost (eXtreme Gradient Boosting) algorithm carries out human action identification.
The present invention is based on automated characterization selection and Ensemble Learning Algorithms one embodiment of human motion recognition method include with
Lower step:
Step A obtains the data information collection of human action: firstly, obtain human action raw data set (such as by
The raw data set of inertial sensor acquisition human action is installed) at human synovial;Then, initial data is pre-processed,
To obtain the data information collection of human action.
Preprocessing process includes: to eliminate the noise that initial data is concentrated using mean filter;And remove raw data set
In preceding 10 second data most started and most end rear 10 second data, to remove movement preparation stage and ending phase nothing
Use data.
Step B, it is dynamic to human body in the sliding window in one sliding window of data information concentrated setting of human action
The data information collection of work carries out resampling, is answered using resampling point data, the corresponding temporal signatures of window data, window data
Frequency domain character constructs sample characteristics space.Wherein, for resampling data, at data information concentrated setting one of human action
The sliding window of 10 seconds length, sliding distance are 5 seconds, carry out resampling to data information collection in 10 seconds sliding windows.
Under the premise of inertial sensor acquisition data frequency is not less than 20 hertz, the time interval of resampling is set as 0.5 second, then often
A reference axis can obtain 19 resampling data at 10 seconds in window.For temporal signatures and frequency domain character, in 10 seconds cunnings
It is calculated in dynamic window.Then, corresponding using resampling point data, the corresponding temporal signatures of window data, window data
Frequency domain character constructs sample characteristics space F, F={ f1,f2,…,fn, wherein fiIt is i-th of primitive character of sample, feature is empty
Between in F each element specifying information it is as shown in table 1.
From f in table 11To f57It is the resampling point of X, Y and Z axis in 10 seconds time windows, XiIt is X-axis in 10 seconds time windows
I-th second data in mouthful, Yi、ZiWith XiIt is identical.f58To f87Represent calculated in time window at 10 seconds X, Y and Z axis when
Characteristic of field, including average value (Mean), minimum value (Min), maximum value (Max), median (Median), variance
(Variance), standard deviation (Std), the degree of bias (Skewness), kurtosis (Kurtosis), zero-crossing values number (Zc) and acceleration are sat
The interrelated index of parameter (Cac).For example, Cac (XY) refers to that X-axis and Y-axis are interrelated in certain 10 seconds time window
Index, calculation formula are as follows:
1 feature space each element information table of table
From f88To f96Represent the frequency domain character for calculating X, Y and Z axis in time window at 10 seconds, including spectral energy
(Se), dominant frequency (Df) and the corresponding amplitude of dominant frequency (Dfm), wherein the calculation formula of spectral energy is as follows:
Step C, using based on Ka Te tree characteristic evaluating algorithm and Ensemble Learning Algorithms to the sample in sample characteristics space
Notebook data is trained, and obtains trained bilayer model --- and RAFs model, wherein the bilayer model includes feature selecting layer
With action recognition layer.
Specifically, the step C includes:
Step C1, if the data information of human action integrates as T, sample characteristics space is F, Characteristic Number K, with loom sky
Between factor alpha, Ka Te tree number be M, feature selection index β;
Step C2 randomly selects α K feature-modeling subcharacter space F in feature space Fsub={ f1,f2,...,fj},
(j=1,2 ..., α K);
Step C3 chooses subcharacter space FsubIn feature data information collection T is mapped, obtain new data set
Close Ti, i=1,2 ..., M, it may be assumed that
Wherein, Π represents map operation,
Step C4 uses new data acquisition system TiTraining card spy tree CTi;
Step C5, for i from 1 to M, repetition step C2~step C4 is M times total, obtains M Ka Te tree CTi;
Step C6, if feature fjNode is in Ka Te tree CTiAmong, then j-th of characteristic node is in i-th Ka Te tree
Depth is dij, then different degree evaluation of estimate E of j-th of feature in i-th Ka Te treeijAre as follows:
If feature fjNode is not in Ka Te tree CTiAmong, then Eij=0;
Step C7, for i from 1 to M, for j from 1 to K, repetition step C6 is M*K times total, obtains each feature in every Ka Te tree
In different degree evaluation of estimate Eij;
Step C8 calculates general comment value E of each characteristic node in all Ka Te trees for j from 1 to Kj:
Step C9, according to EjEach feature in the F of sample characteristics space is ranked up, it is high to select different degree evaluation of estimate
Preceding β K feature forms new feature space Fselected, Fselected={ f1,f2,…,fβK}
Step C10 chooses new feature space FselectedIn feature data information collection T is mapped, obtain new collection
Close Tselected, that is,
Wherein, Π represents map operation,
Step C11, by new data set TselectedIt is input to the second layer of model, i.e. XGBoost algorithm action recognition layer,
For the training of human action identification layer, to obtain final RAFs model.
Step D carries out Classification and Identification to human action with trained RAFs model, realizes automated characterization selection and human body
Action recognition.
To further illustrate implementation process of the invention, beneficial effects of the present invention are now verified using following experiment:
The PAMAP2 data set that the data set that this experiment uses is increased income in UCI machine learning library, the data centralized procurement
Collect the action data of 9 volunteers, including 8 males, 1 women, average age 27.2 years old.The inertia sensing of the data set
Device is distributed in wrist (Wrist), chest (Chest) and the ankle (Ankle) of human body, and the sample frequency of sensor is 100 hertz,
18 kinds of everyday actions and compound action are acquired altogether, and the data of wherein 8 kinds of movements, specifying information such as 2 institute of table is used only in the present invention
Show.
2 experimental data information table of table
1) 2/3rds that each action data is chosen in data set are used as training set, choose the three of each action data
/ mono- is used as test set, and selection operation is to randomly select, and guarantees the data in training set and test set comprising 8 kinds of movements.
2) training set is set as T, and sample characteristics space is set as F, and sample characteristics number is 288, and stochastic subspace coefficient is
0.7, a number for feature selection index 0.7, Ka Te tree is 140.
3) i=0.
4) i=i+1.
5) in sample characteristics space, F kind randomly selects 201 features for constructing subcharacter space Fsub={ f1,f2,...,
f201}。
6) subcharacter space F is chosensubIn feature training set T is mapped, obtain new set Ti, i=1,
2 ..., 140, i.e.,Wherein ∏ represents map operation,
7) new data set T is usediTraining card spy tree CTi。
8) repeat 5), 6), 7), until i be equal to 140, obtain 140 Ka Te tree CTi。
9) i=0.
10) i=i+1.
11) j=0.
12) j=j+1.
13) the different degree evaluation of estimate of the characteristic node in Ka Te tree is calculated.If feature fjNode is in Ka Te tree CTiAmong,
Then depth of j-th of characteristic node in i-th Ka Te tree is dij, then different degree of j-th of feature in i-th Ka Te tree
Evaluation of estimate isIf feature fjNode is not in Ka Te tree CTiAmong, then Eij=0.
14) it repeats 13), is equal to 140 until j is equal to 288, i, finally obtains evaluation of each feature in every Ka Te tree
Value Eij。
15) j=0.
16) j=j+1.
17) general comment value of each characteristic node in all Ka Te trees is calculated
18) it repeats 17), until j is equal to 288, the general comment for finally obtaining each feature is worth Ej, totally 288.
19) according to Ej288 features in feature space F are ranked up, it is higher to select preceding 201 different degree evaluations of estimate
Feature form new feature space Fselected={ f1,f2,…,f201}。
20) new feature space F is chosenselectedIn feature data set T is mapped, obtain new set Tselected,
I.e.Wherein Π represents map operation,
21) by new data set TselectedIt is input to the second layer of model, i.e. XGBoost algorithm action recognition layer, is used for
The training of human action identification layer, to obtain final RAFs model, the final parameter setting of model is as shown in table 3.
Table 3RAFs model parameter sets table
The evaluation index of model of the present invention, selection use F1 value.F1 value (is called together with precision (accuracy rate) and recall
The rate of returning) it is related, calculation formula is as follows:
F1 value is higher, then the recognition effect of model is better.
RAFs model is tested using test set, the average F1 value before feature selecting is 0.9, after feature selecting
Average F1 value rise to 0.94, the results are shown in Table 4 for specific experiment.Experimental data comparison diagram such as Fig. 2 institute before and after feature selecting
Show.
The 4 RAFs aspect of model of table selection front and back experimental data table
According to the experimental data of table 4 and Fig. 2 it is found that RAFs model realization of the invention automated characterization selection, and improve
The discrimination of model, recognition effect are far superior to other models.
The embodiment of the present invention is described with above attached drawing, but the invention is not limited to above-mentioned specific
Embodiment, the above mentioned embodiment is only schematical, rather than limitation, those skilled in the art
Under the inspiration of the present invention, without breaking away from the scope protected by the purposes and claims of the present invention, it can also make very much
Form, within these are all belonged to the scope of protection of the present invention.
Claims (8)
1. it is a kind of based on automated characterization selection and Ensemble Learning Algorithms human motion recognition method, which is characterized in that including with
Lower step:
Step A obtains the data information collection of human action;
Step B, in one sliding window of data information concentrated setting of human action, to human action in the sliding window
Data information collection carries out resampling, utilizes resampling point data, the corresponding temporal signatures of window data, the corresponding frequency of window data
Characteristic of field constructs sample characteristics space;
Step C, using based on Ka Te tree characteristic evaluating algorithm and Ensemble Learning Algorithms to the sample number in sample characteristics space
According to being trained, trained bilayer model is obtained, wherein the bilayer model includes feature selecting layer and action recognition layer;
Step D carries out Classification and Identification to human action with trained bilayer model, realizes automated characterization selection and human action
Identification.
2. the human motion recognition method as described in claim 1 based on automated characterization selection and Ensemble Learning Algorithms, special
Sign is, in the step A, the data information collection process for obtaining human action includes: firstly, obtaining the original number of human action
According to collection;Then, initial data is pre-processed, to obtain the data information collection of human action.
3. the human motion recognition method as described in claim 1 based on automated characterization selection and Ensemble Learning Algorithms, special
Sign is that the step B includes:
In one sliding window of data information concentrated setting of human action, sliding distance and resampling time interval are set,
Resampling is carried out to data information collection in the sliding window;
Temporal signatures and frequency domain character are calculated in the sliding window;
It is special using resampling point data, the corresponding temporal signatures of window data, the corresponding frequency domain character building sample of window data
Levy space.
4. the human motion recognition method as described in claim 1 based on automated characterization selection and Ensemble Learning Algorithms, special
Sign is that the step C includes:
Step C1, if the data information of human action integrates as T, sample characteristics space is F, Characteristic Number K, stochastic subspace system
A number of number α, Ka Te tree are M, feature selection index β;
Step C2 randomly selects α K feature-modeling subcharacter space F in feature space Fsub={ f1,f2,...,fj, (j=
1,2,...,αK);
Step C3 chooses subcharacter space FsubIn feature data information collection T is mapped, obtain new data acquisition system Ti, i
=1,2 ..., M, it may be assumed that
Wherein, Π represents map operation,
Step C4 uses new data acquisition system TiTraining card spy tree CTi;
Step C5, for i from 1 to M, repetition step C2~step C4 is M times total, obtains M Ka Te tree CTi;
Step C6, if feature fjNode is in Ka Te tree CTiAmong, then depth of j-th of characteristic node in i-th Ka Te tree
For dij, then different degree evaluation of estimate E of j-th of feature in i-th Ka Te treeijAre as follows:If feature fjNode
Not in Ka Te tree CTiAmong, then Eij=0;
Step C7, for i from 1 to M, for j from 1 to K, repetition step C6 is M*K times total, obtains each feature in every Ka Te tree
Different degree evaluation of estimate Eij;
Step C8 calculates general comment value E of each characteristic node in all Ka Te trees for j from 1 to Kj:
Step C9, according to EjEach feature in the F of sample characteristics space is ranked up, the high preceding β K of different degree evaluation of estimate is selected
A feature forms new feature space Fselected, Fselected={ f1,f2,…,fβK}
Step C10 chooses new feature space FselectedIn feature data information collection T is mapped, obtain new set
Tselected, that is,
Wherein, Π represents map operation,
Step C11, by new data set TselectedIt is input to XGBoost algorithm action recognition layer, is used for human action identification layer
Training, to obtain final trained bilayer model.
5. the human motion recognition method as claimed in claim 2 based on automated characterization selection and Ensemble Learning Algorithms, special
Sign is, in the step A, preprocessing process includes: to eliminate the noise that initial data is concentrated using mean filter.
6. the human motion recognition method as claimed in claim 2 based on automated characterization selection and Ensemble Learning Algorithms, special
Sign is, in the step A, preprocessing process includes: preceding 10 second data most started and most end for removing initial data concentration
Rear 10 second data of tail.
7. the human motion recognition method as claimed in claim 1 or 3 based on automated characterization selection and Ensemble Learning Algorithms,
It is characterized in that, the temporal signatures include average value, minimum value, maximum value, median, variance, standard deviation, the degree of bias, kurtosis, mistake
One of zero number, the interrelated index of acceleration reference axis are a variety of.
8. the human motion recognition method as claimed in claim 1 or 3 based on automated characterization selection and Ensemble Learning Algorithms,
It is characterized in that, the frequency domain character includes one of spectral energy, dominant frequency, the corresponding amplitude of dominant frequency or a variety of.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910203015.0A CN109934179B (en) | 2019-03-18 | 2019-03-18 | Human body action recognition method based on automatic feature selection and integrated learning algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910203015.0A CN109934179B (en) | 2019-03-18 | 2019-03-18 | Human body action recognition method based on automatic feature selection and integrated learning algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109934179A true CN109934179A (en) | 2019-06-25 |
CN109934179B CN109934179B (en) | 2022-08-02 |
Family
ID=66987359
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910203015.0A Expired - Fee Related CN109934179B (en) | 2019-03-18 | 2019-03-18 | Human body action recognition method based on automatic feature selection and integrated learning algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109934179B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110674875A (en) * | 2019-09-25 | 2020-01-10 | 电子科技大学 | Pedestrian motion mode identification method based on deep hybrid model |
CN110689041A (en) * | 2019-08-20 | 2020-01-14 | 陈羽旻 | Multi-target behavior action recognition and prediction method, electronic equipment and storage medium |
CN111028339A (en) * | 2019-12-06 | 2020-04-17 | 国网浙江省电力有限公司培训中心 | Behavior action modeling method and device, electronic equipment and storage medium |
CN111199187A (en) * | 2019-12-11 | 2020-05-26 | 中国科学院计算机网络信息中心 | Animal behavior identification method based on algorithm, corresponding storage medium and electronic device |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2012006578A2 (en) * | 2010-07-08 | 2012-01-12 | The Regents Of The University Of California | End-to-end visual recognition system and methods |
US20120195495A1 (en) * | 2011-01-31 | 2012-08-02 | Derek Shiell | Hierarchical Tree AAM |
US20160026917A1 (en) * | 2014-07-28 | 2016-01-28 | Causalytics, LLC | Ranking of random batches to identify predictive features |
CN105931224A (en) * | 2016-04-14 | 2016-09-07 | 浙江大学 | Pathology identification method for routine scan CT image of liver based on random forests |
CN108921197A (en) * | 2018-06-01 | 2018-11-30 | 杭州电子科技大学 | A kind of classification method based on feature selecting and Integrated Algorithm |
CN109034398A (en) * | 2018-08-10 | 2018-12-18 | 深圳前海微众银行股份有限公司 | Feature selection approach, device and storage medium based on federation's training |
-
2019
- 2019-03-18 CN CN201910203015.0A patent/CN109934179B/en not_active Expired - Fee Related
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2012006578A2 (en) * | 2010-07-08 | 2012-01-12 | The Regents Of The University Of California | End-to-end visual recognition system and methods |
US20120195495A1 (en) * | 2011-01-31 | 2012-08-02 | Derek Shiell | Hierarchical Tree AAM |
US20160026917A1 (en) * | 2014-07-28 | 2016-01-28 | Causalytics, LLC | Ranking of random batches to identify predictive features |
CN105931224A (en) * | 2016-04-14 | 2016-09-07 | 浙江大学 | Pathology identification method for routine scan CT image of liver based on random forests |
CN108921197A (en) * | 2018-06-01 | 2018-11-30 | 杭州电子科技大学 | A kind of classification method based on feature selecting and Integrated Algorithm |
CN109034398A (en) * | 2018-08-10 | 2018-12-18 | 深圳前海微众银行股份有限公司 | Feature selection approach, device and storage medium based on federation's training |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110689041A (en) * | 2019-08-20 | 2020-01-14 | 陈羽旻 | Multi-target behavior action recognition and prediction method, electronic equipment and storage medium |
CN110674875A (en) * | 2019-09-25 | 2020-01-10 | 电子科技大学 | Pedestrian motion mode identification method based on deep hybrid model |
CN111028339A (en) * | 2019-12-06 | 2020-04-17 | 国网浙江省电力有限公司培训中心 | Behavior action modeling method and device, electronic equipment and storage medium |
CN111028339B (en) * | 2019-12-06 | 2024-03-29 | 国网浙江省电力有限公司培训中心 | Behavior modeling method and device, electronic equipment and storage medium |
CN111199187A (en) * | 2019-12-11 | 2020-05-26 | 中国科学院计算机网络信息中心 | Animal behavior identification method based on algorithm, corresponding storage medium and electronic device |
Also Published As
Publication number | Publication date |
---|---|
CN109934179B (en) | 2022-08-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109934179A (en) | Human motion recognition method based on automated characterization selection and Ensemble Learning Algorithms | |
US6745156B2 (en) | Petroleum exploration and prediction apparatus and method | |
CN105022929B (en) | A kind of cognition accuracy analysis method of personal traits value test | |
Janssen et al. | Recognition of emotions in gait patterns by means of artificial neural nets | |
CN109376636A (en) | Eye ground image classification method based on capsule network | |
Priyanka et al. | Usage of data mining techniques in predicting the heart diseases—Naïve Bayes & decision tree | |
Kumar et al. | Yoga pose detection and classification using deep learning | |
CN108717693A (en) | A kind of optic disk localization method based on RPN | |
CN111728590A (en) | Individual cognitive ability prediction method and system based on dynamic function connection | |
CN111414839A (en) | Emotion recognition method and device based on gestures | |
CN105868532B (en) | A kind of method and system of intelligent evaluation heart aging degree | |
Hameed et al. | Human emotion classification based on respiration signal | |
CN101561881B (en) | Emotion identification method for human non-programmed motion | |
CN110491520A (en) | A kind of construction method of the sclerotin status assessment model based on semi-supervised learning | |
CN110210380A (en) | The analysis method of personality is generated based on Expression Recognition and psychology test | |
CN112401876A (en) | Method and system for human gait detection | |
Amor et al. | Toward sign language handshapes recognition using Myo armband | |
Fourati et al. | Multi-level classification of emotional body expression | |
CN108175426A (en) | A kind of lie detecting method that Boltzmann machine is limited based on depth recursion type condition | |
Senyurek et al. | A comparison of SVM and CNN-LSTM based approach for detecting smoke inhalations from respiratory signal | |
Permatasari et al. | Inertial sensor fusion for gait recognition with symmetric positive definite Gaussian kernels analysis | |
Wang et al. | SwimSense: Monitoring swimming motion using body sensor networks | |
Maekawa et al. | Training data selection with user’s physical characteristics data for acceleration-based activity modeling | |
Seddiqi et al. | Recognition of turkish sign language (TID) using sEMG sensor | |
CN109192315B (en) | Comprehensive age detection system based on weighted kernel regression and packaged deviation search |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220802 |
|
CF01 | Termination of patent right due to non-payment of annual fee |