CN108596100A - A kind of personal identification method based under multi-pose - Google Patents
A kind of personal identification method based under multi-pose Download PDFInfo
- Publication number
- CN108596100A CN108596100A CN201810379291.8A CN201810379291A CN108596100A CN 108596100 A CN108596100 A CN 108596100A CN 201810379291 A CN201810379291 A CN 201810379291A CN 108596100 A CN108596100 A CN 108596100A
- Authority
- CN
- China
- Prior art keywords
- acceleration
- dynamic
- walking
- personnel
- gait cycle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
Abstract
The invention discloses a kind of personal identification methods based under multi-pose.Using mobile phone collector's walking brief acceleration data, in order to further excavate acceleration information, fine-characterization vector, probe into rate of acceleration change, it proposes a kind of information dividing method, according to the size that acceleration information changes, acceleration information is divided into dynamic and two parts of stable state.Respectively from above-mentioned two region extraction feature, dynamic and static two characteristic sets are formed, selected characteristic and new characteristic set is formed in conjunction with peak point line slope from two set with ReliefF algorithms, improves personal identification rate.The method for finally using classifiers combination carries out personal identification.Personal identification is carried out using mobile phone, improves the real-time of identification, it is of low cost.
Description
Technical field
The invention belongs to information security fields, and in particular to a kind of personal identification method based under multi-pose.
Background technology
As the performance of electronic equipment improves the development with social informatization, such as the portable electric of mobile phone, palm PC
The purposes of sub- equipment is increasingly extensive, develops to being integrated with the directions such as electronic payment function, identity authentication function, therefore its
Safety problem is also increasingly paid attention to.Many identification systems, such as fingerprint, password authentification, face have been put into and answer
With, it can be in the identity of start-up phase certification user, but be not suitable for carrying out continuous authentication in use.
In addition, these modes are required for cooperating on one's own initiative for user, excessive verification can reduce the user satisfaction of user.Gait feature
As one kind of biological characteristic, there is uniqueness and stability, be not easy to forge.Gait feature body based on acceleration transducer
Part certification is emerging identity identifying method need not use as the first step of protection portable electronic product information security
Family is cooperated on one's own initiative, and certification can be completed under natural ambulatory status.The Method of Gait Feature Extraction mode of mainstream has based on video, base at present
In foot bottom pressure sensor and it is based on 3 kinds of accelerometer, wherein using accelerometer extraction gait feature, can not dropped
Authentication is realized in the case of low users' satisfaction degree and improves the safety of existing personal portable electronic equipment.
In view of the above-mentioned problems, proposing a kind of personal identification method based under multi-pose.This method is using built in mobile phone
Acceleration transducer records under personnel's difference posture walking data (at a quick pace, slowly, upstairs, downstairs), in order to be more accurate
Information feature when walking is obtained, further investigation rate of acceleration change is divided to accelerating information to do, deeper excavation personnel
Information when walking.And propose the method for your line of extraction peak strip as feature vector introduced feature set.It finally combines and melts
Close the discrimination that people's grader improves personnel.
Invention content
The problem of for personnel are identified currently with human body behavioural characteristic, the present invention utilize the acceleration built in mobile phone
Sensing solves these problems.1) it probes into rate of acceleration change and carries out gait information segmentation, further excavate letter when personnel's walking
Breath, improves the service efficiency of information.2) method for proposing your line of extraction peak strip increases as feature vector introduced feature set
Add discrimination.3) combining classification device blending algorithm increases the robustness of system, nicety of grading, classification effectiveness.
Realize that technical scheme is as follows：
A kind of personal identification method based under multi-pose includes the acquisition segmentation of personnel's walking acceleration information, feature
Extraction model is established and classifiers combination mode.Personal identification step；
The personal identification step of walking acceleration information segmentation includes：
(1) fixation of mobile phone location：Mobile phone fixed position is in trouser pocket position；
(2) data acquisition：Collecting personnel's walking acceleration information using the three axis accelerometer built in mobile phone, it collects number
According to software platform be Sensor Log, Main Analysis mobile phone acceleration sensor Z axis measure acceleration information；
(3) data processing：Removal high frequency is filtered to original acceleration information using 3 rank Butterworth LPFs
Noise；
(4) data are divided：By treated, personnel's walking accelerates information, according to the change rate of acceleration, by speed data
It is divided into dynamic and two parts of stable state；
(5) feature extraction：To whole segment data be subjected to cutting first, and use the acceleration after sliding window mode dividing processing
Degrees of data extracts feature as unit of window.Feature is selected using ReliefF algorithms, outstanding feature is selected to carry out
The model foundation in later stage；
(6) identification model is established：The feature chosen using step (5), is identified the foundation of model.Model foundation uses
The mode of pattern-recognition, using four kinds of classical graders：Bayes (Bayes), multilayer neural network (MLP), random forest
(RF) and K is neighbouring (K-NN), the Method Modeling intersected using ten foldings.Assembled classification is used to further increase personal identification rate
The method of device is combined the single grader in front with the weighted voting algorithm based on recall rate.
Further, the step (4) is as follows：
1) researcher's walking accelerates change rate, according to personnel's walking acceleration information and its change rate, divides dynamic
With stable state section；
2) rate of acceleration change x*={ x are calculated1,x2,x3,….,xm, using rate of acceleration change, can be easier from
The dynamic part of the walking detached in gait cycle.Higher change rate shows the dynamic part of gait cycle；
3) after obtaining rate of acceleration change, threshold value is calculated with the dynamic area to each gait cycle according to rate of acceleration change
Domain is substantially divided with stationary zones；
4) after calculating threshold value, the final step of Method of Gait Feature Extraction is accurately to divide dynamic and steady-state portion.It needs
The segmentation of dynamic and stable state is made to single gait cycle in conjunction with threshold value and limited length.
Further, the step (5) is as follows：
1) when initial, the acceleration degree series after step (4) is divided are obtained, sliding window is set according to the sample frequency of sensor
The size of mouth.In order to reduce the loss of window edge information, windows overlay rate is 50%.
2) extraction feature is vectorial in terms of frequency domain and time domain two respectively as unit of window, and feature vector is respectively standard
Poor (St), average value (Aver), the degree of bias (Sk), root mean square (Rms), energy (En).It is divided into according to above-mentioned gait information dynamic
State and two kinds of regions of stable state, so feature has been also divided into steady state characteristic (S) and behavioral characteristics (D).
3) according to ReliefF algorithms to two kinds of characteristic set D of dynamic and stable state, S carries out feature selecting, unified to feature set
It closes D, S and carries out feature selecting, choose preferable feature and form new feature set NF.
Beneficial effects of the present invention：
(1) audient is extensive
Without other equipment, meets the needs of user mobile phone safeguard protection.
(2) method of use information segmentation further excavates personnel's walking information.
(3) it combines assembled classifier to increase the robustness of system, improves personal identification rate.
(4) it promotes the use of real-time
The mode for being directed to mobile phone safe on the market has very much, and the present invention is passed using the mobile phone movement built in current smart mobile phone
Sensor acquires the walking information of user in real time, and initiative recognition mobile phone user is highly practical, of low cost.
Description of the drawings
Fig. 1 mobile phones placement location acquires direction with acceleration transducer data；
Fig. 2 initial thresholds are set；
Fig. 3 segmentation effects；
Fig. 4 system the general frames.
Specific implementation mode
First, a kind of division methods are proposed according to personnel's walking acceleration change feature, personnel's walking acceleration is drawn
It is divided into relative dynamic and two parts of stable state, respectively from features such as two extracted region standard deviations, mean value, energy.Then, root
These features and peak point line slope are selected to be combined into new characteristic set according to asynchronous scanning frequency rate.Finally, using combination point
The method of class device obtains accuracy of identification.
Concrete scheme：
It is respectively power-walking that user, which carries mobile phone and carries out four kinds of daily behavior activities, slowly walking, upstairs and under
Building.Smart mobile phone is fixed at the right lateral thigh of experimenter, and specific location is as shown in Figure 1, user needs in data acquisition
To be accustomed to walking according to the walking of oneself and obtain stable data for a period of time, the software platform for collecting data is
Sensor Log, user need to record 30 groups of walking data to ensure the accuracy of data.
The invention will be further described below in conjunction with the accompanying drawings.
One, based on and percentage speed variation gait information segmentation：
After the data being collected into first, calculated change rate is needed, as shown in equation (1).
ΔAcct1With Δ Acct2It is two continuous acceleration information samples, Sampling rate are continuous two samplings
Sampling time between point.
Using rate of acceleration change, it can be easier the dynamic part of the walking detached from gait cycle, as shown in Figure 1
Show acceleration change.Higher change rate shows the dynamic part of gait cycle.
After obtaining rate of acceleration change, since it is desired that calculating threshold value with the starting point of the dynamic area of each gait cycle of determination
And end point.Acceleration samples value based on signal period calculates threshold value (2) using following formula.In fig. 2, black is empty
Line shows the threshold value from each axis.In equation (2).
Each gait cycle of same people has fine distinction, for the accurate dynamic for dividing gait and steady-state zone, needs
According to gait cycle, α, calculation formula (3) are set
F in formulai, fi' indicate single gait cycle x respectively, y, in z-axis, peak-peak and second largest peak value.
After calculating threshold value, the final step of Method of Gait Feature Extraction is accurately to divide dynamic and steady-state portion.It needs to tie
Threshold value is closed with limited length to make the segmentation of dynamic and stable state to single gait cycle.When there are one prodigious for dynamic part
When fluctuation is higher or lower than threshold value, it is impossible to threshold value extraction dynamic part be used alone.Accurately to find dynamic part, move
The length of polymorphic segment is confirmed as the half of single gait cycle data sample, such as equation (4).First point higher than threshold value is
The starting point of dynamic acceleration data.From this starting point, last point is calculated by accounting equation (4).When differentiation walking
Steady-state portion and dynamic part.It can be seen that in figure 3 using the result of this equation.The part of high-order horizontal line covering represents
The steady-state portion that dynamic and low level horizontal line represent.
Two, peak point line slope
A kind of new feature vector is proposed according to the walking feature of personnel.The acceleration information distributional difference of different personnel compared with
Greatly, everyone is because of height, and it is different that the difference of athletic posture results in data distribution range.In a gait cycle, walking production
Raw acceleration constantly changes at any time.The acceleration information peak value of signal period is arranged again according to sequence from small to large
After row, the line slope of peak point is calculated.The slope can embody the severe degree of movement, can be used as and distinguish different personnel's
Feature.
Mobile phone coordinate system Z axis is perpendicular to mobile phone screen, by 3-axis acceleration sensor Z axis, n-th window signal period
Peak value according to from small to large be ranked sequentially y={ y1,y2,…yn, i, j indicate peak value yi, yjCorresponding sampling time, yiWith
yjThe respectively peak-peak and second largest peak value of gait cycle, yiWith yjLine slope is defined as：
Two, feature extraction and selection：
The severe degree that standard deviation (St) moves personnel is very sensitive and has reacted the discrete case of acceleration signal
So feature vector that can be important as one.Average value (Aver) can reflect the acceleration information of a gait cycle
Concentrated position, therefore use the average value of each window as a feature vector.The degree of bias (Sk) is for judging acceleration
The index of the distribution regularity of data sequence.Root mean square (RMS), which has proved to be, can preferably distinguish the similar movable feature of gait
Amount.In addition to the extraction of the above temporal signatures, the energy (En) that paper extracts each window is used as frequency domain character, frequency domain energy to exist
Generally reflect the intensity of personnel's acceleration signal.Shown in the expression formula of energy such as formula (6)：
Wherein M indicates sample number, xiFor i-th of FFT Fourier coefficient.
Characteristic set D, S are obtained by characteristic extraction step above.D={ D_Aver, D_St, D_En, D_RMS, D_Sk }, S
={ S_Aver, S_St, S_En, S_RMS, S_Sk } adds peak point line slope conduct in addition to features above in experiment
Feature vector.
Feature selecting is carried out to dynamic and stable state two kinds of characteristic sets D, S based on reliefF algorithms.Selection mode is unified
Feature selecting is carried out to characteristic set D, S, choose preferable feature formed new feature set NF=S_NF, F_NF, U_NF,
D_NF1 }, it is as described above the characteristic set S_NF={ D_En, D_St, S_RMS, S_En, S_St } under different rates； F_NF
={ S_RMS, D_RMS, S_St, D_St, S_Aver }；U_NF={ S_st, S_RMS, D_st, S_Aver, D_RMS }；D_NF=
{ S_Aver, S_st, S_RMS, D_En, D_aver }.
Three, the weight votes method assembled classifier based on recall rate
It is set with classification g={ g1,…,gp}.Test set T={ (x, y)t, t=1 ..., N }, x is feature vector, y ∈ g.
There are base grader G={ G1,…Gj, base grader tag along sort collection is combined into h={ s1,…,sjWherein sj=Gj(x).Assuming that real
Border class label gp, grader Guess gkRecall rate expression formula such as (2), { T in formulap∈ T, i.e. class label is g in TpCollection
It closes.I () is confusion matrix.
Weight votes expression formula based on recall rate such as (3), is divided into two parts, { p according to the classification results of base grader
+ be base grader classification results it is gp, { p- } is that the conjecture of base grader belongs to other classes.Combining form table based on recall rate
Up to formula such as (4).
The series of detailed descriptions listed above only for the present invention feasible embodiment specifically
Bright, they are all without departing from equivalent implementations made by technical spirit of the present invention not to limit the scope of the invention
Or change should be included in protection scope of the present invention it.
Claims (6)
1. a kind of personal identification method based under multi-pose, including the data acquisition of personnel's walking acceleration information, segmentation, spy
Sign extraction, model foundation and Classification and Identification and etc., it specifically includes：
(1) data acquisition：Personnel are collected using the three axis accelerometer built in mobile phone in mobile phone fixed position in trouser pocket position
Walking acceleration information, it is respectively slowly walking, power-walking, upstairs, downstairs that personnel, which distinguish athletic posture, analyzes mobile phone acceleration
Spend the acceleration information that sensor Z axis measures；
(2) data processing：Removal high frequency is filtered using 3 rank Butterworth LPFs to original acceleration information to make an uproar
Sound；
(3) data are divided：By treated, personnel's walking accelerates information to be divided speed data according to the change rate of acceleration
At two parts of dynamic and stable state；
(4) feature extraction：To whole segment data be subjected to cutting first, and use the acceleration number of degrees after sliding window mode dividing processing
According to extracting feature as unit of window；
(5) propose peak point line slope as a kind of new feature vector according to the walking feature of personnel；
(6) the weight votes method assembled classifier based on recall rate, is identified personnel.
2. further, step described in claim 1 (4) data segmentation is as follows：
1) based on and percentage speed variation gait information segmentation：
After the data being collected into first, calculated change rate is needed, as shown in equation (1),
ΔAcct1With Δ Acct2Two continuous acceleration information samples, Sampling rate be continuous two sampled points it
Between sampling time,
Using rate of acceleration change, the dynamic part of the walking detached from gait cycle, higher change rate can be easier
Show the dynamic part of gait cycle,
After obtaining rate of acceleration change, since it is desired that calculating threshold value with the starting point and knot of the dynamic area of each gait cycle of determination
Spot, the acceleration samples value based on signal period calculate threshold value using following formula,
Each gait cycle of same people has fine distinction, for the accurate dynamic for dividing gait and steady-state zone, needs basis
α is arranged in gait cycle, and calculation formula is
F in formulai, fi' indicate single gait cycle x respectively, y, in z-axis, peak-peak and second largest peak value,
After calculating threshold value, the final step of Method of Gait Feature Extraction is accurately to divide dynamic and steady-state portion, needs to combine threshold
Value makes single gait cycle with limited length the segmentation of dynamic and stable state, when there are one prodigious fluctuation height for dynamic part
When threshold value, it is impossible to threshold value extraction dynamic part, accurately to find dynamic part, dynamic part be used alone
Length is confirmed as the half of single gait cycle data sample, and such as equation (4), first point higher than threshold value is dynamic acceleration
The starting point of data calculates last point by accounting equation (4) from this starting point, distinguish steady-state portion when walking with
Dynamic part.
3. further, step described in claim 1 (5) feature extraction is as follows：
The acceleration information distributional difference of different personnel is larger, everyone results in data point because of height, the difference of athletic posture
Cloth range is different, and in a gait cycle, the acceleration that walking generates constantly changes at any time, by the acceleration of signal period
After data peaks are rearranged according to sequence from small to large, the line slope of peak point is calculated, which can embody movement
Severe degree, can be used as the feature for distinguishing different personnel；
Mobile phone coordinate system Z axis is perpendicular to mobile phone screen, by 3-axis acceleration sensor Z axis, the peak value of n-th of window signal period
According to from small to large be ranked sequentially y={ y1,y2,…yn, i, j indicate peak value yi, yjCorresponding sampling time, yiWith yjRespectively
For the peak-peak and second largest peak value of gait cycle, yiWith yjLine slope is defined as：
4. further, step described in claim 1 (6) is as follows：
Weight votes expression formula such as following formula based on recall rate is divided into two parts, { p+ } according to the classification results of base grader
It is g for base grader classification resultsp, { p- } is that the conjecture of base grader belongs to other classes.
5. a kind of personal identification method based under multi-pose according to claim 1, which is characterized in that the step (1)
The middle acceleration sensor using built in mobile phone acquires acceleration information under the different posture of personnel.
6. a kind of personal identification method based under multi-pose according to claim 1, which is characterized in that the step (4)
Personnel's walking acceleration information is divided into two parts by the middle change rate according to acceleration, is as follows：
1) rate of acceleration change when calculated personnel's walking；
2) threshold value T is calculated with the starting point and end point of the dynamic area of each gait cycle of determination；
3) each gait cycle of same people has fine distinction, for the accurate dynamic for dividing gait and steady-state zone, needs root
According to gait cycle, α is set；
4) final step of Method of Gait Feature Extraction is accurately to divide dynamic and steady-state portion, needs to combine threshold value and limited length
To make the segmentation of dynamic and stable state to single gait cycle.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810379291.8A CN108596100A (en) | 2018-04-25 | 2018-04-25 | A kind of personal identification method based under multi-pose |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810379291.8A CN108596100A (en) | 2018-04-25 | 2018-04-25 | A kind of personal identification method based under multi-pose |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108596100A true CN108596100A (en) | 2018-09-28 |
Family
ID=63609217
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810379291.8A Pending CN108596100A (en) | 2018-04-25 | 2018-04-25 | A kind of personal identification method based under multi-pose |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108596100A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080175443A1 (en) * | 2007-01-23 | 2008-07-24 | Fullpower, Inc. | System control via characteristic gait signature |
| CN101695445A (en) * | 2009-10-29 | 2010-04-21 | 浙江大学 | Acceleration transducer-based gait identification method |
| CN107831907A (en) * | 2017-12-07 | 2018-03-23 | 深圳先进技术研究院 | Identity identifying method and device based on Gait Recognition |
-
2018
- 2018-04-25 CN CN201810379291.8A patent/CN108596100A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080175443A1 (en) * | 2007-01-23 | 2008-07-24 | Fullpower, Inc. | System control via characteristic gait signature |
| CN101695445A (en) * | 2009-10-29 | 2010-04-21 | 浙江大学 | Acceleration transducer-based gait identification method |
| CN107831907A (en) * | 2017-12-07 | 2018-03-23 | 深圳先进技术研究院 | Identity identifying method and device based on Gait Recognition |
Non-Patent Citations (3)
| Title |
|---|
| IK-HYUN YOUN等: "New gait metrics for biometric authentication using a 3-axis acceleration", 《IEEE CONSUMER COMMUNICATIONS AND NETWORKING CONFERENCE (CCNC 2014): PEOPLE CENTRIC SENSING AND COMMUNICATIONS (PCSC)》 * |
| LUDMILA I. KUNCHEVA等: "A weighted voting framework for classifiers ensembles", 《KNOWLEDGE AND INFORMATION SYSTEMS》 * |
| 郇战等: "基于步行加速度信息分割的人员识别", 《计算机工程与应用》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Nickel et al. | Authentication of smartphone users based on the way they walk using k-NN algorithm | |
| Jain et al. | Human activity classification in smartphones using accelerometer and gyroscope sensors | |
| Derawi et al. | Gait and activity recognition using commercial phones | |
| Kwon et al. | Unsupervised learning for human activity recognition using smartphone sensors | |
| Kholmatov et al. | SUSIG: an on-line signature database, associated protocols and benchmark results | |
| Ailisto et al. | Identifying people from gait pattern with accelerometers | |
| Derawi et al. | Improved cycle detection for accelerometer based gait authentication | |
| CN104809377B (en) | Network user identity monitoring method based on webpage input behavior feature | |
| CN103530540B (en) | User identity attribute detection method based on man-machine interaction behavior characteristics | |
| Nickel et al. | Classification of acceleration data for biometric gait recognition on mobile devices | |
| CN107831907A (en) | Identity identifying method and device based on Gait Recognition | |
| Muaaz et al. | Influence of different walking speeds and surfaces on accelerometer-based biometric gait recognition | |
| CN108629167A (en) | A kind of more smart machine identity identifying methods of combination wearable device | |
| CN104269025B (en) | Wearable single node feature and the position choosing method of monitoring is fallen down towards open air | |
| Ferrero et al. | On gait recognition with smartphone accelerometer | |
| Wolff | Behavioral biometric identification on mobile devices | |
| Nandy et al. | Novel features for intensive human activity recognition based on wearable and smartphone sensors | |
| CN106910314A (en) | A kind of personalized fall detection method based on the bodily form | |
| Guo et al. | A two-layer and multi-strategy framework for human activity recognition using smartphone | |
| Derawi | Smartphones and biometrics: Gait and activity recognition | |
| Irfan et al. | Anomaly detection in crowds using multi sensory information | |
| CN106971203A (en) | Personal identification method based on characteristic on foot | |
| Permatasari et al. | The MMUISD gait database and performance evaluation compared to public inertial sensor gait databases | |
| CN108596100A (en) | A kind of personal identification method based under multi-pose | |
| CN108596253A (en) | A kind of personal identification method based on the segmentation of walking acceleration information |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180928 |
|
| WD01 | Invention patent application deemed withdrawn after publication |