CN109770912A - A kind of abnormal gait classification method based on depth convolutional neural networks - Google Patents
A kind of abnormal gait classification method based on depth convolutional neural networks Download PDFInfo
- Publication number
- CN109770912A CN109770912A CN201910063488.5A CN201910063488A CN109770912A CN 109770912 A CN109770912 A CN 109770912A CN 201910063488 A CN201910063488 A CN 201910063488A CN 109770912 A CN109770912 A CN 109770912A
- Authority
- CN
- China
- Prior art keywords
- gait
- layer
- abnormal
- recognition methods
- abnormal gait
- 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
Abstract
The invention belongs to biometrics identification technology field, specially a kind of abnormal gait recognition methods based on depth convolutional neural networks.The method of the present invention includes: signal when being walked using the IMU acquisition human normal walking and the typical abnormal gait of simulation that are worn on human body, obtains the 3-axis acceleration information under different gaits；According to target representative row walk cadence by initial data do windowing cutting pre-process and each data queue is stamped by respective labels according to gait classification；CNN depth convolutional neural networks comprising convolutional layer one, convolutional layer two, pond layer one, pond layer two, full articulamentum and softmax output layer；Finally data label is sent into CNN and be trained to training set and test set, training set is divided into, test set assessment models classifying quality is utilized after the completion of training.The present invention eliminates complicated gait cycle division and feature extraction engineering, improves the classification accuracy to a variety of abnormal gaits, reduces data prediction workload, improves classification accuracy.
Description
Technical field
The invention belongs to biometrics identification technology fields, and in particular to a kind of abnormal gait recognition methods.
Background technique
Gait refers to the posture showed when people walks, is one of human body important biomolecule feature.Abnormal gait mostly with disease
Change position is related, as the important feature of reflection human health status and capacity, faces in medical diagnosis, disease prevention etc.
It is concerned in bed research.
The gait recognition method of mainstream mainly divides computer vision scheme based on video and image procossing and is based at present
The sensor plan of pavement and wearable sensor such as IMU.Two schemes all refer to a large amount of professional numbers after obtaining initial data
Data preprocess work and complicated Feature Engineering, to extract each correlated characteristic in gait cycle, although accuracy rate is high in real time
Property it is poor, computation complexity is high.And mainstream Gait Recognition system only provides each numerical index, the identification and classification task of abnormal gait
It is mainly completed by human expert, needs a large amount of professional knowledge of related fields.
Summary of the invention
The purpose of the present invention is to provide a kind of accuracy, and abnormal gait high, that real-time is good, computation complexity is low identifies
Method.
Abnormal gait recognition methods provided by the invention is based on depth convolutional neural networks technology, to the letter of acquisition
Directly classified to abnormal gait by original signal after number being pre-processed, the specific steps are as follows:
Step 1 acquires signal when human normal is walked using the IMU (inertia sensing unit) for being worn on human body, obtains
Normal gait 3-axis acceleration information；
Step 2 acquires signal when human body simulation typical case abnormal gait is walked using the IMU for being worn on human body, obtains different
Normal gait 3-axis acceleration information；
Step 3, according to target representative row walk cadence by initial data do windowing cutting pretreatment and according to gait classification will
Each data queue stamps respective labels；
Step 4 builds depth convolutional neural networks (CNN), which is six layer structure, point
It Wei not convolutional layer one, pond layer one, convolutional layer two, pond layer two, full articulamentum, softmax output layer；Wherein, convolutional layer one
It is connect entirely with pond layer one, pond layer one is fully connected to convolutional layer two, the full connection pool layer two of convolutional layer two, and pond layer two connects
Full connection hidden layer, the full hidden layer that connects are connected to full connection output layer；Define convolutional layer, pond layer, full articulamentum and each sharp
Function living；
Data label obtained in step 3 is sent into step 4 to training set and test set, training set is divided by step 5
In CNN be trained, training after the completion of utilize test set assessment models classifying quality.
The present invention Step 1: in two, the IMU acquisition unit can dispose but be not limited to shank, in shoes, loins, but require
Normal gait is consistent with the mounting means of abnormal gait.
In step 2 of the present invention, it is described typical case abnormal gait include but is not limited to hemiplegic gait, parkinsonian gait, duck step by step
State, steppage gait, diplegia gait.
Described to do windowing cutting to initial data in step 3 of the present invention, data window length representative value is two gaits
In the period, step value takes away a quarter of window length paragraph by paragraph between window, and the label form is but is not limited to one-hot coding.
In step 4 of the present invention, the input matrix Shape of the convolutional layer onecov1It is defined as every segment data window in step 3
(data source port number, current sampling frequency under the sampling number from respective signal data), namely:
Shapecov1: (NIMU, Twindows*fc)
Wherein, NIMUFor IMU data channel number, TwindowsFor data window time domain length, fcFor current sampling frequency.
The output layer nodal point number is defined as gait category number namely a kind of normal gait and all exceptions in "current" model
The sum of gait.
The optimization algorithm of CNN network can be but be not limited to Adam optimizer.
The hidden layer excitation function of the CNN can be but be not limited to ReLU function, Sigmod function, tanh function, export
Layer excitation function is Softmax function.
The invention has the advantages that constructing depth convolutional neural networks, convolution kernel Automatic-searching difference gait is utilized
Lower IMU data time and feature spatially, to complete the Direct Classification to abnormal gait.Initial data only needs to carry out simple
Filtering and windowing intercept and be sent into neural network, thus eliminate complicated period divide with feature extraction engineering, with
This is simultaneously to the available raising of the classification accuracy of abnormal gait, and accuracy rate can reach in more classification task preferred embodiments
96%, reduce data prediction workload, improves classification accuracy and identification effect.
Detailed description of the invention
Fig. 1 is the structure chart for the depth convolutional neural networks that the present invention is built.
Fig. 2 is original signal example under normal gait in present pre-ferred embodiments (interception 10s).
Fig. 3 is original signal example (interception 10s) under the five kinds of typical abnormal gaits chosen in present pre-ferred embodiments,
Wherein, Fig. 3 .1 is hemiplegic gait (drawing circle gait), Fig. 3 .2 is parkinsonian gait (festinating gait), Fig. 3 .3 is gluteus medius myopathy
Gait (duck step by step state), Fig. 3 .4 are steppage gait, Fig. 3 .5 is diplegia gait (scissor-like gait).
Fig. 4 is the ROC curve in embodiment training set.
Fig. 5 is the confusion matrix in embodiment test set.
Specific embodiment
Develop simultaneously preferred embodiment with reference to the accompanying drawing, and the present invention will be described in detail.
The present invention provides a kind of human motion Approach for Gait Classification based on convolutional neural networks, this method is using as follows
Step is realized:
IMU hardware system is fixed on the outside of right leg, wherein Y-axis and horizontal plane, X-axis by step 1 by bandage
Vertical with human coronary face, Z axis is vertical with human body sagittal plane.Setting systematic sampling rate is 512Hz, and setting IMU accelerometer is led
Life degree is ± 2g, and motor message when acquisition human normal is walked, acquisition step number is no less than 100 steps, obtains three axis of normal gait
Acceleration information, original signal example (interception 10s) is as shown in Figure 2 under one section of normal gait.
Step 2, IMU is placed to be accelerated with the same step 1 of collecting flowchart, tri- axis of IMU acquired under target simulation abnormal gait
Spend information.Choose five kinds of allusion quotations in abnormal gait because caused by from common causatives such as myopathy, osteopathy, brain damages in the present embodiment
Type abnormal gait is simulated, respectively hemiplegic gait (drawing circle gait), parkinsonian gait (festinating gait), gluteus medius myopathy step
State (duck step by step state), steppage gait, diplegia gait (scissor-like gait).Original signal example (interception under five kinds of abnormal gaits
10s) as shown in Figure 3, wherein Fig. 3 .1 is hemiplegic gait (drawing circle gait), Fig. 3 .2 is parkinsonian gait (festinating gait), figure
3.3 be gluteus medius myopathic gait (duck step by step state), Fig. 3 .4 is steppage gait, Fig. 3 .5 is diplegia gait (scissor-like gait).
Step 3, according to target representative row walk cadence by initial data do windowing cutting pretreatment and according to gait classification will
Each data queue stamps respective labels.Every segment signal window takes all sampled points in 2000ms in the present embodiment, three at 512Hz
A available 3072 the present embodiment of axis lift six classification tasks, and number label is 0 to 5, and wherein hemiplegic gait (draws circle step
State) label be 0, the label of parkinsonian gait (festinating gait) is 1, and the label of normal gait is 2, gluteus medius myopathic gait
The label of (duck step by step state) is 3, and the label of steppage gait is 4, and the label of diplegia gait (scissor-like gait) is 5.In two classification
In task (normal gait and abnormal gait), it is 1 that normal gait, which can re-flag, and it is 0 that all abnormal gaits, which can re-flag,.
Step 4, builds CNN depth convolutional neural networks, and network structure is shown in Fig. 1.Define convolutional layer one, pond layer one,
Convolutional layer two, pond layer two define full articulamentum and each activation primitive.Wherein, convolutional layer one is connect entirely with pond layer one, Chi Hua
Layer one is fully connected to convolutional layer two, the full connection pool layer two of convolutional layer two, and pond layer two connects full connection hidden layer, connects entirely hidden
Hiding layer is connected to full connection output layer.
Wherein, the input matrix of the convolutional layer one be defined as every segment data window in step 3 (data source port number, when
Sampling number from respective signal data under preceding sample frequency) namely:
Shapecov1: (3,2*512)
The IMU feature is exported in pond layer one, and is connected to convolutional layer two.
The optimization algorithm of the CNN network has been provided in connection with the advantages of two kinds of optimization algorithms of AdaGrad and RMSProp
Adam optimizer.Its gradient updating formula are as follows:
Wherein, default learning rate α is set as 0.001, ε=10^-8, and divisor is avoided to become 0.It is public for gradient mean value
Formula are as follows:
mt=β1mt-1+(1-β1)gt
Wherein, β1Coefficient is exponential decay rate, control weight distribution (momentum and current gradient), takes 0.9 in the present embodiment,
gtFor the gradient of t time step.
For gradient variance, formula are as follows:
vt=β2vt-1+(1-β2)gt 2
Wherein, β2Coefficient is exponential decay rate, the influence situation of the gradient square before controlling, and is taken in the present embodiment
0.999, gtFor the gradient of t time step.
Each layer excitation function is set as the linear correcting unit of ReLU, formula are as follows:
F (x)=max (0, x)
Output layer excitation function is Softmax function, formula are as follows:
Wherein, wj(j=0,1,2,3,4,5) for from full articulamentum to the weight vectors of full connection output layer.
The loss function of the CNN network is multiclass logarithm loss function or multiclass cross entropy cost function, this implementation
The cross entropy cost function concrete form of convolutional neural networks in example are as follows:
Wherein, yi' for i-th of value in label, yiFor the respective components in the vector through softmax normalized output.
The output layer nodal point number be defined as in "current" model gait category number namely a kind of normal gait and all abnormal gaits it
With.The present embodiment is six classification task of abnormal gait, therefore six are defined as output layer number of nodes.
Step 5, by label obtained in step 3 switch to one-hot coding (10000,01000,001000,000100,
000010,000001), and by information window, label 65% is taken to do training set after corresponding, remaining 35% does test set.It will instruction
Practice the CNN that collection is sent into step 4 and carries out more wheel training.
In the present embodiment, epochs namely all training set exercise wheel numbers take 8, batch size namely every wheel training institute
The sequence sets number taken takes 128, and learning rate takes 0.0001 to be trained.Test set assessment models classification effect is utilized after the completion of training
Fruit.Verified, model is 95.35% in six classification accuracy of test set in the present embodiment.Fig. 4 is in this preferred embodiment
Performance of the model on training set and test set after 7 wheel of training, wherein Fig. 4 is the ROC curve in training set, and Fig. 5 is test
The confusion matrix of concentration.
It should be pointed out that above-described embodiment is merely to illustrate the present invention, being achieved in that for each step can be
Variation, various modifications to these embodiments are it will be apparent that therefore all for those skilled in the art
The equivalents and improvement carried out on the basis of theory of the invention general and spirit, should all protection scope of the present invention it
It is interior.
Claims (10)
1. a kind of abnormal gait recognition methods based on depth convolutional neural networks, which is characterized in that carried out to the signal of acquisition
Directly classified to abnormal gait by original signal after pretreatment, the specific steps are as follows:
Step 1 acquires signal when human normal is walked using the IMU for being worn on human body, obtains normal gait 3-axis acceleration
Information；
Step 2 acquires signal when human body simulation typical case abnormal gait is walked using the IMU for being worn on human body, obtains abnormal step
State 3-axis acceleration information；
Step 3 walks cadence according to target representative row and initial data is done windowing cutting pretreatment and according to gait classification by each number
Respective labels are stamped according to queue；
Step 4 builds depth convolutional neural networks CNN, which is six layer structure, respectively convolutional layer one, pond layer one, volume
Lamination two, pond layer two, full articulamentum, softmax output layer；Wherein, convolutional layer one is connect entirely with pond layer one, pond layer one
It is fully connected to convolutional layer two, the full connection pool layer two of convolutional layer two, pond layer two connects full connection hidden layer, connects hidden layer entirely
It is connected to full connection output layer；Define convolutional layer, pond layer, full articulamentum and each activation primitive；
Data label obtained in step 3 is sent into step 4 by step 5 to training set and test set, training set is divided into
CNN is trained, and test set assessment models classifying quality is utilized after the completion of training.
2. abnormal gait recognition methods according to claim 1, which is characterized in that Step 1: IMU described in two is acquired
Unit be placed in shank, in shoes, loins, normal gait is consistent with the mounting means of abnormal gait.
3. abnormal gait recognition methods according to claim 1, which is characterized in that typical case's abnormal gait described in step 2
Including hemiplegic gait, parkinsonian gait, duck state, steppage gait, diplegia gait step by step.
4. abnormal gait recognition methods according to claim 1,2 or 3, which is characterized in that original described in step 3
Data do windowing cutting, and data window length representative value is two gait cycles, and piecewise step value takes away the four of window length between window
/ mono-, the label form is one-hot coding.
5. abnormal gait recognition methods according to claim 4, which is characterized in that convolutional layer one described in step 4 it is defeated
Enter matrix and be defined as every segment data window in step 3: data source port number, comes from respective signal data under current sampling frequency
Sampling number；
The output layer nodal point number is defined as gait category number namely a kind of normal gait and all abnormal gaits in "current" model
The sum of.
6. abnormal gait recognition methods according to claim 5, which is characterized in that the optimization algorithm of CNN network uses
Adam optimizer.
7. according to claim 1, abnormal gait recognition methods described in 2,3,5 or 6, which is characterized in that the hidden layer of the CNN
Excitation function is ReLU function, Sigmod function or tanh function, and output layer excitation function is Softmax function.
8. abnormal gait recognition methods according to claim 6, which is characterized in that the Adam optimizer, gradient is more
New formula are as follows:
Wherein, α is default learning rate, and ε is a decimal, and divisor is avoided to become 0,For gradient mean value, formula are as follows:
Wherein,Coefficient is exponential decay rate, controls the weight distribution of momentum and current gradient,For the gradient of t time step；
For gradient variance, formula are as follows:
Wherein,Coefficient is exponential decay rate, the influence situation of the gradient square before controlling.
9. the abnormal gait recognition methods according to claim 6 or 8, which is characterized in that each layer excitation function is set as
The linear correcting unit of ReLU, formula are as follows:
Output layer excitation function is Softmax function, formula are as follows:
Wherein,(j=0,1,2,3,4,5) for from full articulamentum to the weight vectors of full connection output layer.
10. abnormal gait recognition methods according to claim 9, which is characterized in that the loss function of the CNN network is
Multiclass logarithm loss function or multiclass cross entropy cost function, concrete form are as follows:
Wherein,For i-th of value in label,For the respective components in the vector through softmax normalized output.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910063488.5A CN109770912A (en) | 2019-01-23 | 2019-01-23 | A kind of abnormal gait classification method based on depth convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910063488.5A CN109770912A (en) | 2019-01-23 | 2019-01-23 | A kind of abnormal gait classification method based on depth convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109770912A true CN109770912A (en) | 2019-05-21 |
Family
ID=66501140
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910063488.5A Pending CN109770912A (en) | 2019-01-23 | 2019-01-23 | A kind of abnormal gait classification method based on depth convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109770912A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110232412A (en) * | 2019-05-30 | 2019-09-13 | 清华大学 | A kind of body gait prediction technique based on multi-modal deep learning |
| CN110321897A (en) * | 2019-07-08 | 2019-10-11 | 四川九洲视讯科技有限责任公司 | Divide the method for identification non-motor vehicle abnormal behaviour based on image, semantic |
| CN111067507A (en) * | 2019-12-26 | 2020-04-28 | 常熟理工学院 | Electrocardiosignal denoising method based on generation of countermeasure network and strategy gradient |
| CN112401877A (en) * | 2020-10-27 | 2021-02-26 | 中国电力科学研究院有限公司 | Method and system for monitoring behavior state of target object |
| CN112801929A (en) * | 2021-04-09 | 2021-05-14 | 宝略科技(浙江)有限公司 | Local background semantic information enhancement method for building change detection |
| CN112818927A (en) * | 2021-02-26 | 2021-05-18 | 上海交通大学 | Real-time classification method and system for human body lower limb movement modes |
| CN115019393A (en) * | 2022-06-09 | 2022-09-06 | 天津理工大学 | Exoskeleton robot gait recognition system and method based on convolutional neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104091177A (en) * | 2014-06-30 | 2014-10-08 | 华南理工大学 | Abnormal gait detection method based on determined learning theory |
| WO2016149881A1 (en) * | 2015-03-20 | 2016-09-29 | Intel Corporation | Object recogntion based on boosting binary convolutional neural network features |
| CN108062170A (en) * | 2017-12-15 | 2018-05-22 | 南京师范大学 | Multi-class human posture recognition method based on convolutional neural networks and intelligent terminal |
| CN108345846A (en) * | 2018-01-29 | 2018-07-31 | 华东师范大学 | A kind of Human bodys' response method and identifying system based on convolutional neural networks |
| CN108958482A (en) * | 2018-06-28 | 2018-12-07 | 福州大学 | A kind of similitude action recognition device and method based on convolutional neural networks |
-
2019
- 2019-01-23 CN CN201910063488.5A patent/CN109770912A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104091177A (en) * | 2014-06-30 | 2014-10-08 | 华南理工大学 | Abnormal gait detection method based on determined learning theory |
| WO2016149881A1 (en) * | 2015-03-20 | 2016-09-29 | Intel Corporation | Object recogntion based on boosting binary convolutional neural network features |
| CN108062170A (en) * | 2017-12-15 | 2018-05-22 | 南京师范大学 | Multi-class human posture recognition method based on convolutional neural networks and intelligent terminal |
| CN108345846A (en) * | 2018-01-29 | 2018-07-31 | 华东师范大学 | A kind of Human bodys' response method and identifying system based on convolutional neural networks |
| CN108958482A (en) * | 2018-06-28 | 2018-12-07 | 福州大学 | A kind of similitude action recognition device and method based on convolutional neural networks |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110232412A (en) * | 2019-05-30 | 2019-09-13 | 清华大学 | A kind of body gait prediction technique based on multi-modal deep learning |
| CN110321897A (en) * | 2019-07-08 | 2019-10-11 | 四川九洲视讯科技有限责任公司 | Divide the method for identification non-motor vehicle abnormal behaviour based on image, semantic |
| CN111067507A (en) * | 2019-12-26 | 2020-04-28 | 常熟理工学院 | Electrocardiosignal denoising method based on generation of countermeasure network and strategy gradient |
| CN112401877A (en) * | 2020-10-27 | 2021-02-26 | 中国电力科学研究院有限公司 | Method and system for monitoring behavior state of target object |
| CN112818927A (en) * | 2021-02-26 | 2021-05-18 | 上海交通大学 | Real-time classification method and system for human body lower limb movement modes |
| CN112801929A (en) * | 2021-04-09 | 2021-05-14 | 宝略科技(浙江)有限公司 | Local background semantic information enhancement method for building change detection |
| CN115019393A (en) * | 2022-06-09 | 2022-09-06 | 天津理工大学 | Exoskeleton robot gait recognition system and method based on convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109770912A (en) | A kind of abnormal gait classification method based on depth convolutional neural networks | |
| Xu et al. | InnoHAR: A deep neural network for complex human activity recognition | |
| CN109784412A (en) | The multiple sensor signals fusion method based on deep learning for gait classification | |
| CN110020623B (en) | Human body activity recognition system and method based on conditional variation self-encoder | |
| Huang et al. | Convolutional neural network based on complex networks for brain tumor image classification with a modified activation function | |
| Janarthanan et al. | Optimized unsupervised deep learning assisted reconstructed coder in the on-nodule wearable sensor for human activity recognition | |
| Ahmadi et al. | QAIS-DSNN: tumor area segmentation of MRI image with optimized quantum matched-filter technique and deep spiking neural network | |
| CN109770913A (en) | A kind of abnormal gait recognition methods based on reverse transmittance nerve network | |
| CN109993093A (en) | Road anger monitoring method, system, equipment and medium based on face and respiratory characteristic | |
| Zhang et al. | Human activity recognition based on time series analysis using U-Net | |
| CN110503082B (en) | Model training method based on deep learning and related device | |
| CN111722713A (en) | Multi-mode fused gesture keyboard input method, device, system and storage medium | |
| Huang et al. | The convolutional neural networks training with channel-selectivity for human activity recognition based on sensors | |
| CN106264545B (en) | Step recognition method and device | |
| CN106991409A (en) | A kind of Mental imagery EEG feature extraction and categorizing system and method | |
| Nguyen et al. | IMU-based spectrogram approach with deep convolutional neural networks for gait classification | |
| CN110321827A (en) | A kind of pain level appraisal procedure based on face pain expression video | |
| CN111753683A (en) | Human body posture identification method based on multi-expert convolutional neural network | |
| Tao et al. | Attention-based sensor fusion for human activity recognition using IMU signals | |
| CN115316982A (en) | Muscle deformation intelligent detection system and method based on multi-mode sensing | |
| Liu et al. | An FPGA-based upper-limb rehabilitation device for gesture recognition and motion evaluation using multi-task recurrent neural networks | |
| CN107045624A (en) | A kind of EEG signals pretreatment rolled into a ball based on maximum weighted and sorting technique | |
| CN114241270A (en) | Intelligent monitoring method, system and device for home care | |
| CN112801009B (en) | Facial emotion recognition method, device, medium and equipment based on double-flow network | |
| CN109740418A (en) | A kind of Yoga action identification method based on multiple acceleration transducers |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190521 |
|
| WD01 | Invention patent application deemed withdrawn after publication |