CN109871817A - Walking states recognition methods based on plantar pressure and adaptive integrated study - Google Patents
Walking states recognition methods based on plantar pressure and adaptive integrated study Download PDFInfo
- Publication number
- CN109871817A CN109871817A CN201910144555.6A CN201910144555A CN109871817A CN 109871817 A CN109871817 A CN 109871817A CN 201910144555 A CN201910144555 A CN 201910144555A CN 109871817 A CN109871817 A CN 109871817A
- Authority
- CN
- China
- Prior art keywords
- plantar pressure
- data
- walking states
- classification
- signal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 230000003044 adaptive effect Effects 0.000 title claims abstract description 21
- 230000005021 gait Effects 0.000 claims abstract description 47
- 238000012706 support-vector machine Methods 0.000 claims abstract description 15
- 238000003066 decision tree Methods 0.000 claims abstract description 12
- 239000013598 vector Substances 0.000 claims description 20
- 238000000926 separation method Methods 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 12
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 238000010606 normalization Methods 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 7
- 210000000988 bone and bone Anatomy 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 210000000878 metatarsophalangeal joint Anatomy 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000007599 discharging Methods 0.000 claims description 2
- 230000008439 repair process Effects 0.000 claims 1
- 210000003141 lower extremity Anatomy 0.000 abstract description 7
- 238000010801 machine learning Methods 0.000 abstract description 5
- 238000012549 training Methods 0.000 abstract description 5
- 230000005484 gravity Effects 0.000 description 5
- 238000007635 classification algorithm Methods 0.000 description 4
- 206010049214 Monopodia Diseases 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 3
- 230000003321 amplification Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000003199 nucleic acid amplification method Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000013145 classification model Methods 0.000 description 2
- 230000004069 differentiation Effects 0.000 description 2
- 238000002567 electromyography Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002902 bimodal effect Effects 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 210000003414 extremity Anatomy 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Landscapes
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The present invention is the adaptive walking states recognition methods based on integrated study and plantar pressure, this method be used only plantar pressure sensor to human body be presently in walking states identify, by be placed in vola designated position diaphragm pressure sensor acquire plantar pressure signal.Accurate vola specific position pressure change of the lower limb exoskeleton wearer in a gait cycle is obtained according to filtered signal.Plantar pressure is changed again and is sent into signal handling equipment, provides original data acquisition system for the integrated study walking states Classification and Identification device in machine learning.Signal handling equipment generates the integrated study Classification and Identification model being combined by KNN disaggregated model, multi-class classification support vector machine disaggregated model, multicategory classification decision-tree model according to data acquisition system training.It is directed to different building shape, posture simultaneously, crowd's generation of age and walking habits Classification and Identification device adaptable with it provides accurately control foundation for lower limb exoskeleton control.
Description
Technical field
The present invention relates to human body lower limbs Gait Recognition technical fields, and in particular to a kind of plantar pressure and adaptive set Cheng Xue
The adaptive walking states recognition methods practised.
Background technique
Key in lower limb exoskeleton power-assisting robot technology is to identify walking states in body gait phase, by grinding
Study carefully human body be presently in gait phase so that take corresponding control strategy.In the prior art, patent
CN201610676813 obtains walking information by acquisition electromyography signal, since electromyography signal is more faint, acquisition precision and
Accuracy is difficult to ensure that patent CN201510172243 acquires human body attitude by attitude transducer, this kind of mode needs to acquire
A variety of gait informations, such as angular speed, acceleration, magnetic field strength.Simultaneously as the difference of everyone walking habits, the above knowledge
Other mode does not consider that the accuracy of identification that may cause due to factors such as the aspectual characters of wearer declines problem, has ignored individual
Otherness.Human Sole pressure contains the abundant informations such as human body walking state, walking step state.The existing big portion of patented technology
It point is to be identified for human body in the gait pattern of different conditions, for example patent CN201711291528 is directed to level land
Walk, upstairs, downstairs, jog, hurry up, sidling, jump in place, advancing jump, creep etc. moves gait, however in each step
Walking states identification shortcoming research in the state period, in body gait phase.Accurately identifying for human body walking state determines down
Can limb ectoskeleton power-assisting robot accurately comfortably assist wearer to walk.Therefore, it is capable of providing a kind of more accurate ground
In plantar pressure feedback walking states recognition methods be very necessary.
Summary of the invention
In view of the above technical problems, it is an object of the present invention to provide one kind to be based on plantar pressure and adaptive integrated study
Walking states recognition methods.This method carries out precise classification for walking states respective under every kind of gait pattern, while with
Wearer it is different can adjust automatically control parameters, and then guarantee the accuracy rate of recognition methods.
The present invention solve the technical problem the technical solution adopted is that, provide a kind of based on integrated study and plantar pressure
Adaptive walking states recognition methods, the step of this method, is:
Step 1: human body walking period plantar pressure variable signal obtains and signal condition
Equipment, which is acquired, by plantar pressure acquires multichannel plantar pressure delta data of the human body in a gait cycle,
Then pressure data collected is amplified into filtering processing by signal condition, is carried out using signal handling equipment secondary
Filtering processing;Plantar pressure acquires riding position of the equipment in vola collection point and is at least five, respectively Metatarsophalangeal joint,
Third articulationes metatarsophalangeae, the 5th articulationes metatarsophalangeae, scaphoid and the root position of bone are set;
Step 2: the segmentation and length normalization method of walking cycle
Signal data obtained above is handled in signal handling equipment, will be gone according to the search maximum value method of difference
Walking the period is divided into several gait cycles, while selected first gait cycle passes through as normal period according to normal period
Search variation value maximum algorithm in remaining period redundant data and incomplete data position, and then delete redundant data
And incomplete data is repaired by Lagrange interpolation formula method and completes gait cycle length normalization method;
Step 3: gait cycle plantar pressure delta data separation timing values obtain
After obtaining normalised gait cycle plantar pressure delta data, become according to each channel plantar pressure
The relationship for changing data and curves peak value determines the separation timing of different walking states that is, according to variance rate maximum formula between curve
Value, divides gait cycle according to separation timing values to obtain the timing values range of walking states;
Step 4: establishing multichannel walking states characteristic set
Multichannel plantar pressure delta data after obtaining determining separation according to step 3, Five-channel plantar pressure is become
Changing data fusion becomes five dimension plantar pressure delta data vectors, and the walking states timing values range obtained according to step 3, builds
Base oneself upon the multichannel walking states characteristic set of bottom pressure data variation;
Step 5: establishing integrated study Classification and Identification model
The multichannel walking states data acquisition system obtained according to step 5 establishes KNN disaggregated model, two classification respectively first
Support vector cassification model, two categorised decision tree-models, then pass through one-to-many multi-classification algorithm for two category support vector machines
Disaggregated model and two categorised decision tree-models are converted into multi-class classification support vector machine disaggregated model and multicategory classification decision tree mould
Type, by multi-class classification support vector machine disaggregated model, multicategory classification decision-tree model, KNN disaggregated model according to Nearest Neighbor with Weighted Voting collection
An integrated study Classification and Identification model is integrated at rule.
Step 6: establishing adaptive set into learning classification identification model
User is in use, acquire the plantar pressure delta data of currently used person, and by the data immediate updating step 4
Plantar pressure variation characteristic data acquisition system γ, and it is real-time according to the plantar pressure variation characteristic data acquisition system γ of immediate updating
Amendment updates the integrated study Classification and Identification model of step 5, and self-adapting estimation goes out the walking states of different users, and then improves
Accuracy rate of the integrated study Classification and Identification model to the walking states identification of different users.
Compared with prior art, the beneficial effects of the present invention are:
The present invention be used only plantar pressure sensor to human body be presently in walking states identify, pass through placement
Diaphragm pressure sensor in vola designated position acquires plantar pressure signal.Original plantar pressure signal feeding signal is passed through
The method that software signal filtering processing algorithm and hardware signal filter circuit combine is filtered.It is obtained according to filtered signal
To accurate vola specific position pressure change of the lower limb exoskeleton wearer in a gait cycle.By the above signal acquisition
Signal handling equipment is sent into the plantar pressure variation that equipment obtains, and is the integrated study walking states Classification and Identification in machine learning
Device provides original data acquisition system.Signal handling equipment is generated according to data acquisition system training by KNN disaggregated model, multicategory classification branch
Hold vector machine disaggregated model, the original walking state classification identifier that multicategory classification decision-tree model is combined into, i.e., it is integrated to learn
Practise Classification and Identification model.Synchronous signal processing equipment is walked according to the plantar pressure data point reuse that signal collecting device acquires in real time
State classification model parameter, for different building shape, posture, crowd's generation of age and walking habits classification adaptable with it
Identifier, i.e. adaptive set provide accurately control foundation at learning classification identification model, for lower limb exoskeleton control.Relatively
In only applying a kind of machine learning algorithm, this recognition methods integrates rule according to the Nearest Neighbor with Weighted Voting in integrated learning approach and combines three
Kind machine learning algorithm, algorithm robustness are stronger.
Detailed description of the invention
Fig. 1 signal scaling amplification circuit diagram
Fig. 2 signal filter circuit figure
Fig. 3 plantar pressure sensor position distribution schematic diagram
Fig. 4 plantar pressure and variation diagram
Fig. 5 recognition methods general flow chart
Fig. 6 walking states decision-making technique flow chart
Specific embodiment
The present invention is explained further below with reference to examples and drawings, but not in this, as to the application protection scope
It limits.
A kind of adaptive walking states recognition methods based on integrated study and plantar pressure of the present invention, this method pass through foot
Bottom pressure acquires multichannel plantar pressure variable signal of the equipment acquisition human body in a gait cycle, then by foot collected
Bottom pressure variable signal carries out data processing, is split to walking cycle and plantar pressure delta data length normalization method;Again
The feature key points for extracting each walking states in each gait cycle resettle human body walking gait pattern;It then sets up just
For machine learning classification model (integrated study Classification and Identification model), while the model will training be built in real time to collected data
New self-adapting estimation model is found, until reaching 98% or more discrimination;
Gait cycle refers to that the heel contact of human body side starts the time until the same side heel lands again;It is described
Walking states include: that heelstrike, full foot is laid flat support in gait cycle, and center of gravity Forward, heeloff, tiptoe push off ground, pendulum
Dynamic leg;The molar behavior that gait pattern is taken when referring to side heel contact in people's walking process to the parapodum with landing again
State, for example, quickly race, shanking, being careful is different gait pattern;Walking states refer to that people completes a gait cycle
When certain movement of required completion, spatial pose state at human body lower limbs.For example, being heelstrike and heeloff
Different walking states.Walking cycle refers to that people completes time consumed by several gait cycles, the purpose divided here
It is to split each gait cycle from walking cycle.
The plantar pressure acquisition equipment is the equipment that can acquire plantar pressure variable signal, as plantar pressure acquires
Plate, plantar pressure sensor etc..
Plantar pressure sensor such as Fig. 3 arranges vola, and in the process of walking, vola generates pressure to pressure sensor to people.
According to pressure output electric signal, electric signal is passed to signal condition and carries out signal condition plantar pressure sensor.Signal acquisition is set
Standby acquisition plantar pressure variable signal process finishes.Plantar pressure variable signal is acquired by the above signal collecting device, will be believed
It number is transferred in signal handling equipment.Signal handling equipment carries out processing to signal and is converted into plantar pressure delta data;Again will
Delta data processing generates adaptive set and is stored in storage equipment at learning classification identification model, such as hard disk.
The specific steps of which are as follows:
(1) acquisition of human body walking period plantar pressure variable signal and signal condition
The fixation position in vola synchronizes and obtains user's plantar pressure variable signal under natural walking states, first in Fig. 3
1. pressure sensor is located at Metatarsophalangeal joint, 2. second pressure sensor is located at third articulationes metatarsophalangeae, third pressure sensor
3. being located at the 5th articulationes metatarsophalangeae, 4. the 4th pressure sensor is located at scaphoid, 5. the 5th pressure sensor is located at the root position of bone and sets.
The signal condition includes signal scaling amplifying circuit and signal filter circuit, and signal scaling amplifying circuit is with integrated
Amplifier chip MCP6004 is amplifier element, and signal filter circuit constitutes low-pass filter circuit using integrated transporting discharging chip OP207,
Passband is 1~100Hz.
The signal scaling amplifying circuit (referring to Fig. 1) includes slide rheostat R5, slide rheostat R5 and plantar pressure
(plantar pressure sensor RX-D1016 is the sensor of pressure resistance type to sensor RX-D1016, and with the variation of pressure, resistance value becomes
Change.The resistance value size of RX-D1016 is as the external world applies change in pressure) composition bleeder circuit, slide rheostat R5
Fixing end ground connection, the fixing end and resistance R1 of the sliding end connection plantar pressure sensor RX-D1016 of slide rheostat R5
One end, the reverse input end connection of one end, integrated operational amplifier MCP6004 of another terminating resistor R3 of resistance R1, integrates
One end of the noninverting input connection resistance R2 of operational amplifier, the other end ground connection of resistance R2;The sliding of RX-D1016 terminates
Supply voltage VCC, one end of the output end connection current-limiting resistance R4 of integrated operational amplifier, another termination electricity of current-limiting resistance R4
One end of R3 is hindered, and draws output voltage Vout;The size of current-limiting resistance R4 is determined according to output voltage.VSS+ is indicated in figure
Integrated operational amplifier (MCP6004 or OPA207) power supply positive supply, what VSS- was indicated is amplifier power supply negative supply.
Plantar pressure sensor output calibration is arrived by same level by bleeder circuit first, by adjusting slide rheostat
All plantar pressure sensors are unified for same voltage value in the voltage value exported under no-load condition by R5.
The plantar pressure sensor output voltage Vin in all channels is uniformly demarcated as 2.5v in the present embodiment, then passes through collection
Calibrated signal is amplified at operational amplifier MCP6004, formula is as follows:
Wherein VoutFor amplified voltage, VinFor plantar pressure sensor output voltage, Vcc is supply voltage.
Its amplification factor is set as 4 times by R3=40K, R1=10K in the present embodiment.
Usually there is a large amount of noise in obtained original plantar pressure variable signal, will affect subsequent data processing.It adopts
The method that algorithm and hardware signal filter circuit combine is filtered with software signal to handle signal.The hardware letter
Number filter circuit, as shown in Fig. 2, the filter circuit is made of First-order Rc Circuit and operational amplifier OP207.In order to filter frequency
The noise of rate 100Hz or more, the cutoff frequency for designing low-pass filter is 100Hz, and cutoff frequency and plantar pressure here is believed
Number acquisition equipment frequency acquisition it is related.
The signal amplification factor of signal scaling amplifying circuit and the selection of signal collecting device are related.Signal collecting device institute
The voltage signal intensity of output meets the needs of signal handling equipment, for example signal collecting device output voltage is 1v, letter
Number processing equipment lower voltage limit is 5v, then needs output signal at least amplifying 5 times at this time, otherwise signal handling equipment cannot respond to
It receives.Signal handling equipment refers to that any general purpose computing device that can carry out calculation processing, such as DSP, single-chip microcontroller calculate
Machine etc..
According to formula:
fl=1/2 π R6C1
Wherein flIt refers to cutoff frequency, selectes R6=1.6k Ω, C1=100uF.
The plantar pressure variable signal in five channels after hardware signal filter circuit is turned by signal handling equipment
Turn to plantar pressure delta data.Algorithm is filtered by software signal to filter plantar pressure delta data using first-order lag
Wave device carries out software secondary filtering to the collected Five-channel plantar pressure delta data of institute in each period, there is following formula:
Wherein, i refers to that the timing values of data, value range are i=1,2,3.......n (n random natural numbers).Here to adopt
The timing values of the first plantar pressure delta data collected are as benchmark 1, the timing of the vola pressure delta data acquired below
Value is labeled as 2,3,4 etc. in order.(for example the first plantar pressure delta data acquired is 2.12v, second foot of acquisition
Bottom pressure delta data is 2.46v, then collecting the pressure data of 2.12v at the moment " 1 ", the moment " 2 " collects 2.46v's
Pressure data;It is equivalent at 10 points and collects first data in 10 seconds 10 minutes;10 points collect second data in 11 seconds 10 minutes.Here
Timing values " 1 " be equivalent to " 10 minutes and 10 seconds " at 10 points, timing values " 2 " are equivalent to " 10 minutes and 11 seconds " at 10 points) xi,jIt is the i-th timing values,
The original plantar pressure delta data in jth channel;It is the i-th timing values, plantar pressure of the jth channel after software filtering
Delta data;It is the (i-1)-th timing values, plantar pressure delta data of the jth channel after software filtering;λ be ratio because
Son takes 0.9.
(2) segmentation and length normalization method of walking cycle
Signal acquisition plantar pressure delta data in a collected walking cycle be several gait cycles
Plantar pressure delta data.It needs to come out each individually period divisions therein.According in obtained gait cycle
Plantar pressure data and change curve, such as Fig. 4, abscissa is timing values in figure, and ordinate is number pressure, two peak values occurs
With the bimodal waveform of a paddy.Wherein first peak, which comes across, pushes off when the rear monopodia support phase starts parapodum, at this time monopodia
Weight bearing it is larger.Valley come across weight bearing monopodia Forward during because opposite side push off movement generate a body to
On acceleration so that plantar pressure at this time is slightly less than constitution magnitude.It is fully heelstrike same that second peak comes across opposite side
When parapodum pushes off, this brief acceleration reduces, and unilateral plantar pressure is also at higher level.
Period divisions with the following method, select plantar pressure data here and to refer to, choose in each walking cycle
Plantar pressure and the significant the lowest point of curve mark off several gait cycles then using the lowest point as division points.According to aobvious
Work experience (is verified to obtain according to the experience of life and experimental data;It is the foot and ground of people when the contact force on foot and ground is 0
When face is discontiguous, that is, at the time of people will just be lifted away from ground and just set foot on ground) it can must start in gait cycle
When, sole is leaving ground completely, plantar pressure delta data and be minimum.Using first occur peak value timing values as
Gait cycle starting point, while the point is also used as the end point of a cycle, is made with the timing values of the peak value of next appearance
For end cycle point, while the point is also used as the starting point of next cycle.This is a gait cycle.Gait cycle below is cut
It takes and herewith manages.Period divisions time difference calculation formula is as follows:
Tm=is+1-is(s≤1,2.....), s are integer,
TmIt is m-th of the gait cycle obtained after walking cycle is divided, isAnd is+1Be respectively plantar pressure delta data and
S-th of middle appearance and timing values when s+1 peak value.
Data are normalized: here using the gait cycle that obtains for the first time as normal period, gait later
Plantar pressure delta data is using search variation value maximum algorithm process in period.For redundant data, such as normal period
Length is T1, the cycle length in this section is T1+100.It is changed rate to each data here to calculate, to change rate
It takes absolute value and sorts by size, delete before change rate absolute value 100 redundant data.Similarly, for incomplete data, change is found out
100 plantar pressure delta data timing values, choose former and later two volas of the plantar pressure delta data after rate absolute value
Pressure change data are referred to as interpolation, are filled up using the method for Lagrange's interpolation and are repaired incomplete data, are normalized
Plantar pressure delta data afterwards.Lagrange's interpolation formula is as follows:
In formula, PjBeing that j-th of channel is to be inserted supplies bottom pressure delta data, iP′jBeing that j-th of channel is to be inserted supplies bottom pressure
The previous timing values of delta data, iP″jIt is j-th of channel the latter timing values to be inserted for supplying bottom pressure delta data.P′j
It is the pressure data of j-th of channel previous timing values to be inserted for supplying bottom pressure delta data, P "jIt is that j-th of channel is to be inserted
Supply the pressure data of the latter timing values of bottom pressure delta data.
Above-mentioned search variation value maximum algorithm and Lagrange's interpolation collectively form normalized.
(3) gait cycle plantar pressure delta data separation timing values obtain
After obtaining normalised gait cycle plantar pressure delta data, become according to each channel plantar pressure
Change the relationship of data and curves peak value, i.e., variance rate maximum formula between curve determines the separation timing values of different walking states.Root
Gait cycle is divided according to separation timing values to obtain the timing values range of walking states.Variance rate maximum formula between curve
It is as follows:
HereIt is the i-th timing values, the vola change pressure after the normalization in the collected all channels in j-th of channel
Statistical average;xi,jIt is i-th of timing values, the plantar pressure delta data after the normalization in j-th of channel, j=1,2,3,4,
5.. it is determined according to pressure acquisition equipment institute's acquisition position number, position number is 5 in the present embodiment.By above formula, obtain not
With the separation timing values between walking states.According to priori knowledge, Five-channel sensor is that pressure is most in initial landing
Greatly, it is liftoff pedal ground when pressure it is minimum, for the 1st, 2,3,4,5 channel sensor at " leading leg ", plantar pressure is all the smallest.
And so on have following table:
By in upper table gained pressure law and second step obtain curve between maximum difference rate compare (in table
Data be example, 1 shows that the position pressure value is greater than 0, specifically how many to be determined according to applicable cases) obtain
The timing values of each walking states separation obtain the timing values range of walking states.
(4) multichannel walking states characteristic set is established
Since the plantar pressure delta data in all channels is all synchronous acquisition, each channel obtained above is not gone together
State separation timing values are walked all to be consistent.The plantar pressure delta data fusion in 5 channels of acquisition is become into 5 dimension vola pressures
Power delta data vector, the walking states timing values range obtained according to upper step, establishes plantar pressure data variation characteristic
Set:
Wherein, γ is plantar pressure variation characteristic data acquisition system,The plantar pressure variation characteristic data of i-th timing values to
Amount,Indicate five dimension plantar pressure delta data vectors of the i-th timing values;It is the i-th timing values walking states label confidence level vector, i.e. front foot is worked as in first vector element expression
A possibility that bottom pressure delta data vector is " heelstrike " weight is 1.Second vector element indicates current vola
A possibility that pressure change data vector is " full foot is laid flat support " weight is 0, below similarly.According to 5 dimension foot obtained above
Bottom pressure delta data vector is matched with corresponding walking states label confidence level vector.Here plantar pressure changes number
It is corresponded according to vector and walking states possibility vector, it is as follows that the relationship illustrated corresponds to table:
Plantar pressure variation characteristic data acquisition system γ is stored in the storage unit of signal handling equipment.
(5) integrated study Classification and Identification model is established
According to above data set γ, signal handling equipment carries out classification based training, specifically includes the following steps:
1) multi-class classification support vector machine is established, K closes on, multicategory classification decision-tree model
The first step establishes multi-class classification support vector machine disaggregated model
According to data acquisition system γ, non-linear multi-class classification support vector machine disaggregated model is established.M in data acquisition system γi
Containing there are six classification (it is corresponding to be respectively defined as 1,2,3,4,5,6 with above-mentioned 6 kinds of states): heelstrike, full foot is laid flat support, weight
Heart Forward, heeloff, tiptoe pushes off ground, leads leg;Six two category support vector machines are respectively created to classify.
It is that one-to-many multi-classification algorithm is explained below:
Firstly, being distinguished with two category support vector machines 1 heelstrike with other walking states, being defined as y=+1 indicates foot
Heelstrike, other walking states are indicated with y=-1, will heelstrike distinguished.Then it is distinguished with two category support vector machines 2
Full foot is laid flat support and other walking states.And so on, realize the differentiation to six kinds of walking states.When a test data is defeated
When entering, six two category support vector machines respectively determine the data classification, obtain six decision probabilities.To wherein highest
Probability results as multi-category support vector machines disaggregated model final result, be denoted as h1。
Second step establishes KNN disaggregated model
Firstly, according to data acquisition system γ, by pressure data feature vectorThe 5 European near spaces of dimension are constituted, formula is as follows:
WhereinFor the plantar pressure delta data vector at the i-th moment, u is plantar pressure delta data to be sorted, i.e.,
The plantar pressure delta data currently acquired.Here k=5 is chosen, before distance data to be sorted are nearest after acquisition distance-taxis
Five data acquisition systems.The highest classification of the frequency of occurrences is exported into result as KNN disaggregated model and is denoted as h2。
Third step establishes multicategory classification decision-tree model
According to the six of data acquisition system γ classifications: heelstrike, full foot is laid flat support, and center of gravity moves forward, heeloff, tiptoe
Ground is pushed off, is led leg;Create six two categorised decision tree-models.
Using one-to-many multi-classification algorithm, more categorised decision trees are established.It is one-to-many multi-classification algorithm process interpretations below:
Firstly, being distinguished with two categorised decision tree-models 1 heelstrike with other walking states, being defined as y=+1 indicates heelstrike, to use
Y=-1 indicates other walking states, will heelstrike distinguish.Then full foot is distinguished with two categorised decision tree-models 2 to be laid flat
Support and other walking states.And so on, realize the differentiation to six kinds of walking states.When a test data inputs
It waits, six two categorised decision tree-models respectively determine the data classification, obtain six decision probabilities.To wherein highest probability
As a result it is used as multicategory classification decision-tree model output category result and is denoted as h2。
The process for establishing two category support vector machines and two categorised decision trees among the above is the prior art.
2) weighted voting algorithm establishes integrated study Classification and Identification model
Wherein weight wl, respectively 0.3,0.4,0.3, which voluntarily selects according to the confidence level to base classifier.
hlPresentation class device l output is as a result, that i.e. each basic classification device is classified as a result, l=1 indicates that KNN disaggregated model, l=2 indicate
Multi-class classification support vector machine disaggregated model, l=3 indicate multicategory classification decision-tree model.Specific ballot rule follows following
Principle: when three kinds of disaggregated models here respectively export a certain input data for classification a, when a, b, confidence level is obtained according to weight
Weight table:
Recognition result according to the available integrated study Classification and Identification model of upper table is a.By testing and verification this method
Obtained walking states recognition result success rate has reached 98.5%, has reached higher recognition accuracy.
(6) adaptive set is established into learning classification identification model
Adaptively refer to: user is in use, the plantar pressure delta data of acquisition can the variation of immediate updating plantar pressure
Characteristic set γ, and corrected in real time according to immediate updating plantar pressure variation characteristic data acquisition system γ and update above-mentioned collection
At learning classification identification model, and then improves adaptive set and identified accurately at walking states of the learning model to different users
Rate.
Specific as follows: with the walking of wearer, newest training data enters new foot by the way of first in, first out
Bottom pressure variation characteristic data acquisition systemThe update amendment of walking states label confidence level vector is basis
The walking states of the walking states and current integrated study disaggregated model identification of current integrated study Classification and Identification model identification
Next stage walking states joint determines.For example, one section of continuous data that original classification device detects all is " leading leg "
When, the walking states that next stage integrated study disaggregated model is predicted should be " heelstrike ".If the row detected
Not this state of state is walked, other states such as " center of gravity Forward ", then by the original tag value of this data " center of gravity Forward "
Confidence level weight reduces, and " center of gravity Forward " confidence level weight (the confidence level weight table i.e. in the 5th step of adjustment) is increased.Simultaneously
Walking states label confidence level vector is updated using first in, first out mode, until some element increases to 1 in vector.Here with
" leading leg " walking states are as reference standard, under this kind of walking states, everyone plantar pressure and be " 0 ".In this way,
Avoid the influence due to individual differences such as weight, walking habits for walking states identification model.
Integrated study Classification and Identification model often receives a new plantar pressure delta data, plantar pressure variation characteristic
Data acquisition system γ just deletes an element, until its identification error rate ε=f (x)-y < 0.015.Wherein, f (x) prediction is defeated
Out, y is really exported.
The present invention does not address place and is suitable for the prior art.
Claims (7)
1. a kind of the step of adaptive walking states recognition methods based on integrated study and plantar pressure, this method, is:
Step 1: human body walking period plantar pressure variable signal obtains and signal condition
Equipment is acquired by plantar pressure and acquires multichannel plantar pressure delta data of the human body in a gait cycle, then
Pressure data collected is amplified into filtering processing by signal condition, carries out secondary filtering using signal handling equipment
Processing;Plantar pressure acquires riding position of the equipment in vola collection point and is at least five, respectively Metatarsophalangeal joint, third
Articulationes metatarsophalangeae, the 5th articulationes metatarsophalangeae, scaphoid and the root position of bone are set;
Step 2: the segmentation and length normalization method of walking cycle
Signal data obtained above is handled in signal handling equipment, will be walked week according to the search maximum value method of difference
Phase is divided into several gait cycles, while selected first gait cycle passes through search according to normal period as normal period
Variation value maximum algorithm in remaining period redundant data and incomplete data position, and then delete and redundant data and lead to
It crosses Lagrange interpolation formula method and repairs incomplete data completion gait cycle length normalization method;
Step 3: gait cycle plantar pressure delta data separation timing values obtain;
Step 4: establishing multichannel walking states characteristic set
Multichannel plantar pressure delta data after obtaining determining separation according to step 3, changes number for Five-channel plantar pressure
Become five dimension plantar pressure delta data vectors, and the walking states timing values range obtained according to step 3 according to fusion, establishes foot
The multichannel walking states characteristic set of bottom pressure data variation;
Step 5: establishing integrated study Classification and Identification model
The multichannel walking states data acquisition system obtained according to step 5, establish KNN disaggregated model respectively, multicategory classification support to
Amount machine disaggregated model, multicategory classification decision-tree model, then above three model is integrated into rule according to Nearest Neighbor with Weighted Voting and is integrated into one
A integrated study Classification and Identification model;
Step 6: establishing adaptive set into learning classification identification model
User is in use, acquire the plantar pressure delta data of currently used person, and by the foot of the data immediate updating step 4
Bottom pressure variation characteristic data acquisition system γ, and corrected in real time according to the plantar pressure variation characteristic data acquisition system γ of immediate updating
The integrated study Classification and Identification model of step 5 is updated, self-adapting estimation goes out the walking states of different users.
2. the adaptive walking states recognition methods according to claim 1 based on integrated study and plantar pressure, special
Sign is that the detailed process of step 3 is: after obtaining normalised gait cycle plantar pressure delta data, according to
The relationship of each channel plantar pressure delta data peak of curve, i.e., according to variance rate maximum formula between curve, determination is not gone together
The separation timing values for walking state divide gait cycle according to separation timing values to obtain the timing values model of walking states
It encloses.
3. the adaptive walking states recognition methods according to claim 1 based on integrated study and plantar pressure, special
Sign is that the signal condition includes signal scaling amplifying circuit and signal filter circuit, and signal scaling amplifying circuit is with integrated
Amplifier chip MCP6004 is amplifier element, and signal filter circuit constitutes low-pass filter circuit using integrated transporting discharging chip OP207,
Passband is 1~100Hz.
4. the adaptive walking states recognition methods according to claim 3 based on integrated study and plantar pressure, special
Sign is that the signal scaling amplifying circuit includes slide rheostat R5, and slide rheostat R5 and plantar pressure sensor form
Bleeder circuit, the fixing end ground connection of slide rheostat R5, the sliding end of slide rheostat R5 connect plantar pressure sensor RX-
The fixing end of D1016 and one end of resistance R1, one end of another terminating resistor R3 of resistance R1, integrated operational amplifier
The reverse input end of MCP6004 connects, and one end of the noninverting input connection resistance R2 of integrated operational amplifier, resistance R2's is another
One end ground connection;The sliding of RX-D1016 terminates supply voltage VCC, the output end connection current-limiting resistance R4's of integrated operational amplifier
One end, one end of another terminating resistor R3 of current-limiting resistance R4, and draw output voltage Vout.
5. the adaptive walking states recognition methods according to claim 1 based on integrated study and plantar pressure, special
Sign is, the formula of secondary filtering are as follows:
Wherein, i refers to that the timing values of data, value range are i=1, and 2,3.......n, n is natural number;xi,jIt is the i-th timing values,
The original plantar pressure delta data in jth channel;It is the i-th timing values, plantar pressure of the jth channel after software filtering
Delta data;It is the (i-1)-th timing values, plantar pressure delta data of the jth channel after software filtering;λ be ratio because
Son.
6. the adaptive walking states recognition methods according to claim 1 based on integrated study and plantar pressure, special
Sign is, with plantar pressure data and is reference when step 2 period divisions, choose plantar pressure in each walking cycle and
The significant the lowest point of curve marks off several gait cycles then using the lowest point as division points.
7. the adaptive walking states recognition methods according to claim 1 based on integrated study and plantar pressure, special
Sign is, the ballot rule are as follows:
Wherein weight wl, respectively 0.3,0.4,0.3, hlPresentation class device l output is as a result, l=1 indicates KNN disaggregated model, l=
2 indicate that multi-class classification support vector machine disaggregated model, l=3 indicate multicategory classification decision-tree model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910144555.6A CN109871817B (en) | 2019-02-27 | 2019-02-27 | Walking state identification method based on plantar pressure and self-adaptive integrated learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910144555.6A CN109871817B (en) | 2019-02-27 | 2019-02-27 | Walking state identification method based on plantar pressure and self-adaptive integrated learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109871817A true CN109871817A (en) | 2019-06-11 |
CN109871817B CN109871817B (en) | 2021-08-27 |
Family
ID=66919419
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910144555.6A Expired - Fee Related CN109871817B (en) | 2019-02-27 | 2019-02-27 | Walking state identification method based on plantar pressure and self-adaptive integrated learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109871817B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110893100A (en) * | 2019-12-16 | 2020-03-20 | 广东轻工职业技术学院 | Device and method for monitoring posture change based on plantar pressure sensor |
CN111329484A (en) * | 2020-02-24 | 2020-06-26 | 华南理工大学 | Diabetic foot risk early warning device based on plantar pressure information time-space domain characteristics |
CN113658707A (en) * | 2021-08-26 | 2021-11-16 | 华南理工大学 | Foot varus angle detection modeling method and system |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060224246A1 (en) * | 2004-02-12 | 2006-10-05 | Clausen Arinbjorn V | Systems and methods for adjusting the angle of a prosthetic ankle based on a measured surface angle |
CN102670207A (en) * | 2012-05-15 | 2012-09-19 | 北京大学 | Gait analysis method based on plantar pressure |
CN102968986A (en) * | 2012-11-07 | 2013-03-13 | 华南理工大学 | Overlapped voice and single voice distinguishing method based on long time characteristics and short time characteristics |
CN103268500A (en) * | 2013-05-29 | 2013-08-28 | 山东大学 | Gait identifying method with robustness to walking gait changes |
CN104729507A (en) * | 2015-04-13 | 2015-06-24 | 大连理工大学 | Gait recognition method based on inertial sensor |
CN106308809A (en) * | 2016-08-15 | 2017-01-11 | 河北工业大学 | Method for recognizing gait of thigh amputation subject |
CN106778509A (en) * | 2016-11-23 | 2017-05-31 | 清华大学 | A kind of Gait Recognition device and method |
-
2019
- 2019-02-27 CN CN201910144555.6A patent/CN109871817B/en not_active Expired - Fee Related
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060224246A1 (en) * | 2004-02-12 | 2006-10-05 | Clausen Arinbjorn V | Systems and methods for adjusting the angle of a prosthetic ankle based on a measured surface angle |
CN102670207A (en) * | 2012-05-15 | 2012-09-19 | 北京大学 | Gait analysis method based on plantar pressure |
CN102968986A (en) * | 2012-11-07 | 2013-03-13 | 华南理工大学 | Overlapped voice and single voice distinguishing method based on long time characteristics and short time characteristics |
CN103268500A (en) * | 2013-05-29 | 2013-08-28 | 山东大学 | Gait identifying method with robustness to walking gait changes |
CN104729507A (en) * | 2015-04-13 | 2015-06-24 | 大连理工大学 | Gait recognition method based on inertial sensor |
CN106308809A (en) * | 2016-08-15 | 2017-01-11 | 河北工业大学 | Method for recognizing gait of thigh amputation subject |
CN106778509A (en) * | 2016-11-23 | 2017-05-31 | 清华大学 | A kind of Gait Recognition device and method |
Non-Patent Citations (2)
Title |
---|
RU MA 等: "A speed一independent feedback index for walking pattern recognition for a walking assistive robotic suit", 《2017 IEEE 8TH TNTERNATIONAL CONFERENCE ON CTS&RAM》 * |
陈贵亮 等: "适应个体差异的下肢康复机器人步态规划", 《机械设计与制造》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110893100A (en) * | 2019-12-16 | 2020-03-20 | 广东轻工职业技术学院 | Device and method for monitoring posture change based on plantar pressure sensor |
CN111329484A (en) * | 2020-02-24 | 2020-06-26 | 华南理工大学 | Diabetic foot risk early warning device based on plantar pressure information time-space domain characteristics |
CN113658707A (en) * | 2021-08-26 | 2021-11-16 | 华南理工大学 | Foot varus angle detection modeling method and system |
Also Published As
Publication number | Publication date |
---|---|
CN109871817B (en) | 2021-08-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109871817A (en) | Walking states recognition methods based on plantar pressure and adaptive integrated study | |
CN103426438B (en) | Method and system for analyzing baby crying | |
CN107092894A (en) | A kind of motor behavior recognition methods based on LSTM models | |
CN107219924B (en) | A kind of aerial gesture identification method based on inertial sensor | |
US20080288493A1 (en) | Spatio-Temporal Self Organising Map | |
CN105224104B (en) | Pedestrian movement's state identification method based on smart mobile phone grip mode | |
CN107398912B (en) | Domestic robot user behavior statistical system | |
CN108764123A (en) | Intelligent recognition human body sleep posture method based on neural network algorithm | |
CN104809469A (en) | Indoor scene image classification method facing service robot | |
CN111611859B (en) | Gait recognition method based on GRU | |
CN111126280B (en) | Gesture recognition fusion-based aphasia patient auxiliary rehabilitation training system and method | |
CN107688790A (en) | Human bodys' response method, apparatus, storage medium and electronic equipment | |
CN108703824A (en) | A kind of bionic hand control system and control method based on myoelectricity bracelet | |
CN110977961A (en) | Motion information acquisition system of self-adaptive power-assisted exoskeleton robot | |
CN107609501A (en) | The close action identification method of human body and device, storage medium, electronic equipment | |
CN110532898A (en) | A kind of physical activity recognition methods based on smart phone Multi-sensor Fusion | |
CN113314209A (en) | Human body intention identification method based on weighted KNN | |
CN111046655B (en) | Data processing method and device and computer readable storage medium | |
CN109948640A (en) | Electromyographic signal classification method based on two-parameter core Optimization-type extreme learning machine | |
CN114492634B (en) | Fine granularity equipment picture classification and identification method and system | |
CN108433728A (en) | A method of million accidents of danger are fallen based on smart mobile phone and ANN identification construction personnel | |
CN110132276B (en) | Self-adaptive step length estimation method based on pedestrian motion state | |
WO2022100187A1 (en) | Mobile terminal-based method for identifying and monitoring emotions of user | |
Ergene et al. | Robotic hand grasping of objects classified by using support vector machine and bag of visual words | |
CN112949676B (en) | Self-adaptive motion mode identification method of flexible lower limb assistance exoskeleton robot |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210827 |