CN108596074A - A kind of human body lower limbs action identification method based on inertial sensor - Google Patents
A kind of human body lower limbs action identification method based on inertial sensor Download PDFInfo
- Publication number
- CN108596074A CN108596074A CN201810354823.2A CN201810354823A CN108596074A CN 108596074 A CN108596074 A CN 108596074A CN 201810354823 A CN201810354823 A CN 201810354823A CN 108596074 A CN108596074 A CN 108596074A
- Authority
- CN
- China
- Prior art keywords
- human body
- inertial sensor
- lower limbs
- body lower
- data
- 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.)
- Pending
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
- G06F18/24—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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
- G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The invention discloses a kind of human body lower limbs action identification method based on inertial sensor.Acceleration, angular speed and the angle signal at human body knee joint are collected first with the inertial sensor of built-in acceleration meter and gyroscope；Then to signal usage time window method, allow window with time continuous intercept signal segment.For noise problem existing for signal itself, removed it using Wavelet-denoising Method.Next to the signal segment after interception using Fourier transform and using its coefficient as characteristic value.The method for finally using support vector machines (SVM) carries out model training and the road conditions identification of lower limb movement.Present method be advantageous in that, the characteristic value of action behavior can more be showed by being obtained using series of preprocessing and feature extraction algorithm, then it uses SVM identification human body lower limbs manner of execution and road conditions, this method not only to reduce the training time of model, and improves the recognition speed and accuracy rate of model.
Description
Technical field
The present invention relates to human body lower limbs action recognition field more particularly to it is a kind of based on inertial sensor by small wavelength-division
The human body lower limbs action identification method that analysis, principal component analysis and support vector machines are combined.
Background technology
Lower limb movement identification at present is one of the critical issue of intelligent artificial limb human-computer fusion.If people, artificial limb and environment three
Effective information exchange cannot be carried out between person and coordinates to control, then cannot achieve in various road conditions, different behaviors, not going together
The ideal control effect under the stage is walked, this allows for artificial limb and is difficult to keep good stability and comfort.In recent years, lower main drive
Make identification technology and at home and abroad obtains extensive concern.In current Research on Gait Recognition, the information for describing body gait
Include mainly mechanical information, video image information, bio-electrical information and Inertia information.But human body lower limbs Activity recognition at present
Major part is the identification about video image class.Although pretty good based on image recognition effect, due to the recognition methods of image class
It is easy to be influenced by place and background.Although mechanical information mainly uses suitable algorithm according to the ground reaction force in vola
Action is identified.Although this signal can be applied to normal person, since the plantar pressure of intelligent artificial limb is with normal
People or distinguishing, is not suitable in this way.Analysis method based on bioelectrical signals is current new research field,
Various researchs are still within theoretical level, can not temporarily use actual scene.So in conclusion best signal source is just at present
It is inertial signal, inertial sensor signal stabilization, precision be high, compact these advantages determine that the sensor is to be most suitable for answering
With the sensor on intelligent artificial limb.
Therefore, those skilled in the art is dedicated to developing a kind of human body lower limbs action recognition side based on inertial sensor
Method, and it is an object of the present invention to provide it is a kind of based on inertial signal (acceleration, angle and angular speed) and with wavelet analysis, principal component analysis and
The human body lower limbs action recognition that support vector machines scheduling algorithm is combined.
Invention content
In view of the drawbacks described above of the prior art, the technical problem to be solved by the present invention is to solve human body lower limbs action to know
The technical issues of other precision is high, jitter.
To achieve the above object, the present invention provides a kind of human body lower limbs action identification method based on inertial sensor,
Include the following steps：
Step 1 acquires acceleration, angular speed and the angle-data at lower limb knee joint using inertial sensor, and acquisition is run
Step takes a walk, cycles, the test data of this 7 kinds actions of climb and fall and stair activity, and the acquisition time of each action is 8 minutes；
Step 2, to collected data use slip window sampling data intercept segment, then to the signal in time window into
Row noise reduction；
Step 3 carries out dimension elimination using normalized method to angle of acceleration speed and angle, is then directed to normalization
Rear data carry out feature extraction, to acceleration, angular speed and the angle signal on three directions of x, y, z under each action into
4 rank Fourier transformation of row, and Fourier Transform Coefficients are extracted as eigenmatrix, the eigenmatrix after transformation is 9 × 9 original spies
Levy matrix；
Step 4, using the method for principal component analysis by 9 × 9 eigenmatrix dimensionality reduction to 6 × 6, then 6 × 6 feature
The one-dimensional characteristic vector that matrix conversion is 1 × 36；
Step 5, the Nonlinear separability feature vector for being extracted step 1-4 using support vector machines, are reflected by kernel function
It is mapped in higher dimensional space, an optimal planar is then found out in higher dimensional space, different classifications is distinguished from, construction is supported
Vector machine (SVM) grader, obtains the corresponding action attributes of feature vector；
The initial data newly inputted is extracted 1 × 36 one-dimensional vector using the method for step 1-4, then inputted by step 6
To among the trained model of step 5, the corresponding action of vector to be measured is identified, to realization inertial sensor pair
The detection of human body lower limbs action.
Further, the inertial sensor uses and is integrated with three-axis gyroscope, 3-axis acceleration and magnetometer
MPU6050。
Further, the inertial sensor is fixed on human body lower limbs knee, the inertial sensor local Coordinate System
Z-axis towards immediately ahead of human body, x-axis is perpendicular to the ground upwards, and direction of rotation meets the right-hand rule.
Further, the sensor can in real time be transferred data to upper during measurement by bluetooth approach
Machine.
Further, the data overlap rate of the data intercept segment is 30%-50%, and the time span of time frame is 1
Second to 3 seconds, interior each window includes 700 groups of data.
Further, the method for the noise reduction uses wavelet analysis method, and the number of plies of wavelet decomposition is 4 layers, the small echo
Fundamental wave is more Bei Xi (Daubechies) wave.
Further, the method for normalizing of the removal dimension is Z score method (Z-score), as：
Here X indicates that former data, μ indicate that signal mean value, σ indicate signal variance.
Further, the kernel function uses radial basis function (RBF).
Further, the function of the support vector machines is defined as：
Wherein φi(xi) indicating kernel function, selection is RBF cores, ωiIt is the weights of model, N is the total number of sample,
{xi,liIt is specified training set, xiIt is i-th of sample of training set, liIt is corresponding class label.
Further, the support vector machines grader is constituted by 7, and identification is taken a walk, and running cycles, stair activity
With 7 kinds of actions of climb and fall, i.e. fj=sng (y (xi)), wherein j=1,2 ... 7, j-th of grader f when trainingjIn training set
Jth class data are divided into positive class, are indicated with+1, and the remaining negative class use -1 that is classified as indicates；
If stroll acts, then the SVM classifier taken a walk will recognise that+1, others action can all be classified as -1；
If data are running actions, then the SVM classifier run will recognise that+1, others, which act, can all be classified as-
1；
And so on, until that will cycle, and go upstairs, and go downstairs, totally 7 kinds of states will all be divided into+1 or -1 to ascents and descents；
If the output result for multiple SVM classifiers occur is+1, the y (x of more different SVM are selectedi), selective value is most
Big is test sample classification.
Present method be advantageous in that for interference and the more lower limb inertial signal of noise use window technique, wavelet analysis,
A series of this Preprocessing Algorithm of normalization method make signal become more smooth regular.Secondly, this method is become by Fourier
It changes and extracts its coefficient as characteristic value, this feature extracting method extracts the characteristic value that can more show action behavior.Separately
The typical SVM algorithm used in external recognizer improves the identification speed of model in addition to reducing the training time of model
It spends and discrimination is also very high.The Activity recognition that can be good at completing human body lower limbs is applied in combination in this series of method,
With good application prospect.
The technique effect of the design of the present invention, concrete structure and generation is described further below with reference to attached drawing, with
It is fully understood from the purpose of the present invention, feature and effect.
Description of the drawings
Fig. 1 is the flow chart of the human body lower limbs action identification method of the preferred embodiment of the present invention；
Fig. 2 is the sensor of the preferred embodiment of the present invention in human body placement location figure；
Fig. 3 is the schematic diagram of the window technique extraction signal of the preferred embodiment of the present invention；
Fig. 4 is the identification process figure of the human body lower limbs action of the preferred embodiment of the present invention.
Specific implementation mode
Multiple preferred embodiments that the present invention is introduced below with reference to Figure of description, keep its technology contents more clear and just
In understanding.The present invention can be emerged from by many various forms of embodiments, and protection scope of the present invention not only limits
The embodiment that Yu Wenzhong is mentioned.
In the accompanying drawings, the identical component of structure is indicated with same numbers label, everywhere the similar component of structure or function with
Like numeral label indicates.The size and thickness of each component shown in the drawings are to be arbitrarily shown, and there is no limit by the present invention
The size and thickness of each component.In order to keep diagram apparent, some places suitably exaggerate the thickness of component in attached drawing.
As shown in Figure 1, the present invention, which mainly faces intelligent artificial limb, proposes a kind of human body lower limbs action based on inertial sensor
Recognition methods.First with the acceleration at the inertial sensor of built-in acceleration meter and gyroscope collection human body knee joint, angle
Speed signal.Then to signal usage time window method, allow window with time continuous intercept signal.Exist for signal itself
Noise problem, removed it using wavelet algorithm.Next to the signal after interception using Fourier transform and by its coefficient
As characteristic value.The method for finally using support vector machines (SVM) carries out lower limb movement and road conditions identification.
It elaborates to the case study on implementation of the present invention with reference to experimental data, the step implementation steps of this method are such as
Under：
The first step：Inertial sensor is integrated with three-axis gyroscope, 3-axis acceleration and magnetometer using MPU6050, can be with
Kneed acceleration, angle and angular velocity signal are effectively acquired as original signal source.Inertial sensor puts direction such as
Shown in Fig. 2, the z-axis of the local Coordinate System of sensor is towards human body front, and x-axis is perpendicular to the ground upwards, the rotation of x-axis to y-axis
Turn direction and meets the right-hand rule.During test data collection, subject only need according to complete to run and walk, cycle, on
This 7 kinds actions of descending and stair activity, and to ensure that training data is abundant, each action lasts about 8 minutes.Then the sensing
Device can transfer data to host computer in real time during measurement by bluetooth approach.
Second step：Slip window sampling data intercept segment is used to collected data, as shown in figure 3, in figure, w1, w2…
wnIndicate window.Window technique can intercept more signal segments and significantly more efficient can utilize CPU.In interception segment
Between data overlap rate about 30% to 50%.Between size selection 1s to the 3s of time frame, and the data in time window are about
700 or so.The global feature of the segment signal only could be preferably embodied according to such time window.Next it is directed to the time
Signal in window carries out noise reduction.The method of noise reduction uses wavelet analysis method, and the number of plies of wavelet decomposition is 4 layers, and the base of small echo
Wave is Daubechies waves.By the good time domain locality and multiresolution analysis ability of wavelet analysis, remove in signal
HF noise signal so that signal is purer.This method makes next signal characteristic abstraction that can more show signal
Essence.
Third walks：Because sensor acquires acceleration on three directions of x, y, z, angle speed according to local Coordinate System respectively
The difference of degree and angle signal, dimension can generate feature extraction greatly interference, so this method is normalized using Z-score
Method dimension elimination is carried out to angle of acceleration speed and angle.Removal dimension method be：
Here X indicates that former data, μ indicate that signal mean value, σ indicate signal variance.
Then the data after normalization are directed to and carry out feature extraction.Since waveform is in integrally during extracting feature
A kind of existing periodicity, so the matched curve by the way of Fourier transformation, the exponent number of fitting is 4 layers, and extracts Fourier's change
Coefficient is changed as eigenmatrix.Here 9 kinds of signals are shared so the primitive character matrix that the eigenmatrix after transformation is 9*9.
4th step：The matrix of 9*9 can make the calculation amount of computer greatly increase, to reduce the real-time of COMPUTER DETECTION
Property, thus this method using the method for principal component analysis by the eigenmatrix dimensionality reduction of 9*9 to 6*6.This method not only reduces
Calculation amount and the feature for having taste for remaining primitive character matrix.In model training, due to SVM model training mistakes
Training data in journey is all one-dimensional, so the eigenmatrix of 6*6 is converted to the feature vector of 1*36.
5th step：Since support vector machines generalization ability is stronger, be used herein support vector machines has as grader
Preferable effect.The Nonlinear separability feature vector that this method is extracted step 1~4 using support vector machines, passes through one
Kernel function is mapped in higher dimensional space, and this method is used as kernel function using radial basis function (RBF), is then sought in higher dimensional space
An optimal planar is found out to be distinguished from different classifications.So for specified training set { xi,li, wherein xiIt is trained
I-th of sample of collection, and liIt is corresponding class label.The function of so support vector machines is defined as：
Wherein φi(xi) indicating kernel function, what is selected in context of methods is RBF cores；ωiIt is the weights of model；N is sample
Total number.The purpose of model training is exactly to find a hyperplane to be distinguished from different classifications.Due to support vector machines
Itself it is the method for discrimination of two classification problems, so cannot be directly used to more classification processing.This method is wanted for this problem
Identification is taken a walk, running, is cycled, the 7 kinds of actions of stair activity and climb and fall, therefore construct 7 binary classifiers altogether.Construction 7
SVM classifier fj=sng (y (xi)), wherein j=1,2 ... 7.J-th of grader f when trainingjThe jth class data in training set
It is divided into positive class, is indicated with+1, the remaining negative class use -1 that is classified as indicates.Sorting technique if it is taking a walk as shown in figure 4, go
For then the SVM classifier taken a walk will recognise that+1, other behaviors can all be classified as -1.If same data are running rows
For then the SVM classifier run will recognise that+1, other behaviors can all be classified as -1.And so on, until that will cycle,
It goes upstairs, go downstairs, totally 7 kinds of states are all divided into+1 or -1 to ascents and descents.But if there is the output of multiple SVM classifiers
Result is+1, then selects the y (x of more different SVMi), maximum selective value is test sample classification.
6th step：During identification, also 1*36's will be extracted according to step 1~4 to the original signal newly inputted
One-dimensional vector, among this one-dimensional vector to be identified is then input to the trained model of step 5, whole process is as schemed
Shown in 4.It can identify that vector to be measured is corresponding under this model and what behavior belonged to, to realization inertial sensor
Detection to human body lower limbs behavior.
The preferred embodiment of the present invention has been described in detail above.It should be appreciated that the ordinary skill of this field is without wound
The property made labour, which according to the present invention can conceive, makes many modifications and variations.Therefore, all technician in the art
Pass through the available technology of logical analysis, reasoning, or a limited experiment on the basis of existing technology under this invention's idea
Scheme, all should be in the protection domain being defined in the patent claims.
Claims (10)
1. a kind of human body lower limbs action identification method based on inertial sensor, which is characterized in that include the following steps：
Step 1 acquires acceleration, angular speed and the angle-data at lower limb knee joint using inertial sensor, and acquisition running dissipates
Step cycles, the test data of this 7 kinds actions of climb and fall and stair activity, and the acquisition time of each action is 8 minutes；
Step 2 uses slip window sampling data intercept segment to collected data, is then dropped to the signal in time window
It makes an uproar；
Step 3 carries out dimension elimination using normalized method to angle of acceleration speed and angle, after being then directed to normalization
Data carry out feature extraction, to acceleration, angular speed and the angle signal on three directions of x, y, z reference axis under each action
4 rank Fourier transformations are carried out, and extract Fourier Transform Coefficients as eigenmatrix, the eigenmatrix after transformation is 9 × 9 original
Eigenmatrix；
Step 4, using the method for principal component analysis by 9 × 9 eigenmatrix dimensionality reduction to 6 × 6, then 6 × 6 eigenmatrix
Be converted to 1 × 36 one-dimensional characteristic vector；
Step 5, the Nonlinear separability feature vector for being extracted step 1-4 using support vector machines, are mapped to by kernel function
In higher dimensional space, an optimal planar is then found out in higher dimensional space, different classifications is distinguished from, construct supporting vector
Machine grader obtains the corresponding action attributes of feature vector；
The initial data newly inputted is extracted 1 × 36 one-dimensional vector using the method for step 1-4, then is input to step by step 6
Among rapid 5 trained model, the corresponding action of vector to be measured is identified, to which realization inertial sensor is to human body
The detection of lower limb movement.
2. the human body lower limbs action identification method based on inertial sensor as described in claim 1, which is characterized in that described
Inertial sensor is using the MPU6050 for being integrated with three-axis gyroscope, 3-axis acceleration and magnetometer.
3. the human body lower limbs action identification method based on inertial sensor as described in claim 1, which is characterized in that described
Inertial sensor is fixed on human body lower limbs knee, and the z-axis of the inertial sensor local Coordinate System is towards immediately ahead of human body, x
Axis is perpendicular to the ground upwards, and direction of rotation meets the right-hand rule.
4. the human body lower limbs action identification method based on inertial sensor as described in claim 1, which is characterized in that described
Sensor can transfer data to host computer in real time during measurement by bluetooth approach.
5. the human body lower limbs action identification method based on inertial sensor as described in claim 1, which is characterized in that described
The data overlap rate of data intercept segment is 30%-50%, and the time span of time frame is 1 second to 3 seconds, includes in each window
700 groups of data.
6. the human body lower limbs action identification method based on inertial sensor as described in claim 1, which is characterized in that described
The method of noise reduction uses wavelet analysis method, and the number of plies of wavelet decomposition is 4 layers, and the fundamental wave of the small echo is more Bei Xibo.
7. the human body lower limbs action identification method based on inertial sensor as described in claim 1, which is characterized in that described
The method for normalizing for removing dimension is Z score method, as：
Here X indicates that former data, μ indicate that signal mean value, σ indicate signal variance.
8. the human body lower limbs action identification method based on inertial sensor as described in claim 1, which is characterized in that described
Kernel function uses radial basis function (RBF).
9. the human body lower limbs action identification method based on inertial sensor as described in claim 1, which is characterized in that described
The function of support vector machines is defined as：
Wherein φi(xi) indicating kernel function, selection is RBF cores, ωiIt is the weights of model, N is the total number of sample, { xi,li}
It is specified training set, xiIt is i-th of sample of training set, liIt is corresponding class label.
10. the human body lower limbs action identification method based on inertial sensor as described in claim 1, which is characterized in that institute
It states support vector machine classifier to be constituted by 7, identification is taken a walk, running, is cycled, the 7 kinds of actions of stair activity and climb and fall, i.e. fj=
sng(y(xi)), wherein j=1,2 ... 7, j-th of grader f when trainingjJth class data in training set are divided into positive class, are used
+ 1 indicates, the remaining negative class use -1 that is classified as indicates；
If stroll acts, then the support vector machine classifier taken a walk will recognise that+1, others action can all be classified as -1；
If data are running actions, then the support vector machine classifier run will recognise that+1, others action can all be sorted out
It is -1；
And so on, until that will cycle, and go upstairs, and go downstairs, totally 7 kinds of states will all be divided into+1 or -1 to ascents and descents；
If the output result for multiple SVM classifiers occur is+1, the y (x of more different support vector machines are selectedi), selective value
Maximum is test sample classification.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810354823.2A CN108596074A (en) | 2018-04-19 | 2018-04-19 | A kind of human body lower limbs action identification method based on inertial sensor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810354823.2A CN108596074A (en) | 2018-04-19 | 2018-04-19 | A kind of human body lower limbs action identification method based on inertial sensor |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108596074A true CN108596074A (en) | 2018-09-28 |
Family
ID=63611306
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810354823.2A Pending CN108596074A (en) | 2018-04-19 | 2018-04-19 | A kind of human body lower limbs action identification method based on inertial sensor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108596074A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109978001A (en) * | 2019-02-21 | 2019-07-05 | 上海理工大学 | Karate moving state identification device based on multilayer Hybrid Clustering Algorithm |
| CN110084286A (en) * | 2019-04-10 | 2019-08-02 | 武汉理工大学 | A kind of human motion recognition method of sensor-based ECOC technology |
| CN110377159A (en) * | 2019-07-24 | 2019-10-25 | 张洋 | Action identification method and device |
| CN111178155A (en) * | 2019-12-10 | 2020-05-19 | 中国科学院深圳先进技术研究院 | Gait feature extraction and gait recognition method based on inertial sensor |
| CN112294295A (en) * | 2020-11-18 | 2021-02-02 | 王健 | Human body knee motion posture identification method based on extreme learning machine |
| CN112733631A (en) * | 2020-12-28 | 2021-04-30 | 华南理工大学 | Lower limb dynamic action mode identification method |
| CN114239724A (en) * | 2021-12-17 | 2022-03-25 | 中南民族大学 | Cuball motion recognition and skill evaluation method based on inertial sensor |
Citations (5)
| 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 |
| CN104571506A (en) * | 2014-12-25 | 2015-04-29 | 西安电子科技大学 | Smart watch based on action recognition and action recognition method |
| CN106176149A (en) * | 2016-09-08 | 2016-12-07 | 电子科技大学 | A kind of ectoskeleton gait analysis system based on multi-sensor fusion and method |
| 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 |
| CN107085246A (en) * | 2017-05-11 | 2017-08-22 | 深圳合优科技有限公司 | A kind of human motion recognition method and device based on MEMS |
-
2018
- 2018-04-19 CN CN201810354823.2A patent/CN108596074A/en active Pending
Patent Citations (5)
| 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 |
| CN104571506A (en) * | 2014-12-25 | 2015-04-29 | 西安电子科技大学 | Smart watch based on action recognition and action recognition method |
| 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 |
| CN106176149A (en) * | 2016-09-08 | 2016-12-07 | 电子科技大学 | A kind of ectoskeleton gait analysis system based on multi-sensor fusion and method |
| CN107085246A (en) * | 2017-05-11 | 2017-08-22 | 深圳合优科技有限公司 | A kind of human motion recognition method and device based on MEMS |
Non-Patent Citations (3)
| Title |
|---|
| AHMET TURAN ÖZDEMIR: "An Analysis on Sensor Locations of the Human Body for Wearable Fall Detection Devices: Principles and Practice", 《SENSORS》 * |
| 张欣: "基于分布式微惯性传感器组合的人体行为识别研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 樊瑜波，张明著: "《生物力学研究前沿系列 康复工程生物力学》", 31 December 2017 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109978001A (en) * | 2019-02-21 | 2019-07-05 | 上海理工大学 | Karate moving state identification device based on multilayer Hybrid Clustering Algorithm |
| CN110084286A (en) * | 2019-04-10 | 2019-08-02 | 武汉理工大学 | A kind of human motion recognition method of sensor-based ECOC technology |
| CN110377159A (en) * | 2019-07-24 | 2019-10-25 | 张洋 | Action identification method and device |
| CN110377159B (en) * | 2019-07-24 | 2023-06-09 | 张洋 | Action recognition method and device |
| CN111178155A (en) * | 2019-12-10 | 2020-05-19 | 中国科学院深圳先进技术研究院 | Gait feature extraction and gait recognition method based on inertial sensor |
| CN111178155B (en) * | 2019-12-10 | 2024-03-29 | 中国科学院深圳先进技术研究院 | Gait feature extraction and gait recognition method based on inertial sensor |
| CN112294295A (en) * | 2020-11-18 | 2021-02-02 | 王健 | Human body knee motion posture identification method based on extreme learning machine |
| CN112733631A (en) * | 2020-12-28 | 2021-04-30 | 华南理工大学 | Lower limb dynamic action mode identification method |
| CN112733631B (en) * | 2020-12-28 | 2023-06-13 | 华南理工大学 | Method for identifying dynamic action mode of lower limb |
| CN114239724A (en) * | 2021-12-17 | 2022-03-25 | 中南民族大学 | Cuball motion recognition and skill evaluation method based on inertial sensor |
| CN114239724B (en) * | 2021-12-17 | 2023-04-18 | 中南民族大学 | Cuball motion recognition and skill evaluation method based on inertial sensor |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108596074A (en) | A kind of human body lower limbs action identification method based on inertial sensor | |
| Ogbuabor et al. | Human activity recognition for healthcare using smartphones | |
| Yang et al. | Using acceleration measurements for activity recognition: An effective learning algorithm for constructing neural classifiers | |
| US11096593B2 (en) | Method for generating a personalized classifier for human motion activities of a mobile or wearable device user with unsupervised learning | |
| He et al. | A low power fall sensing technology based on FD-CNN | |
| Lin et al. | A novel multichannel dilated convolution neural network for human activity recognition | |
| Jensen et al. | Classification of kinematic swimming data with emphasis on resource consumption | |
| Yhdego et al. | Towards musculoskeletal simulation-aware fall injury mitigation: transfer learning with deep CNN for fall detection | |
| Liu et al. | An analysis of segmentation approaches and window sizes in wearable-based critical fall detection systems with machine learning models | |
| CN106210269A (en) | A kind of human action identification system and method based on smart mobile phone | |
| Choi et al. | Deep learning-based near-fall detection algorithm for fall risk monitoring system using a single inertial measurement unit | |
| CN105868779A (en) | Method for identifying behavior based on feature enhancement and decision fusion | |
| Guo et al. | Human activity recognition by fusing multiple sensor nodes in the wearable sensor systems | |
| Bayat et al. | Human gait recognition using bag of words feature representation method | |
| Coelho et al. | Human activity recognition based on convolutional neural network | |
| Coelho et al. | A lightweight model for human activity recognition based on two-level classifier and compact CNN model | |
| Jung et al. | OPTIMIZING WEARABLE SYSTEM WITH SEQUENTIAL CLASSIFIER FOR CHILDREN WITH DEVELOPMENTAL DISABILITIES | |
| Avadut et al. | A Deep Learning based IoT Framework for Assistive Healthcare using Gesture Based Interface | |
| Tasjid et al. | Leveraging smartphone sensors for detecting abnormal gait for smart wearable mobile technologies | |
| İsmail et al. | Human activity recognition based on smartphone sensor data using cnn | |
| CN112861679A (en) | Transfer learning method and system for behavior recognition | |
| Fataniya et al. | Comprehensive analysis of deep learning-based human activity recognition approaches based on accuracy | |
| Chen et al. | An Alarm System Based on BP Neural Network Algorithm for the Detection of Falls to Elderly Person | |
| Açış et al. | Classification of human movements by using Kinect sensor | |
| Putra et al. | Genetic-algorithm-based feature-selection technique for fall detection using multi-placement wearable sensors |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180928 |
|
| RJ01 | Rejection of invention patent application after publication |