CN109086698A - A kind of human motion recognition method based on Fusion - Google Patents
A kind of human motion recognition method based on Fusion Download PDFInfo
- Publication number
- CN109086698A CN109086698A CN201810803749.8A CN201810803749A CN109086698A CN 109086698 A CN109086698 A CN 109086698A CN 201810803749 A CN201810803749 A CN 201810803749A CN 109086698 A CN109086698 A CN 109086698A
- Authority
- CN
- China
- Prior art keywords
- sensor node
- matrix
- indicate
- data
- carried out
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The present invention relates to human actions to identify field, provides a kind of human motion recognition method based on Fusion, comprising: using the N number of inertial sensor node for being individually fixed in human body different parts, acquires human action data;Window segmentation is carried out to each sensor node human action data collected using sliding window cutting techniques, obtains multiple action data segments of each sensor node;Feature extraction is carried out to the action data segment of each sensor node, obtains corresponding feature vector;Feature Dimension Reduction is carried out using feature vector of the RLDA algorithm to each sensor node of acquisition;Parameter training and modeling are carried out using the feature vector after each sensor node dimensionality reduction as training data, obtains corresponding hierarchical fusion model;Using obtained hierarchical fusion model, human action identification is carried out.The present invention can effectively overcome drawback of the single classifier in identification process, can effectively improve human action accuracy of identification.
Description
Technical field
The present invention relates to human action identification field more particularly to a kind of human actions based on Fusion
Recognition methods.
Background technique
Human action identification technology is a kind of new man-machine interaction mode risen in recent decades, has been increasingly becoming
The hot issue that domestic and foreign scholars study.Human action be primarily referred to as human body action mode and people to environment or
The reaction of object, human body is by the compound movements of limbs, to describe or express complicated human action.It can be said that human body is dynamic
Making majority is to need to embody by the movement of human body limb.By the movement to human body come the movement of research and probe human body
Just become a very effective approach of analysis human action.Human action identification based on inertial sensor is pattern-recognition
One emerging research field in field, essence generate when obtaining people's movement by one or more inertial sensors first
Then motor message pre-processes data, feature extraction and selection, is finally classified according to the feature of extraction to movement
And identification.
During using inertial sensor research human action identification, it can not be suitable for using single sorting algorithm
The identification of all people's body movement.This is because there are certain decision errors for each single classifier, single classification is utilized
Device not can solve all practical problems centainly;In addition, movement has random and randomness, this is just under physical condition
Considerably increase the difficulty of identification.Many researchers generally use multiple Classifiers Combination technology pair when identifying some compound actions
Mankind's activity in practical application is monitored.Decision (or being Decision fusion) is carried out to raising identity in conjunction with multi-categorizer
Can have a great impact.Decision fusion can effectively improve the classification performance of identifying system, improve the robustness of identifying system.
Summary of the invention
Present invention mainly solves the single sorting algorithms of the prior art can not be suitable for the identification that all people's body acts,
Single classifier proposes that a kind of human action based on Fusion is known there are technical problems such as certain decision errors
Other method can effectively overcome drawback of the single classifier in identification process, utilize hierarchical fusion mould proposed by the invention
Type recognition result obtained is substantially better than traditional recognition method.
The present invention provides a kind of human motion recognition methods based on Fusion, comprising the following steps:
Step 100, using the N number of inertial sensor node for being individually fixed in human body different parts, human action number is acquired
According to;
Step 200, each sensor node human action data collected is carried out using sliding window cutting techniques
Window segmentation, obtains multiple action data segments of each sensor node;
Step 300, feature extraction is carried out to the action data segment of each sensor node, obtain corresponding feature to
Amount;
Step 400, Feature Dimension Reduction is carried out using feature vector of the RLDA algorithm to each sensor node of acquisition;
Step 500, using the feature vector after each sensor node dimensionality reduction as training data carry out parameter training and
Modeling, obtains corresponding hierarchical fusion model, including step 501 is to 506:
Step 501, cross validation method is rolled over by k, the feature vector after the dimensionality reduction of each sensor node is carried out
Verifying obtains each movement for the contribution rate of each classifier;
Step 502, the evaluations matrix of following Multiple Classifier Fusion layer is established according to contribution rate:
Wherein, Y indicates that evaluations matrix, c indicate that action classification, k presentation class device quantity, i indicate i-th of inertial sensor
Node, mijIndicate j-th of classifier relative to i-th of inertial sensor node, yqjQ-th of movement is expressed as at j-th point
Contribution rate under class device;
Step 503, the evaluations matrix obtained according to step 502 obtains the Shannon of each classifier using following formula
Entropy:
Wherein, ejIndicate that Shannon entropy, η are a constants, and η=1/log2(c);
The amount of redundancy of this classifier is obtained according to Shannon entropy, and using following formula:
rj=1-ej
Wherein, rjIndicate amount of redundancy;
The weighted value of i-th of sensor node, j-th classifier is obtained by following formula:
Wherein,Indicate the weighted value of i-th of sensor node, j-th classifier;
The output result of i-th of sensor node is obtained by following formula:
Wherein, λi,qIndicate that test sample x has been assigned to q class;
Step 504, it for the feature vector after the dimensionality reduction in i-th of sensor node, is obtained q-th by following formula
Act the discrimination of class:
Wherein,Indicate the discrimination of q-th of movement class;
Step 505, the evaluations matrix such as lower sensor fused layer is established according to discrimination:
Step 506, the evaluations matrix obtained according to step 505 obtains the Shannon of each sensor using following formula
Entropy:
The amount of redundancy of this sensor is obtained according to Shannon entropy, and using following formula:
Pass through the output weight of each sensor node of following formula:
Obtain following hierarchical fusion model:
Wherein, λqIndicate that test sample is assigned to q class;
Step 600, using obtained hierarchical fusion model, human action identification is carried out.
Preferably, window is carried out to each sensor node human action data collected using sliding window cutting techniques
Mouth segmentation, comprising:
For i-th of sensor node, enabling the size of split window is l, if the length of exercise data matrix isThen
Exercise data matrix AiIt can be divided intoA data window, the segmentation data matrix size in each window are
(l × 6) dimension and every two adjacent data window have 50% repetitive rate.
Preferably, feature extraction is carried out to the action data segment of each sensor node, the feature of extraction includes: three axis
Acceleration information and the root mean square of three axis angular rate data, absolute mean square deviation, kurtosis, covariance, zero-crossing rate and energy.
Preferably, Feature Dimension Reduction is carried out using feature vector of the RLDA algorithm to each sensor node of acquisition, including
Following steps:
Step 401, it for characteristic vector space corresponding to i-th of sensor node, obtains spreading square in corresponding class
Battle array and between class scatter matrix:
Wherein, SωIndicate within-class scatter matrix, SbIndicate between class scatter matrix, μaIndicate in a class all feature vectors and
Mean value, μ indicate characteristic vector space XiAll feature vector sums be averaged;
Step 402, invertible matrix is solved according to contract matrix theorem and matrix basic transformation, obtains following formula:
PTSωP=In
Wherein, P indicates invertible matrix,For SbCharacteristic value, and InIndicate that n ties up unit matrix;
Step 403, according to step 402 acquired results, following optimal projection matrix is obtained using Fei Sheer decision criteria:
φopt=KPT
Wherein, φoptIndicate optimal projection matrix, K=φ (PT)-1,Indicate projection matrix to be solved;
Step 404, Feature Dimension Reduction is carried out using optimal projection matrix.
A kind of human motion recognition method based on Fusion provided by the invention, is mainly utilized matrix
Contract theorem in improves traditional LDA algorithm, improved Feature Dimension Reduction algorithm can effectively reduce by
In the huge disturbance that the inverse time of the characteristic value in estimation within-class scatter matrix is generated due to small deviation, to be conducive to mention
High algorithm performance;In action recognition level, the present invention mainly proposes a kind of new hierarchical fusion algorithm, and the blending algorithm is main
Including two layers, first layer is Multiple Classifier Fusion layer, and the second layer is sensor fused layer, and each layer of the main entropy of output weight
Method obtains.Algorithm proposed by the present invention carries out the determination in complete by Information Entropy, can effectively improve the robust of disaggregated model
Property, by design layered can effectively raising movement accuracy of identification.
Detailed description of the invention
Fig. 1 is the implementation flow chart of the human motion recognition method the present invention is based on Fusion;
Fig. 2 is hierarchical fusion model schematic of the invention.
Specific embodiment
To keep the technical problems solved, the adopted technical scheme and the technical effect achieved by the invention clearer, below
The present invention is described in further detail in conjunction with the accompanying drawings and embodiments.It is understood that specific implementation described herein
Example is used only for explaining the present invention rather than limiting the invention.It also should be noted that for ease of description, attached drawing
In only some but not all of the content related to the present invention is shown.
Fig. 1 is the implementation flow chart of the human motion recognition method the present invention is based on Fusion.Such as Fig. 1 institute
Show, the human motion recognition method provided in an embodiment of the present invention based on Fusion, detailed process is as follows:
Step 100, using the N number of inertial sensor node for being individually fixed in human body different parts, human action number is acquired
According to.
Specifically, N number of inertial sensor node is individually fixed in first N number of position of human body, each sensor node
A three axis accelerometer and a three-axis gyroscope are respectively included, and the action data of acquisition is uploaded by means of receiving node
To host computer data processing platform (DPP);Then human action data, such as station, race, upstairs human body are acquired using N number of sensor node
The data of movement.Human action data includes the 3-axis acceleration data and three axis angular rate data of each sensor node.
(i ∈ { 1,2, N }) a sensor node for i-th, the human action data of acquisition include x-axis, y-axis and z-axis
Acceleration information ai=[aix,aiy,aiz] and x-axis, the angular velocity data ang of y-axis and z-axisi=[angix,angiy,angiz], then
For i-th of sensor node, 3-axis acceleration data and three axis angular rate data composition have 6 the original motion data arranged
Matrix Ai=[ai,angi]=[aix,aiy,aiz,angix,angiy,angiz]。
Step 200, each sensor node human action data collected is carried out using sliding window cutting techniques
Window segmentation, obtains multiple action data segments of each sensor node.
After action data collected in obtaining step 100, window segmentation is carried out to action data.The present embodiment is main
Window division is carried out to action data using sliding window cutting techniques: selecting the window size of regular length first, then moves
Dynamic window is split action data, and two adjacent windows have 50% repetitive rate.
Particularly, for i-th of sensor node, enabling the size of split window is l, if the length of exercise data matrix isThen exercise data matrix AiIt can be divided intoA data window, the segmentation data matrix in each window
Size is the repetitive rate that (l × 6) dimension and every two adjacent data window have 50%.
Step 300, feature extraction is carried out to the action data segment of each sensor node, obtain corresponding feature to
Amount.
After the initial data corresponding to each sensor is divided into multiple data windows, need in each window
Divide data matrix and carry out feature extraction, the feature of extraction mainly includes following 6 kinds:
1, the root mean square (Root mean square, RMS) of 3-axis acceleration data and three axis angular rate data is equal
Root, expression formula are respectivelyWherein T={ x, y, z } is indicated
Three axis directions, i indicate i-th of sensor;
2, the absolute mean square deviation (Mean absolute deviation, MAD) of 3-axis acceleration data and three shaft angles speed
The absolute mean square deviation of degree evidence, expression formula are respectively
Wherein, T={ x, y, z } indicates three axis directions, and i indicates i-th of biography
Sensor;
3, the covariance of the covariance (Covariance) of 3-axis acceleration data and three axis angular rate data, difference
It is expressed asWherein T1,T2={ x, y, z } indicates three axis directions, i table
Show i-th of sensor;
4, the kurtosis of the kurtosis (Kurtosis) of 3-axis acceleration data and three axis angular rate data, formula mainly indicate
ForWherein T
={ x, y, z } indicates three axis directions, and i indicates i-th of sensor,Indicate the mean value of acceleration information in window,Table
Show the variance of acceleration information in window,Indicate the mean value of angular velocity data in window,Indicate window interior angle
The variance of speed data;
5, the zero-crossing rate of 3-axis acceleration dataAnd three axis angular rate
The zero-crossing rate of dataWherein T={ x, y, z } indicates three axis directions, and i indicates i-th of sensor;
6, the energy of 3-axis acceleration data and the energy of three axis angular rate data, formula are expressed as
WithWherein T={ x, y, z } indicates three axis sides
To i indicates i-th of sensor, xang,T,jIndicate raw data matrix aiTObtained coefficient after Fast Fourier Transform,
xang,T,jIndicate raw data matrix angiTObtained coefficient after Fast Fourier Transform.
After to above 6 kinds of feature extractions, corresponding to j-th of data window of i-th of sensing node, it can get as follows
Feature vector:
Step 400, Feature Dimension Reduction is carried out using feature vector of the RLDA algorithm to each sensor node of acquisition.
Specifically, the invention proposes RLDA after having extracted feature for the action data of each sensor node acquisition
Algorithm is used to carry out dimensionality reduction to feature space.Its key step is as follows:
Step 401, for characteristic vector space X corresponding to i-th of sensor nodei, calculate in corresponding class and spread
Matrix SωAnd between class scatter matrix Sb, calculation formula is respectively as follows:
Wherein, SωIndicate within-class scatter matrix, SbIndicate between class scatter matrix, μaIndicate in a class all feature vectors and
Mean value, μaIndicate the mean value of all feature vector sums in a class, μ indicates characteristic vector space XiAll feature vector sums
It is average.
Step 402, invertible matrix P is solved according to contract matrix theorem and matrix basic transformation, so that following formula is set up:
PTSωP=In
Wherein, P indicates invertible matrix,For SbCharacteristic value, and InIndicate that n ties up unit matrix.
Step 403, according to step 402 acquired results, elementary variation is carried out to Fei Sheer decision criteria, obtains solving optimal
Projection matrix.
Specifically, Fei Sheer, which maximizes criterion, to be indicated are as follows:
Wherein,Projection matrix to be solved is indicated, due to SbAnd SωAll be it is positive semi-definite, in addition, again according to linear discriminant
Analyze S known to (Linear discriminant analysis, LDA) theoryωFor positive definite matrix.It is then fixed according to contract matrix
An invertible matrix P is certainly existed known to reason, so thatPTSωP=In, here,It is matrix SbCharacteristic value, InIndicate that n ties up unit matrix.Then Fei Sheer maximizes criterion and can convert
For following form:
It enablesK=φ (PT)-1, then Fei Sheer maximizes criterion and can convert are as follows:
Then the optimal projection matrix of linear discriminant analysis can obtain φ by following formulaopt:
φopt=KPT。
Wherein, φoptIndicate optimal projection matrix.
Step 404, Feature Dimension Reduction is carried out using optimal projection matrix.
Step 500, using the feature vector after each sensor node dimensionality reduction as training data carry out parameter training and
Modeling, obtains corresponding hierarchical fusion model.
Fig. 2 is hierarchical fusion model schematic of the invention.Referring to Fig. 2, the hierarchical fusion model of the embodiment of the present invention is used
To identify human body various motion, which mainly includes two layers.First layer can be referred to as Multiple Classifier Fusion layer, basic
Thought is exactly the classification results that multiple classifiers are merged using the thought of majority voting method then, and the strategy of fusion is mainly in conjunction with power
The thought of weight, and the weight of the corresponding result of decision is mainly obtained by Information Entropy.The second layer is referred to as sensor fused layer, main
It to be exactly the output for the sensor that fusion is bundled in the multiple positions of body as a result, the strategy of fusion is still and is obtained by Information Entropy
Weighted value carries out decision.The specific steps of which are as follows:
Step 501, cross validation method is rolled over by k, the feature vector after the dimensionality reduction of each sensor node is carried out
Verifying obtains each movement for the contribution rate of each classifier.
For rolling over cross validation method, training data by k from the training dataset after the dimensionality reduction that a sensor obtains
Collection can be divided into k parts by random.Then by this k group data, we can be clipped to k basic classification device relative to c with score
The identification accurate rate of a action classification, this k identification accurate rate are marked as c, yqjQ-th of movement is expressed as to classify at j-th
Identification accurate rate under device.Here yqjQ-th of movement is considered for the contribution rate of j-th of classifier.
Step 502, the evaluations matrix Y of following Multiple Classifier Fusion layer is established according to contribution rate:
Wherein, Y indicates that evaluations matrix, c indicate that action classification, k presentation class device quantity, i (0≤i≤N) indicate i-th
Inertial sensor node, mijIndicate jth (0≤j≤k) a classifier relative to i-th of inertial sensor node, yqjIt indicates
For contribution rate of a movement of q (0≤q≤c) under j-th of classifier;
Step 503, the evaluations matrix obtained according to step 502 obtains the Shannon of each classifier using following formula
Entropy:
Wherein, ejIndicate Shannon entropy, j indicates that j-th of classifier, η are a constants, and η=1/log2(c)。
Then according to Shannon entropy, and the amount of redundancy r of this classifier is obtained using following formulaj:
rj=1-ej
And the weighted value of i-th of sensor node, j-th classifier is obtained by following formula
The output result of i-th of sensor node is obtained by following formula:
Wherein, λi,qIndicate that test sample x has been assigned to q class.
Step 504, for the feature vector (training data) after the dimensionality reduction in i-th of sensor node, pass through following public affairs
Formula obtains the discrimination of q-th of movement class
Wherein,Indicate the discrimination of q-th of movement class.
Step 505, the evaluations matrix such as lower sensor fused layer is established according to discrimination
Step 506, the evaluations matrix obtained according to step 505 obtains the Shannon of each sensor using following formula
Entropy
Then according to Shannon entropy, and the amount of redundancy of this sensor is obtained using following formula
And pass through the output weight of each sensor node of following formula
Obtain following hierarchical fusion model:
Wherein, λqIndicate that test sample is assigned to q class.For test sample, last result of decision λq, can be melted by layer
Molding type obtains.
Step 600, using obtained hierarchical fusion model, human action identification is carried out.
When test data is input in corresponding hierarchical fusion model, corresponding classification results, Jin Ershi can be obtained
Existing human action identification.
Explanation that the present invention will be further explained by way of example below:
For example, acquiring human action data by five sensor nodes, each sensor node includes that three axis add
Speedometer and a three-axis gyroscope, sample frequency 50HZ.Experimental subjects has 8 people altogether, and the age is between 24-34 years old.Five
Sensor node is individually placed to the right hand wrist of experimental subjects, left hand arm, waist, right crus of diaphragm ankle, at left foot thigh.In addition,
The designed movement of this experiment includes: to walk (to execute on the treadmill of gymnasium, the speed of setting is respectively 3km/h, 5km/h, is held
The about 3 minutes every time row time);Run (it being executed on the treadmill of gymnasium, the speed of setting is respectively 6km/h, 8km/h,
12km/h executes about 3 minutes every time time);Rope skipping, (practical to execute);Cycling (executes, executing the time is 3 in campus
Minute);It goes upstairs and (is executed in campus);It goes downstairs and (is executed in campus);Gymnastics (practical to execute).In addition, the original number of acquisition
It is handled in MATLAB according to meeting, and combine the recognizer write, obtain recognition result to the end.This example is adopted altogether
(8 people × 5 sensors × 10 movements) 400 action sequences are collected, each action sequence is about 10000 sampled points.It is each dynamic
Making sequence all includes 3-axis acceleration data and three axis angular rate data.
Then, window segmentation is carried out to the action sequence of acquisition.For example, for i-th (i=1,2,3,4,5) a sensor
Node, taking the size of split window is 256, i.e., every 256 sampled points are a data window.If the length of exercise data matrix
ForThen each exercise data sequence can be divided intoA data window, the segmentation data in each window
Matrix size is the repetitive rate that 256 × 6 dimensions and every two adjacent data window have 50%.
After obtaining data window, each data window is needed to carry out feature extraction, as previously described, the feature of extraction
It include: root mean square, absolute mean square deviation, kurtosis, covariance, zero-crossing rate and energy.For extracting three in each data window
Axle acceleration degree accordingly and the feature of three axis angular rate data, extraction specially for:
The dimension of feature in each data window is (36 × 1) dimension, and each feature vector is counted as a data sample
This, is then identified and is classified to data sample again.
After having extracted feature, no matter for test data and training data, data set is needed to carry out dimension-reduction treatment.Drop
The method of dimension is RLDA algorithm proposed by the invention.Data sample obtained for each sensor, each sample
Intrinsic dimensionality is 36, and the intrinsic dimensionality of each sensing data data sample obtained is reduced to 9 dimensions using RLDA algorithm
Below.
For the exercise data after dimensionality reduction, test data is identified using the hierarchical fusion algorithm proposed in the present invention
Classification, to assess algorithm performance proposed in the present invention.Single classifier used in blending algorithm specifically includes that recently
Adjacent classifier (KNN), Naive Bayes Classifier (NB), decision tree C4.5 classifier (C4.5), support vector machines (SVM) are hidden
Markov classifier (HMM).Used classifier is all identical in blending algorithm.
This example is mainly using staying a proof method to evaluate algorithm.The data conduct of 1 experimental subjects is taken out first
Then test data, the data of remaining 7 people are circuited sequentially as training data.Last experimental result is to 8 test datas
Results are averaged to obtain final result.
Table 1 gives the classification accuracy obtained under different recognition methods.Wherein this example gives direct benefit
Recognition result obtained when identification test sample is directly carried out with single classifier;In addition this example gives classical classification
Device blending algorithm-majority voting method (MV) recognition result obtained.As can be seen that method proposed by the present invention from result
Highest accuracy of identification can be obtained, reaches 96.96%.
Classification accuracy obtained under the different recognition methods of table 1
Method | KNN | NB | SVM | C4.5 | HMM | MV | The present invention |
Average recognition rate | 84.55% | 84.79% | 87.59% | 82.48% | 89.17% | 94.77% | 96.96% |
Human motion recognition method provided in this embodiment based on Fusion, first result contragradient transformation
Traditional linear discriminant analysis algorithm is improved, a kind of new feature selecting algorithm-RLDA, the algorithms are proposed
The main characteristic value inverse to within-class scatter matrix using contract matrix transformation reevaluates, thus reduce due to compared with
The error disturbance that small eigenvalue estimate is inaccurate and generates, improves arithmetic accuracy.The invention also provides a kind of layerings to melt
For molding type to identify a variety of human actions, which mainly includes two layers.First layer is Multiple Classifier Fusion layer, the second layer
It is mainly obtained by Information Entropy for each layer of sensor fused layer of decision weights.There are two excellent for blending algorithm proposed by the present invention
Gesture, using for first layer structure can make final output result more accurate, can play to a certain extent strong
The advantage of classifier, in addition the use of upper Information Entropy can guarantee that this sorting algorithm has stronger robustness in algorithm.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations;To the greatest extent
Present invention has been described in detail with reference to the aforementioned embodiments for pipe, those skilled in the art should understand that: its is right
Technical solution documented by foregoing embodiments is modified, or is equally replaced to some or all of the technical features
It changes, the range for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution.
Claims (4)
1. a kind of human motion recognition method based on Fusion, which comprises the following steps:
Step 100, using the N number of inertial sensor node for being individually fixed in human body different parts, human action data is acquired;
Step 200, window is carried out to each sensor node human action data collected using sliding window cutting techniques
Segmentation, obtains multiple action data segments of each sensor node;
Step 300, feature extraction is carried out to the action data segment of each sensor node, obtains corresponding feature vector;
Step 400, Feature Dimension Reduction is carried out using feature vector of the RLDA algorithm to each sensor node of acquisition;
Step 500, parameter training and modeling are carried out using the feature vector after each sensor node dimensionality reduction as training data,
Obtain corresponding hierarchical fusion model, including step 501 is to 506:
Step 501, cross validation method is rolled over by k, the feature vector after the dimensionality reduction of each sensor node is verified,
Each movement is obtained for the contribution rate of each classifier;
Step 502, the evaluations matrix of following Multiple Classifier Fusion layer is established according to contribution rate:
Wherein, Y indicates that evaluations matrix, c indicate that action classification, k presentation class device quantity, i indicate i-th of inertial sensor section
Point, mijIndicate j-th of classifier relative to i-th of inertial sensor node, yqjQ-th of movement is expressed as to classify at j-th
Contribution rate under device;
Step 503, the evaluations matrix obtained according to step 502 obtains the Shannon entropy of each classifier using following formula:
Wherein, ejIndicate that Shannon entropy, η are a constants, and η=1/log2(c);
The amount of redundancy of this classifier is obtained according to Shannon entropy, and using following formula:
rj=1-ej
Wherein, rjIndicate amount of redundancy;
The weighted value of i-th of sensor node, j-th classifier is obtained by following formula:
Wherein,Indicate the weighted value of i-th of sensor node, j-th classifier;
The output result of i-th of sensor node is obtained by following formula:
Wherein, λi,qIndicate that test sample x has been assigned to q class;
Step 504, for the feature vector after the dimensionality reduction in i-th of sensor node, q-th of movement is obtained by following formula
The discrimination of class:
Wherein,Indicate the discrimination of q-th of movement class;
Step 505, the evaluations matrix such as lower sensor fused layer is established according to discrimination:
Step 506, the evaluations matrix obtained according to step 505 obtains the Shannon entropy of each sensor using following formula:
The amount of redundancy of this sensor is obtained according to Shannon entropy, and using following formula:
Pass through the output weight of each sensor node of following formula:
Obtain following hierarchical fusion model:
Wherein, λqIndicate that test sample is assigned to q class;
Step 600, using obtained hierarchical fusion model, human action identification is carried out.
2. the human motion recognition method according to claim 1 based on Fusion, which is characterized in that benefit
Window segmentation is carried out to each sensor node human action data collected with sliding window cutting techniques, comprising:
For i-th of sensor node, enabling the size of split window is l, if the length of exercise data matrix isThen move number
According to matrix AiIt can be divided intoA data window, the segmentation data matrix size in each window are (l × 6)
Dimension and every two adjacent data window have 50% repetitive rate.
3. the human motion recognition method according to claim 1 based on Fusion, which is characterized in that right
The action data segment of each sensor node carries out feature extraction, and the feature of extraction includes: 3-axis acceleration data and three axis
The root mean square of angular velocity data, absolute mean square deviation, kurtosis, covariance, zero-crossing rate and energy.
4. the human motion recognition method according to claim 1 based on Fusion, which is characterized in that benefit
Feature Dimension Reduction is carried out with feature vector of the RLDA algorithm to each sensor node of acquisition, comprising the following steps:
Step 401, for characteristic vector space corresponding to i-th of sensor node, obtain corresponding within-class scatter matrix with
And between class scatter matrix:
Wherein, SωIndicate within-class scatter matrix, SbIndicate between class scatter matrix, μaIndicate the equal of all feature vector sums in a class
Value, μ indicate characteristic vector space XiAll feature vector sums be averaged;
Step 402, invertible matrix is solved according to contract matrix theorem and matrix basic transformation, obtains following formula:
PTSωP=In
Wherein, P indicates invertible matrix,For SbCharacteristic value, and InIndicate that n ties up unit matrix;
Step 403, according to step 402 acquired results, following optimal projection matrix is obtained using Fei Sheer decision criteria:
φopt=KPT
Wherein, φoptIndicate optimal projection matrix, K=φ (PT)-1,Indicate projection matrix to be solved;
Step 404, Feature Dimension Reduction is carried out using optimal projection matrix.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810803749.8A CN109086698B (en) | 2018-07-20 | 2018-07-20 | Human body action recognition method based on multi-sensor data fusion |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810803749.8A CN109086698B (en) | 2018-07-20 | 2018-07-20 | Human body action recognition method based on multi-sensor data fusion |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109086698A true CN109086698A (en) | 2018-12-25 |
CN109086698B CN109086698B (en) | 2021-06-25 |
Family
ID=64838379
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810803749.8A Active CN109086698B (en) | 2018-07-20 | 2018-07-20 | Human body action recognition method based on multi-sensor data fusion |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109086698B (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109784418A (en) * | 2019-01-28 | 2019-05-21 | 东莞理工学院 | A kind of Human bodys' response method and system based on feature recombination |
CN109902565A (en) * | 2019-01-21 | 2019-06-18 | 深圳市烨嘉为技术有限公司 | The Human bodys' response method of multiple features fusion |
CN109919034A (en) * | 2019-01-31 | 2019-06-21 | 厦门大学 | A kind of identification of limb action with correct auxiliary training system and method |
CN110058699A (en) * | 2019-04-28 | 2019-07-26 | 电子科技大学 | A kind of user behavior recognition method based on Intelligent mobile equipment sensor |
CN110377159A (en) * | 2019-07-24 | 2019-10-25 | 张洋 | Action identification method and device |
CN110796188A (en) * | 2019-10-23 | 2020-02-14 | 华侨大学 | Multi-type inertial sensor collaborative construction worker work efficiency monitoring method |
CN112016430A (en) * | 2020-08-24 | 2020-12-01 | 郑州轻工业大学 | Hierarchical action identification method for multi-mobile-phone wearing positions |
CN112434669A (en) * | 2020-12-14 | 2021-03-02 | 武汉纺织大学 | Multi-information fusion human behavior detection method and system |
CN113057628A (en) * | 2021-04-04 | 2021-07-02 | 北京泽桥传媒科技股份有限公司 | Inertial sensor based motion capture method |
CN113065581A (en) * | 2021-03-18 | 2021-07-02 | 重庆大学 | Vibration fault migration diagnosis method for reactance domain adaptive network based on parameter sharing |
CN114241603A (en) * | 2021-12-17 | 2022-03-25 | 中南民族大学 | Shuttlecock action recognition and level grade evaluation method and system based on wearable equipment |
CN114832277A (en) * | 2022-05-20 | 2022-08-02 | 广东沃莱科技有限公司 | Rope skipping mode identification method and rope skipping |
CN115731602A (en) * | 2021-08-24 | 2023-03-03 | 中国科学院深圳先进技术研究院 | Human body activity recognition method, device, equipment and storage medium based on topological representation |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104268577A (en) * | 2014-06-27 | 2015-01-07 | 大连理工大学 | Human body behavior identification method based on inertial sensor |
CN105868779A (en) * | 2016-03-28 | 2016-08-17 | 浙江工业大学 | Method for identifying behavior based on feature enhancement and decision fusion |
CN106210269A (en) * | 2016-06-22 | 2016-12-07 | 南京航空航天大学 | A kind of human action identification system and method based on smart mobile phone |
-
2018
- 2018-07-20 CN CN201810803749.8A patent/CN109086698B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104268577A (en) * | 2014-06-27 | 2015-01-07 | 大连理工大学 | Human body behavior identification method based on inertial sensor |
CN105868779A (en) * | 2016-03-28 | 2016-08-17 | 浙江工业大学 | Method for identifying behavior based on feature enhancement and decision fusion |
CN106210269A (en) * | 2016-06-22 | 2016-12-07 | 南京航空航天大学 | A kind of human action identification system and method based on smart mobile phone |
Non-Patent Citations (3)
Title |
---|
ORESTI BANOS ET AL.: "Multi-sensor Fusion Based on Asymmetric Decision Weighting for Robust Activity Recognition", 《NEURAL PROCESS LETTERS》 * |
姜鸣,王哲龙: "基于无线传感网络的人体动作识别系统", 《2009中国自动化大会暨两化融合高峰会议论文集》 * |
陈野 等: "基于BSN和神经网络的人体日常动作识别方法", 《大连理工大学学报》 * |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109902565A (en) * | 2019-01-21 | 2019-06-18 | 深圳市烨嘉为技术有限公司 | The Human bodys' response method of multiple features fusion |
CN109784418B (en) * | 2019-01-28 | 2020-11-17 | 东莞理工学院 | Human behavior recognition method and system based on feature recombination |
CN109784418A (en) * | 2019-01-28 | 2019-05-21 | 东莞理工学院 | A kind of Human bodys' response method and system based on feature recombination |
CN109919034A (en) * | 2019-01-31 | 2019-06-21 | 厦门大学 | A kind of identification of limb action with correct auxiliary training system and method |
CN110058699B (en) * | 2019-04-28 | 2021-04-27 | 电子科技大学 | User behavior identification method based on intelligent mobile device sensor |
CN110058699A (en) * | 2019-04-28 | 2019-07-26 | 电子科技大学 | A kind of user behavior recognition method based on Intelligent mobile equipment sensor |
CN110377159A (en) * | 2019-07-24 | 2019-10-25 | 张洋 | Action identification method and device |
CN110796188A (en) * | 2019-10-23 | 2020-02-14 | 华侨大学 | Multi-type inertial sensor collaborative construction worker work efficiency monitoring method |
CN110796188B (en) * | 2019-10-23 | 2023-04-07 | 华侨大学 | Multi-type inertial sensor collaborative construction worker work efficiency monitoring method |
CN112016430B (en) * | 2020-08-24 | 2022-10-11 | 郑州轻工业大学 | Hierarchical action identification method for multi-mobile-phone wearing positions |
CN112016430A (en) * | 2020-08-24 | 2020-12-01 | 郑州轻工业大学 | Hierarchical action identification method for multi-mobile-phone wearing positions |
CN112434669A (en) * | 2020-12-14 | 2021-03-02 | 武汉纺织大学 | Multi-information fusion human behavior detection method and system |
CN112434669B (en) * | 2020-12-14 | 2023-09-26 | 武汉纺织大学 | Human body behavior detection method and system based on multi-information fusion |
CN113065581A (en) * | 2021-03-18 | 2021-07-02 | 重庆大学 | Vibration fault migration diagnosis method for reactance domain adaptive network based on parameter sharing |
CN113065581B (en) * | 2021-03-18 | 2022-09-16 | 重庆大学 | Vibration fault migration diagnosis method for reactance domain self-adaptive network based on parameter sharing |
CN113057628A (en) * | 2021-04-04 | 2021-07-02 | 北京泽桥传媒科技股份有限公司 | Inertial sensor based motion capture method |
CN115731602A (en) * | 2021-08-24 | 2023-03-03 | 中国科学院深圳先进技术研究院 | Human body activity recognition method, device, equipment and storage medium based on topological representation |
CN114241603A (en) * | 2021-12-17 | 2022-03-25 | 中南民族大学 | Shuttlecock action recognition and level grade evaluation method and system based on wearable equipment |
CN114832277A (en) * | 2022-05-20 | 2022-08-02 | 广东沃莱科技有限公司 | Rope skipping mode identification method and rope skipping |
CN114832277B (en) * | 2022-05-20 | 2024-02-06 | 广东沃莱科技有限公司 | Rope skipping mode identification method and rope skipping |
Also Published As
Publication number | Publication date |
---|---|
CN109086698B (en) | 2021-06-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109086698A (en) | A kind of human motion recognition method based on Fusion | |
CN104268577B (en) | Human body behavior identification method based on inertial sensor | |
CN110245718A (en) | A kind of Human bodys' response method based on joint time-domain and frequency-domain feature | |
CN110334573B (en) | Human motion state discrimination method based on dense connection convolutional neural network | |
Tao et al. | Gait based biometric personal authentication by using MEMS inertial sensors | |
WO2023035093A1 (en) | Inertial sensor-based human body behaviour recognition method | |
Whelan et al. | Leveraging IMU data for accurate exercise performance classification and musculoskeletal injury risk screening | |
Wang et al. | A2dio: Attention-driven deep inertial odometry for pedestrian localization based on 6d imu | |
CN108717548A (en) | A kind of increased Activity recognition model update method of facing sensing device dynamic and system | |
Lu et al. | MFE-HAR: multiscale feature engineering for human activity recognition using wearable sensors | |
Tu et al. | A Review of Human Motion Monitoring Methods using Wearable Sensors. | |
CN111259956A (en) | Rapid identification method for unconventional behaviors of people based on inertial sensor | |
Zhou et al. | A self-supervised human activity recognition approach via body sensor networks in smart city | |
CN115273237B (en) | Human body posture and action recognition method based on integrated random configuration neural network | |
Zhang et al. | Multi-STMT: Multi-level Network for Human Activity Recognition Based on Wearable Sensors | |
Ambati et al. | A comparative study of machine learning approaches for human activity recognition | |
Srivastava et al. | Hierarchical human activity recognition using GMM | |
Zhang et al. | PCA & HMM based arm gesture recognition using inertial measurement unit | |
Arshad et al. | Gait-based human identification through minimum gait-phases and sensors | |
İsmail et al. | Human activity recognition based on smartphone sensor data using cnn | |
Kusuma et al. | Health Monitoring with Smartphone Sensors and Machine Learning Techniques | |
Zainudin et al. | Hybrid relief-f differential evolution feature selection for accelerometer actions | |
Yang et al. | Recognition of human activities based on decision optimization model | |
Raziff et al. | Gait identification using one-vs-one classifier model | |
Wang et al. | A smartphone location independent activity recognition method based on the angle feature |
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 |