CN103049741A - Foot-to-ground acting force-based gait feature extraction method and gait identification system - Google Patents

Abstract

The invention discloses a foot-to-ground acting force-based gait feature extraction method and a foot-to-ground acting force-based gait identification system. The system comprises a gait data acquisition module and a gait data processing and identification module, wherein the gait data acquisition module acquires the foot-to-ground acting force during walking in a hidden and non-intrusive way; and the gait data processing and identification module extracts gait features and performs gait identification from the foot-to-ground acting force. The gait feature extraction method comprises the following steps of: denoising the foot-to-ground acting force by a wavelet transform hard threshold method; detecting a key point and calculating gait time-domain features by a first-order difference algorithm; and representing the gait frequency-domain features by adopting a wavelet decomposition coefficient after performing waveform alignment on the foot-to-ground acting force. When the system is used for identifying, an optimal gait feature set is selected by a fuzzy C-means algorithm; and the classification or the identification can be performed on the gait based on a support vector machine. The gait information acquired by the method and the system is natural, true and not easy to simulate; the extracted gait features are full and accurate; and the requirements on gait analysis, classification and long-distance gait identification can be met better.

Description

Gait feature abstracting method and Gait Recognition system based on foot ground acting force

Technical field

The invention belongs to mode identification technology, particularly based on gait feature abstracting method and the Gait Recognition system of foot ground acting force.

Background technology

The posture that Gait Recognition utilizes the people to walk realizes the identification of personal identification or classification, the identification of physiology, pathology and psychological characteristics gait, it is a new and developing branch in the biometrics identification technology field, has the total characteristic of other biological feature identification technique, be that biological characteristic can be carried, do not need memory, have uniqueness and be difficult to the characteristics such as replicability.Simultaneously, Gait Recognition has the unique advantage that the other biological feature identification technique is not had, and namely gait information is can be at a distance hidden and collect non-contactly, and gait is difficult to hide and camouflage, is a kind of remote biometrics identification technology.Mat is in this, and the Gait Recognition application prospect is extensive, especially has more potentiality and advantage in the identification in security sensitive place or authentication application.Because the biometrics identification technologies such as fingerprint, people's face, iris, palmmprint, palm shape, people's ear, retina, DNA, smell, signature, voice and keystroke when practical application, all need checked object closely close fit just can finish information acquisition, do not possess disguise, easily cause discovering and then taking to forge and the counterreconnaissance means such as imitation of offender, not too be applicable to the safety certification demand in security sensitive place; And the physiological characteristics such as used fingerprint, people's face, iris, palmmprint, palm shape, people's ear, retina and smell all are static natures, are easy to forge; This three behaviors feature of keystroke, voice and signature is also easily imitated; Although DNA authentication technique reliability is high, what sampling neither hidden, and analytic process is longer, requires high-end instrument and equipment, only be confined at present that the laboratory is used and the law evaluation in.

Because have unique advantage, Gait Recognition is in recent ten years by extensive concern and research, and gait feature abstracting method and gait recognition method, system are the focuses of paying close attention to.Comparatively speaking, the Research on Gait Recognition of computer vision field the earliest, and is the most active, and achievement is also maximum, but because the restriction of all objective factors has brought many challenges for the practical application of Gait Recognition.Which people is the gait that can't accurately identify deliberately imitation such as, the Gait Recognition of computer vision field really be, and in fact people also can be on the vision pattern others' gait be imitated the spitting image of, produce gait image information about the same; May be subject to the interference of complex background and factors such as self blocking in gait image or the video acquisition process, the gait feature that is difficult to detect exactly human body target He extracts robust.Therefore, need research and development based on personal identification method and the system of other gait information, to adapt to the safety certification demand of cyberage.In addition, in using based on the identification of gait and based on the classification of physiology, pathology and the psychological characteristics gait of gait or identification, gait feature is crucial, current gait feature abstracting method concentrates on the kinematics character of the sign gaits such as Fourier transform, wavelet transformation and the Radon conversion with image sequence mostly, and in fact gait also has dynamic characteristic, but owing to prior art fails to extract at many levels gait feature from multi-angle, thereby can not extract all-sidedly and accurately the gait feature that reflects gait kinematics and dynamic characteristic.

China Patent No. ZL01144157.7 discloses a kind of method of identifying personnel by the gait image sequence of human locomotion, but the method is owing to using common camera to take gait image, can't ideally solve complex background, block the problem with noise, can't be applied to the outdoor of unglazed photograph at night, also not have to consider deliberately to imitate the problem of gait.Chinese patent application publication number CN101251894A disclosed a kind of can around-the-clock gait recognition method and system because still be based on the recognition methods of gait image, still can't process deliberately imitation and occlusion issue.The disclosed a kind of multiaxis power plane matrix that is consisted of by four six axle power platforms of China Patent No. 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 test and the 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.US Patent No. 2002/0107649 discloses a kind of gait detection device and gait detection method of the sound oscillation signal when detecting walking, and as personal identification system, this personal identification system requires to place microphone (acoustic-electric converter) at human body, microphone obtains the electric signal subset by the sound oscillation energy that gathers in the walking movement process, detect walking period with this electric signal subset as an index that goes on foot, the signal subset of the sound oscillation when collection expression pin lands, system extracts the specific gait feature of human body based on this signal subset, and uses it for individual's identification.But this personal identification system has two deficiencies: the one, the sound oscillation when landing along with microphone on human body the placement position and change; The 2nd, owing to being subjected to the major effect of microphone ambient noise and electrical noise, can not determine exactly the sound oscillation electric signal when the expression pin lands, also just be difficult to extract exactly the feature of gait waveform.Patented claim international publication number WO/2004/040501 disclose a kind of with the waveform peak amplitude in the allocated frequency band as the reference index, and determine the gait waveform in a step in the electric signal that from the electric field displacement detector, detects and extract method and the personal identification system of gait waveform feature, the method not interfered by the electric charge between the left and right leg and the installation position of sensor on health affects, but because electromagnetic interference (EMI) can have a strong impact on the accuracy that detects electric signal, be not suitable in the environment that has complicated electromagnetic interference (EMI).And, in US Patent No. 2002/0107649A1 and the disclosed technical scheme of international publication number WO/2004/040501, its gait information remains and relies on the sensor collection that is placed on the health, remain the contact type measurement mode, the measured can perceive and obtains, and the offender is easy to take the counterreconnaissance means to get by under false pretences or be stung to fury so that the event that wakes a snake in the crowd because perceiveing.

Summary of the invention

The objective of the invention is to propose a kind of gait feature abstracting method of kinematics and dynamic characteristic can characterize more all-sidedly and accurately the walking of people's nature the time, and take the method as the basis, propose a kind of can eliminate complex background and the extraneous factor such as block disturb, can avoid to greatest extent again simultaneously the Gait Recognition system based on foot ground acting force of the situation of getting by under false pretences, with hidden and without personnel being carried out identification with invading and harassing, avoid causing human rights dispute, can realize round the clock remote monitoring, avoid to greatest extent the situation of getting by under false pretences, be specially adapted to customs, the airport, the museum, party is on-the-spot, the identification in the security sensitive such as bank and national treasury place also is applicable to automatic classification and the discriminating of all kinds of gaits.

Gait feature abstracting method based on foot ground acting force of the present invention is characterized in that: comprise that the gait frequency domain character extracts and the gait temporal signatures extracts;

Described gait frequency domain character extracts and refers to: the foot ground acting force during with the walking of obtaining carries out denoising and waveform alignment as gait data to gait data, adopts the WAVELET PACKET DECOMPOSITION method to extract the gait frequency domain character from the gait data of processing;

Described gait temporal signatures extracts and refers to: the foot ground acting force during with the walking of obtaining is as gait data, gait data is carried out denoising, detect key point and reference point on the foot ground acting force curve, with acting force rate of change and the momentum sign gait temporal signatures of the timely phase of power value at key point place, adjacent key point;

Described foot ground acting force comprises left and right sides shearing force, front and back shearing force and vertical support power;

The step that described gait frequency domain character extracts is as follows:

Frequency domain first step S11: the foot ground acting force when adopting the three-dimensional force force plate to obtain people's normal gait by the gait passage, with once without stop without return normal consistently from the gait passage walking by the time foot ground acting force as an effective gait data, transmit and be saved in the computer PC;

Frequency domain second step S12: adopt the wavelet transformation hard threshold method that foot ground acting force is carried out denoising, obtain the foot ground acting force after the denoising;

Frequency domain the 3rd step S13: select the vertical support power in the foot ground acting force after the denoising, adopt the first order difference algorithm to detect trough point on the vertical support force curve, with the reference point of the trough point on the vertical support force curve as sufficient ground acting force curve;

Frequency domain the 4th step S14: take the reference point of foot ground acting force curve as benchmark, the foot ground acting force after adopting linear interpolation algorithm to denoising carries out the waveform alignment, obtains the foot ground acting force after the waveform alignment;

The 5th step of frequency domain S15: the complexity according to the application places gait data acquisition judges whether the needs exptended sample, if in the place that personnel fix, can multi collect everyone respectively with the different speeds of travel with wear the normally gait data by the gait passage of different shoes, then do not need exptended sample, carry out the 8th step of frequency domain S18; If on the airport, the strong places of mobility of people such as market and customs, can not multi collect everyone respectively with the different speeds of travel with wear different shoes normally by the gait data of gait passage, then need exptended sample, carry out the 6th step of frequency domain S16;

Frequency domain the 6th step S16: when needing exptended sample, take the reference point of foot ground acting force curve as benchmark, the foot ground acting force after the waveform alignment is carried out sample split, split into a plurality of segmentations foots ground acting force, the quantity that expands the gait data sample;

The 7th step of frequency domain S17: adopt L layer wavelet packet decomposition algorithm from the segmentation foot ground acting force that splits out, to extract the gait frequency domain character of segmentation;

The 8th step of frequency domain S18: when not needing exptended sample, adopt the foot ground acting force of L layer wavelet packet decomposition algorithm after the waveform alignment and extract omnidistance gait frequency domain character;

The step that described gait temporal signatures extracts is as follows:

Time domain first step S21: the foot ground acting force when obtaining people's normal gait by the gait passage by the three-dimensional force force plate, with once without stop without return normal consistently from the gait passage walking by the time foot ground acting force as an effective gait data, transmit and be saved in the computer PC;

Time domain second step S22: adopt the wavelet transformation hard threshold method that foot ground acting force is carried out denoising, obtain the foot ground acting force after the denoising;

The 3rd step of time domain S23: adopt front and back shearing force curve and the wave crest point on the vertical support force curve and trough point in the foot ground acting force after the first order difference algorithm detects respectively denoising, with wave crest point and the trough point key point as foot ground acting force curve, with the trough point on the vertical support force curve as sufficient reference point of acting force curve;

The 4th step of time domain S24: the complexity according to the application places gait data acquisition judges whether the needs exptended sample, if in the place that personnel fix, can multi collect everyone respectively with the different speeds of travel with wear the normally gait data by the gait passage of different shoes, then do not need exptended sample, carry out the 7th step of time domain S27; If on the airport, the strong places of mobility of people such as market and customs, can not multi collect everyone respectively with the different speeds of travel with wear different shoes normally by the gait data of gait passage, then need exptended sample, carry out the 5th step of time domain S25;

Time domain the 5th step S25: when needing exptended sample, take the reference point of foot ground acting force curve as benchmark, the foot ground acting force after the denoising is carried out sample split, split into a plurality of segmentations foots ground acting force, the quantity that expands the gait data sample;

Time domain the 6th step S26: with the power value at the key point place on the vertical support force curve in each segmentation foot ground acting force and power value occur the time on the acting force rate of change of phase, adjacent key point and momentum and the corresponding front and back shearing force curve the key point place the power value, drive the gait temporal signatures that momentum and braking momentum characterize segmentation;

Time domain the 7th step S27: when not needing exptended sample, with the power value at the key point place on the vertical support force curve in the foot ground acting force after the denoising and power value occur the time on the acting force rate of change of phase, adjacent key point and momentum and the corresponding front and back shearing force curve the key point place the power value, drive momentum and the braking momentum characterizes omnidistance gait temporal signatures;

The step of described waveform alignment is as follows:

The waveform alignment first step: the dimension of the foot ground force data after adopting linear interpolation algorithm with denoising normalizes to same value G * Np, Np is the sum of the three-dimensional force force plate of use, G is the step pitch parameter relevant with sample frequency, G is set as 50 integral multiple;

Waveform alignment second step: go out n trough point on the vertical support force curve in the foot ground acting force after the dimension normalization by the first order difference algorithm search, n trough point is as a reference point;

Waveform alignd for the 3rd step: the reference point on the vertical support force curve is as reference, by linear interpolation left and right sides shearing force curve, front and back shearing force curve and vertical support force curve in the foot ground acting force are alignd, so that the trough point of the n on the vertical support force curve snaps to respectively the position of appointment, n trough point snaps to G * (n-0.5);

The step that described sample splits is as follows:

Sample splits the first step: go out n trough point on the vertical support force curve in the foot ground acting force by the first order difference algorithm search, and n trough point is as a reference point;

Sample splits second step: the reference point on the vertical support force curve is as cut-point, respectively the part between first reference point and last reference point in left and right sides shearing force curve, front and back shearing force curve and the vertical support force curve in the foot ground acting force is split into the n-1 section, foot ground force data between the two adjacent reference point is extracted, as a new sample, split out altogether n-1 sample.

Simultaneously, the present invention is take above-mentioned gait feature abstracting method based on foot ground acting force as the basis, also propose a kind of Gait Recognition system based on foot ground acting force, it is characterized in that, comprise the processing of gait data acquisition module MD1 and gait data and identification module MD2:

Described gait data acquisition module MD1, formed by Np three-dimensional force force plate MD11 and 1 data collection and transmission unit MD12, Np three-dimensional force force plate MD11 is electrically connected, they are closely concordant and be installed in snugly and consist of gait passage GW under the Flo of ground, data acquisition is electrically connected with Np three-dimensional force force plate MD11 that is electrically connected by Ethernet with the end of transmission unit MD12, data acquisition is electrically connected with computer PC by usb data line DL with the other end of transmission unit MD12, realizes Real-time Collection and the processing of foot ground force data;

Described gait data is processed and identification module MD2, be mounted in the software module in the computer PC, this module is at first carried out denoising by the wavelet transformation hard threshold method to the foot ground acting force that collects, then characterize gait by gait feature abstracting method extraction gait frequency domain character and gait temporal signatures, merge by gait feature again and set up the gait feature collection, adopt at last support vector machine classifier that gait is identified;

Described gait data is processed and identification module MD2, is the gait data that obtains is processed with feature extraction and realized the module of remote hidden Gait Recognition, comprises with lower unit:

Gait data denoising unit MD21, the noises such as the industrial frequency noise in the foot ground acting force that the removal of employing wavelet transformation hard threshold method collects and electromagnetic interference (EMI) obtain the high gait data of signal to noise ratio (S/N ratio);

Waveform alignment unit MD22, the step that adopts the alignment of described waveform will foot ground acting force automatic aligning according to the reference point on the vertical support force curve in the foot ground acting force with the foot ground acting force of all personnel's sample of collecting;

Sample split cells MD23, the foot ground acting force that the step that adopts described sample to split will collect splits into a plurality of segmentation foots ground acting force according to the reference point on the vertical support force curve in the foot ground acting force;

Gait frequency domain character extraction unit MD24, the step that adopts described gait frequency domain character to extract is alignd from process denoising and waveform, or further characterizes the gait frequency domain character through extraction WAVELET PACKET DECOMPOSITION coefficient in the foot ground acting force of sample deconsolidation process again;

Gait temporal signatures extraction unit MD25, adopt step that described gait temporal signatures extracts from through denoising even the power value of calculating key point through the foot ground acting force curve of sample deconsolidation process with the time mutually, rate of change and the momentum of adjacent key point characterize the gait temporal signatures;

Gait feature integrated unit MD26 adopts FCM Algorithms to pick out minimum optimum gait frequency domain character subset from the gait frequency domain character that extracts, and directly makes up the gait feature collection after obtaining merging with the gait temporal signatures again;

Gait Recognition unit MD27 adopts support vector machine classifier that the gait feature that extracts from foot ground acting force is carried out gait and identification.

Gait is biological behavior characteristic with the obvious advantage in the remote identification demand, is particularly suitable for safety certification and the safety precaution demand of cyberage.Prior art discloses the method and system that walking electric field signal that walking sound oscillation signal that the gait image that obtains by video camera or microphone obtain or electric field displacement detector obtain is differentiated individual identity, gait feature abstracting method and Gait Recognition system based on foot ground acting force that the present invention proposes are with respect to the advantage of prior art:

One, foot and the foot ground acting force that ground interaction produces can get access to easily and accurately by being hidden in underground force plate/platform when used gait information was human body walking among the present invention.With respect to sensor being placed in measured's method of obtaining walking sound oscillation signal and walking electric field signal with it, gait information collecting device among the present invention is hidden in subsurface, be not placed in it the measured, gait information acquisition process is hidden to the measured fully, on naturally walking without any impact of measured.That is to say, gait information acquisition among the present invention is noncontact and hidden measuring method, can obtain the true gait information when being identified the walking of object nature, be identified object and can not realize identification apparatus is arranged and to take measures to get by under false pretences, can also can monitor identifying in far place at the scene, can not cause human rights dispute yet, can satisfy better that cyberage airport, customs, museum, party are on-the-spot, bank and

The identity in the security sensitive places such as national treasury is differentiated and the safety precaution demand.

They are two years old, since among the present invention used gait information be force plate/platform obtain human body walking the time foot and the ground interaction foot ground acting force that produces, with respect to the method for obtaining gait image in the computer vision field by video camera, recognition result is not subjected to that complex background, clothes and health block etc. to be affected.What more have superiority a bit is, situation when the present invention can tackle mimicker's gait deliberately, and computer vision field can't better be tackled based on the personal identification method of gait, a because people mimicker's the gait posture that often others walks from vision pattern learning, can imitate the spitting image of, but because the gait mechanical information of foot and ground interaction can't observe and record others and walk the time, also just can't learn, imitate mechanical information, produce surely similar gait mechanical information even if similar walking posture also differs.

They are three years old, the gait feature that extracts the foot ground acting force of the inventive method during from walking comprises gait temporal signatures and gait frequency domain character, with respect to the gait image sequence, the global features such as the kinematics of the gait temporal signatures reflection walking of from foot ground acting force, extracting and dynamics, the motion characteristic that can reflect all-sidedly and accurately walking, the minutia of the gait frequency domain character reflection walking of extracting, when two kinds of gait features are used for Gait Recognition simultaneously, because utilized the complementation of whole and minutia, its recognition performance is more excellent.

Its four, the inventive method has proposed the waveform alignment schemes foot ground acting force has been alignd at the reference point place in gait frequency domain character leaching process, to improve contrast and the classification capacity of gait frequency domain character; Simultaneously, the sample method for splitting has also been proposed, in the situation that do not increase the effective expansion of data acquisition number of times for the quantity of the sufficient ground force data sample of training, the sample that provides as much as possible is provided for sorter, thus the accuracy rate of the robustness of the sorter of raising training and raising Gait Recognition.

Its five, the gait feature abstracting method that designs among the present invention and Gait Recognition system except can being used for identification, can also promote the use of a lot of fields.For example, in the medical aided diagnosis field, the gait temporal signatures of extraction and gait frequency domain character can be used for analysis, automatic classification and the identification of the clinical gaits such as parkinsonian gait, cerebral apoplexy gait, diabetes gait, outer measured gait with the toes pointing outwards, interior measured gait with the toes pointing outwards; At sports research area, utilize the dynamic time sequence parameter of walking and execution process mesopodium ground acting force, the athletic locomitivity of assessment and analysis and the effect etc. of having an effect are as the foundation of improving training patterns, examination training effect and injury gained in sports assessment; In the medical rehabilitation field, by with the contrast of normal person's gait feature, can the evaluating patient skeletal muscle etc. the degree that suffers damage of tissue and the ability of control balance, the patient is classified and classification, determine the therapy rehabilitation scheme, and assess the rehabilitation degree after can in therapeutic process, verifying curative effect, treatment; In addition, gait temporal signatures and gait frequency domain character can also be applied to the fields such as the design of insurance assessment, Claims Resolution (insurance company is used for wounded or disabled assessment, therapy rehabilitation assessment) and human lives's means of production (furniture, shoes, artificial limb, office accommodations etc.) and body-building, rehabilitation material and measures of effectiveness.

Description of drawings

Fig. 1 is based on the Method of Gait Feature Extraction algorithm flow chart of foot ground acting force, and component wherein (a) extracts process flow diagram for the gait frequency domain character, and component (b) extracts process flow diagram for the gait temporal signatures.

Fig. 2 is the waveform alignment of foot ground acting force curve and the reference point schematic diagram that sample splits.

Fig. 3 is left and right sides shearing force Fx curve, front and back shearing force Fy curve and the vertical support power Fz curve in the front foot ground acting force of denoising.

Fig. 4 is left and right sides shearing force Fxd curve, front and back shearing force Fyd curve and the vertical support power Fzd curve in the foot ground acting force after the denoising.

Fig. 5 is left and right sides shearing force Fxr curve, front and back shearing force Fyr curve and the vertical support power Fzr curve in the foot ground acting force after the alignment of denoising and waveform.

Fig. 6 is through the vertical support power Fwz curve in the foot ground acting force after denoising, waveform alignment and the amplitude normalization.

Fig. 7 carries out 4 layers of ground floor yardstick sequence after the WAVELET PACKET DECOMPOSITION to vertical support power Fwz.

Fig. 8 carries out 4 layers of ground floor wavelet sequence after the WAVELET PACKET DECOMPOSITION to vertical support power Fwz.

Fig. 9 is that key point and the gait temporal signatures of foot ground acting force curve characterizes schematic diagram.

Figure 10 is based on the Gait Recognition system chart of foot ground acting force.

Figure 11 is that gait data is processed and the identification module block diagram.

Embodiment

Embodiment 1:

The present invention is further illustrated below in conjunction with drawings and Examples.

The present invention is take people's gait as starting point, to obtain people's natural reality and comprehensive gait feature and to realize that the Gait Recognition of automatic monitoring is applied as purpose round the clock, a kind of gait feature abstracting method and the Gait Recognition system based on foot ground acting force that propose relate to the extraction of gait temporal signatures and gait frequency domain character, the problems such as Design and Features realization of Gait Recognition system.

Foot ground acting force when being subject to people's walking is difficult to imitated and can be without getting access to and the inspiration of time frequency analysis combination with invading and harassing, in the implementation process of the inventive method, can adopt the foot ground acting force that vola and ground interaction produce when being hidden in underground force plate collection people walking, replace the easily gait image sequence of imitation with the gait mechanical information that can not imitate, and from foot ground acting force, extract gait temporal signatures and gait frequency domain character, two kinds of Fusion Features are used for Gait Recognition, and then have realized the Gait Recognition system based on foot ground acting force.

Figure 10 is based on the Gait Recognition system chart of foot ground acting force, comprise that based on the Gait Recognition system of foot ground acting force gait data acquisition module MD1 and gait data among Figure 10 process and identification module MD2 two large modules, wherein: the foot ground acting force when gait data acquisition module MD1 obtains the walking of people's nature with being used for the nothing invasion, formed by Np three-dimensional force force plate MD11 and 1 data collection and transmission unit MD12, Np three-dimensional force force plate MD11 is electrically connected, they are closely concordant and be installed in snugly and consist of gait passage GW under the Flo of ground, data acquisition is electrically connected with Np three-dimensional force force plate MD11 that is electrically connected by Ethernet with the end of transmission unit MD12, data acquisition is electrically connected with computer PC by usb data line DL with the other end of transmission unit MD12, realizes Real-time Collection and the processing of foot ground force data.Np represents is the number of the three-dimensional force force plate that adopts, and for the people's that can collect different heights foot ground force data, Np is preferably and is not less than 3 integer.Three-dimensional force force plate of the present invention is that China Patent No. is six axle power platforms described in the patent of invention " a kind of multiaxis power plane matrix and human body walking gait information getting method " of ZL200410014352.9.In actual applications, as measured when walking is passed through on the gait passage shown in the GW from Figure 10, because three-dimensional force force plate MD11 is installed under the Flo of ground, the measured does not just know to have instrument and equipment gathering his/her gait data like this, be that gait data acquisition of the present invention is hidden and without invading and harassing, just can guarantee that also the gait information that collects is true nature, thereby avoid the appearance of the situation of deliberately imitating; Gait data processing and identification module MD2 are mounted in the software module in the computer PC, this module is at first carried out denoising by the wavelet transformation hard threshold method to the foot ground force data that collects, then characterize gait by gait feature abstracting method extraction gait frequency domain character and gait temporal signatures, merge by gait feature again and set up the gait feature collection, adopt at last support vector machine classifier that gait is identified.

Figure 11 is the block diagram of gait data processing and identification module MD2, it is that the gait data that obtains is processed with feature extraction and realized the module of remote hidden Gait Recognition that gait data among the present invention is processed with identification module MD2, by gait data denoising unit MD21, waveform alignment unit MD22, sample split cells MD23, gait frequency domain character extraction unit MD24, gait temporal signatures extraction unit MD25, gait feature integrated unit MD26 and Gait Recognition unit MD27 coordinate to finish gait data and process and identification work, wherein: gait data denoising unit MD21, the noises such as the industrial frequency noise in the foot ground acting force that the removal of employing wavelet transformation hard threshold method collects and electromagnetic interference (EMI) obtain the high gait data of signal to noise ratio (S/N ratio); Waveform alignment unit MD22, the step that adopts the waveform alignment will foot ground acting force automatic aligning according to the reference point on the vertical support force curve in the foot ground acting force with the foot ground acting force of all personnel's sample of collecting; Sample split cells MD23, the foot ground acting force that the step that adopts described sample to split will collect splits into a plurality of segmentation foots ground acting force according to the reference point on the vertical support force curve in the foot ground acting force; Gait frequency domain character extraction unit MD24, the step that adopts described gait frequency domain character to extract is alignd from process denoising and waveform, or further characterizes the gait frequency domain character through extraction WAVELET PACKET DECOMPOSITION coefficient in the foot ground acting force of sample deconsolidation process again; Gait temporal signatures extraction unit MD25, adopt step that the gait temporal signatures extracts from through denoising even the power value of calculating key point through the foot ground acting force curve of sample deconsolidation process with the time mutually, rate of change and the momentum of adjacent key point characterize the gait temporal signatures; Gait feature integrated unit MD26 adopts FCM Algorithms to pick out minimum optimum gait frequency domain character subset from the gait frequency domain character that extracts, and directly makes up the gait feature collection after obtaining merging with the gait temporal signatures again; Gait Recognition unit MD27 adopts support vector machine classifier that the gait feature that extracts from foot ground acting force is carried out gait and identification.

Foot ground acting force when utilization is obtained pedestrian's natural reality and is difficult for imitated walking with can being hidden in the hidden nothing invasion of underground force plate in the inventive method implementation process is as gait data, adopt again the wavelet transformation hard threshold method that gait data is carried out denoising, adopt wavelet packet decomposition algorithm and difference algorithm from foot ground acting force, to extract respectively gait frequency domain character and gait temporal signatures sign gait, use support vector machine that the pedestrian is carried out gait classification or identification, propose simultaneously waveform alignment and sample method for splitting and improve recognition accuracy, consist of the Gait Recognition system of unique advantage.In the technical scheme of Gait Recognition of the present invention system, Method of Gait Feature Extraction is one of core, it also is the basis of gait analysis, the present invention proposes from comprising left and right sides shearing force, the gait feature abstracting method of kinematics and dynamic characteristic when characterizing all-sidedly and accurately the walking of people's nature in the foot ground acting force of front and back shearing force and vertical support power, this Method of Gait Feature Extraction comprises that the gait frequency domain character extracts and the gait temporal signatures extracts, foot ground acting force when the extraction of gait frequency domain character refers to the walking of obtaining is as gait data, gait data is carried out denoising and waveform alignment, adopt the WAVELET PACKET DECOMPOSITION method from the gait data of processing, to extract the gait frequency domain character; Foot ground acting force when the extraction of gait temporal signatures refers to the walking of obtaining is as gait data, gait data is carried out denoising, detect key point and reference point on the foot ground acting force curve, with acting force rate of change and the momentum sign gait temporal signatures of the timely phase of power value at key point place, adjacent key point.

Fig. 1 is based on the Method of Gait Feature Extraction algorithm flow chart of foot ground acting force, and component wherein (a) extracts process flow diagram for the gait frequency domain character, and component (b) extracts process flow diagram for the gait temporal signatures.

Implementation step based on the gait feature abstracting method technical scheme of foot ground acting force among the present invention is as follows:

Referring to the component among Fig. 1 (a), the step that this gait frequency domain character extracts is as follows:

Frequency domain first step S11: the foot ground acting force when adopting the three-dimensional force force plate to obtain people's normal gait by the gait passage, with once without stop without return normal consistently from the gait passage walking by the time foot ground acting force as an effective gait data, transmit and be saved in the computer PC;

Frequency domain second step S12: adopt the wavelet transformation hard threshold method that foot ground acting force is carried out denoising, obtain the foot ground acting force after the denoising;

Frequency domain the 3rd step S13: select the vertical support power in the foot ground acting force after the denoising, adopt the first order difference algorithm to detect trough point on the vertical support force curve, with the reference point of the trough point on the vertical support force curve as sufficient ground acting force curve;

Frequency domain the 4th step S14: take the reference point of foot ground acting force curve as benchmark, the foot ground acting force after adopting linear interpolation algorithm to denoising carries out the waveform alignment, obtains the foot ground acting force after the waveform alignment;

The 5th step of frequency domain S15: the complexity according to the application places gait data acquisition judges whether the needs exptended sample, if in the place that personnel fix, can multi collect everyone respectively with the different speeds of travel with wear the normally gait data by the gait passage of different shoes, then do not need exptended sample, carry out the 8th step of frequency domain S18; If on the airport, the strong places of mobility of people such as market and customs, can not multi collect everyone respectively with the different speeds of travel with wear different shoes normally by the gait data of gait passage, then need exptended sample, carry out the 6th step of frequency domain S16;

Frequency domain the 6th step S16: when needing exptended sample, take the reference point of foot ground acting force curve as benchmark, the foot ground acting force after the waveform alignment is carried out sample split, split into a plurality of segmentations foots ground acting force, the quantity that expands the gait data sample;

The 7th step of frequency domain S17: adopt L layer wavelet packet decomposition algorithm from the segmentation foot ground acting force that splits out, to extract the gait frequency domain character of segmentation;

The 8th step of frequency domain S18: when not needing exptended sample, adopt the foot ground acting force of L layer wavelet packet decomposition algorithm after the waveform alignment and extract omnidistance gait frequency domain character.

Referring to the component among Fig. 1 (b), the step that this gait temporal signatures extracts is as follows:

Time domain first step S21: the foot ground acting force when obtaining people's normal gait by the gait passage by the three-dimensional force force plate, with once without stop without return normal consistently from the gait passage walking by the time foot ground acting force as an effective gait data, transmit and be saved in the computer PC;

Time domain second step S22: adopt the wavelet transformation hard threshold method that foot ground acting force is carried out denoising, obtain the foot ground acting force after the denoising;

The 3rd step of time domain S23: adopt front and back shearing force curve and the wave crest point on the vertical support force curve and trough point in the foot ground acting force after the first order difference algorithm detects respectively denoising, with wave crest point and the trough point key point as foot ground acting force curve, with the trough point on the vertical support force curve as sufficient reference point of acting force curve;

The 4th step of time domain S24: the complexity according to the application places gait data acquisition judges whether the needs exptended sample, if in the place that personnel fix, can multi collect everyone respectively with the different speeds of travel with wear the normally gait data by the gait passage of different shoes, then do not need exptended sample, carry out the 7th step of time domain S27; If on the airport, the strong places of mobility of people such as market and customs, can not multi collect everyone respectively with the different speeds of travel with wear different shoes normally by the gait data of gait passage, then need exptended sample, carry out the 5th step of time domain S25;

Time domain the 5th step S25: when needing exptended sample, take the reference point of foot ground acting force curve as benchmark, the foot ground acting force after the denoising is carried out sample split, split into a plurality of segmentations foots ground acting force, the quantity that expands the gait data sample;

Time domain the 6th step S26: with the power value at the key point place on the vertical support force curve in each segmentation foot ground acting force and power value occur the time on the acting force rate of change of phase, adjacent key point and momentum and the corresponding front and back shearing force curve the key point place the power value, drive the gait temporal signatures that momentum and braking momentum characterize segmentation;

Time domain the 7th step S27: when not needing exptended sample, with the power value at the key point place on the vertical support force curve in the foot ground acting force after the denoising and power value occur the time on the acting force rate of change of phase, adjacent key point and momentum and the corresponding front and back shearing force curve the key point place the power value, drive momentum and the braking momentum characterizes omnidistance gait temporal signatures;

Here in fact the momentum of indication is exactly the integration of power and time, and driving momentum is exactly the power and the integration of time that occupy on the force-time curve more than 0, brakes momentum and be exactly the power that occupy on the force-time curve below 0 and the integration of time.

The gait feature that extracts by above Time-Frequency Analysis Method can characterize people's natural walking feature all-sidedly and accurately, the global features such as the kinematics of the gait temporal signatures reflection walking of from foot ground acting force, extracting and dynamics, the motion characteristic that can reflect all-sidedly and accurately walking, the minutia of the gait frequency domain character reflection walking of extracting, from these gait features, both can analyze the periodicity of gait, stability and dynamic property, can also analyze subtly the spectral property of gait, when two kinds of gait features are used for Gait Recognition simultaneously, because utilized the complementation of whole and minutia, its recognition performance is more excellent, and great many of experiments has also confirmed the validity of the inventive method.The below is elaborated to the details that relates in gait feature abstracting method of the present invention and the Gait Recognition system.

1. gait data---foot ground acting force obtains

Used gait data among the present invention, i.e. foot ground acting force during walking by as shown in figure 10 gait data acquisition module MD1 without obtaining with invading and harassing.When the gait passage GW of Np three-dimensional force force plate MD11 composition shown in Figure 10 is passed through in measured's walking, the foot ground acting force when gait data acquisition module MD1 just gathers his/her walking in real time.In the present embodiment Np is set as 5.Among the present invention, the foot ground acting force that gathers is three-dimensional force, comprise as shown in Figure 2 left and right sides shearing force Fx, front and back shearing force Fy and vertical support power Fz, they are difficult to imitated, and can be hidden without getting access to with invading and harassing, remote collection does not need the measured to cooperate during collection round the clock, can guarantee that like this gait data that gathers is natural reality.

2. gait data denoising

In order to make more accurate stable of the gait feature that extracts, the gait data that requires to be used for feature extraction has higher quality, and the data that collect often contain noise.Therefore, before carrying out Method of Gait Feature Extraction, need to carry out denoising to the gait data that collects.

The software filtering method is host computer denoising method commonly used outside the hardware filtering denoising.Because wavelet transformation has the multiresolution analysis characteristic of being described as " school microscop ", makes it have band-pass filtering function, Wavelet noise-eliminating method is used more and more wider.Wavelet noise-eliminating method carries out wavelet decomposition with signal under different scale, just the mixed signal that comprises the multi-frequency composition can be decomposed different frequency range, then presses frequency band according to the different characteristic of each seed signal on frequency domain and processes.Foot ground force signals during walking mainly be frequency at 40Hz with interior low frequency signal, and be the 50Hz industrial frequency noise introduced of working power and the electromagnetic interference (EMI) in the Acquisition Circuit by the overriding noise in the gait passage acquisition system, the electromagnetic interference (EMI) major part is high frequency noise.In addition, the singular point of the foot ground acting force that collects is more, a lot of flex points, wave crest point P1, the P2 among similar Fig. 2, P3, P4 and trough point V1, V2, V3, V4, V5 can occur at its curve.And based on wavelet modulus maxima method and Wavelet Transform Threshold method relatively are fit to carry out denoising to containing the more signal of singular point or data, again because the Wavelet Transform Threshold method is applicable to the denoising of most of signals, and computing velocity is fast, is very suitable for the foot ground force data among the present invention is carried out the requirement that real-time de-noising is processed.Therefore, the present invention selects the Wavelet Transform Threshold method that foot ground force data is carried out denoising.Wavelet Denoising Method comprises wavelet decomposition → to the operation of high frequency wavelet coefficient processing → wavelet reconstruction three basic, for time-domain signal f (t), first with its discrete f that turns to _k, make c _{0, k}=f _kThe gait data that collects among the present invention can be considered as discretize data, i.e. f _k=f (t), the orthogonal wavelet decomposition formula of signal f (t) is:

$\left\{\begin{array}{c}{c}_{j,k}=\underset{n}{\mathrm{\Σ}}{c}_{j-1,k}{h}_{n-2k}\\ d_{j,k}=\underset{n}{\mathrm{\Σ}}{d}_{j-1,k}{g}_{n-2k}\end{array}\right.(k=\mathrm{0,1,2},...,N-1)---\left(1\right)$

The wavelet reconstruction formula is:

$c_{j-1,n}=\underset{n}{\mathrm{\Σ}}{c}_{j,n}{h}_{k-2n+}\underset{n}{\mathrm{\Σ}}{d}_{j,n}{g}_{k-2n}---\left(2\right)$

In the formula, c _{J, k}Be scale coefficient, d _{J, k}Be wavelet coefficient, h, g are pair of orthogonal mirror filter group, and j is for decomposing the number of plies, and N is discrete sampling count (data length).

For the Wavelet Transform Threshold method, estimate wavelet coefficient c according to threshold value λ _{J, k}Normal adopt two kinds of functions: soft-threshold function and hard-threshold function are called wavelet transformation soft-threshold method and wavelet transformation hard threshold method based on the Wavelet Transform Threshold method of these two kinds of functions.Carry out signal after wavelet reconstruction can obtain denoising by the wavelet coefficient of estimating.The soft-threshold function is:

$c_{j,k}=\left\{\begin{array}{cc}\mathrm{sgn}\left(c_{j,k}\right)\left(\right|c_{j,k}|-\mathrm{\&lambda;})& \left|c_{j,k}\right|\&GreaterEqual;\mathrm{\&lambda;}\\ 0& \left|c_{j,k}\right|<\mathrm{\&lambda;}\end{array}\right.---\left(3\right)$

The hard-threshold function is:

$c_{j,k}=\left\{\begin{array}{cc}{c}_{j,k}& \left|c_{j,k}\right|\&GreaterEqual;\mathrm{\&lambda;}\\ 0& \left|c_{j,k}\right|<\mathrm{\&lambda;}\end{array}\right.---\left(4\right)$

In the formula, sgn () is sign function, and λ is threshold value.The choosing method of threshold value λ is more, and is larger on the denoising effect impact, should select according to pretreated signal.Unified threshold value by Donoho and Johnstone proposition

More with getting, σ is the standard deviation of noise.

The present invention is by analyzing and experimental verification, find that the wavelet transformation hard threshold method is better to the denoising effect of foot ground acting force than wavelet transformation soft-threshold method, also be determined by experiment and adopt 5 layers of Daubechies wavelet function as the wavelet basis function of wavelet transformation hard threshold method foot ground acting force to be carried out denoising.Fig. 4 has showed the effect that adopts the wavelet transformation hard threshold method that foot ground acting force is carried out denoising, Fig. 3 is the left and right sides shearing force Fx curve in the front foot ground acting force of denoising, front and back shearing force Fy curve and vertical support power Fz curve, Fig. 4 is the left and right sides shearing force Fxd curve in the foot ground acting force after the denoising, front and back shearing force Fyd curve and vertical support power Fzd curve, contrast is found, the low noise while is falling in the wavelet transformation hard threshold method, the unique point of original signal has all been preserved, do not change details and the configuration feature of original signal, prove that its denoising effect is better.

3. Method of Gait Feature Extraction

Gait feature is the basis of gait analysis and Gait Recognition, therefore, the good gait feature of feature extraction algorithm Extraction and discrimination performance that need to design, the design feature selection algorithm reduces intrinsic dimensionality again, improves efficiency of algorithm.

Comprise abundant gait information in foot ground acting force during walking, the temporal signatures of existing reflection walking period and globality also has the frequency domain character that reflects walking spectrum signature and local detail.From data Layer, foot ground acting force curve has typical morphological feature, is the external embodiment of walking movement and dynamic characteristic, and the unique point on the curve is related with the foot motion in each stage of gait cycle very large.The gait temporal signatures has clear and definite physical significance, and interpretation is strong, and deficiency is to take into account the overall situation and details, may omit some have fine effect to classification minutia.Frequency-domain analysis method is the analytical approach of signal common, can go out to excavate the feature that signal can't be observed on time domain.Wavelet transformation is as a kind of outstanding signal time frequency analyzing tool, the local feature of the characteristic of its multiresolution analysis gait so that we have the ability to characterize in time-frequency two territories, and wavelet package transforms more is 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 is used for extracting the cancer classification feature from gene expression profile data.And gait temporal signatures and gait frequency domain character are merged use, and will make gait feature complementary on whole and details, will improve recognition performance.Therefore, design feature extraction algorithm of the present invention extracts frequency domain gait feature and time domain gait feature simultaneously, and they are used for identification simultaneously.

3.1 the gait frequency domain character extracts

Get access to effective foot ground acting force by above-mentioned frequency domain first step S11, adopt again the wavelet transformation hard threshold method that foot ground acting force is carried out denoising, obtain the foot ground acting force after the denoising.Foot ground acting force after the denoising is carried out the step of waveform alignment, is namely carried out following steps:

The waveform alignment first step: the dimension of the foot ground force data after adopting linear interpolation algorithm with denoising normalizes to same value G * Np, Np is the sum of the three-dimensional force force plate of use, G is the step pitch parameter relevant with sample frequency, G is set as 50 integral multiple, G is set as 500 in the present embodiment;

Waveform alignment second step: go out n trough point on the vertical support force curve in the foot ground acting force after the dimension normalization by the first order difference algorithm search, n trough point is as a reference point; N is the number of the trough point on the vertical support force curve, is limited positive integer; Fig. 2 is the waveform alignment of foot ground acting force curve and the reference point schematic diagram that sample splits, by waveform alignment second step, can search out V1, V2, V3, V4 and V5 such as crisis-cross among Fig. 2 and be the trough point on the vertical support power Fz curve, with the reference point of these five trough points as the present embodiment;

Waveform alignd for the 3rd step: the reference point on the vertical support force curve is as reference, by linear interpolation left and right sides shearing force curve, front and back shearing force curve and vertical support force curve in the foot ground acting force are alignd, so that the trough point of the n on the vertical support force curve snaps to respectively the position of appointment, n trough point snaps to G * (n-0.5); In the present embodiment, five trough point V1, V2, V3, V4 and V5 on the vertical support power Fz curve as shown in Figure 2 are as reference point, need to be by linear interpolation with the left and right sides shearing force Fx curve in the foot ground acting force, front and back shearing force Fy curve and the alignment of vertical support power Fz curve, so that five trough point V1, V2, V3, V4, V5 on the vertical support power Fz curve snap to respectively 250,750,1250,1750 and 2250 position.Fig. 4 and Fig. 5 have showed the curve of the foot ground acting force before and after the waveform alignment.

Fig. 4 is left and right sides shearing force Fxd curve, front and back shearing force Fyd curve and the vertical support power Fzd curve in the foot ground acting force after the denoising, foot ground acting force is not carried out the waveform alignment, five trough point V of crisis-cross do not snap to the position of appointment on the vertical support power Fzd curve in the foot ground acting force; Fig. 5 is left and right sides shearing force Fxr curve, front and back shearing force Fyr curve and the vertical support power Fzr curve in the foot ground acting force after the alignment of denoising and waveform, and five trough point V on the vertical support power Fzr curve in the foot ground acting force among Fig. 5 have snapped to respectively 250,750,1250,1750 and 2250 position.

After by above step foot ground acting force being carried out the waveform alignment, complexity according to the application places gait data acquisition judges whether the needs exptended sample again, if in the place that personnel fix, can multi collect everyone respectively with the different speeds of travel with wear different shoes normally by the gait data of gait passage, then do not need exptended sample; If on the airport, the strong places of mobility of people such as market and customs, can not multi collect everyone respectively with the different speeds of travel with wear different shoes normally by the gait data of gait passage, then need exptended sample, carries out the step of following sample fractionation:

Sample splits the first step: go out n trough point on the vertical support force curve in the foot ground acting force by the first order difference algorithm search, and n trough point is as a reference point; P0, the P1, P2, P3, P4 and the P5 that indicate such as Fig. 2 hollow core circle are the wave crest point of vertical support power Fz curve, the V1 of crisis-cross, V2, V3, V4 and V5 are five trough points of vertical support power Fz curve, and be in the present embodiment that these five trough point V1, V2, V3, V4 and V5 are as a reference point;

Sample splits second step: the reference point on the vertical support force curve is as cut-point, respectively the part between first reference point and last reference point in left and right sides shearing force curve, front and back shearing force curve and the vertical support force curve in the foot ground acting force is split into the n-1 section, foot ground force data between the two adjacent reference point is extracted, as a new sample, split out altogether n-1 sample.In the present embodiment, be in part between reference point (being the trough point) V1 and the V5 in left and right sides shearing force Fx curve, front and back shearing force Fy curve and the vertical support power Fz curve with foot shown in Figure 2 ground acting force take V2, V3 and V4 as cut-point, they are split into 4 sections, [V1 will be in respectively, V2], [V2, V3], the foot ground acting force between [V3, V4] and [V4, V5] is as 4 new samples.As being in respectively [250 in left and right sides shearing force Fxr curve, front and back shearing force Fyr curve and the vertical support power Fzr curve in the foot ground acting force shown in Figure 5,750], [750,1250], 4 sections between [1250,1750] and [1750,2250] are as 4 new samples.

The step that above sample splits also can be used for the leaching process of gait temporal signatures.

Foot ground acting force is carried out can adopting wavelet packet decomposition algorithm to extract the gait frequency domain character after denoising and the waveform alignment, and the operating process that the WAVELET PACKET DECOMPOSITION of the segmentation foot ground acting force after the foot ground acting force of the whole process walking before sample splits or sample split is extracted the gait frequency domain character is the same.Be split as example not need carrying out sample in the present embodiment, adopt wavelet packet decomposition algorithm from characterizing the gait frequency domain character through extracting the WAVELET PACKET DECOMPOSITION coefficient the foot ground acting force after denoising and the waveform alignment.

When carrying out WAVELET PACKET DECOMPOSITION, determine in advance wavelet basis function and decompose the number of plies.The selection of wavelet basis function usually needs by virtue of experience to determine with experiment, the wavelet basis function commonly used that can be used at present not only continuous wavelet transform but also can be used for wavelet transform has five kinds of Haar, Daubechies, Biorthgonal, Coiflets and Symlets, and it is higher that the present invention found through experiments when selecting the Coif1 small echo to carry out WAVELET PACKET DECOMPOSITION discrimination.Decompose the number of plies and usually determined by signal low-limit frequency and highest frequency, the highest frequency of the foot ground force signals that collects is in 400Hz, and the eigenfrequency of the foot ground acting force of gait is decomposed the number of plies about 30Hz

We select 4 layers of WAVELET PACKET DECOMPOSITION.The foot ground force data that collects among the present invention can be considered as discretize data, be fn=f (t), the present invention carries out 4 layers of WAVELET PACKET DECOMPOSITION to front and back shearing force, left and right sides shearing force and vertical support power in the foot ground acting force successively by WAVELET PACKET DECOMPOSITION formula (5), preserve the WAVELET PACKET DECOMPOSITION coefficient of each subspace of every one deck, the WAVELET PACKET DECOMPOSITION coefficient is namely as the gait frequency domain character.

The WAVELET PACKET DECOMPOSITION formula:

${\mathrm{<figure-callout\; id="0"\; label="W"\; filenames="HDA00002628656000021.png,HDA00002628656000022.png"\; state="\{\{state\}\}">W</figure-callout>}}_0^0\left(n\right)=f_n,$ n=0,1,…，N-1

$W_j^{2p}\left(n\right)=\underset{k\&Element;Z}{\mathrm{\&Sigma;}}{g}_{k-2n}{W}_{j-1}^l\left(k\right),$ n=0,1,…，N ^2-j-1 (5)

$W_j^{2p+1}\left(n\right)=\underset{k\&Element;Z}{\mathrm{\&Sigma;}}{h}_{k-2n}{W}_{j-1}^l\left(k\right),$ n=0,1,…，N ^2-j-1

Wherein, j=1,2 ..., J, J=log ₂N is scale parameter,

Be the WAVELET PACKET DECOMPOSITION coefficient, l, p, n are respectively and decompose the number of plies, frequency and time point.

Fig. 6, Fig. 7 and Fig. 8 have showed the example that adopts 4 layers of WAVELET PACKET DECOMPOSITION method extraction gait frequency domain character from foot ground acting force.With the wavelet transformation hard threshold method original foot ground acting force is carried out denoising first, again foot ground acting force is carried out the amplitude normalized and data dimension is normalized to 2500 dimensions divided by body weight, trough point take the vertical support force curve aligns as reference point foot ground acting force carries out waveform again, adopt at last 4 layers of wavelet packet decomposition algorithm that the foot ground acting force after processing through these is decomposed, the WAVELET PACKET DECOMPOSITION coefficient that obtains every one deck characterizes the gait frequency domain character.Fig. 6 is through the vertical support power Fwz curve after denoising, waveform alignment and the amplitude normalized, Fig. 7 carries out 4 layers of ground floor yardstick sequence after the WAVELET PACKET DECOMPOSITION to vertical support power Fwz, Fig. 8 carries out 4 layers of ground floor wavelet sequence after the WAVELET PACKET DECOMPOSITION to vertical support power Fwz, characterizes the gait frequency domain character from general picture and details respectively.

3.2 the gait temporal signatures extracts

The gait temporal signatures extracts mainly and constructs the characteristic parameter relevant with dynamics with walking movement (periodically) with calculating by the key point on the foot ground acting force curve, and key is the detection of key point and definition, the calculating of characteristic parameter.At first get access to effective foot ground acting force by above-mentioned time domain first step S21, then adopt the wavelet transformation hard threshold method that foot ground acting force is carried out denoising, obtain the foot ground acting force after the denoising, and then detect key point and calculated characteristics parameter on the foot ground acting force curve.The same for described in whether needing to carry out step that judgement that sample splits and sample split and the gait frequency domain character extracting, before and after sample split, the detection of sufficient ground acting force curve key point and the computing method of characteristic parameter were the same.The present invention is defined as key point with wave crest point and the trough point on the foot ground acting force curve, because the pedaling of they and walking, step on actions such as dual-grippers certain corresponding relation arranged, can intuitively reflect the features such as the kinematics (cycle, time phase, rhythm) of walking and dynamics (action intensity, the mild property of moving), these features also are the gait features of essence.Therefore, the present invention defines and calculates the gait temporal signatures with the front and back shearing force curve in the foot ground acting force and the wave crest point on the vertical support force curve and trough point as key point, wave crest point on the curve and trough point are the differential variation points, be the difference value opposite in sign of its left and right sides point of proximity, and the wave crest point on the mesopodium of the present invention ground acting force curve and trough point have periodically simultaneously.So the present invention adopts the first order difference algorithm can detect wave crest point and trough point in the foot ground acting force curve.

Fig. 9 is that key point and the gait temporal signatures of foot ground acting force curve characterizes schematic diagram.In the present embodiment, wave crest point P0 with open circles sign on the vertical support power Fz curve in the foot shown in Figure 9 ground acting force, P1, P2, P3, P4, the trough point V1 of P5 and crisis-cross, V2, V3, V4, the detection of V5 is example, the single order forward difference of node-by-node algorithm vertical support power Fz, local wave crest point or trough point appear when the symbol of single order forward difference value changes, if the single order forward difference value of this some left side point of proximity is for just, and the single order forward difference value of right side point of proximity is for negative, then be judged to be local wave crest point, then be judged to be on the contrary local trough point; If do not have the point larger than its value in 100 points that close on about this point, judge that then this part wave crest point is overall wave crest point P0, P1, P2, P3, P4, P5; If do not have the point less than its value in 100 points that close on about this point, judge that then this part trough point is overall trough point V1, V2, V3, V4, V5.Wave crest point on the shearing force Fy curve of front and back and the detection method of trough point are analogous to the operation on the vertical support power Fz curve.With the power value of detected all wave crest points and trough point and power value occur the time define mutually and calculate the gait temporal signatures.For the time phase definition, Fz is as the criterion with vertical support power.The time that trough point Vi on the vertical support power Fz is corresponding is Ti, and corresponding power value is Fzvi; The time that wave crest point Pi is corresponding is Ti, and the time that i+1(P0 is corresponding is set as T0, and the time that P5 is corresponding is set as T6), corresponding power value is Fzpi; The lasting cycle T a of vertical support power Fz; The time T i that the time phase Ti ' that the corresponding force value occurs is occurred by the power value accounts for the ratio that continues cycle T a and calculates, i.e. Ti '=Ti/Ta; Power value corresponding to trough point on the shearing force Fy curve of front and back is Fyvi, and the power value that wave crest point is corresponding is Fypi; Acting force rate of change Rzps, Rzpe, Rvpi and Rpvi on the vertical support power Fz curve represents to move mild property, calculated by power value and the ratio of time of adjacent key point, that is:

Rvpi＝(Fzpi-Fzvi)/(T _i,i+1-T _i)

Rpvi＝(Fzpi-Fzvi ₊₁)/(T _i+1-T _i，i+1) （6）

Rzps＝Fzp0/T0

Rzpe＝Fzp5/Ta-T6

Momentum I represents the action intensity of gait, calculates with section internal force and the integration of time sometime, that is:

$I={\&Integral;}_{\mathrm{ti}}^{\mathrm{tj}}\mathrm{Fdt}---\left(7\right)$

Here in fact the momentum of indication is exactly the integration of power and time, and driving momentum is exactly the power and the integration of time that occupy on the force-time curve more than 0, brakes momentum and be exactly the power that occupy on the force-time curve below 0 and the integration of time.

In the present embodiment, can obtain wave crest point shown in Figure 9 corresponding power value Fzp0, Fzp1, Fzp2, Fzp3, Fzp4, Fzp5, Fyp1, Fyp2, Fyp3, Fyp4, Fyp5, power value Fzv1, Fzv2, Fzv3, Fzv4, Fzv5, Fyv0, Fyv1, Fyv2, Fyv3, Fyv4 that trough point is corresponding, the time T i that the time phase Ti ' that the corresponding force value occurs is occurred by the power value accounts for the ratio that continues cycle T a and calculates, be Ti'=Ti/Ta, phase T0 ', T1, T12', T2', T23', T3, T34', T4', T45', T5 ', T6' when obtaining; Acting force rate of change Rzps, Rzpe, Rvp1, Rvp2, Rvp3, Rvp4, Rpv1, Rpv2, Rpv3, Rpv4 and momentum Izpvs, Izvp1, Izvp2, Izvp3, Izvp4, Izpv1, Izpv2, Izpv3, Izpv4, the Izpve of adjacent key point on the vertical support power Fz curve; Driving momentum Iyvp1, Iyvp2, Iyvp3, Iyvp4 and braking momentum Iypv1, Iypv2, Iypv3, Iypv4 on the shearing force Fy curve of front and back; Characterize the gait temporal signatures with these 60 characteristic parameters.

4. gait feature merges

This unit is in order to obtain the gait feature collection for training and testing from gait frequency domain character and gait temporal signatures, adopt FCM Algorithms from the gait frequency domain character that extracts, to pick out minimum optimum gait frequency domain character subset, directly make up again the gait feature collection after obtaining merging with the gait temporal signatures.The step of selecting of minimum optimum gait frequency domain character subset is to adopt first FCM Algorithms to pick out the set of optimal wavelet bag from a plurality of wavelet packets that decomposite, and selects optimal wavelet bag coefficient of dissociation with FCM Algorithms from the set of optimal wavelet bag again.

FCM Algorithms is selected optimal wavelet bag and WAVELET PACKET DECOMPOSITION coefficient and is based on fuzzy membership and sorts to realize, for c class classification problem, and definition u _Ik∈ [0,1] is the degree of membership of k sample in the i class:

$u_{\mathrm{ik}}={\left[\underset{j=1}{\overset{c}{\mathrm{\&Sigma;}}}{({\left|\right|x_k-x_i\left|\right|}^2/{\left|\right|x_k-v_j\left|\right|}^2)}^{1/(b-1)}\right]}^{-1}---\left(8\right)$

Wherein,

Be the mean value of i class training sample, A _iThe set of i class training sample, N _iBe the sample number of i class training sample, || || be Euclidean distance, b〉1 fuzzy factor of degree of membership for a change, we get b=2.Work as x _k=v _iThe time, make u _Ik=1; Work as x _k=v _j, during i ≠ j, make u _Ik=0.

To the training sample that in feature space X, is labeled, define a fuzzy membership function F (x) ∈ (0, N] estimate the classification capacity of X,

The value of F (x) is larger, and the classification capacity of feature space X is stronger.

$F\left(x\right)=\underset{i=1}{\overset{c}{\mathrm{\&Sigma;}}}\underset{k\&Element;A_i}{\mathrm{\&Sigma;}}{u}_{\mathrm{ik}}---\left(9\right)$

In WAVELET PACKET DECOMPOSITION, the wavelet packet S (j, k) that decomposition is obtained judges the height of its classification capacity as a feature space by calculating its membership function.Definition S (j, k) is S (l), l=2 ^j-1+k, k=0,1 ..., 2 ^j-1.Make x ^(l)={ x ₁, x ₂... x _NBe the training sample of S (j, k) subspace, and calculating F (S (l)) by formula (8) and (9), F (S (l)) is larger, and the classification capacity of proper subspace S (j, k) is just stronger.Optimal wavelet bag collection to select step as follows:

1) to all l=2 ^j-1+k, k=0,1 .., 2 ^j-1, calculate degree of membership F (S (l)).

2) from big to small S is sorted according to the degree of membership value, establish new

Order

3) with first element S (j of S ₁) move on to X ^*

4) to any one S (k) ∈ S, if S (k) is S (j ₁) former generation or offspring, S (k) is shifted out S.

5) if

Go to step 6); Otherwise, go to step 3) and continue.

6) selected X ^*Be the set of optimal wavelet bag.

After choosing the optimal wavelet bag, also need all the WAVELET PACKET DECOMPOSITION coefficients in the selected wavelet packet set are further sorted by its classification capacity.At X ^*In sort method as follows:

(1) with X ^*Whole WAVELET PACKET DECOMPOSITION coefficients of middle all elements are as the feature of sample, mark training sample Ψ={ x ₁, x ₂..., x _N.

(2) calculate sample x in the i class by formula (10) _kThe degree of membership of j feature.Work as x _Kj=v _IjThe time, make u _Ij(x _Kj)=1; Work as x _Kj=v _Mj, during m ≠ j, make u _Ij(x _Kj)=0.

$u_{\mathrm{ik}}\left(x_{\mathrm{kj}}\right)={\left[\underset{m=1}{\overset{c}{\mathrm{\&Sigma;}}}\right[{(x_{\mathrm{kj}}-v_{\mathrm{ij}})}^2/{(x_{\mathrm{kj}}-v_{\mathrm{mj}})}^2\left]\right]}^{-1}---\left(10\right)$

(3) feature j is calculated the degree of membership F (j) of all samples.

$F\left(j\right)=\underset{i=1}{\overset{c}{\mathrm{\&Sigma;}}}\underset{k\&Element;A_i}{\mathrm{\&Sigma;}}{u}_{\mathrm{ij}}\left(x_{\mathrm{kj}}\right)---\left(11\right)$

(4) all features are carried out descending sort according to its degree of membership value, its classification capacity of the larger explanation of degree of membership value is stronger, dimension as required pick out come the front the WAVELET PACKET DECOMPOSITION coefficient as optimal wavelet bag coefficient of dissociation, consist of minimum optimum gait frequency domain character subset.

5. based on the Gait Recognition of support vector machine

In pattern recognition system, feature extraction is the basis, and the design of sorter and selection have important impact to the system identification performance.The present invention adopts the support vector machine (SVM:Support Vector Machines) that has outstanding behaviours in solving small sample, non-linear and higher-dimension pattern classification problem to come gait is identified.The unique distinction of support vector machine just is 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, that is to say that kernel function is crucial, should select suitable kernel function according to problem.At present kernel function commonly used has three kinds of polynomial kernel function, radial basis kernel function and Sigmoid kernel functions, compares with the Sigmoid kernel function, and the parameter that polynomial kernel function and radial basis kernel function need to be inputted is few; For Nonlinear Classification, the radial basis kernel function is best selection, choice for use radial basis kernel function among the present invention.In addition, when selecting the radial basis kernel function, the selection of a pair of parameter of kernel function---penalty factor and kernel functional parameter γ also has certain influence to classification performance, need to be with the parameter of radial basis kernel function to (C during use, γ) adjust to suitable value, adopt the grid search algorithm to determine that the kernel functional parameter of support vector machine is to (C, γ) among the present invention.

When adopting support vector machine to carry out Gait Recognition, first training classifier re-uses trained listening group and identifies unknown gait sample.Before the training, at first gather gait Sample Establishing gait data storehouse, suppose to have registered in the gait data storehouse n class (individual) gait sample, these gait sample input support vector machine classifiers are trained, obtain output and a classification model corresponding to 1～n.In the identifying, the real-time gait sample that gets access to is input in the trained support vector machine classifier, namely mate with the classification model of having trained, judge that according to output valve it is which kind of or which individual's gait, if output valve exceeds the scope of 1～n, then be judged to be unregistered gait sample.

In order to test the validity of gait feature abstracting method proposed by the invention and Gait Recognition system, sufficient ground force data when having implemented to gather in the experiment 61 people's self-selected speeds and wearing footwear and naturally walk by the gait passage shown in GW among Figure 10, with a people once naturally the foot ground force data of walking by the gait passage as a gait sample, everyone is gathered 20 effective gait samples, amount to 1220 effective gait samples, set up 61 people's gait data storehouse, with the reference index of correct recognition rata as evaluation recognition performance quality.At first adopt the foot ground acting force denoising of wavelet transformation hard threshold method to collecting, adopt respectively again the foot ground acting force of above-mentioned gait feature abstracting method after denoising and extract gait frequency domain character and gait temporal signatures, when extracting the gait frequency domain character, the waveform alignment schemes that adopts the present invention to propose is carried out the waveform alignment to foot ground acting force, adopt FCM Algorithms to pick out 120 optimum gait frequency domain characters, constitute again the gait feature vector sign gait of one 180 dimension with 60 gait temporal signatures, adopt respectively the support vector machine of different IPs function as sorter, carry out 4 repeated tests, correct recognition rata when find adopting support vector machine based on the radial basis kernel function as sorter is the highest, when kernel functional parameter to (C, γ) be set as (10,0.2) time, average correct recognition rata reaches 98.1%, proves that the present invention is feasible and effective.

Embodiment 2:

Being applied as example with the Gait Recognition system in the entrance guard management of certain bank vault in the present embodiment is illustrated implementation process of the present invention.

Bank vault is the focused protection place that a designated person just has the turnover authority; also be that the offender thinks into the place of stealing very much; even just as describe in the film, they are by snatching password and key, disguising oneself or the means such as fingerprint, iris of copying enter national treasury and steal.The convenience that the foot ground acting force of mat when walking obtaining and very difficult imitated, adopt when of the present invention, can press gait passage of 3 meters long that is formed by 5 three-dimensional force force plates of tight concordant erection method installation shown in the GW among Figure 10 by the subsurface before national treasury entrance antitheft door, thereby utilize the inventive method execution based on identification and the entrance guard management work of foot ground acting force.If the visitor walks the out-of-date personnel that enter of allowing that can be identified as appointment from the gait passage, then the antitheft door of encrypted pinning is opened automatically, allows the visitor to enter; Otherwise, will be denied access to, and superior management and control unit gives the alarm, automatically snap simultaneously and transmit image scene to higher level management and control unit.In the entrance guard management application of adopting the inventive method to bank vault, at first need to be in training process with designated person's sample training, the classification model of having been trained, identifying is identified the visitor by template again, identify visiting person and whether allow the people that enters, order or signal that whether recognition result is opened as the control anti-theft door for national treasury.

In the training process, on-demand creation designated person's gait data storehouse at first, foot ground acting force when utilization is installed in underground gait passage collection designated person before the bank vault entrance antitheft door with free constant speed, at a slow speed with fast by the gait passage, to foot ground acting force as gait data, gather 2 times under everyone the every kind self-selected speed, everyone has the sample of 6 gait datas; Utilize above-mentioned wavelet transformation hard threshold method to all gait data denoisings, with gait data storehouse of Sample Establishing of the gait data after all designated persons' the denoising, the later stage regularly or irregularly upgrades or increase gait data storehouse; Then adopt the above-mentioned gait frequency domain character extracting method based on WAVELET PACKET DECOMPOSITION from gait data, to extract the gait frequency domain character, adopt the above-mentioned gait temporal signatures extracting method based on the curve critical point detection from gait data, to extract the gait temporal signatures; Adopt again FCM Algorithms from the gait frequency domain character, to pick out minimum optimum gait frequency domain character subset, directly make up again the optimum gait feature collection after obtaining merging and the gait feature template of having trained with the gait temporal signatures; Adopt again support vector machine classifier to the training of gait feature collection, the classification model of having been trained.

In the identifying, at first adopt gait data acquisition module MD1 shown in Figure 10 to gather the visitor and walk out-of-date foot ground acting force from the gait passage is up, as the gait data of test sample book; Adopt again the method the same with training process to the gait data denoising and extract the gait feature collection, the gait feature template that utilization has been trained is carried out Feature Mapping to visitor's gait feature collection, pick out the gait feature subset of answering with the optimum gait feature set pair of training process, adopt support vector machine test visitor's gait feature subset whether to mate with the classification model of having trained, if mate fully with certain already present classification model in the gait data storehouse, the visitor is the personnel that enter of allowing of appointment so, give an order or signal is opened the antitheft door of encrypted pinning automatically, allow the visitor to enter; If do not mate with all classification models that existed in the gait data storehouse, the visitor is considered to not allow the personnel that enter so, do not send order or the signal of opening antitheft door, refusing it enters, and superior management and control unit gives the alarm, and automatically snaps simultaneously and transmit image scene to arrive higher level management and control unit.

Embodiment 3

In the present embodiment take the Gait Recognition system in the customer classification statistics application in certain market as example, implementation process of the present invention is done further concrete example explanation.

For the market of being engaged in the marketing activity of general merchandise class, should select throw in suitable commodity or kind, the grade of institute's merchandising adjusted according to client's classification.Wherein, the reasonable input of carrying out commodity according to client's age bracket is important, such as according to the turnover this market be young man, young women, middle-aged male, female middle-aged, elderly men or the old women commodity of selecting mainly to throw in which age bracket customer need in the majority.For realizing that client is automatically added up and classifies, at market entrance subsurface as shown in Figure 10 gait passage GW is installed, the Gait Recognition system classifies to the client in turnover market and adds up, and namely simply saves trouble, and also can not invade client's privacy.At first utilize the gait data acquisition module MD1 in the Gait Recognition system to gather young man, young women, middle-aged male, female middle-aged, elderly men and the old women foot ground acting force when each 50 people's walking of totally 6 class crowds are by the gait passage, gait data is processed with identification module MD2 the foot ground acting force of 6 class gait samples is processed and classification based training, be that gait data denoising unit MD21 carries out denoising to foot ground acting force, sample split cells MD23 splits into the foot ground force data of 1 gait sample the foot ground force data of 4 gait samples, extract the gait temporal signatures the foot ground acting force of gait temporal signatures extraction unit MD25 after processing, waveform alignment unit MD22 carries out the waveform alignment to the foot ground acting force after processing, extract the gait frequency domain character the foot ground acting force of gait frequency domain character extraction unit MD24 after sample fractionation and waveform registration process, gait feature integrated unit MD26 obtains the gait feature collection for training from gait frequency domain character and gait temporal signatures, Gait Recognition unit MD27 obtains 6 classification models to 6 class crowds' gait feature collection training.After training is finished, can carry out the classification of age bracket to the personnel in turnover market, namely when client when being installed in the underground gait passage in gateway, market, gait data acquisition module MD1 obtains client's foot ground acting force, gait data is processed with identification module MD2 and extract the gait feature collection from client's foot ground acting force, and mate with 6 classification models that training stage training obtains, be young man with customer identification, young women, middle-aged male, female middle-aged, a certain class people in elderly men and the old women, add up the probability that such crowd occurs, according to the in time input of adjustment commodity of probability size that such crowd occurs, accomplished the purpose input.

In a word, foot ground acting force during based on walking, the present invention proposes a kind of simple and effective gait feature abstracting method and remote hidden Gait Recognition system, this system comprises gait data acquisition module and gait data processing and identification module, foot ground acting force when the hidden nothing of gait data acquisition module is obtained walking with invading and harassing, gait data is processed and identification module therefrom extracts gait feature and carries out Gait Recognition.At first, the foot ground acting force that utilization can be hidden in gait passage under the ground/floor when obtaining walking, then carry out the Method of Gait Feature Extraction step: adopt the wavelet transformation hard threshold method to the acting force denoising of foot ground, adopt again first order difference algorithm detection key point and calculate the gait temporal signatures, foot ground acting force is carried out the rear employing of waveform alignment WAVELET PACKET DECOMPOSITION coefficient characterize the gait frequency domain character.When being used for identification, adopting take FCM Algorithms and pick out optimum gait feature collection as main Feature Fusion Algorithm, based on support vector machine gait is classified again or identify.Gait Recognition result verification on 61 self-built people gait data storehouse feasibility of the present invention and validity, the present invention is convenient and easy, has robustness.The foot ground acting force of the present invention during with walking is used for gait analysis and identification, and Gait Recognition of the present invention system has disguise and non-invasion in the information acquisition process, can obtain the real gait mechanical information of nature; Gait feature abstracting method of the present invention extracts gait feature from time domain and frequency domain, kinematics and dynamic characteristic in the time of can characterizing more all-sidedly and accurately the walking of people's nature propose contrast and classification capacity that waveform alignment and sample method for splitting improve gait feature simultaneously; Gait recognition method with respect to computer vision field, the recognition result of Gait Recognition of the present invention system is not subjected to that complex background, clothes and health block etc. to be affected, and the gait mechanical information is very difficult imitated, and native system can be avoided the problem of getting by under false pretences to greatest extent; The present invention combines these advantages, can satisfy better gait analysis and classification for the demand of gait feature, what is more important can satisfy the identity in the security sensitive places such as cyberage airport, customs, museum, party scene, bank and national treasury and differentiate and the safety precaution demand.

Claims (3)

1. the gait feature abstracting method based on foot ground acting force is characterized in that, comprises that the gait frequency domain character extracts and the gait temporal signatures extracts:

Described gait frequency domain character extracts and refers to: the foot ground acting force during with the walking of obtaining carries out denoising and waveform alignment as gait data to gait data, adopts the WAVELET PACKET DECOMPOSITION method to extract the gait frequency domain character from the gait data of processing;

Described gait temporal signatures extracts and refers to: the foot ground acting force during with the walking of obtaining is as gait data, gait data is carried out denoising, detect key point and reference point on the foot ground acting force curve, with acting force rate of change and the momentum sign gait temporal signatures of the timely phase of power value at key point place, adjacent key point;

Described foot ground acting force comprises left and right sides shearing force, front and back shearing force and vertical support power;

The step that described gait frequency domain character extracts is as follows:

The frequency domain first step (S11): the foot ground acting force when adopting the three-dimensional force force plate to obtain people's normal gait by the gait passage, with once without stop without return normal consistently from the gait passage walking by the time foot ground acting force as an effective gait data, transmit and be saved in the computing machine (PC);

Frequency domain second step (S12): adopt the wavelet transformation hard threshold method that foot ground acting force is carried out denoising, obtain the foot ground acting force after the denoising;

Frequency domain the 3rd step (S13): select the vertical support power in the foot ground acting force after the denoising, adopt the first order difference algorithm to detect trough point on the vertical support force curve, with the reference point of the trough point on the vertical support force curve as sufficient ground acting force curve;

Frequency domain the 4th step (S14): take the reference point of foot ground acting force curve as benchmark, the foot ground acting force after adopting linear interpolation algorithm to denoising carries out the waveform alignment, obtains the foot ground acting force after the waveform alignment;

The 5th step (S15) of frequency domain: the complexity according to the application places gait data acquisition judges whether the needs exptended sample, if in the place that personnel fix, can multi collect everyone respectively with the different speeds of travel with wear the normally gait data by the gait passage of different shoes, then do not need exptended sample, carry out the 8th step (S18) of frequency domain; If on the airport, the strong places of mobility of people such as market and customs, can not multi collect everyone respectively with the different speeds of travel with wear different shoes normally by the gait data of gait passage, then need exptended sample, carry out the 6th step (S16) of frequency domain;

Frequency domain the 6th step (S16): when needing exptended sample, take the reference point of foot ground acting force curve as benchmark, the foot ground acting force after the waveform alignment is carried out sample split, split into a plurality of segmentations foots ground acting force, the quantity that expands the gait data sample;

The 7th step (S17) of frequency domain: adopt L layer wavelet packet decomposition algorithm from the segmentation foot ground acting force that splits out, to extract the gait frequency domain character of segmentation;

The 8th step (S18) of frequency domain: when not needing exptended sample, adopt the foot ground acting force of L layer wavelet packet decomposition algorithm after the waveform alignment and extract omnidistance gait frequency domain character;

The step that described gait temporal signatures extracts is as follows:

The time domain first step (S21): the foot ground acting force when obtaining people's normal gait by the gait passage by the three-dimensional force force plate, with once without stop without return normal consistently from the gait passage walking by the time foot ground acting force as an effective gait data, transmit and be saved in the computing machine (PC);

Time domain second step (S22): adopt the wavelet transformation hard threshold method that foot ground acting force is carried out denoising, obtain the foot ground acting force after the denoising;

The 3rd step (S23) of time domain: adopt front and back shearing force curve and the wave crest point on the vertical support force curve and trough point in the foot ground acting force after the first order difference algorithm detects respectively denoising, with wave crest point and the trough point key point as foot ground acting force curve, with the trough point on the vertical support force curve as sufficient reference point of acting force curve;

The 4th step (S24) of time domain: the complexity according to the application places gait data acquisition judges whether the needs exptended sample, if in the place that personnel fix, can multi collect everyone respectively with the different speeds of travel with wear the normally gait data by the gait passage of different shoes, then do not need exptended sample, carry out the 7th step (S27) of time domain; If on the airport, the strong places of mobility of people such as market and customs, can not multi collect everyone respectively with the different speeds of travel with wear different shoes normally by the gait data of gait passage, then need exptended sample, carry out the 5th step (S25) of time domain;

Time domain the 5th step (S25): when needing exptended sample, take the reference point of foot ground acting force curve as benchmark, the foot ground acting force after the denoising is carried out sample split, split into a plurality of segmentations foots ground acting force, the quantity that expands the gait data sample;

Time domain the 6th step (S26): with the power value at the key point place on the vertical support force curve in each segmentation foot ground acting force and power value occur the time on the acting force rate of change of phase, adjacent key point and momentum and the corresponding front and back shearing force curve the key point place the power value, drive the gait temporal signatures that momentum and braking momentum characterize segmentation;

Time domain the 7th step (S27): when not needing exptended sample, with the power value at the key point place on the vertical support force curve in the foot ground acting force after the denoising and power value occur the time on the acting force rate of change of phase, adjacent key point and momentum and the corresponding front and back shearing force curve the key point place the power value, drive momentum and the braking momentum characterizes omnidistance gait temporal signatures;

The step of described waveform alignment is as follows:

The waveform alignment first step: the dimension of the foot ground force data after adopting linear interpolation algorithm with denoising normalizes to same value G * Np, Np is the sum of the three-dimensional force force plate of use, G is the step pitch parameter relevant with sample frequency, G is set as 50 integral multiple;

Waveform alignment second step: go out n trough point on the vertical support force curve in the foot ground acting force after the dimension normalization by the first order difference algorithm search, n trough point is as a reference point;

Waveform alignd for the 3rd step: the reference point on the vertical support force curve is as reference, by linear interpolation left and right sides shearing force curve, front and back shearing force curve and vertical support force curve in the foot ground acting force are alignd, so that the trough point of the n on the vertical support force curve snaps to respectively the position of appointment, n trough point snaps to G * (n-0.5);

The step that described sample splits is as follows:

Sample splits the first step: go out n trough point on the vertical support force curve in the foot ground acting force by the first order difference algorithm search, and n trough point is as a reference point;

Sample splits second step: the reference point on the vertical support force curve is as cut-point, respectively the part between first reference point and last reference point in left and right sides shearing force curve, front and back shearing force curve and the vertical support force curve in the foot ground acting force is split into the n-1 section, foot ground force data between the two adjacent reference point is extracted, as a new sample, split out altogether n-1 sample.

2. the Gait Recognition system of described gait feature abstracting method based on foot ground acting force according to claim 1 is characterized in that, comprises that gait data acquisition module (MD1) and gait data process and identification module (MD2):

Described gait data acquisition module (MD1), formed by Np three-dimensional force force plate (MD11) and 1 data collection and transmission unit (MD12), Np three-dimensional force force plate (MD11) is electrically connected, they are closely concordant and be installed in snugly formation gait passage (GW) ground (Flo) under, data acquisition is electrically connected with Np three-dimensional force force plate (MD11) that is electrically connected by Ethernet with an end of transmission unit (MD12), data acquisition is electrically connected with computing machine (PC) by usb data line (DL) with the other end of transmission unit (MD12), realizes Real-time Collection and the processing of foot ground force data;

Described gait data is processed and identification module (MD2), be mounted in the software module in the computing machine (PC), this module is at first carried out denoising by the wavelet transformation hard threshold method to the foot ground acting force that collects, then characterize gait by gait feature abstracting method extraction gait frequency domain character and gait temporal signatures, merge by gait feature again and set up the gait feature collection, adopt at last support vector machine classifier that gait is identified.

3. by the Gait Recognition system of the described gait feature abstracting method based on foot ground acting force of claim 2, it is characterized in that described gait data processing and identification module (MD2), be that the gait data that obtains is processed with feature extraction and realized the module of remote hidden Gait Recognition, comprise:

Gait data denoising unit (MD21), the noises such as the industrial frequency noise in the foot ground acting force that the removal of employing wavelet transformation hard threshold method collects and electromagnetic interference (EMI) obtain the high gait data of signal to noise ratio (S/N ratio);

Waveform alignment unit (MD22), the step that adopts the alignment of described waveform will foot ground acting force automatic aligning according to the reference point on the vertical support force curve in the foot ground acting force with the foot ground acting force of all personnel's sample of collecting;

The foot ground acting force that sample split cells (MD23), the step that adopts described sample to split will collect splits into a plurality of segmentation foots ground acting force according to the reference point on the vertical support force curve in the foot ground acting force;

Gait frequency domain character extraction unit (MD24), the step that adopts described gait frequency domain character to extract is alignd from process denoising and waveform, or further characterizes the gait frequency domain character through extraction WAVELET PACKET DECOMPOSITION coefficient in the foot ground acting force of sample deconsolidation process again;

Gait temporal signatures extraction unit (MD25), adopt step that described gait temporal signatures extracts from through denoising even the power value of calculating key point through the foot ground acting force curve of sample deconsolidation process with the time mutually, rate of change and the momentum of adjacent key point characterize the gait temporal signatures;

Gait feature integrated unit (MD26) adopts FCM Algorithms to pick out minimum optimum gait frequency domain character subset from the gait frequency domain character that extracts, and directly makes up the gait feature collection after obtaining merging with the gait temporal signatures again;

Gait Recognition unit (MD27) adopts support vector machine classifier that the gait feature that extracts from foot ground acting force is carried out gait and identification.

Images