CN110084286A - A kind of human motion recognition method of sensor-based ECOC technology - Google Patents
A kind of human motion recognition method of sensor-based ECOC technology Download PDFInfo
- Publication number
- CN110084286A CN110084286A CN201910285366.0A CN201910285366A CN110084286A CN 110084286 A CN110084286 A CN 110084286A CN 201910285366 A CN201910285366 A CN 201910285366A CN 110084286 A CN110084286 A CN 110084286A
- Authority
- CN
- China
- Prior art keywords
- ecoc
- data
- classifier
- class
- sensor
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention belongs to Human bodys' response fields, a kind of human motion recognition method of sensor-based ECOC technology is provided, this method measures X by two acceleration transducers and gyroscope, Y, the nine column exercise datas such as acceleration and angular speed on tri- directions Z, Wavelet-denoising Method pretreatment is carried out to initial data, and the time window for carrying out Duplication 50% to signal is handled, characteristic value is extracted by temporal analysis and frequency analysis method, more classification tasks are divided into limited two classification task by M times, mainly pass through ECOC encoder matrix, it finds an optimal planar from higher dimensional space using Laplce's kernel function of support vector machines (SVM) two category divisions come, by calculate Hamming distance from, export classification corresponding to minimum range.The method of the present invention can effectively improve the Accuracy and high efficiency of identification classification task.
Description
Technical field
The invention belongs to Human bodys' response fields, are related to a kind of human action identification of sensor-based ECOC technology
Method.
Background technique
All extensive concern the elderly, child and disabled person, Human bodys' responses such as the fields such as present life, medical treatment have
Important meaning.By accurately identifying their human body items behavior, many safety measures are realized, such as daily monitoring even exists
Carry out first time rescue when falling down equal hazardous acts, medically, by the identification of the human body behavior classification to patient and
Analysis, can provide foundation for condition-inference and treatment.
Currently, mostly using camera acquisition data to carry out image processing techniques Human bodys' response, at application image
The disadvantages of there are data volumes for reason technology greatly, it is complicated to calculate and cannot carry out real-time monitoring to human body, and view-based access control model identifies
Technology can not be accurately identified when light is bad, not adapt to varying environment, and carry out at analysis by video acquisition data
Reason is easy leakage privacy of user.
Summary of the invention
The purpose of the present invention is to overcome the above deficiencies in the prior art, provides a kind of sensor-based
The human motion recognition method of ECOC technology is that one kind is based on support vector machines (SVM) in conjunction with error correcting output code (ECOC) technology
More classification tasks Human bodys' response method, realize identification to a variety of behaviors such as daily behavior and tumble, improve and know
The Accuracy and high efficiency of other classification task.
To achieve the above object, the technical solution adopted by the present invention is as follows.
A kind of human motion recognition method of sensor-based ECOC technology, comprising the following steps:
(1) data are acquired using acceleration transducer and gyroscope, measures the acceleration and angle of every kind of movement
Three number of axle evidence of x, y, z;The z-axis of the local Coordinate System of sensor is towards human body front, and x-axis is perpendicular to the ground upwards, x-axis to y
The direction of rotation of axis meets the right-hand rule, in data acquisition tester need to complete according to specified time to be careful,
It hurries up, sits slowly, sit, stand up fastly, improper on foot, gently jump, walking slip, laterally slip, vertically falling, dumping forward, laterally
Topple over, jog, going upstairs, 15 class behaviors of going downstairs.
(2) slip window sampling data intercept segment is used to collected data, noise reduction is carried out to the data in window, use is small
Wave conversion eliminates jittering noise and the interference for measuring signal;Due to efficiently analyzing and handling non-stationary signal, so that wavelet packet
Noise reduction effectively reduces high-frequency signal noise signal, and retains the main feature of data.
Then data processing is carried out to the data slot in time window: by f=200Hz, it is assumed that when the acquisition of every kind of movement
Between be 12s, therefore 2400 groups of data are acquired in 12s, windows overlay rate is set as 50%, and the span of time window is 1s.
(3) time and frequency domain analysis carried out to processed acceleration and angular speed data, two acceleration transducers and
The acceleration of x-axis direction and the value of angular speed that angular speed generates are respectively ax、bx、ωx;The acceleration in y-axis direction and angle
The value of speed is respectively ay、by、ωy;The acceleration in z-axis direction and the value of angular speed are respectively az、bz、ωz。
The characteristic quantity of calculating includes: resultant acceleration
Resultant acceleration
Close angle speed
Then, the mean value in each window are as follows:
Variance in each window are as follows:
Wherein k is number of samples in a window, to the acceleration on three directions of x, y, z under every kind of movement, angle speed
Degree carries out Fast Fourier Transform (FFT), and the m for extracting the window that length is k ties up Fourier Transform Coefficients.
(4) the Strategies Training classifier for using multi-to-multi, using ECOC encoder matrix to 15 classes described in step (1)
It carry out not divide for M times, using three-unit code, i.e., (classifier does not use such for the specified class that is positive of classifier, anti-class or " deactivated class "
Not), the row of encoder matrix is to need 15 kinds of classifications classifying, and column are the classifier f of two classification based trainings each of after M divisionj。
(5) model training is carried out using multiple two classification tasks, is trained using support vector machines, passes through one
The maps feature vectors of Nonlinear separability are found an optimal planar into higher dimensional space and open two category divisions by kernel function
Come.
(6) among trained model, ECOC decoding operate is carried out, output Hamming distance is short corresponding from most
Classification is prediction result.
In the above-mentioned technical solutions, acceleration transducer uses ADXL345, MMA8451Q acceleration sensing in step (1)
Device, gyroscope use ITG3200, and taking data frequency is 200Hz, and the acquisition time of every kind of movement is 12s either 15s.
In the above-mentioned technical solutions, the kernel function used in step (5) is general Laplce's kernel function:
Wherein xi,xjFor any two training data, σ > 0 is the standard deviation of data,
By constructing support vector machine classifier:Wherein j=1,2 ... L, fjIt is every
One classifier,For the coefficient of kernel function.
J-th of classifier f divides the jth class data in training set and is indicated when being positive class with+1 when training, anti-class use -1
It indicates, 0 presentation class device f does not use such sample.L two category support vector machines SVM classifiers are obtained, can be classified slow
Walk, hurry up, jogging, going upstairs, going downstairs, sit slowly, sit, stand up fastly, it is improper walk, it is light jump, walking slips, laterally slips,
It vertically falls, dump forward, laterally toppling over etc. movement.
In the above-mentioned technical solutions, the ECOC decoding operate that step (6) uses tests sample and classifier to each to calculate
The distance of the division result of class, calculate Hamming distance from;
L is classifier number, f in formulaiIndicate the corresponding encoded radio of the i-th class classifier, MiPresentation class device is to test sample
As a result, dijIndicate fiWith MiBetween Hamming distance from.
The present invention is based on the human motion recognition methods of the ECOC technology of sensor, compared with prior art, have following
Advantage:
1. combining by temporal aspect and frequecy characteristic, the characteristic value of movement human body behavior can be more showed.
2. taking ECOC to encode for more classification tasks, there can be certain tolerance in mistake of the test phase to classifier
And capability for correcting.
3. two classification tasks resolved into reduce the training time of model, and improve with simple typical SVM algorithm
The accuracy rate of model.
Detailed description of the invention
Fig. 1 is the flow chart of human motion recognition method of the present invention.
Fig. 2 is error correcting output code of the invention (ECOC) figure.
Specific embodiment
In order to make those skilled in the art better understand method of the invention, invention is carried out below in conjunction with attached drawing
It is further to be described in detail.
As shown in Figure 1, the embodiment of the present invention provides a kind of human motion recognition method of sensor-based ECOC technology,
The following steps are included:
(1) two acceleration transducers of ADXL345, MMA8451Q and ITG3200 gyroscope are used, with the frequency of 200Hz
Rate acquires data, and acceleration transducer ADXL345 measurement range is ± 16g, and resolution ratio is 13;MMA8451Q measurement range is
± 8g, resolution ratio are 14.ITG3200 gyroscope range is ± 2000 °/s, and resolution ratio is 16.The own coordinate of sensor
The z-axis of system is towards human body front, and x-axis is perpendicular to the ground upwards, and the direction of rotation of x-axis to y-axis meets the right-hand rule.It is testing
In data acquisition tester need to complete according to specified time to be careful, hurry up, sit slowly, sitting fastly, standing up, it is improper walk
15 classes such as road, gently jump, walking slip, laterally slip, vertically falling, dumping forward, transverse direction topples over, jogs, going upstairs, going downstairs
Behavior.
These acceleration informations measured (AD) are converted to acceleration of gravity by following conversion formula:
Acceleration [g]=(2 × range)/(2Resolution ratio)]×AD
The spin data measured (RD) is converted into angular speed by following conversion formula:
Angular speed [rad/s]=[(2 × range)/(2Resolution ratio)]×RD
(2) pretreatment of data: slip window sampling data intercept segment is used to collected data, is being taken segment
Between, and between the about data overlap rate 50% of the data in time window, between size selection 1s to the 3s of time frame.
Noise reduction is carried out to the signal in time window, jittering noise and the interference for measuring signal are eliminated by wavelet transformation, by
Non-stationary signal is efficiently analyzed and handles in wavelet analysis, by wavelet analysis and reconfiguration technique, so that wavelet-packet noise reduction has
The reduction high-frequency signal noise signal of effect, and retain the main feature of data.
(3) acceleration and angular speed data collected carry out time and frequency domain analysis: two acceleration transducers and angle
The acceleration of x-axis direction and the value of angular speed that speed generates are respectively ax、bx、ωx;The acceleration in y-axis direction and angle speed
The value of degree is respectively ay、by、ωy;The acceleration in z-axis direction and the value of angular speed are respectively az、bz、ωz。
The characteristic quantity of calculating includes: resultant acceleration
Resultant acceleration
Close angle speed
Then, the mean value in each window, variance are as follows:
Wherein k is number of samples in a window, to the acceleration on three directions of x, y, z under every kind of movement, angle speed
Degree carries out Fast Fourier Transform (FFT), and the m for extracting the window that length is k ties up Fourier Transform Coefficients;
(4) when carrying out classification task for the training set of step (3), for more classification tasks, mainly by multiple classification
Task Switching be multiple two classification tasks, conversion regime has very much, but one-to-one and one-to-many cannot be promoted very well point
The effect of class device, so the Strategies Training classifier of multi-to-multi is used here, using error correcting output code (ECOC) encoder matrix pair
This 15 classifications carry out M division, using three-unit code as shown in Fig. 2, i.e. classifier specifies the class that is positive, anti-class or " deactivated class "
(classifier does not use the category).The row of encoder matrix is to need 15 kinds of classifications classifying, and column are two points each of after M division
The classifier f of class trainingj.In cataloged procedure in ECOC, encoder matrix should meet following two condition:
A. uncorrelated between the row of encoder matrix;
B. uncorrelated and not complementary between the column of encoder matrix, for k class classification problem, code length L be must satisfy:
log2 k<L<2k-1-1
In general, coding is longer, error correcting capability is stronger.But coding is longer, and the classifier of required training is more, for
Limited assortment number, code length be more than a certain range after just lose meaning.So must expire here mainly by random search algorithm
Hope that length is the k random coded sequence of L.
(5) multiple two classification tasks obtained for step 4 carry out model training, are instructed using support vector machines
Practice, the maps feature vectors of Nonlinear separability are found into higher dimensional space for two by an optimal planar by a kernel function
Category division comes, and the kernel function used here is general Laplce's kernel function:
Wherein xi,xjFor any two training data, σ > 0 is the standard deviation of data
By constructing support vector machine classifier:Wherein j=1,2 ... L, fjIt is every
One classifier,For the coefficient of kernel function.
J-th of classifier f divides the jth class data in training set and is indicated when being positive class with+1 when training, negative class use -1
It indicates, 0 presentation class device f does not use such sample.Such as will be careful, hurrying up is divided into+1, another part will jog and fastly
Race is divided into -1.Successively result is exported into ECOC three-unit code encoder matrix.
(6) by among the trained model of step (5), ECOC decoding operate is carried out: main by calculating test
Sample and classifier by a distance to training data classification results, calculate here Hamming distance from;
L is classifier number, f in formulaiIndicate the corresponding encoded radio of the i-th class classifier, MiPresentation class device is to test sample
As a result, dijIndicate fiWith MiBetween Hamming distance from.
By calculating Hamming distance from rear, that classification corresponding to output distance minimum is prediction result.
Error correcting output code (ECOC) biggest advantage is can to correct the mistake generated when classification, and error correcting capability is used
Hamming distance is from differentiation, when the minimum Hamming distance of calculating is from for d, then at leastClassification error in a code bit can be with
It is corrected.
To sum up, a kind of human motion recognition method of sensor-based ECOC technology of the present invention is, it can be achieved that each to human body
Effective differentiation of class behavior posture.It is measured on three directions of x, y, z by two acceleration transducers and gyroscope first
The nine column exercise data such as acceleration and angular speed.Wavelet-denoising Method pretreatment is carried out to initial data, and signal is overlapped
The time window of rate 50% is handled, and extracts characteristic value by temporal analysis and frequency analysis method.More classification tasks are passed through M times
It is divided into limited two classification task, mainly by ECOC encoder matrix, using the laplace kernel letter of support vector machines (SVM)
Number finds an optimal planar from higher dimensional space and two category divisions comes, by calculating Hamming distance from output is minimum
Apart from corresponding classification.Present method be advantageous in that the division error correcting output code (ECOC) that more classification tasks use when dividing,
The mistake generated when classification can be corrected, and the recognition speed of model can be improved using classical algorithm of support vector machine.
It is according to the present invention to be contemplated that main contents, pass through the suitable specific reality for the invention that logic analysis is described in detail
Apply example.It should be appreciated that the present invention makes many modifications and variations reasonings or the limited available technology of experiment.Therefore,
All technician in the art carry out logic analysis, reasoning on the basis of relying on this technology, all should be by claims
Identified protection scope in.
Claims (4)
1. a kind of human motion recognition method of sensor-based ECOC technology, it is characterised in that this method includes following step
It is rapid:
(1) data are acquired using acceleration transducer and gyroscope, measures the acceleration of every kind of movement and the x, y, z of angle
Three number of axle evidences;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 meet the right-hand rule, tester needs to complete to be careful according to the specified time, hurry up, slowly in data acquisition
It sits, sits, stands up fastly, improper on foot, gently jump, walking are slipped, laterally slipped, vertically falling, dumping forward, laterally toppling over, slowly
It runs, go upstairs, 15 class behaviors of going downstairs;
(2) slip window sampling data intercept segment is used to collected data, noise reduction is carried out to the data in window, is become with small echo
Change jittering noise and the interference eliminated and measure signal;
(3) time and frequency domain analysis is carried out to processed acceleration and angular speed data, calculates mean value in each window, side
Difference and covariance, and Fast Fourier Transform (FFT) is carried out, and extract the Fast Fourier Transform (FFT) coefficient for the window that length is k;
(4) use multi-to-multi Strategies Training classifier, using ECOC encoder matrix to 15 classifications described in step (1) into
Row M times division, using three-unit code, i.e. classifier specified be positive class, anti-class or deactivated class, the row of encoder matrix is to need to classify
15 kinds of classifications, column be by M times divide after each of two classification based trainings classifier fj;
(5) classification based training is carried out to multiple two classification tasks using support vector machines method and obtains L classifier, pass through one
The maps feature vectors of Nonlinear separability are found an optimal planar into higher dimensional space and divide two classifications by a kernel function;
(6) among trained model, ECOC decoding operate is carried out, using Hamming distance from as evaluation index, prediction is tied
Fruit output is the most short corresponding classification of distance.
2. the human motion recognition method of sensor-based ECOC technology according to claim 1, it is characterised in that: step
Suddenly acceleration transducer uses ADXL345, MMA8451Q acceleration transducer in (1), and gyroscope uses ITG3200, takes number
It is 200Hz according to frequency, is the sensing data that 12s or 15s collects every kind of movement according to acquisition time.
3. the human motion recognition method of sensor-based ECOC technology according to claim 1, it is characterised in that: step
Suddenly the kernel function used in (5) is general Laplce's kernel function:
Wherein xi,xjFor any two training data, σ > 0 is the standard deviation of data,
By constructing support vector machine classifier:Wherein j=1,2 ... L, fjFor each point
Class device,For the coefficient of kernel function;
J-th of classifier f divides the jth class data in training set and is indicated when being positive class with+1 when training, and negative class use -1 indicates,
0 presentation class device f does not use such sample, obtains L two category support vector machines SVM classifiers, is careful, fastly for classifying
It walks, jog, going upstairs, going downstairs, sit slowly, sit, stand up fastly, improper on foot, gently jump, walking slip, laterally slip, vertically falling
, it dumps forward, laterally topple over movement.
4. the human motion recognition method of sensor-based ECOC technology according to claim 1, it is characterised in that: step
Suddenly the ECOC decoding operate that (6) use for calculate test sample and classifier to all kinds of division results at a distance from, calculate
Hamming distance is from calculation formula is as follows:
L is classifier number, f in formulaiIndicate the corresponding encoded radio of the i-th class classifier, MiKnot of the presentation class device to test sample
Fruit, dijIndicate fiWith MiBetween Hamming distance from.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910285366.0A CN110084286A (en) | 2019-04-10 | 2019-04-10 | A kind of human motion recognition method of sensor-based ECOC technology |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910285366.0A CN110084286A (en) | 2019-04-10 | 2019-04-10 | A kind of human motion recognition method of sensor-based ECOC technology |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110084286A true CN110084286A (en) | 2019-08-02 |
Family
ID=67414628
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910285366.0A Pending CN110084286A (en) | 2019-04-10 | 2019-04-10 | A kind of human motion recognition method of sensor-based ECOC technology |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110084286A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110555413A (en) * | 2019-09-05 | 2019-12-10 | 第四范式(北京)技术有限公司 | method and device for processing time sequence signal, equipment and readable medium |
CN111681763A (en) * | 2020-04-16 | 2020-09-18 | 北京大学 | Total knee replacement prosthesis model prediction method based on error correction output codes and electronic device |
CN112311630A (en) * | 2020-11-04 | 2021-02-02 | 国网北京市电力公司 | Network equipment identification method and device |
CN112504295A (en) * | 2020-07-14 | 2021-03-16 | 华为技术有限公司 | Riding detection method, electronic device and computer readable storage medium |
CN112582063A (en) * | 2019-09-30 | 2021-03-30 | 长沙昱旻信息科技有限公司 | BMI prediction method, device, system, computer storage medium, and electronic apparatus |
CN112699744A (en) * | 2020-12-16 | 2021-04-23 | 南开大学 | Fall posture classification identification method and device and wearable device |
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 |
CN116058298A (en) * | 2023-03-06 | 2023-05-05 | 北京市农林科学院信息技术研究中心 | Livestock behavior monitoring method and device |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105787434A (en) * | 2016-02-01 | 2016-07-20 | 上海交通大学 | Method for identifying human body motion patterns based on inertia sensor |
CN107169415A (en) * | 2017-04-13 | 2017-09-15 | 西安电子科技大学 | Human motion recognition method based on convolutional neural networks feature coding |
CN108596074A (en) * | 2018-04-19 | 2018-09-28 | 上海理工大学 | A kind of human body lower limbs action identification method based on inertial sensor |
US20180314897A1 (en) * | 2017-05-01 | 2018-11-01 | Sensormatic Electronics, LLC | Surveillance System with Human Behavior Prediction by Human Action Recognition |
-
2019
- 2019-04-10 CN CN201910285366.0A patent/CN110084286A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105787434A (en) * | 2016-02-01 | 2016-07-20 | 上海交通大学 | Method for identifying human body motion patterns based on inertia sensor |
CN107169415A (en) * | 2017-04-13 | 2017-09-15 | 西安电子科技大学 | Human motion recognition method based on convolutional neural networks feature coding |
US20180314897A1 (en) * | 2017-05-01 | 2018-11-01 | Sensormatic Electronics, LLC | Surveillance System with Human Behavior Prediction by Human Action Recognition |
CN108596074A (en) * | 2018-04-19 | 2018-09-28 | 上海理工大学 | A kind of human body lower limbs action identification method based on inertial sensor |
Non-Patent Citations (4)
Title |
---|
FANG DENG 等: "Sensor Multifault Diagnosis With Improved Support Vector Machines", 《IEEE》 * |
张春祥 等: "《基于短语评价的翻译知识获取》", 29 February 2012 * |
都明宇 等: "基于改进支持向量机的人手动作模式识别方法", 《浙江大学学报(工学版)》 * |
雷阳 等: "《直觉模糊核匹配追踪理论及应用》", 31 March 2019 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110555413A (en) * | 2019-09-05 | 2019-12-10 | 第四范式(北京)技术有限公司 | method and device for processing time sequence signal, equipment and readable medium |
CN112582063A (en) * | 2019-09-30 | 2021-03-30 | 长沙昱旻信息科技有限公司 | BMI prediction method, device, system, computer storage medium, and electronic apparatus |
CN111681763A (en) * | 2020-04-16 | 2020-09-18 | 北京大学 | Total knee replacement prosthesis model prediction method based on error correction output codes and electronic device |
CN111681763B (en) * | 2020-04-16 | 2023-01-17 | 北京大学 | Total knee arthroplasty prosthesis model prediction method based on error correction output code and electronic device |
CN112504295A (en) * | 2020-07-14 | 2021-03-16 | 华为技术有限公司 | Riding detection method, electronic device and computer readable storage medium |
CN112504295B (en) * | 2020-07-14 | 2022-04-12 | 荣耀终端有限公司 | Riding detection method, electronic device and computer readable storage medium |
CN112311630A (en) * | 2020-11-04 | 2021-02-02 | 国网北京市电力公司 | Network equipment identification method and device |
CN112699744A (en) * | 2020-12-16 | 2021-04-23 | 南开大学 | Fall posture classification identification method and device and wearable device |
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 |
CN116058298A (en) * | 2023-03-06 | 2023-05-05 | 北京市农林科学院信息技术研究中心 | Livestock behavior monitoring method and device |
CN116058298B (en) * | 2023-03-06 | 2023-09-12 | 北京市农林科学院信息技术研究中心 | Livestock behavior monitoring method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110084286A (en) | A kind of human motion recognition method of sensor-based ECOC technology | |
Zhu et al. | Efficient human activity recognition solving the confusing activities via deep ensemble learning | |
CN110245718A (en) | A kind of Human bodys' response method based on joint time-domain and frequency-domain feature | |
Le et al. | Robust hand detection in vehicles | |
CN110659677A (en) | Human body falling detection method based on movable sensor combination equipment | |
Chatterjee et al. | A quality prediction method for weight lifting activity | |
CN106725495A (en) | A kind of fall detection method, apparatus and system | |
CN108021888A (en) | A kind of fall detection method | |
CN105869354B (en) | A kind of Falls Among Old People detection method based on attractor propagation algorithm | |
Guo et al. | Human activity recognition by fusing multiple sensor nodes in the wearable sensor systems | |
Wang et al. | Motion recognition for smart sports based on wearable inertial sensors | |
Lu et al. | MFE-HAR: multiscale feature engineering for human activity recognition using wearable sensors | |
Tran et al. | Robust classification of functional and nonfunctional arm movement after stroke using a single wrist-worn sensor device | |
Zhu et al. | Deep ensemble learning for human activity recognition using smartphone | |
Zhou et al. | Motion recognition by using a stacked autoencoder-based deep learning algorithm with smart phones | |
Pajak et al. | Sports activity recognition with UWB and inertial sensors using deep learning approach | |
CN111582361A (en) | Human behavior recognition method based on inertial sensor | |
Supanich et al. | Machine learning-based exercise posture recognition system using mediapipe pose estimation framework | |
Georgakopoulos et al. | Change detection and convolution neural networks for fall recognition | |
Eyobu et al. | A real-time sleeping position recognition system using IMU sensor motion data | |
Goh et al. | Multilayer perceptron neural network classification for human vertical ground reaction forces | |
Kraft et al. | Wrist-worn accelerometer based fall detection for embedded systems and IoT devices using deep learning algorithms | |
Straczkiewicz et al. | A systematic review of smartphone-based human activity recognition for health research | |
Liu et al. | Preimpact fall detection for elderly based on fractional domain | |
Low et al. | Lower extremity kinematics walking speed classification using long short-term memory neural frameworks |
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: 20190802 |
|
RJ01 | Rejection of invention patent application after publication |