CN103093234B - Based on the personal identification method of ground reaction force during walking - Google Patents
Based on the personal identification method of ground reaction force during walking Download PDFInfo
- Publication number
- CN103093234B CN103093234B CN201210559665.7A CN201210559665A CN103093234B CN 103093234 B CN103093234 B CN 103093234B CN 201210559665 A CN201210559665 A CN 201210559665A CN 103093234 B CN103093234 B CN 103093234B
- Authority
- CN
- China
- Prior art keywords
- wavelet
- gait
- force
- reaction force
- ground reaction
- 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.)
- Active
Links
Landscapes
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses a kind of personal identification method based on ground reaction force during walking, feature utilizes the gait information by when being hidden in gait channel acquisition walking that ground or underfloor three-dimensional force force plate/platform form, and data prediction is carried out to it, wavelet packet decomposition algorithm is adopted to extract WAVELET PACKET DECOMPOSITION coefficient again, wavelet-packet energy, the average of WAVELET PACKET DECOMPOSITION coefficient and variance characterize gait feature, feature selecting algorithm and support vector machine classifier is utilized to set up the gait feature template and classification model of having trained in training process, the gait feature of gait feature template to the object to be identified extracted of having trained is utilized to carry out Feature Mapping in identifying, recycle the classification model of having trained and template matches and identification are carried out to it, export recognition result.Gait information acquisition of the present invention and identifying have disguise, can not cause and discover and human rights dispute, can meet identity verify and the safety precaution demand in cyberage security sensitive place better.
Description
Technical field
The invention belongs to mode identification technology, particularly based on the personal identification method of ground reaction force during walking.
Background technology
The world today faces the severe attack of terrorism and finance theft waits safety difficulties, and identification or certification become more and more important, especially on airport, customs, museum, party scene, the security sensitive place such as bank and national treasury.Biometrics identification technology be utilize human body intrinsic physiological characteristic and behavioural characteristic to realize individual identity identification or certification, because selected biological characteristic is that human body is intrinsic, can carry with, do not need memory, there is uniqueness and be difficult to the features such as replicability, being more suitable for the identification under new demand for security or certification.At present, the existing kind more than ten of the biological characteristic for identification, wherein, physiological characteristic mainly contains fingerprint, face, iris, palmmprint, palm shape, people's ear, retina, DNA and smell etc., and behavioural characteristic mainly contains signature, voice, keystroke and gait etc.The recognition technologies such as fingerprint, palmmprint, palm shape, iris, face, DNA, voice and signature have comparatively ripe product, and they all serve positive effect in respective application.But regrettably, the biometrics identification technologies such as fingerprint, face, iris, palmmprint, palm shape, people's ear, retina, DNA, smell, signature, voice and keystroke all need checked object close fit just can complete information acquisition when practical application, do not possess disguise, easily cause discovering and then taking counterreconnaissance means of offender.Meanwhile, physiological characteristic used is all static nature, is easy to forge; This three behaviors feature of keystroke, voice and signature is also easily imitated; Except senior recognition of face, above-mentioned biometrics identification technology is all contact collection and closely identifies, easily causes the human rights to dispute on and diplomatic disturbance, is not too applicable to the anti-terrorism safety check demand of entry and exit port.Although DNA authentication technique reliability is high, sampling neither hidden, and its analytic process is longer, requires high-end instrument and equipment, to be only confined to laboratory at present and to use and during law identifies.That is, the biometrics identification technology based on these physiological characteristics and this three behaviors feature of keystroke, voice and signature is closely identity recognizing technology, can not adapt to the identification requirement under current safety situation well.Therefore, need development and apply remote identity recognizing technology.
At present, the biological characteristic that can be used for remote identification has face and gait.What still adopt due to remote recognition of face is the human face image information that video camera or camera obtain, its identify and closely recognition of face equally by illumination, people or thing block with background influence comparatively greatly, and the face tracking a great problem especially under crowded crowd or complex environment.The posture that Gait Recognition is walked by people and/or the footprint stayed identify and certification identity, people can have such cognitive experience: we can walk farly just recognize familiar people by observing his/her in crowd, or hear that when not seeing its people walking sound just can recognize familiar people, this cognitive experience is also by the experimental verification of psychophysics family expenses, they point out: even if in very poor environment of observation, and the mankind are the capable identity distinguishing walker according to gait also.At computer vision field, Gait Recognition has more advantage relative to recognition of face, and it utilizes the walking behavioural characteristic of people, and unlike face, the means such as easy face mould or cosmetic just can be got by under false pretences; Also can play a role in some special event, such as in the events such as bank raid and Cultural relics in museum theft, suspect puts on face shield or cap just can be hidden face picture and escape from the tracking of video camera, but gait remains visible, or can carry out the discriminating of gait to judge whether it is real criminal, precondition is that suspect does not deliberately change its gait.In fact, people on vision pattern, others' gait can be imitated the spitting image of, produce gait image information about the same, thus cause the gait recognition method in computer vision correctly to distinguish.Therefore, need research and development based on the personal identification method of other gait information.
Chinese patent ZL01144157.7 discloses a kind of method that gait image sequence by human locomotion identifies personnel, the method information acquisition has the advantages such as untouchable and disguised, the methods such as space profiles segmentation, principal component analysis (PCA) and personalized corporal characteristic auxiliary examination that take are to improve discrimination and to reduce calculation cost, but this method based on gait image is owing to using common camera shooting gait image, can't ideally solve complex background, block the problem with noise, also not consider the problem of deliberately imitating gait.A kind of multi-axial forces plane matrix be made up of four six axle power platforms disclosed in Chinese patent ZL200410014352.9, and the method for dynamic gait information when obtaining human body walking by this plane matrix, this plane matrix can be used for the testing and analysis of balanced ability of human body, the coordination ability and nervous function, but does not relate to the method for Method of Gait Feature Extraction and Gait Recognition.Disclosed a kind of " the gait detecting system of US Patent No. 2002/0107649A1, pick-up unit, equipment and gait detection method " (Gaitdetectionsystem, gaitdetectionapparatus, device, andgaitdetectionmethod), sound oscillation signal when detecting walking is also used as personal identification system, this system requirements places microphone (acoustic-electric converter) on human body, microphone obtains electric signal subset by the sound oscillation energy gathered in walking movement process, use this electric signal subset as the index of a step to detect walking period, collect the signal subset of sound oscillation when expression pin lands, system extracts the specific gait feature of human body based on this signal subset, and use it for individual's identification.This personal identification system has two deficiencies: one is sound oscillation when landing changes along with the placement position of microphone on human body; Two is due to the major effect by microphone ambient noise and electrical noise, can not determine sound oscillation electric signal when expression pin lands exactly, also just be difficult to the feature extracting gait waveform exactly.Disclosed " gait waveform feature extracting method and personal identification system " (the Gaitwaveformfeatureextractingmethodandindividualidentifi cationsystem) of patented claim international publication number WO/2004/040501 is using the waveform peak amplitude in allocated frequency band as reference index, and determine the gait waveform of a step in the electric signal detected from electric field displacements detecting device and extract method and the personal identification system of gait waveform feature, can not by left, the installation position impact on health of charge interference between right leg and sensor, but be not still suitable in the environment that there is complicated electromagnetic interference (EMI).And, in the technical scheme that US Patent No. 2002/0107649A1 and international publication number WO/2004/040501 opens, its gait information remains the sensor collection relying on and be placed on health, remain contact type measurement mode, measured can perceive and obtains, thus offender may be caused because perceiveing to take counterreconnaissance means to get by under false pretences, even may be stung to fury so that the event that wakes a snake in crowd.
Summary of the invention
The object of the invention is to propose a kind of personal identification method based on ground reaction force during walking, with hidden and non-contact measurement method, identifying is not easily realized, avoid being identified object to get by under false pretences, to be applicable to the identification in the security sensitive places such as customs, airport, museum, party scene, bank and national treasury.
The present invention is based on the personal identification method of ground reaction force during walking, it is characterized in that comprising training process and identifying;
Described training process comprises: utilize by gait information when being hidden in gait channel acquisition walking that ground or underfloor three-dimensional force force plate/platform form and set up gait data storehouse, carry out data prediction, Method of Gait Feature Extraction, gait feature selects and sorter is trained, the gait feature template of finally having been trained and classification model;
Described identifying comprises: obtain the real-time gait information of object to be identified to obtain test sample book, data prediction, Method of Gait Feature Extraction are carried out to test sample book, the gait feature template of having trained is utilized to carry out Feature Mapping to the gait feature of object extraction to be identified, recycle the classification model of having trained, according to nearest neighbouring rule, template matches and identification are carried out to test sample book, export recognition result;
Described gait information is the ground reaction force of the single step gathered by three-dimensional force force plate/platform, or the ground reaction force of continuous multiple single step cumulative ground reaction force obtaining synthesizing in sequential;
The ground reaction force of described single step or the ground reaction force of synthesis are all made up of these three components of left and right directions shearing force, fore-and-aft direction shearing force and vertical direction anchorage force;
Described data prediction comprises denoising and effective sample is selected; Described denoising refers to and adopts Wavelet Transform Threshold method to carry out level discharge rating process to ground reaction force data; Described effective sample selects the data dimension that refers to vertical direction anchorage force in the ground reaction force after according to denoising and whether peak point numerical value judges in effective range, selects ground reaction force data in effective range as effective sample; Described effective range refers to the numerical range relevant with sample frequency Fp to the number Np of adopted three-dimensional force force plate/platform, for the ground reaction force of single step, if wherein the data dimension of vertical direction anchorage force is lower than 0.4Fp or higher than 0.8Fp, or the crest of vertical direction anchorage force and the ordinate value of trough differ by more than 300, namely be judged to be invalid sample, otherwise be judged to be effective sample; For the ground reaction force that the continuous multiple single steps gathered by Np three-dimensional force force plate/platform are synthesized, if wherein the data dimension of vertical direction anchorage force is lower than 0.4Fp × Np or higher than 0.8Fp × Np, or the number of the trough of vertical direction anchorage force is less than Np, or the maximin of the ordinate value of all troughs of vertical direction anchorage force differs by more than 300, namely be judged to be invalid sample, otherwise be judged to be effective sample;
Described Method of Gait Feature Extraction refers to and adopts Wavelet Packet Transform Method from ground reaction force, to extract WAVELET PACKET DECOMPOSITION coefficient, wavelet-packet energy, the average of WAVELET PACKET DECOMPOSITION coefficient and variance to characterize gait feature, and step is:
First adopt piecewise linear interpolation algorithm to carry out dimension normalization to the ground reaction force data through data prediction, the data dimension of effective sample is normalized to same value;
Adopt L layer wavelet packet decomposition algorithm to carry out WAVELET PACKET DECOMPOSITION to the effective sample after denoising and dimension normalization again, decompose and obtain 2
l+1-2 sub-wavelet packets, characterize gait feature with the average of the WAVELET PACKET DECOMPOSITION coefficient of every sub-wavelet packet, wavelet-packet energy, WAVELET PACKET DECOMPOSITION coefficient and variance;
Then feature selecting algorithm is adopted to select WAVELET PACKET DECOMPOSITION coefficient, pick out wavelet packets coefficient of dissociation, the average of wavelet packets coefficient of dissociation and wavelet-packet energy, WAVELET PACKET DECOMPOSITION coefficient and variance are combined, the gait feature template obtaining gait feature collection and trained;
Described sorter training refers to and adopts support vector machine classifier to train training sample, the classification model of having been trained; The training of this sorter also first can adopt the gait feature of maximum-minimum requirement method to input support vector machine to be normalized in amplitude before described employing support vector machine classifier is trained training sample.
Compared with prior art, the remarkable advantage that the present invention is based on the personal identification method of ground reaction force during walking has:
One, the ground reaction force that when being human body walking for the gait information of identification in the present invention, foot and ground interaction produce, can by being hidden in ground or underfloor force plate/platform convenience, get exactly, relative to by sensor placement in measured's acquisition walking sound oscillation signal with it and the method for walking electric field signal, under the gait information collecting device in the present invention can be hidden in ground or floor, be not placed in it measured, gait information acquisition process is hidden to measured completely, naturally walking without any impact on measured.That is, gait information acquisition of the present invention is noncontact and hidden measuring method, true gait information when being identified the walking of object nature can be obtained, be identified object and can not realize there is identification apparatus and takes measures to get by under false pretences, also identifying can be monitored round the clock in far place at the scene, also can not cause human rights dispute, identity verify and the safety precaution demand in the security sensitive places such as cyberage airport, customs, museum, party scene, bank and national treasury can be met better.
They are two years old, due to the ground reaction force that foot when gait information used in the present invention is the human body walking of force plate/platform acquisition produces with ground interaction, obtain the method for gait image relative in computer vision field by video camera, recognition result not to block etc. by complex background, clothes and health to be affected; Advantageously be a bit, the present invention can tackle situation during deliberately mimicker's gait, and computer vision field cannot better be tackled based on the personal identification method of gait, because the gait of a people mimicker often learns from vision pattern the posture that others walks, can imitate the spitting image of, but because cannot observe and record the gait mechanical information of foot and ground interaction when others walks, also just cannot learn, imitate mechanical information, even if similar walking posture also differs produce similar gait mechanical information surely.
Its three, adopt the inventive method, single force plate/platform both can have been used to obtain the ground reaction force of single step, the ground reaction force of single step is used for identification; Also multiple force plate/platform can be used to obtain the ground reaction force of continuous multiple single step, then the cumulative ground reaction force obtaining synthesizing in sequential, then the ground reaction force of synthesis is used for identification; The feature extraction adopted and sorting algorithm can be distinguished and process both of these case well, facilitate the application of different places.
Its four, the gait mechanical information acquisition method, Method of Gait Feature Extraction and the identification algorithm that adopt in the present invention can promote the use of a lot of field.Such as, in medical rehabilitation field, by the contrast with normal person's gait feature, evaluating patient skeletal muscle etc. can organize the degree suffered damage and the ability controlling balance, patient is classified and classification, determine therapy rehabilitation scheme, and curative effect, treatment later evaluation rehabilitation degree can be verified over the course for the treatment of; In addition, this gait mechanical information can also be applied to the fields such as insurance assessment, Claims Resolution (insurance company is used for disability assessment, therapy rehabilitation assessment) and human lives's means of production (furniture, shoes, artificial limb, office accommodations etc.) and body-building, the design of rehabilitation material and measures of effectiveness.
Accompanying drawing explanation
Fig. 1 is the algorithm flow chart of the identification based on ground reaction force during walking;
Fig. 2 is the staggered at equal intervals assembled gait access diagram of ground reaction force when being hidden in subsurface collection walking;
Fig. 3 is the tight concordant assembled gait access diagram of ground reaction force when being hidden in subsurface collection walking;
The force-time curve of the left and right directions shearing force Fx in the ground reaction force of single step when Fig. 4 is walking, fore-and-aft direction shearing force Fy and vertical direction anchorage force Fz;
Left and right directions shearing force Fhx, fore-and-aft direction shearing force Fhy in the ground reaction force of continuous multiple single step synthesis when Fig. 5 is walking and the force-time curve of vertical direction anchorage force Fhz;
Fig. 6, Fig. 7 and Fig. 8 are left and right directions shearing force curve, fore-and-aft direction shearing force curve and vertical direction anchorage force curve in the ground reaction force of single step before denoising respectively;
Fig. 9, Figure 10 and Figure 11 are left and right directions shearing force curve, fore-and-aft direction shearing force curve and vertical direction anchorage force curve in the ground reaction force of single step after adopting the denoising of Wavelet Transform Threshold method respectively;
Figure 12 is the force-time curve of left and right directions shearing force Fhxd, fore-and-aft direction shearing force Fhyd in the ground reaction force of synthesis after denoising and dimension normalization and vertical direction anchorage force Fhzd;
Figure 13 is through the vertical direction anchorage force Fwz curve after denoising, weight normalized and dimension normalized;
Figure 14 carries out the second layer metric space after 4 layers of WAVELET PACKET DECOMPOSITION to vertical direction anchorage force Fwz;
Figure 15 carries out the second layer wavelet space after 4 layers of WAVELET PACKET DECOMPOSITION to vertical direction anchorage force Fwz.
Embodiment
Embodiment 1:
Remote biometrics identification technology is the important identity recognizing technology means of high-tech crime epoch protection public safety and personal security; can be used for the biological characteristic of remote identification at present except face, the human body behavior biological characteristic that gait is with the largest potentiality at last.The present invention with the gait of people for starting point, adopt the ground reaction force that when being hidden in ground or underfloor force plate/platform collection people's walking, foot and ground interaction produce, replace the gait image sequence of comparatively easily imitating with not imitable gait mechanical information, and propose the acquisition of gait mechanical information, disposal route and personal identification method.Below in conjunction with drawings and Examples, the present invention is further illustrated, it is pointed out that described embodiment is for the ease of understanding the present invention, not playing any restriction effect to the present invention.
What the present invention proposed comprises training process and identifying based on the counteractive personal identification method in ground during walking, and algorithm flow as shown in Figure 1.In the present embodiment, no matter be the gait information of training process or identifying, it is all the ground reaction force during walking adopting the gait channel acquisition being hidden in ground or underfloor three-dimensional force force plate/platform composition to arrive.The present invention utilizes the gait channel acquisition gait information be made up of force plate/platform, set up gait data storehouse, then denoising and effective sample are carried out to the gait information collected and the data prediction such as to select, bamboo product feature extraction algorithm extracts gait feature, feature selecting algorithm is adopted to pick out optimum gait feature in the training process, the gait feature template of having been trained, select suitable sorter to train training sample simultaneously, the classification model of having been trained, the algorithm flow of training process is as shown in the first half in Fig. 1; Adopt in identifying the feature extraction algorithm the same with training process from Real-time Obtaining to object to be identified gait information extract gait feature, the gait feature template of having trained obtained in training process is utilized to carry out Feature Mapping to the gait feature of object to be identified, the classification model of having trained obtained in recycling training process carries out template matches and identification according to nearest neighbouring rule to test sample book, finally export recognition result, the algorithm flow of identifying is as shown in Fig. 1 the latter half.
Below gait information acquisition of the present invention, Method of Gait Feature Extraction and recognition methods are described in more detail.
1. gait information acquisition
From current technology realizability, the gait information that can be used for Gait Recognition has gait image/video information, gait acoustic information, gait electromagnetic field signal and gait mechanical information, gait image/video information has had a lot of research in Gait Recognition application, but still exists and blocked and the problem such as illumination effect; Gait acoustic information and the collection of gait electromagnetic field signal need sensor placement with it the measured, remain contact type measurement mode, and measured can perceive and obtains, and is easy to occur being taked the means such as imitation to get by under false pretences because perceive by discriminating one's identification.The present invention is directed to Gait Recognition these deficiencies on an information layer, propose gait mechanical information, the ground reaction force that namely during walking, foot and ground interaction produce is used for identification.In order to collect gait mechanical information as much as possible when object is walked naturally, under multiple three-dimensional force force plate/platform is arranged on ground or floor by certain erection method by the present invention snugly, form a gait passage, gather human body walking by three-dimensional ground reaction force during gait passage.Adopt China Patent No. for six axle power platforms described in ZL200410014352.9 " a kind of multi-axial forces plane matrix and human body walking gait information getting method " are to gather three-dimensional force in the present invention, be referred to as three-dimensional force force plate/platform in the present invention.In specific implementation process, by multiple three-dimensional force force plate/platform according to the place size of application places and application demand, different erection methods is adopted to form gait passage, Fig. 2 is the staggered at equal intervals assembled gait access diagram of ground reaction force when being hidden in subsurface collection walking, and multiple three-dimensional force force plate/platform Plt is assemblied to form a gait passage alternately with identical the first from left right side, interval one; Fig. 3 is the tight concordant assembled gait access diagram of ground reaction force when being hidden in subsurface collection walking, and multiple three-dimensional force force plate/platform Plt is closely evenly assemblied to form a gait passage one by one.Staggered at equal intervals assembled gait passage shown in Fig. 2 can distinguish the ground reaction force of left and right pin simply relative to the tight concordant assembled gait passage shown in Fig. 3, but require that the left foot Lf of measured and right crus of diaphragm Rf correctly sets foot on corresponding force plate/platform by the left and right corresponded manner indicated in figure, be a three-dimensional force force plate/platform Plt below region as black box sign each in Fig. 2, on direct of travel, require in the black box that measured right crus of diaphragm Rf steps on the right side of dotted line in the drawings, left foot Lf steps in the black box on the left of dotted line in the drawings, so just can ensure that measured correctly sets foot on corresponding force plate/platform, gait passage shown in Fig. 3 do not have this requirement.
In the present invention, the ground reaction force gathered is three-dimensional ground reaction force, comprise left and right directions shearing force Fx, fore-and-aft direction shearing force Fy and vertical direction anchorage force Fz, the force-time curve of the left and right directions shearing force Fx in the ground reaction force of single step when Fig. 4 is walking, fore-and-aft direction shearing force Fy and vertical direction anchorage force Fz, vertical direction anchorage force Fz mainly reflects the load-bearing process of walking, its force-time curve generally in horse-saddle, namely generally has two crests and clips a trough; Shearing force mainly reflects the process of friction catch, driving and balance health.Therefore, three-dimensional ground reaction force during walking is adopted will intuitively to characterize the pattern feature of walking from multi-angle.
2. gait data pre-service
In order to make the gait feature more accurate stable extracted, require that the gait data being used for feature extraction has higher quality, and the data collected are often containing noise, some data even may because the invalid data of acquisition operations generation lack of standardization.Therefore, before carrying out Method of Gait Feature Extraction, need to carry out pre-service to the gait data collected, comprise denoising and effective sample is selected.
For noises such as the association interference between the Hz noise existed in data acquisition circuit, circuit, on circuit, adopted hardware filtering method to carry out preliminary denoising, in order to eliminate various noise as far as possible, host computer adopt again software filtering method carry out further denoising.Be described as the multiresolution analysis characteristic of " school microscop " because wavelet transformation has, make it have band-pass filtering function, Wavelet noise-eliminating method application is more and more wider.Signal is carried out wavelet decomposition by Wavelet noise-eliminating method under different scale, just the mixed signal comprising multi-frequency composition can be decomposed different frequency range, then presses frequency band process according to the different characteristic of each seed signal on frequency domain.Ground reaction force signal during walking is the low frequency signal of frequency within 40Hz mainly, and by the overriding noise in gait channel acquisition system be working power introduce 50Hz industrial frequency noise and Acquisition Circuit in electromagnetic interference (EMI), electromagnetic interference (EMI) major part is high frequency noise.In addition, the singular point of the ground reaction force collected is more, and its force-time curve there will be a lot of flex point; And wavelet modulus maxima method and Wavelet Transform Threshold method are relatively applicable to carrying out denoising containing the more signal of singular point or data, again because Wavelet Transform Threshold method is applicable to the denoising of most of signal, and computing velocity is fast, is very suitable for the requirement ground reaction force data in the present invention being carried out to real-time de-noising process.Therefore, the present invention selects Wavelet Transform Threshold method to carry out noise suppression preprocessing to ground reaction force data.Wavelet Denoising Method comprises wavelet decomposition → to three basic operations of high-frequency wavelet coefficient process → wavelet reconstruction, for time-domain signal f (t), first it is discretely turned to f
k, make c
0, k=f
k.The gait data collected in the present invention can be considered as the data of discretize, i.e. f
k=f (t), the orthogonal wavelet decomposition formula of signal f (t) is:
Wavelet reconstruction formula is:
In formula, c
j,kfor scale coefficient, d
j,kfor wavelet coefficient, h, g are pair of orthogonal mirror filter group, and j is Decomposition order, and N is that (data length) is counted in discrete sampling.
For Wavelet Transform Threshold method, estimate wavelet coefficient c according to threshold value λ
j,knormal employing two kinds of functions: soft-threshold function and hard threshold function, the Wavelet Transform Threshold method based on these two kinds of functions is called wavelet transformation Soft thresholding and wavelet transformation hard threshold method.Carry out wavelet reconstruction by the wavelet coefficient estimated and can obtain the signal after denoising.Soft-threshold function is:
Hard threshold function is:
In formula, sgn () is sign function, and λ is threshold value.The choosing method of threshold value λ is more, comparatively large on denoising effect impact, should select according to pretreated signal.The uniform threshold proposed by Donoho and Johnstone
more with obtaining, σ is the standard deviation of noise.
By analyzing and experimental verification, in the present embodiment, determine that employing 5 layers of Daubechies wavelet function carry out denoising as the wavelet basis function of Wavelet Transform Threshold method to gait data.Fig. 6, Fig. 7 and Fig. 8 are the example of left and right directions shearing force curve, fore-and-aft direction shearing force curve and vertical direction anchorage force curve in the ground reaction force of single step before denoising respectively.Wherein, Fig. 6 is the force-time curve of the left and right directions shearing force before denoising, and Fig. 7 is the force-time curve of the fore-and-aft direction shearing force before denoising, and Fig. 8 is the force-time curve of the vertical direction anchorage force before denoising; Fig. 9, Figure 10 and Figure 11 adopt the ground reaction force data of Wavelet Transform Threshold method to the single step shown in Fig. 6, Fig. 7 and Fig. 8 to carry out the force-time curve example after denoising.Wherein, Fig. 9 is the force-time curve of the left and right directions shearing force after denoising, and Figure 10 is the force-time curve of the fore-and-aft direction shearing force after denoising, and Figure 11 is the force-time curve of the vertical direction anchorage force after denoising.By finding the contrast of these curves, the unique point of original signal, while reduction noise, all saves by Wavelet Transform Threshold method, and does not change details and the configuration feature of original signal, proves that its denoising effect is better.
Effective sample in the present invention is selected the data dimension of the vertical direction anchorage force in the ground reaction force after according to denoising and peak point numerical value in effective range, whether is judged that gathered gait data whether can as effective sample, selects gait data in effective range as effective sample according to term of reference.Described effective range refers to the numerical range relevant with sample frequency Fp to the number Np of adopted three-dimensional force force plate/platform, for the ground reaction force of single step, if wherein the data dimension of vertical direction anchorage force is lower than 0.4Fp or higher than 0.8Fp, or the crest of vertical direction anchorage force and the ordinate value of trough differ by more than 300, namely be judged to be invalid sample, otherwise be judged to be effective sample; For the ground reaction force that the continuous multiple single steps gathered by Np three-dimensional force force plate/platform are synthesized, if wherein the data dimension of vertical direction anchorage force is lower than 0.4Fp × Np or higher than 0.8Fp × Np, or the number of the trough of vertical direction anchorage force is less than Np, or the maximin of the ordinate value of all troughs of vertical direction anchorage force differs by more than 300, namely be judged to be invalid sample, otherwise be judged to be effective sample.In the present embodiment, the sample frequency Fp of three-dimensional force force plate/platform is 800Hz.Gone by the staggered at equal intervals assembled gait passage shown in Fig. 2 is up by 5 three-dimensional force force plate/platforms for a people, when being used for identification with the ground reaction force of single step, if the data dimension of the vertical direction anchorage force Fz of ground reacting force is lower than 320 or higher than 640 as shown in Figure 4, or the crest of vertical direction anchorage force Fz and the ordinate value of trough differ by more than 300, namely be judged to be invalid sample, otherwise be judged to be effective sample, when being used for identification with the ground reaction force of continuous multiple single step synthesis, if the data dimension of vertical direction anchorage force Fhz is lower than 1600 or higher than 3200 in the ground reaction force curve of synthesis as shown in Figure 5, or the trough of vertical direction anchorage force Fhz (as cross curve in Fig. 5 indicate these points of V) number be less than 5, or all troughs of vertical direction anchorage force Fhz (as cross curve in Fig. 5 indicate these points of V) the maximin of ordinate value differ by more than 300, as long as occur a kind of in these three kinds of situations, this gait data is invalid sample, contrary, be effective sample, the gait data of effective sample is set up gait data storehouse.
3. Method of Gait Feature Extraction and selection
Gait feature is the basis of gait analysis and Gait Recognition, based on the quality directly affecting classification results in the identification of gait for the quality of the gait feature of classifying, therefore, the gait feature that the feature extraction algorithm Extraction and discrimination performance that needs design is good, bamboo product feature selecting algorithm reduces intrinsic dimensionality, improves efficiency of algorithm.
First, as the scheme of alternative, the ground reaction force of single step that described for the ground reaction force of training and identify can be, or the ground reaction force of the ground reaction force of continuous multiple single step cumulative synthesis obtained in sequential.Abundant gait information is comprised in ground reaction force during walking, except the time domain charactreristic parameter of the reflection gait cycle that can see from ground reaction force curve and globality, the feature of more reflection walking frequency spectrums and local detail is hidden on frequency domain.Frequency-domain analysis method is the analytical approach of signal common, can go out to excavate the feature that signal cannot be observed in time domain.Wavelet transformation is as a kind of outstanding signal time frequency analyzing tool, the characteristic of its multiresolution analysis makes us have the ability to characterize in time-frequency two territory the local feature of gait, and wavelet package transforms is more better than time of wavelet transformation and frequency resolution with it and more can be competent at our demand.WAVELET PACKET DECOMPOSITION has been widely used in signal decomposition, de-noising, coding and compression etc., also for extracting cancer classification feature from gene expression profile data.Therefore, wavelet packet decomposition algorithm is adopted to carry out multiscale analysis to ground reaction force in the present invention, with the average of the WAVELET PACKET DECOMPOSITION coefficient of every sub-wavelet packet, wavelet-packet energy, WAVELET PACKET DECOMPOSITION coefficient and variance not only entirety but also characterize gait feature subtly from time-frequency domain, and they are used for identification.
The step of Method of Gait Feature Extraction comprises:
(1) first adopt piecewise linear interpolation algorithm to carry out dimension normalization to selecting pretreated ground reaction force data through denoising and effective sample, the data dimension of effective sample is normalized to same value.When forming a gait passage by Np three-dimensional force force plate/platform, the same value that the gait data dimension of the ground reaction force of single step in the present embodiment normalizes to calculates by 500, and the same value that the gait data dimension of the ground reaction force of synthesis normalizes to calculates by 500 × Np.For the gait passage be made up of 5 three-dimensional force force plate/platforms, Figure 12 is the force-time curve of left and right directions shearing force Fhxd, fore-and-aft direction shearing force Fhyd in the ground reaction force of synthesis after denoising and dimension normalization and vertical direction anchorage force Fhzd, be adopt the ground reaction force data of piecewise linear interpolation algorithm to the synthesis of Fig. 5 to carry out interpolation, data dimension normalized to the force-time curve after 2500 dimensions.
(2) adopt wavelet packet decomposition algorithm to carry out L layer WAVELET PACKET DECOMPOSITION to the ground reaction force after dimension normalization again, decompose and obtain 2
l+1-2 sub-wavelet packets, characterize gait feature with the average of the WAVELET PACKET DECOMPOSITION coefficient of every sub-wavelet packet, wavelet-packet energy, WAVELET PACKET DECOMPOSITION coefficient and variance.When carrying out WAVELET PACKET DECOMPOSITION, determine the wavelet basis function Sum decomposition number of plies in advance.The selection of wavelet basis function usually needs by virtue of experience to determine with experiment, and the conventional wavelet basis function that not only may be used for continuous wavelet transform at present but also can be used for wavelet transform has Haar, Daubechies, Biorthgonal, Coiflets and Symlets five kinds.Adopt the inventive method, found through experiments, when selecting Coif1 small echo to carry out WAVELET PACKET DECOMPOSITION, discrimination is higher.Decomposition order is determined by signal low-limit frequency and highest frequency usually, the highest frequency of the ground reaction force signal collected within 400Hz, the eigenfrequency of the ground reaction force of gait at about 30Hz, Decomposition order
4 layers of WAVELET PACKET DECOMPOSITION can be selected.In the present embodiment, 4 layers of WAVELET PACKET DECOMPOSITION are carried out successively for the ground reaction force of each sample, obtain 2
l+1-2 sub-wavelet packets, preserve the WAVELET PACKET DECOMPOSITION coefficient of all sub-wavelet packets, using the average of the WAVELET PACKET DECOMPOSITION coefficient of each sub-wavelet packet, wavelet-packet energy, WAVELET PACKET DECOMPOSITION coefficient and variance as initial gait feature collection.Figure 13 is through the vertical direction anchorage force Fwz curve after denoising, weight normalized and dimension normalized, Figure 14 carries out the second layer metric space after 4 layers of WAVELET PACKET DECOMPOSITION to vertical direction anchorage force Fwz, and Figure 15 carries out the second layer wavelet space after 4 layers of WAVELET PACKET DECOMPOSITION to vertical direction anchorage force Fwz.Wherein, Figure 14 and Figure 15 illustrates employing 4 layers of WAVELET PACKET DECOMPOSITION method extract gait feature example from the vertical direction anchorage force the ground reaction force of synthesis.First with wavelet transformation hard threshold method, denoising is carried out to original gait data, again data are carried out standardization divided by body weight and data dimension is normalized to 2500 dimensions, the force-time curve of the vertical direction anchorage force Fwz after these pretreatment operation as shown in figure 13; And then 4 layers of WAVELET PACKET DECOMPOSITION are carried out to it, vertical direction anchorage force Fwz is resolved into metric space and wavelet space; Metric space is the compression of original signal, comprises the low-frequency information of original signal, and Figure 14 carries out the second layer metric space after 4 layers of WAVELET PACKET DECOMPOSITION to vertical direction anchorage force Fwz; Wavelet space is the complementary space of metric space, comprises the high-frequency information of original signal, and Figure 15 carries out the second layer wavelet space after 4 layers of WAVELET PACKET DECOMPOSITION to vertical direction anchorage force Fwz; Each space is a sub-wavelet packet, the coefficient of every sub-wavelet packet is WAVELET PACKET DECOMPOSITION coefficient, again integration is carried out on frequency band to the WAVELET PACKET DECOMPOSITION coefficient of every sub-wavelet packet and obtain wavelet-packet energy, ask average and the variance of the WAVELET PACKET DECOMPOSITION coefficient of every sub-wavelet packet, finally using the average of the WAVELET PACKET DECOMPOSITION coefficient of all sub-wavelet packets and wavelet-packet energy, WAVELET PACKET DECOMPOSITION coefficient and variance as initial gait feature collection.
(3) then in the training process, FCM Algorithms is adopted to carry out Feature Dimension Reduction to the WAVELET PACKET DECOMPOSITION coefficient that initial gait feature is concentrated, pick out the few and wavelet packets coefficient of dissociation that classification capacity is strong of redundancy, preserve by wavelet packets coefficient of dissociation and corresponding to the sequence number (call number) of initial gait feature collection, and every the wavelet-packet energy of sub-wavelet packet, the average of WAVELET PACKET DECOMPOSITION coefficient and a variance combine the gait feature template obtaining gait feature collection and trained;
(4) carrying out Feature Mapping according to the initial gait feature collection of call number to test sample book in the gait feature template of having trained in identifying, picking out the gait feature for identifying.
4. the training of sorter
In pattern recognition system, feature extraction and selection are bases, and the selection of sorter and training are also and very crucial.Traditional statistical pattern recognition method sets up disaggregated model under abundant sample, and when only having sample number to be tending towards infinity, its classification performance just can have theoretic guarantee, and in practical application, sample number is but limited retrievable.Being based upon the learning algorithm on structuring principle of minimization risk basis---support vector machine (SVM:SupportVectorMachines) has outstanding performance in solution small sample, non-linear and high dimensional pattern classification problem, therefore, the present invention adopts support vector machine as sorter.
Support vector machine proposes for two points of problems, for multi-class classification problem, needs to carry out PROBLEM DECOMPOSITION and re-uses support vector machine and classify.Suppose there is M class training sample in the present invention, the present invention adopts the many classification problems of one-to-many combined strategy process, and by the 1 pair of M-1 combination of M class training sample, training obtains N number of support vector machine respectively, determines classification results according to nearest neighbouring rule.Suppose in gait data storehouse, to have have registered n class (individual) gait sample, these samples input support vector machine classifier is trained, obtains the output and the classification model that correspond to 1 ~ n.In identifying, the real-time gait sample got is input in trained support vector machine classifier, namely mates with the classification model of training, if output valve is between 1 ~ n, then assert that unknown object is registered competent person, otherwise be judged to be unregistered personnel.
The present invention, before employing support vector machine classifier is trained training sample, first adopts the normalization that the gait feature of maximum-minimum requirement method to input support vector machine carries out in amplitude.The unique distinction of support vector machine is just to adopt the kernel function of Nonlinear Mapping the lower dimensional space of linearly inseparable to be mapped to the higher dimensional space of linear separability, and should select suitable kernel function according to problem, kernel function comparatively conventional at present has three kinds:
Polynomial kernel function K (x, x')=((xx')+1)
d, d>1, need determine (5) in advance
Radial basis kernel function K (x, x')=exp (-γ || x-x'||
2), γ >0, need determine (6) in advance
Sigmoid core K (x, x')=tanh (β
0(xx')+β
1) (7)
When the d in Polynomial kernel function is 1, Polynomial kernel function becomes linear kernel function.With Sigmoid nuclear phase ratio, Polynomial kernel function and Radial basis kernel function need the parameter of input few.For Nonlinear Classification, Radial basis kernel function is best selection, choice for use Radial basis kernel function of the present invention.In addition, when selecting Radial basis kernel function, a pair parameter of kernel function---the selection of penalty factor and kernel functional parameter γ also has certain influence to classification performance.Penalty factor is used for Controlling model divides sample punishment degree to mistake, decision model complexity and generalization ability.Kernel functional parameter γ determines the complexity (dimension) mapping rear feature space, thus affects the minimum experience error of classifying face.Therefore, need the parameter of Radial basis kernel function to adjust to suitable value to (C, γ) during use, the present invention adopts grid-search algorithms to determine the kernel functional parameter of support vector machine to (C, γ).
5. the identification of unknown object and recognition result
After completing the training of support vector machine classifier, in identifying, complete the identification to unknown object.In identifying, the real-time gait sample of the unknown object got is input in trained support vector machine classifier, namely mate with the classification model of training, if output valve is between the training result 1 ~ n of registered sample, then assert that unknown object is registered competent person, otherwise be judged to be unregistered personnel.
In order to test the validity of institute of the present invention extracting method, acquire 103 people's self-selected speeds in the present embodiment experiment and wear ground reaction force data when footwear are walked naturally, everyone gathers 12 effective samples, amounts to 1236 effective samples, establishes the gait data storehouse of 103 people.For evaluating recognition performance, use correct recognition rata CRR(CorrectRecognitionRate) as evaluation index.Gait feature is extracted from the ground reaction force of synthesis, adopt the support vector machine of different IPs function as sorter respectively, carry out 4 repeated tests, find that employing is the highest as correct recognition rata CRR during sorter based on the support vector machine of Radial basis kernel function, reach the average correct recognition rata of 95.3%.Therefore, the present invention is feasible and effective.What is more important, the present invention is disguised and anti-pretends to be ability good, recognition result the disturbing factor such as not to block by complex background, clothes and health to be affected, can well Integrated predict model in the identity verify and security access control system in cyberage security sensitive place.
Embodiment 2:
In order to further describe the specific embodiment of the present invention, the present embodiment is applied as example with the entrance guard management of certain bank vault and is illustrated.
Bank vault is the focused protection place that a designated person just has turnover authority; also be the place that offender thinks into theft very much; even describe in film, they by snatch password and key, disguise oneself or the means such as fingerprint, iris of copying enter national treasury steal.Therefore, by the tight concordant erection method shown in Fig. 3, a gait passage of 3 meters long be made up of 6 triaxial residual stresses can be installed by subsurface before national treasury entrance antitheft door, utilize the present invention to perform identification based on ground reaction force during walking and entrance guard management work.If visitor walks the out-of-date personnel allowing to enter that can be identified as specifying from gait passage, then the antitheft door of encrypted pinning is opened automatically, allows visitor to enter; Otherwise, will be denied access to, and superior management and control unit gives the alarm, automatically snaps simultaneously and transmit image scene to higher level's management and control unit.
In the entrance guard management application of bank vault, first the sample training with designated person is needed, the gait feature template of having been trained and classification model, identifying identifies visitor by template again, whether be the people that allow enter, recognition result is as controlling the anti-theft door for national treasury order of whether opening or signal if identifying visitor.
For training process, first the gait data storehouse of on-demand creation designated person, utilize the ground reaction force of some when underground gait channel acquisition designated person repeatedly walks on passage before being arranged on bank vault entrance antitheft door, the ground reaction force data of everyone continuous multiple single step walking on gait passage of each collection, the ground reaction force of synthesis is as shown in Figure 5 obtained by superposition calculation, and as gait data sample; For each designated person, above-mentioned effective sample selection method is utilized to pick out effective gait data sample from the gait data sample collected, by gait data Sample Establishing gait data storehouse of all designated persons, the later stage regularly or irregularly upgrades or increases gait data storehouse.For the gait data be collected, the ground reaction force data of the effective synthesis namely picked out, first Wavelet Transform Threshold method is adopted to carry out noise suppression preprocessing to it, then above-mentioned gait feature abstracting method is adopted to extract gait feature, FCM Algorithms is adopted to select gait feature again, the gait feature template of having been trained; Support vector machine classifier is adopted to train training sample again, the classification model of having been trained, before employing support vector machine classifier is trained training sample, first can adopt the normalization that the gait feature of maximum-minimum requirement method to input support vector machine carries out in amplitude.
For identifying, first to utilize before being arranged on bank vault entrance antitheft door underground gait channel acquisition visitor from gait passage up walk out-of-date ground reaction force, the ground reaction force of synthesis is as shown in Figure 5 obtained fast by superposition calculation, and as test sample book; Adopt said method to carry out effective sample again to select and the data prediction such as denoising, adopt said method to carry out Method of Gait Feature Extraction, utilize the gait feature template of having trained to carry out Feature Mapping (i.e. feature selecting) to gait feature to be identified; Then maximum-minimum requirement method is carried out in amplitude normalization to gait feature to be identified is adopted, adopt support vector machine to test gait feature to be identified to mate with the classification model of training, if mated completely with certain already present classification model in gait data storehouse, so visitor is the personnel allowing to enter specified, to give an order or signal opens the antitheft door of encrypted pinning automatically, allow visitor to enter; If do not mated with all classification models existed in gait data storehouse, so visitor is considered to the personnel that do not allow to enter, do not send the order or signal of opening antitheft door, refuse it to enter, and superior management and control unit gives the alarm, automatically snap simultaneously and transmit image scene to higher level's management and control unit.
In a word, the present invention is based on ground reaction force during walking, propose a kind of simple and effective remote hidden personal identification method.First, utilizing can freely assembled and gait passage under can being hidden in ground/floor ground reaction force when obtaining walking, adopts Wavelet Transform Threshold method to carry out noise suppression preprocessing to ground reaction force; Adopt the feature extracting method of proposition to extract gait feature again, and adopt feature selecting algorithm to select optimum gait feature; Then, gait data storehouse adopt support vector machine to train classification model; Finally, in identifying, according to nearest neighbouring rule, template matches and identification are carried out to test sample book, export recognition result.Test result on the gait data storehouse of 103 self-built people demonstrates the feasibility of the inventive method and the validity of algorithm, and the present invention is convenient and easy, has robustness.Because the gait information that the present invention utilizes can by being hidden in the force-sensing sensor collection under ground/floor, gatherer process has untouchable and disguised, be identified object and can not realize there is identification apparatus and takes measures to get by under false pretences, also can not cause human rights dispute; Meanwhile, relative to the gait identification method of computer vision field, recognition result not to block etc. by complex background, clothes and health to be affected, and vola force information also hardly may be imitated; The present invention combines these advantages, can meet identity verify and the safety precaution demand in the security sensitive places such as cyberage airport, customs, museum, party scene, bank and national treasury better.Meanwhile, the present invention can also provide supplementary and analytical approach for the apparatus Computer Aided Designs such as sports and scientific research, medical rehabilitation, productive life body-building and criminal investigation.
Claims (3)
1., based on a personal identification method for ground reaction force during walking, it is characterized in that comprising training process and identifying;
Described training process comprises: utilize by gait information when being hidden in gait channel acquisition walking that ground or underfloor three-dimensional force force plate/platform form and set up gait data storehouse, carry out data prediction, Method of Gait Feature Extraction, gait feature selects and sorter is trained, the gait feature template of finally having been trained and classification model;
Described identifying comprises: obtain the real-time gait information of object to be identified to obtain test sample book, data prediction, Method of Gait Feature Extraction are carried out to test sample book, the gait feature template of having trained is utilized to carry out Feature Mapping to the gait feature of object extraction to be identified, recycle the classification model of having trained, according to nearest neighbouring rule, template matches and identification are carried out to test sample book, export recognition result;
Described gait information is the ground reaction force of the single step gathered by three-dimensional force force plate/platform, or the ground reaction force of continuous multiple single step cumulative ground reaction force obtaining synthesizing in sequential;
The ground reaction force of described single step or the ground reaction force of synthesis are all made up of these three components of left and right directions shearing force, fore-and-aft direction shearing force and vertical direction anchorage force;
Described data prediction comprises denoising and effective sample is selected; Described denoising refers to and adopts Wavelet Transform Threshold method to carry out level discharge rating process to ground reaction force data; Described effective sample selects the data dimension that refers to vertical direction anchorage force in the ground reaction force after according to denoising and whether peak point numerical value judges in effective range, selects ground reaction force data in effective range as effective sample; Described effective range refers to the numerical range relevant with sample frequency Fp to the number Np of adopted three-dimensional force force plate/platform, for the ground reaction force of single step, if wherein the data dimension of vertical direction anchorage force is lower than 0.4Fp or higher than 0.8Fp, or the crest of vertical direction anchorage force and the ordinate value of trough differ by more than 300, namely be judged to be invalid sample, otherwise be judged to be effective sample; For the ground reaction force that the continuous multiple single steps gathered by Np three-dimensional force force plate/platform are synthesized, if wherein the data dimension of vertical direction anchorage force is lower than 0.4Fp × Np or higher than 0.8Fp × Np, or the number of the trough of vertical direction anchorage force is less than Np, or the maximin of the ordinate value of all troughs of vertical direction anchorage force differs by more than 300, namely be judged to be invalid sample, otherwise be judged to be effective sample;
Described Method of Gait Feature Extraction refers to and adopts Wavelet Packet Transform Method from ground reaction force, to extract WAVELET PACKET DECOMPOSITION coefficient, wavelet-packet energy, the average of WAVELET PACKET DECOMPOSITION coefficient and variance to characterize gait feature, that is: first adopt piecewise linear interpolation algorithm to carry out dimension normalization to the ground reaction force data through data prediction, the data dimension of effective sample is normalized to same value; Adopt L layer wavelet packet decomposition algorithm to carry out WAVELET PACKET DECOMPOSITION to the effective sample after denoising and dimension normalization again, decompose and obtain 2
l+1-2 sub-wavelet packets, characterize gait feature with the average of the WAVELET PACKET DECOMPOSITION coefficient of every sub-wavelet packet, wavelet-packet energy, WAVELET PACKET DECOMPOSITION coefficient and variance; Then feature selecting algorithm is adopted to select WAVELET PACKET DECOMPOSITION coefficient, pick out wavelet packets coefficient of dissociation, the average of wavelet packets coefficient of dissociation and wavelet-packet energy, WAVELET PACKET DECOMPOSITION coefficient and variance are combined, the gait feature template obtaining gait feature collection and trained; Described wavelet-packet energy carries out integral and calculating by WAVELET PACKET DECOMPOSITION coefficient and obtains on the frequency band of correspondence;
Described sorter training refers to and adopts support vector machine classifier to train training sample, the classification model of having been trained.
2., by the personal identification method based on ground reaction force during walking described in claim 1, be characterised in that the step of described Method of Gait Feature Extraction is:
First adopt piecewise linear interpolation algorithm to carry out dimension normalization to the ground reaction force data through data prediction, the data dimension of effective sample is normalized to same value;
Adopt L layer wavelet packet decomposition algorithm to carry out WAVELET PACKET DECOMPOSITION to the effective sample after denoising and dimension normalization again, decompose and obtain 2
l+1-2 sub-wavelet packets, characterize gait feature with the average of the WAVELET PACKET DECOMPOSITION coefficient of every sub-wavelet packet, wavelet-packet energy, WAVELET PACKET DECOMPOSITION coefficient and variance; When carrying out WAVELET PACKET DECOMPOSITION, determine the wavelet basis function Sum decomposition number of plies in advance, by the logarithm determination Decomposition order L of signal highest frequency and low-limit frequency ratio; After determining wavelet basis function Sum decomposition number of plies L, the ground reaction force for each sample carries out L layer WAVELET PACKET DECOMPOSITION successively, obtains 2
l+1-2 sub-wavelet packets, preserve the WAVELET PACKET DECOMPOSITION coefficient of all sub-wavelet packets, the WAVELET PACKET DECOMPOSITION coefficient of every sub-wavelet packet is carried out the wavelet-packet energy that integral and calculating obtains this sub-wavelet packet on the frequency band of its correspondence, calculate average and the variance of the WAVELET PACKET DECOMPOSITION coefficient of every sub-wavelet packet, using the average of the WAVELET PACKET DECOMPOSITION coefficient of each sub-wavelet packet, wavelet-packet energy, WAVELET PACKET DECOMPOSITION coefficient and the variance initial gait feature subset as sub-wavelet packet; Finally using the average of the WAVELET PACKET DECOMPOSITION coefficient of all sub-wavelet packets and wavelet-packet energy, WAVELET PACKET DECOMPOSITION coefficient and variance as initial gait feature collection;
Then in the training process, FCM Algorithms is adopted to carry out Feature Dimension Reduction to the WAVELET PACKET DECOMPOSITION coefficient that initial gait feature is concentrated, pick out wavelet packets coefficient of dissociation, preserve by wavelet packets coefficient of dissociation and corresponding to the call number of initial gait feature collection, and every the wavelet-packet energy of sub-wavelet packet, the average of WAVELET PACKET DECOMPOSITION coefficient and a variance combine the gait feature template obtaining gait feature collection and trained;
Carrying out Feature Mapping according to the initial gait feature collection of call number to test sample book in the gait feature template of having trained in identifying, picking out the gait feature for identifying.
3. by the personal identification method based on ground reaction force during walking described in claim 1, be characterised in that the training of described sorter first adopts the gait feature of maximum-minimum requirement method to input support vector machine to be normalized in amplitude, support vector machine classifier is adopted to train training sample again, the classification model of having been trained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210559665.7A CN103093234B (en) | 2012-12-21 | 2012-12-21 | Based on the personal identification method of ground reaction force during walking |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210559665.7A CN103093234B (en) | 2012-12-21 | 2012-12-21 | Based on the personal identification method of ground reaction force during walking |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103093234A CN103093234A (en) | 2013-05-08 |
| CN103093234B true CN103093234B (en) | 2016-01-06 |
Family
ID=48205782
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210559665.7A Active CN103093234B (en) | 2012-12-21 | 2012-12-21 | Based on the personal identification method of ground reaction force during walking |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103093234B (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104434128B (en) * | 2014-12-25 | 2016-08-24 | 中国科学院合肥物质科学研究院 | A kind of left and right based on plantar nervous arch information foot dynamic identifying method |
| KR102459677B1 (en) * | 2015-11-05 | 2022-10-28 | 삼성전자주식회사 | Method and apparatus for learning algorithm |
| TWI615129B (en) * | 2016-02-19 | 2018-02-21 | 財團法人資訊工業策進會 | Gait analysis system and method thereof |
| CN107066945B (en) * | 2017-03-10 | 2019-06-18 | 清华大学 | A kind of quick identity checking method of big flow clearance and system |
| CN106953810A (en) * | 2017-03-13 | 2017-07-14 | 成都育芽科技有限公司 | The call method and device of template in a kind of WTC communication protocols |
| CN107028427B (en) * | 2017-06-08 | 2020-02-21 | 翼猫科技发展（上海）有限公司 | Novel sensing method of sleep system capable of automatically and intelligently adjusting |
| CN108053587A (en) * | 2018-01-08 | 2018-05-18 | 京东方科技集团股份有限公司 | Personal identification method and its device, carpet |
| CN109274930B (en) * | 2018-07-26 | 2020-06-26 | 华为技术有限公司 | Safety monitoring method and device based on gait recognition |
| CN109359580B (en) * | 2018-10-12 | 2021-05-11 | 中国科学院福建物质结构研究所 | Footprint identification and gait detection method and device based on deep learning |
| WO2020077482A1 (en) * | 2018-10-15 | 2020-04-23 | 王长贵 | Composite ceramic tile and gait detection system having same |
| CN109447199A (en) * | 2018-10-16 | 2019-03-08 | 山东大学 | A kind of multi-modal criminal's recognition methods and system based on step information |
| CN109740423B (en) * | 2018-11-22 | 2023-04-07 | 霍尔果斯奇妙软件科技有限公司 | Ethnicity identification method and system based on face and wavelet packet analysis |
| CN112198222B (en) * | 2020-09-17 | 2022-04-05 | 美智纵横科技有限责任公司 | Ground material identification method, system, equipment and storage medium |
| CN115553757B (en) * | 2022-12-06 | 2023-04-07 | 深圳市第二人民医院(深圳市转化医学研究院) | Method and equipment for intelligently screening effective ground reaction force in gait process |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8041513B2 (en) * | 2005-12-19 | 2011-10-18 | Analiza, Inc. | Systems and methods involving data patterns such as spectral biomarkers |
-
2012
- 2012-12-21 CN CN201210559665.7A patent/CN103093234B/en active Active
Non-Patent Citations (2)
| Title |
|---|
| 一种改进的机遇地面反作用力的步态识别方法;林尔东，姚志明，郑重，周旭，孙向阳，孙怡宁;《模式识别与人工智能》;20110630;第24卷(第3期);全文 * |
| 基于步态触觉信息的身份识别研究;姚志明;《中国博士学位论文全文数据库》;20110630;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103093234A (en) | 2013-05-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103093234B (en) | Based on the personal identification method of ground reaction force during walking | |
| CN103049741A (en) | Foot-to-ground acting force-based gait feature extraction method and gait identification system | |
| Wan et al. | A survey on gait recognition | |
| Pandya et al. | Ambient acoustic event assistive framework for identification, detection, and recognition of unknown acoustic events of a residence | |
| Rong et al. | A wearable acceleration sensor system for gait recognition | |
| CN104102915B (en) | Personal identification method based on ECG multi-template matching under a kind of anomalous ecg state | |
| Mubashir et al. | A survey on fall detection: Principles and approaches | |
| CN104573458A (en) | Identity recognition method, device and system based on electrocardiogram signals | |
| US10614299B2 (en) | Indoor identification of individuals through footstep induced structural vibration | |
| Abhyankar et al. | Integrating a wavelet based perspiration liveness check with fingerprint recognition | |
| CA2494491C (en) | Method and apparatus for electro-biometric identity recognition | |
| CN111178155B (en) | Gait feature extraction and gait recognition method based on inertial sensor | |
| He et al. | Intrusion detection based on stacked autoencoder for connected healthcare systems | |
| Moctezuma et al. | Subjects identification using EEG-recorded imagined speech | |
| CN107422643A (en) | Smart home monitoring method and system based on vibration detection | |
| Ghofrani et al. | Reliable features for an ECG-based biometric system | |
| CN113609977B (en) | Pedestrian gait recognition method based on channel state information quotient distance | |
| CN105139029A (en) | Activity recognition method and activity recognition device for persons serving sentences | |
| Hestbek et al. | Biometric gait recognition for mobile devices using wavelet transform and support vector machines | |
| He | Activity recognition from accelerometer signals based on wavelet-ar model | |
| Choi et al. | User Authentication System Based on Baseline-corrected ECG for Biometrics. | |
| CN105184254A (en) | Identity authentication method and system | |
| El_Rahman | Biometric human recognition system based on ECG | |
| CN109447199A (en) | A kind of multi-modal criminal's recognition methods and system based on step information | |
| Ning et al. | Fall detection algorithm based on gradient boosting decision tree |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |