CN105279411A - Gait bio-feature based mobile device identity recognition method - Google Patents

Abstract

The invention provides a gait bio-feature based mobile device identity recognition method. The method is combined with a mobile terminal and is perfection and expansion of an existing bio-identification technology. The method consists of two parts of a training module and an identification module. The training module comprises three steps of data acquisition, feature extraction and model establishment; and the identification module comprises to-be-identified user data collection, model matching and notification response. According to the method, by utilizing self functions and convenience of the mobile terminal, data is completely acquired, a scientific mathematic basic model is constructed, and gait information of a user is obtained, so that user uniqueness is ensured. A system adopts a national SMS4 symmetric encryption algorithm to perform encryption, so that the data transmission security is well ensured. The method has the advantages that rich sensor and network functions of the mobile device are fully utilized, a modeling process is combined with daily gaits of the user, and the bio-identification technology can be continuously perfected without the need for a deliberate operation. Compared with other identification modes, the security system is relatively high in concealing property and difficult to target. Meanwhile, no contact is required, the concealing is difficult, and remote identification can be performed, so that the user security guarantee is greatly improved.

Description

A mobile equipment identity authentication method based on gait biological characteristic

Technical field

The invention belongs to biotechnology identification and security fields, mobile terminal, be intended to work out a mobile equipment identity authentication method based on gait biological characteristic, make mobile device become more intelligent and safety.

Background technology

Along with the raising of people's living standard, the consumer that Intelligent mobile equipment becomes people gradually to.According to market survey report, continuation expands by international mobile equipment market scale.Smart machine physique is petite, and it is convenient to carry, and therefore, many lawless persons can carry out pilferage activity to it, obtains illicit income, or uses the pilferage important information obtained in equipment to carry out unlawful activities.Moreover, mobile device is as the personal effects of people, and function is numerous, not only stores the phone of contact person, perhaps also has a series of sensitive information such as mailbox, birthday, even also comprises the privacy informations such as the photo of user, note.When equipment is lost accidentally, the leakage of privacy will become allows people's urgent problem.

Guard technology for smart machine becomes and becomes more and more important.From existing fail-safe software, they protect based on the mode such as password or gesture mostly, but the defect that this mode has it intrinsic, as easily copied, easily revealing, there is no specificity etc., and protection is very passive, so bio-identification becomes the security protection means of main flow.Relative to other living things feature recognition, Gait Recognition has self characteristics and advantages.In recent years, along with the research and development of computer vision technique, utilize computer vision technique to analyze and identify that the gait of people becomes possibility.The requirement of people to social safety is more and more higher simultaneously, and need a kind of method to identify the identity of people quickly and easily, Gait Recognition is paid attention to as a kind of new security means.As behavioural characteristic gait feature have untouchable, be difficult to hide and camouflage, be easy to gather, can the advantage such as perception at a distance.

Summary of the invention

System adopts C/S framework, and client and server adopts the close SM4 of state to be encrypted communication.Client comprises Data Collection, feature extraction, identification response and remote control module.Server comprises gait modeling and control order generation module.

Gait data under data collection module adopts acceleration transducer to collect the different behavior of user, comprises the behaviors such as walking, running, stair activity.

The characteristic extracting module first technology such as level and smooth, the position correction of window service time and filters filter carries out noise reduction to raw data, then extract proper vector according to mathematical formulae to the data after optimization, these features comprise time, discretize distribution etc. between mean value, standard deviation, mean absolute difference, average resultant acceleration, peak value.

Gait modeling utilizes neural network algorithm to carry out modeling to the proper vector that characteristic extracting module produces, and generates behavioral value model and user's detection model respectively, then model is transferred to client.

Identify that first respond module utilizes behavioral value model inspection user current state, then whether invoke user detection model identification active user is legal and respond, if user is illegal, then automatically lock equipment and send unlocking pin to equipment legitimate holder, unlocking pin uses close 128 ciphertexts be encrypted of state.Can decipher in server web page input necessary information after validated user receives ciphertext and obtain expressly.

Control command generation module generates corresponding control command according to subscriber mailbox, password, current time and a random value, comprises jingle bell, vibrations, locates, clears data, sets unlocking pin.When user learns device losses, positioning function determination equipment approximate location can be used, if equipment nearby, jingle bell or vibration function can be used to find equipment; If when determining that equipment cannot be given for change, the function that clears data can be used to remove equipment total data, loss is dropped to minimum.When user forgets Password, the unlocking pin that setting unlocking pin function setting is new can be used.Above-mentioned functions is realized by remote control module opertaing device.

Accompanying drawing explanation

Fig. 1 is data acquisition flow figure

Fig. 2 is feature extraction process flow diagram

Fig. 3 is model Establishing process figure

Fig. 4 is Model Matching process flow diagram

Fig. 5 is push-notification-answer process flow diagram

Embodiment

Below in conjunction with accompanying drawing, the specific embodiment of the present invention is described.

One, data acquisition

Because the gait feature of people under different walking states is not identical, so data acquisition subsystem is envisioned for different walking states stamp different labels, so set up the gait pattern stage in data process subsystem, according to different labels, different models is set up to same person, therefore, the identification rate of native system can improve greatly.

The gait data collected at us is divided into following several types: usual walks, and runs, goes upstairs and go downstairs, other.

Two, data processing

The process of data is included that time window is level and smooth, position correction, filters filter and feature selecting four part, wherein, time window is the basic operation to basic data, reorientation coordinate system is carried out afterwards by position correction, carry out further noise reduction by wave filter again, finally obtain the eigenwert of these data.

(1) time window: for time window, is set to 10 seconds, i.e. the data in time window record 10 second, i.e. T _{a_Window}=10.Adopt windows overlay technology simultaneously, used herein be 50% windows overlay, namely when previous time window Ci record proceeds to 5 seconds, next time window Ci+1 just opens, i.e. T _{new_Window}=5.When smoothing operation time, choose current time window Ci, previous time window Ci-1 and a rear time window Ci+1, get its mean value, H (T _current)=(H (T _current-T _{a_Window})+H (T _current)+H (T _current+ T _{a_Window}))/3, so just can smooth out some noises in data and don't as the too many data of loss.Just can be further processed afterwards.

(2) position correction process: used space geometry knowledge here.Because user puts the direction difference to some extent of mobile phone, the vertical direction of such acceleration transducer cannot be consistent with acceleration of gravity, and the data of sensor will be caused to produce error, so will carry out the calibration operation of position.When tester is static, mobile phone only can be subject to the impact of acceleration of gravity, but in practical situations both, all directions all create acceleration.We can suppose that the direction of g and the vertical direction angle of sensor be the angle of projection on sensor levels face of α, g and human body working direction are β.When measuring, the numerical value of sensor three axles is A, B, C, then real accekeration is:

A′＝Asinβ-Bcosβ+Ccosβ

B′＝Bsinβsinα-Csinβcosα+Ccosβsinα

C′＝Bsinβcosα+Csinβsinα-Ccosβcoaα

Just can be calibrated to the coordinate system of standard after operating above under.

(3) filters filter: utilize the direction of travel reprojection of complementary filter in three dimensions orthogonal to plane here.Afterwards with reference to gait pattern, in the maximum deceleration time period, high frequency noise and other influences factor are removed by a low-pass filter by square place, and the vector obtained the most at last, by median filter, then exports.

Three, feature extraction

Although the data collected can not embody the gait of a people intuitively, this biological characteristic can well be embodied by the eigenwert extracted.The cadence that everyone walks, paces size etc. are not quite similar, and these can be embodied by following characteristics, such as: the time between peak value can embody a people normally walk frequency.

During feature extraction, suppose there are n bar data, three axles are x-axis, y-axis, z-axis; Mean value is avg, and standard deviation is sd, and mean absolute difference is aad, and average resultant acceleration is ara.

Mean value (3): mean value (x, y, the z) avg of each axle _x=(x ₁+ x ₂+ ... + x _n)/n.

Standard deviation (3): the standard deviation (x, y, z) of each axle

${\mathrm{sd}}_x=\sqrt{\frac{1}{n}\left(\right|x_1-{\mathrm{avg}}_x{|}^2+|x_2-{\mathrm{avg}}_x{|}^2+\mathrm{...}+|x_i-{\mathrm{avg}}_x{|}^2)}.$

Mean absolute difference (3): ${\mathrm{aad}}_x=\frac{1}{n}\left({\Σ}_{i=1}^n\right|x_i-{\mathrm{avg}}_x\left|\right).$

Average resultant acceleration (1): $ara=\frac{1}{n}\left({\Σ}_{i=1}^n\sqrt{x_i^2+y_i^2+z_i^2}\right).$

Time (3) between peak value: by data fitting function try to achieve ω.

Discretize distribution (30): take out each number of axle according to inner maximal value and minimum value, the difference of subtracting each other divided by 10 result as interval, calculate the number percent shared by number put in each interval.

Four, Modling model

First, utilize all data collected, use weka Data Mining Tools to set up User Status model.WEKA, as a disclosed data mining workbench, has gathered a large amount of machine learning algorithm bearing data mining task, has comprised and carry out pre-service to data, classification, recurrence, cluster, correlation rule and visual on new interactive interface.Used neural network respectively in our experimentation, Naive Bayes Classification, J48 algorithm and our improve after neural network algorithm carry out Modling model, the result of experiment shows that the neural network effect after improving is optimum.

Five, user data to be identified is collected

The data acquisition of this step and training stage is similar, unlike, the data collected by this step do not have label and name, and these data only have x, and the data of y, z tri-axles, for next step.

Six, Model Matching

After getting the data of previous step, first carry out state coupling, namely mate with the User Status model of previous step, coupling obtains result.The process of coupling is similar to the process of classification, and after model is set up, by the data of real-time collection user, the process of feature extraction is consistent with above-mentioned feature extraction, finally using the feature that extracts as the input of model, obtain recognition result.Result is walked, a state in running, stair activity and other four kinds of states.Then, its data mated with user's gait pattern, whether the user that can obtain producing these data is validated user, and analysis result is passed to next stage.

Seven, push-notification-answer

From previous step, get the result whether user is legal, be divided into two kinds of response mechanisms.

If 1. user is legal, do not respond;

If 2. user is illegal, client-side lock equipment, stochastic generation 6 identifying codes, and generate encryption key according to the password that user uses, be encrypted with this double secret key identifying code, send the identifying code after encryption by mail and note two kinds of modes.The password used in the input of server web page after validated user receives ciphertext can be deciphered and be verified code.

Claims (4)

1., based on a mobile equipment identity authentication method for gait biological characteristic, main employing is made up of client and service end, contains three subsystems:

(1) data acquisition subsystem: the gait data being gathered user by the sensor on equipment.

(2) data process subsystem: carrying out noise reduction process by collecting the data obtained, extracting eigenwert afterwards, then comparing process by sorting algorithm, obtaining net result.

(3) push-notification-answer subsystem: user is fed back to obtained result notice.

2. the mobile equipment identity authentication method based on gait biological characteristic according to claim 1, it is characterized in that, Intelligent mobile equipment application as client service data acquisition subsystem, a station server as service end, service data processing subsystem and push-notification-answer subsystem.Client, by network and server communication, makes three subsystems mutually can exchange data.

3. the mobile equipment identity authentication method based on gait biological characteristic according to claim 1, is characterized in that, collects gait data by the application on mobile device, is sent to the server of specifying after calculating eigenwert by internet.If these data are for training, so from data, reading user name and state tag, adding database and being used for training pattern; If these data are for identifying, so read from database and generate user behavior characteristic model, identifying user behavior, after end of identification, from database, read the data of respective behavior again and the user identity model under generating respective behavior, identify user identity.If user identity is legal, so client application normally works; If user is illegal, notice client application locking mobile device, and notify validated user at once.

4. the mobile equipment identity authentication method based on gait biological characteristic according to claim 1, is characterized in that, the operation of biometric identity Verification System is divided into two stages: training stage and cognitive phase.In each stage, subsystem runs in a different manner.In the training stage, user is according to own situation, and first initialization push-notification-answer subsystem sets necessary parameter, and then reminder-data acquisition subsystem gathers dissimilar data; After collecting a given data, notification data processing subsystem goes out acceleration transducer x from the extracting data recorded, y, the data of z tri-axles, carry out the pre-service of data, namely series of computation is carried out to data, subsequently the numerical value calculated is used neural network, decision tree scheduling algorithm generates the gait pattern of user, for next stage.At cognitive phase, the data of data acquisition subsystem Real-time Collection acceleration transducer also allow data process subsystem be processed by the gait pattern generated previous stage, identify whether the user producing these data is validated user, if user is illegal, then trigger notice response subsystem.