Dominant shape position cluster and sub- morpheme cluster is proposed, specifically in embodiment:Described dominant shape position clusters:According to specific
Criterion, by all of dynamic respiration parameter(Vector)Several the non-cross set being formed, this set is complete to be covered
All dynamic respiration parameter vectors, and the respiratory model based on broad sense, provide the description for periodic breathing process and identification.
Note:One group of dynamic respiration parameter for several the continuous breathing cycles belonging to someone(Vector)Set,
The first time cluster analyses being carried out.According to specific criterion, all dynamic respiration parameters are divided into a broad sense(Or
Say " normalized ")Multiple particular states in breathing cycle(In other words " stage "), wherein each state(Distribution)All contain
Multiple dynamic respiration parameters(Vector).In addition, belonging to different conditions(Stage)Vector be non-cross.It is right that dominant shape position provides
Description and identification in the cyclically-varying process of breathing.This description and identification are general for different people, that is, each
This model is all followed in the breathing of people.
Described sub- morpheme clusters and is:For by all main breathing morpheme obtained by first time cluster analyses, carried out
Second cluster analysis.The all dynamic respiration parameter that each dominant shape position is comprised(Vector), according to specific criterion, will
They are divided into several non-cross classifications, these classifications completely cover all dynamic parameters in this breathing morpheme to
Amount, and the description of numerical statistic feature and identification for this breathing morpheme are provided.
Note:On the basis of the multiple dominant shape positions obtained by first time cluster analyses, second cluster being carried out is divided
Analysis.The all dynamic respiration parameter that each dominant shape position is comprised(Vector), based on the numerical value of each dynamic parameter vector, press
According to specific criterion, they are divided into non-cross multiple classifications, each classification comprises multiple dynamic respiration parameters(To
Amount), these vectors have same or analogous numerical value.This will be completely covered according to all categories obtained by a breathing morpheme
All dynamic parameter vectors in breathing morpheme, and the vector belonging to a different category is non-cross.Sub- morpheme provide for
The fine description of statistical nature of each breathing morpheme and identification.For different people, even the main breathing of identical
Morpheme, the numeric distribution of its sub- breathing morpheme is also different.For same person, its main breathing morpheme and son
The numerical distribution characteristic of breathing morpheme has long stability again, can be used to carry out personal identification.
The present embodiment is related to thermal infrared image characteristics based on the nose in the respiratory method to identify personnel.Real
The ultimate principle applying the identification in example is:In multiple breathing cycles of a people, the temperature of its nose in time domain and sky
Between domain characteristic distributions, comprise physiology and the behavior characteristicss of this personnel.Static temperature distribution according to ad-hoc location on nose
With dynamic temperature change, respiratory characteristic parameter specific to this personnel can be extracted, and then identify the identity of corresponding personnel.On
The respiratory characteristic parameter stated comprises the content of two aspects:In a stationary situation, the temperature of the diverse location of the nose of personnel
Degree has differences, and this Temperature Distribution of different personnel has its own feature, and the extraction of this feature is relatively easy, Ke Yitong
Cross infrared thermography to directly obtain;One people's dynamic change characterization that nostril position is showed in breathing, this needs
Obtained by investigating the statistical nature of breathing morpheme in multiple breathing cycles for this personnel and son breathing morpheme.To breathing shape
Position, the detailed description of sub- breathing morpheme, will be carried out below.
The physiological Foundations of the different personnel of identification in the present embodiment, be different people respiratory activity can show different
Characteristic.Explanation to the breathing of people, contributes to understanding that this is true.Human body passes through lasting respiratory activity, sucks extraneous oxygen,
Give off carbon dioxide, to ensure the normal activities of itself.The respiratory system of human body is mainly made up of respiratory tract and lung, human body
Respiratory activity have periodically.In the different breathing cycles, lung is experienced the physiological statuss of multiple fixations successively.Air-breathing
When, drive the skeleton in thoracic cavity and diaphragm to move out by respiratory muscle, thoracic cavity expands, lung expands therewith, and air is inhaled into.Air-breathing
During release, intrapulmonic pressure and atmospheric pressure balance each other.Hereafter, each composition structure in thoracic cavity bounces back because of elasticity, and thoracic cavity reduces,
Intrapulmonary gas flows out in vitro, and this process is exhalation process.After exhalation process terminates, human body will start next breathing process.
Human body is to be completed by related nervous organ to the regulation of lung activity.Such nervous organ includes respiratory center
With intrapulmonary respiratory reflex device.Wherein, respiratory center is located at oblongata, by inspiratory center and expiratory center two parts, inspiratory center
With expiratory center alternating excitation and suppression, thus forming regular respiratory movement.Another adjusts the nerve of the activity of lung
Organ is the respiratory reflex device of intrapulmonary.The respiratory reflex of intrapulmonary is completed by vagus nerve and pulmonary receptor.Intrapulmonary
There is pulmonary stretch receptor, J sensor, the pulmonary receptor of stimulation three kinds of forms of sensor, they are each responsible for the rhythm and pace of moving things of lung
Property relax contracting, and the in particular cases regulation to the activity of lung.
Because the respiratory apparatus of different people and the physiological structure of nervous tissue are different with dynamic response characteristic and different
People there are different breathings customs, the breathing of therefore different people has different characteristics.These characteristics have following several respects
Performance:1., under conditions of identical or essentially identical, the breathing of different people has the characteristic of its own, and observer is permissible
Perceive the difference between them, condition here potentially includes external environment, kinestate, breathing pattern etc.;2. for same
Personally for one, under different conditions, its respiratory activity can change, but because the physiological structure of its own is fixing
, therefore the respiratory activity under its different mode still has certain similarity, and this similarity can be identified and extract
Come;3. the biological property of the uniqueness that the breathing of same person is showed possesses long stability, will not be with year
The growth in age and weaken.According to the above fact, by extracting the respiratory characteristic of the distinctive essence of someone, can effectively carry out
Personal identification.
Above-mentioned respiratory characteristic includes static temperature distribution and dynamic temperature changes two aspects, and this can be by breathing
The thermal infrared images of the nasal area in journey reflects.This phenomenon can be by the physiological structures of nose, and thermal infrared
Physical principle two aspect of imaging is understanding.
The nose of people can be divided into external nose, nasal cavity, three parts of nasal sinuses, and the part relevant with the present invention is external nose and nose
Chamber.External nose therein be nose be evident that part, it be located at mid-face, shape such as substrate is in three side cones of lower section.
External nose has following recognizable part:The nasion, nose, the bridge of the nose, bridge of the nose, the wing of nose, nasolabial fold, nose bottom, nose pillar, left and right before
Nostril.External nose is mainly made up of cartilage frame and external nose muscle, has abundant blood vessel, nerve and lymph around external nose simultaneously.Outward
The shapes and sizes of nose have obvious individual variation, race and regional difference, as an example, be given white people, black race and
The difference of the external nose of yellow.White people's bridge of the nose is high, and the nasion is narrow, and nose is little, in olecranon sample nose;Black race's bridge of the nose is low, nasal root breadth,
Nose is big, as flat upturned nose;Yellow occupy therebetween.The nose of different people has different shapes, physiological structure, and
There is different heat production and heat-transfer character, lead to the nose of different people to show obvious difference in thermal infrared images.In addition,
For same person, when his remains stationary or microinching, shape in thermal infrared images for its nose(Temperature Distribution)Tool
There are the long-time stability of inherence.This vary with each individual, but same person is had to the Characteristics of The Distribution of Temperature of stability, Ke Yizuo
It is characterized and be applied to personal identification.This Characteristics of The Distribution of Temperature is referred to as respiration indices feature, the mark sheet extracting herein
It is now one group of respiration indices characteristic parameter.
Another nose structure relevant with the present embodiment is nasal cavity.In the composition structure of the nasal cavity of people, nasal cavity be divided into for
Left and right two chambeies, the outside of every side nasal cavity(Anterior)Opening is referred to as prenarises.In addition, with nose threshold as boundary, nasal cavity be divided into nasal vestibule and
Cavum nasi proprium.It can be seen that nasal vestibule when prenarises are inwardly seen, and part nasal membrane.Similar with the activity of lung, each
In the individual breathing cycle, nose also periodically can experience multiple fixing physiologically states successively.Further, since nose is human body with outward
First link of portion's exchanging gas, therefore in breathing, the state change of nose is obvious.The air sucking from the external world, with
And the air from pulmonary's exhalation, all can flow through nasal cavity and prenarises, in turn result in this two parts temperature significantly reduce or
Raise.Temperature change in respiratory for the nostril position of different people has the characteristics that different, and this is also different people
A kind of performance of the uniqueness of breathing.
The physical principle of thermal infrared imaging is:Any object in real world(Including human body)All constantly to around
Radiated electromagnetic energy(Spontaneous radiation).At normal temperatures, the spontaneous radiation of object is mainly infra-red radiation, and also known as infrared ray, it is
Light invisible to the human eye, has strong heat effect.Infrared ray and visible ray, ultraviolet, X-ray, gamma-rays and microwave etc.
Radio magnetic wave together, constitutes an infinitely continuous electromagnetic spectrum, infrared imaging system utilize target and environment it
Between due to different hot contrasts produced by the difference of temperature radiation and emissivity, detect infra red energy density and be distributed simultaneously
Shown.In various thermal infrared imaging occasions, it usually needs detection be and object itself radiation and body surface temperature
The relevant information of degree distribution.This detection mode does not need any type of artificial lighting, is, therefore, commonly referred to as imaging and passive imaging.
In terms of state-detection data acquisition, infrared imaging system has the advantage that:1) infrared imaging belongs to non-
Contact measurement, can obtain longer-distance target information, realize large range of monitoring;2) can be by the data detecting essence
Really quantify, generally within ± 2 DEG C, temperature resolution is in 0.01 DEG C of rank for certainty of measurement;3) intelligence degree is high, it is possible to achieve
In the case of unmanned long lasting for operation;4) rely on object itself Radiation work it is not necessary to auxiliary source, will not be right
Human body produces radiation;5)Infrared ray passes through the ability of mist, haze than visible light intensity, therefore thermal infrared imaging possess a certain degree of complete
It when, all weather operations ability;6) there is much solid, light and handy, the thermal infrared imaging equipment that is easy to carry about with one and disposes at present.Mesh
Before, thermal infrared imaging equipment is all widely used in terms of military and civilian.With the maturation of thermal infrared imaging technology, occur
A lot of low costs are applied to civilian thermal infrared imager, and the personnel that these equipment all can apply to proposed in the present invention know
Not.
As it was previously stated, there is individual variation in the breathing of different people.One of this individual variation is presented as, the exhaling of different people
During suction, the change of the state of nose has different qualities.When being observed using thermal infrared imaging equipment, can be in nostril portion
Position sees that the periodicity of some states substitutes.In infrared image, the periodic change procedure of nostril region includes:First stage,
The thoracic cavity expansion of human body, breathing process starts, and outside air enters nasal cavity.Because the temperature of extraneous air is relatively low, therefore this
The temperature making nasal cavity is reduced by individual process.In thermal infrared images, show as nostril position and continue dimmed, the area of dark areas
Increase, this change is continued until that thoracic cavity is expanded to and stops during maximum(Now air-breathing stops);Second stage, chest compression,
Exhalation process starts, and gas flows out human body by the pulmonary of people via nose, because the temperature of these gases is higher, therefore makes nasal cavity
Interior temperature rises, and in thermal infrared images, shows as nostril position and persistently brightens, and the area of dark areas reduces, this change
It is continued until that chest compression arrives to stop during minimum(Now exhale and stop).The above-mentioned nostril region change in respiratory activity
, there is similarity for different people in process.But because the breathing of people has individual variation, therefore above-mentioned change has again
There is the characteristic varying with each individual.
In order to be described in the change of the nostril region in thermal infrared images, the evolution process in nostril can be summarized as some
Plant typicalness, every kind of typicalness is referred to as a kind of breathing morpheme, every kind of breathing morpheme corresponds to a kind of image of nostril region
State(Temperature Distribution).In a breathing cycle, the image of nostril region several different state successively, and every kind of
State both corresponds to a special breathing morpheme.For example, when in thermal infrared images nostril position the darkest(Represent nasal cavity temperature
Minimum, air-breathing just stops)When, correspond to " air-breathing stopping " morpheme.When nostril position is the brightest(Represent nasal cavity temperature highest, exhale
Gas just stops)When, correspond to " exhale and stop " morpheme.In addition to " air-breathing stopping ", " exhale and stop " morpheme, a people exhales
Multiple different breathing morphemes are also comprised in suction.The all of breathing cycle of one people all can be regarded as by similar to above-mentioned
The multiple different breathing morpheme composition of " air-breathing stopping ", " exhale and stop " morpheme.And in the different breathing cycles, exist
It is the different instances of same breathing morpheme.In a breathing cycle, the example of morpheme of not sharing a common fate with fixing order according to
Secondary appearance.The quantity of the breathing morpheme included in the breathing cycle of different people is identical, and the breathing reflecting different people is lived
The similar composition of disorder of internal organs.Meanwhile, the same breathing morpheme of different people has different statistical natures, which reflects difference
The composition varying with each individual in the respiratory activity of people.
Even in addition, the different instances of the same breathing morpheme of same person, being also typically present some between them
Technicality, the probability that the example of different qualities occurs is also different, in order to more subtly investigate the respiratory activity of a people, needs
The example of same breathing morpheme is segmented further, such subdivision is referred to as sub- breathing morpheme.It is similar to breathing morpheme,
The quantity of the son breathing morpheme in the same breathing morpheme of different people is the same, but the same breathing morpheme of different people
In son breathing morpheme characterisitic parameter different.
As noted previously, as there is individual variation in the breathing of people, the dynamic change of nostril region therefore in thermal infrared images
Change and also show the characteristic varying with each individual.In the dynamic variation characteristic that these vary with each individual, exist personally for one relatively
Stable respiratory characteristic, is referred to here as dynamic respiratory characteristic parameter, to be different from aforesaid respiration indices characteristic parameter.This
The stability of dynamic respiratory characteristic shows, they are not subject to the physiological statuss of external condition, main body a bit, and concrete breathing pattern
Impact.Dynamic respiratory characteristic is to show in multiple breathing cycles of personnel, and the breathing shape with this personnel
Position and son breathing morpheme in the statistical nature in multiple breathing cycles has relation.Exhaled multiple by investigating this personnel
Breathing morpheme in the suction cycle and the statistical nature of son breathing morpheme, it is possible to obtain independent of the dynamic respiratory characteristic of breathing cycle
Parameter.
This method is the different personnel that identified based on respiratory characteristic parameter, therefore extract respiratory characteristic parameter be one very
Important process.The process extracting respiratory characteristic parameter comprises the steps of:Obtain the sample breath data of identified person;According to
Sample breath data obtains respiration indices characteristic vector and dynamic respiratory characteristic vector;Generate breathing dominant shape position cluster and son breathing
Morpheme clusters;Generate the respiratory characteristic parameter for identifying this personnel.This process will be described in detail below.
In order to extract the respiratory characteristic parameter of a people it is necessary first to catch the sample breath data of this personnel.Exhale for one
Inhaling data sample is the one group of breath data frame being caught by breathing recording equipment.These breath data frames correspond to the company of a people
Continue multiple breathing cycles, can be used for the respiratory characteristic parameter extracting this personnel.In the present embodiment, original the exhaling of someone
Inhale data to be obtained by some form of thermal infrared imaging equipment, the form of data is the Temperature Distribution at nostril position.Herein
Equipment for obtaining breath data can have various ways, is referred to as breathing recording equipment.The breath data being obtained be
The thermal infrared images of the nasal area of each sampling instant, referred to as breath data frame.Hereinafter will be to breathing recording equipment and breathing
Frame is described in detail.When carrying out specific people's registration it may be necessary to catch multiple sample breath data of this personnel,
And the process extracting respiratory characteristic parameter is implemented to each sample, to reduce the randomness of an independent sample breath data.Separately
Outward, when entering administrative staff's registration it may be necessary to catch the multiple sample breath data being registered when personnel breathe in different modalities,
And extract the common respiratory characteristic parameter under different breathing patterns of this personnel, to reduce during later identification for breathing mould
The dependence of formula.
Hereafter, the sample breath data of this personnel is processed.Data handling procedure may relate to same personnel
One or more sample breath data.Process multiple sample breath data of same personnel simultaneously, will improve for this people
The recognition performance of member, because this will extract multigroup respiratory characteristic parameter, and multigroup respiratory characteristic parameter can be more accurate
Really reflect the respiratory characteristic of this personnel.But, even only one of which sample breath data, the system remains on and can complete
The registration of personnel or identification.In addition, for the process of an independent sample breath data, being that multiple sample breath data are carried out
The basis processing.Therefore, the process in only one of which sample breath data will be introduced here.
For a sample breath data from certain personnel, extract a static state for wherein each breath data frame and exhale
Inhale parameter vector and a dynamic respiration parameter vector.Each obtained respiration indices parameter vector comprises multiple respiration indices
Parameter, each dynamic respiration parameter vector comprises multiple dynamic respiration parameters.In one example, above-mentioned respiration indices parameter
Potentially include:The temperature of the left side wing of nose, the temperature of the right side wing of nose, the temperature of left side nasolabial fold, the temperature of right side nasolabial fold, nose are little
The temperature of post, the temperature of nose, the temperature of bridge of the nose the latter half, air themperature of current environment etc..Above-mentioned dynamic breathing ginseng
Number is possibly including, but not limited to:Temperature minimum in left nose hole(Or peak)Frequency, the minimum temperature in left nose hole, left nose
Temperature minimum in the maximum temperature in hole, right nostril(Or peak)Frequency, the minimum temperature in right nostril, right nostril
High-temperature, the compactness of the maximum low-temperature region in left nose hole(Circularity), right nostril maximum low-temperature region compactness(Circularity)、
Left nose hole low-temperature region(When maximum)The ratio of width and nose pillar width, right nostril low-temperature region(When maximum)Width and nose
The ratio of pillar width, left and right sides nostril reach maximum temperature(Minimum temperature)Time difference etc..In a sample breath data
In, the distribution of all respiration indices parameter vectors and dynamic respiration parameter vector presents certain randomness, can be regarded as
The point cloud with certain probability density characteristicses in higher dimensional space.
Hereafter, based on certain predetermined clustering method, all dynamic respiration parameter vector of this sample breath data is entered
Row cluster analyses, formation multiple breathing morphemes cluster, wherein each breathing morpheme cluster are exhaled corresponding to making a reservation for be identified one
Inhale morpheme.As it was previously stated, one breathes a kind of special state that morpheme represents the thermal infrared images at nostril position, and then instead
Reflect one of breathing cycle moment.By carrying out to all dynamic respiration parameter vector in a sample breath data
Cluster analyses, it is possible to achieve the summary to respiratory characteristic, and therefrom extract the inherent feature of the breathing of this personnel.Cluster completes
Afterwards, all of dynamic respiration parameter vector is divided into several predefined breathing morpheme clusters.These breathing morphemes are mutually not
Intersect, that is, each dynamic respiration parameter vector broadly falls into and is pertaining only to a kind of breathing morpheme cluster.The division of breathing morpheme, provides
The basic description of one dynamic process to breathing.
As it was previously stated, a breathing morpheme may be made up of many height breathing morpheme.Therefore, to belonging to a breath data
All breathing morpheme clusters in sample all proceed as follows:Based on a predetermined clustering method, morpheme is breathed to each
All dynamic respiration parameter vector in cluster carries out cluster analyses again, forms many height breathing morphemes clusters, wherein each
Son breathing morpheme cluster is corresponding to a predetermined son breathing morpheme to be identified.So, one that belongs to certain breathing morpheme is moved
State respiration parameter vector, belongs to certain height breathing morpheme further.Every kind of breathing morpheme is subdivided into after sub- breathing morpheme, obtains
Finer description to the feature of personnel's breathing.For above-mentioned breathing morpheme cluster, a breathing in the middle of the side of taking off
Morpheme clusters and does further cluster analyses, obtains some son breathing morpheme clusters.
Hereafter, the statistical nature based on breathing morpheme and son breathing morpheme, determines one group of respiratory characteristic of personnel to be identified
Parameter.This group respiratory characteristic parameter comprises following index:Every height breathes all dynamic respiration parameter that morpheme cluster is comprised
The center of vector, and the average of all respiration indices parameter vectors that each breathing morpheme cluster is comprised.By this way
The respiratory characteristic parameter obtaining varies with each individual, and has the stability of inherence for same person.
In the present embodiment, complete personal identification process is divided into registration and two stages of identification.This two stages are respectively
It is to increase new personal information in systems, and the respiratory characteristic parameter of personnel to be identified is carried out with information existing in system
The process of contrast.
In registration phase, system gathers one or more sample breath data of certain personnel according to above-mentioned flow process, so
Therefrom extract corresponding respiratory characteristic parameter afterwards, and then generate the model of this personnel based on these respiratory characteristic parameters.Model is
The some form of expression of respiratory characteristic.Different according to the recognition method that system adopts, the model in system is probably explicit
Or implicit expression.Correspondingly, the model library of system is also likely to be explicit or implicit expression(Multiple models being registered personnel in system
Constitute model library).Explicit model is compared with implicit model, when keeping identical response speed, the calculating that explicit model requires
Ability is significantly larger than implicit model, and therefore the system more favours the recognition method using implicit expression.A kind of recognition method of implicit expression
It is using certain grader(Such as one grader based on neutral net, belongs to the known technology in computer science, referring to
[pattern classification]. (beautiful) Richard O.Duda etc. writes, Li Hongdong Deng Yi .2003. China Machine Press .p230-p283)Come
The respiratory characteristic parameter of identification Real-time Collection.Now, the model of a people may be presented as one group of weights of grader, is provided with
The grader of this group weights, can provide very high output valve, and provide low output valve in identity mistake when identity is correct.
Above-mentioned weights may be by obtaining using belonging to many personal multigroup respiratory characteristic parameters this grader is trained, its
In everyone at least one group respiratory characteristic parameter.Training process may relate to the respiratory characteristic parameter of new addition, and is
Existing respiratory characteristic parameter in system.The target of training is so that grader is had for the respiratory characteristic parameter from different personnel
Good separating capacity, such as, for the respiratory characteristic parameter from personnel A, the class other places in personnel A are given relatively by grader
High output valve, provides relatively low output valve at personnel B, and for the respiratory characteristic parameter from personnel B, grader is then given
Go out the result of contrast.The process of training may relate to various methods, the such as back-propagation algorithm on basis, radial direction base letter
Number, matched filter, deep learning etc..The grader training will be used in this personnel of later identification.
In registration phase, in order to improve recognition effect it is sometimes desirable to use the multigroup respiratory characteristic belonging to same personnel
Parameter is training above-mentioned grader.Its reason is, due to the presence of noise, even by same breathing recording equipment institute
The different sample breath data of the same person obtaining, the respiratory characteristic parameter being obtained also likely to be present certain difference.
Such difference may reduce the performance of personal identification.And(Obtained according to the multiple sample breath data from same person)
Multigroup respiratory characteristic parameter then more accurately intactly reflects the respiratory characteristic of a Man's Nature, special with such multigroup breathing
Levy parameter to complete above-mentioned classifier training, effect of noise can be weakened, fully reflected the individual character breathing of a people
The model of feature.
The identification process of system comprises the steps of:Gather a sample breath data of personnel to be identified, and based on this
Sample breath data extracts the respiratory characteristic parameter of this personnel;By the new respiratory characteristic parameter obtaining and existing model in system
(Usually multiple)Contrasted, obtained the matching degree of the quantification corresponding to each model;It is then based on these and meet journey
Degree, makes decisions analysis by certain decision logic, determines the identity of identified personnel.The basic principle of judgment analysis, is to calculate
The high model of matching degree in result, it is more likely that real identity.When being identified using grader, first will be from certain
The grader that the respiratory characteristic parameter input of individual personnel has trained, this grader all produces to personnel identity existing in system
The response value of a raw quantification, and the response value of corresponding correct personnel identity, are higher than the response value to other identity.
Here identification has discriminating and confirms two kinds of forms:Differentiate to refer to there is no any vacation in the identity to identified person
If in the case of identify its identity, this recognition method needs to check the new respiratory characteristic parameter obtaining and all models in system
Meet situation, and with matching degree highest model as identification result.And confirm then to refer to first do body by personnel to be identified
Part statement, then system obtain the respiratory characteristic parameter of this personnel, and check that it meets situation with the model claimed, if
Matching degree is higher than certain threshold value then it is assumed that this personnel has claimed identity.Typically, differentiate identifying system is wanted
Ask higher.
In identification, test approaches that the decision-making of system is carried out according to system and different:In closed set test, known
Others' member must be registered in systems, and that is, system default does not have personator it is only necessary to find out in model library and current
The most like people of people is as the result of identification;And in opener test, speaker may not register in systems, that is, be not excluded for
The probability that people assumes another's name, not only most like therewith in model library the to be found out people of system, also will investigate this similarity degree, to sentence
Whether disconnected speaker is personator.
Being had the advantage that based on the personal identification method of breathing in the present embodiment:The respiratory characteristic parameter of one people is anti-
Reflect is the substitutive characteristics of the breathing of a people, and these features are independent of specific breathing pattern, and not with external environment and are somebody's turn to do
The temporary physiological reaction of personnel and change, the therefore strong robustness of this recognition methods.In addition, in the present system, personnel exhale
Suction state is that the thermal infrared images based on nose areas obtains, and gathers thermal infrared images directly connecing without people and equipment
Touch, can carry out in distance relatively far away from, therefore identification process is less for the interference of identified person.
Personal identification system in the present embodiment can be based on different computing devices(Any amount), transmission environment and/
Or configuration realization.Computing device therein can be notebook computer, desktop computer, work station, large scale computer, server, flat board
Computer, smart mobile phone, intelligent appliance etc..System, according to the respiratory characteristic of personnel, is identified to it.
System includes a breathing recording equipment.For the personnel being in its visual field, breathe recording equipment and catch it
One or more sample breath data.Wherein each sample breath data comprises a number of breath data frame, covers one
The temperature changing process of nose areas in continuously multiple breathing cycles for the individual personnel.The Respiration Rate being caught by breathing recording equipment
According to sample, will be processed by system.Breathing recording equipment mentioned here can have multiple implementations, but different realization side
Result obtained by formula is the same, obtains breath data frame.
First realization of above-mentioned breathing recording equipment, is a thermal infrared imaging equipment.It is located at by the collection of this equipment
The breath data of one static personnel of dead ahead, and then identify this personnel.By adjusting the position of this thermal infrared imaging equipment
Put, direction and focal length, it can be made to capture the nose of this personnel with predetermined angle, and the nose in image is had relatively
High resolution.The process of identification comprises the steps of:The first step, by the initialization in certain apparatus for initializing and/or computer
Routine, starts the sequential image acquisition of this thermal infrared imaging device, contains the nose of personnel to be identified in acquired image.
This apparatus for initializing or routine possess low power consumption or the low feature of CPU usage, can continuously and uninterruptedly run;Second step,
In all thermal infrared images, determine the nasal area of personnel, and by image cropping to this region, the thermal infrared images after cutting
It is above-mentioned breath data frame.Hereafter, as the method previously described, all breath data frames of this personnel are processed,
Obtain one group of respiratory characteristic parameter of this personnel eventually, and complete personal identification based on these characteristic parameters.
Position nose in thermal infrared images, and the process by image cropping to this region, it is to rely on computer automatic
Complete.Basic operation is to determine the state of nose(Position, size, direction etc.), this can be using in computer vision
Target following technology is realized, and target following is one of computer vision field basic technology.Thermal infrared images is followed the tracks of
The process of nose comprises the steps of:The first step, in the first frame thermal infrared images, manually or automatically detects nose.For
In the case of automatic detection nose, system obtains the sample of the thermal infrared images of nose of different personnel in a large number in advance, and from this
Extract the general feature of nose in a little samples, and then realize the nose detection based on this general feature.Mentioned here
General feature, can be surface structure feature, two sub-circulars such as at nostril, the semicircle phaeodium at the wing of nose etc.,
Can be other any suitable features.Second step, using the state of nose that detects in the first two field picture as initial shape
State, obtains the feature of nose, the descriptive model of construction nose.Here feature can be global characteristics(Such as object module, system
Meter rectangular histogram, profile etc. or local feature(As angle point, edge, Harr feature etc.).3rd step, in follow-up image
In, using the descriptive model of nose, using some way, such as statistical filtering or density estimation, to estimate the current state of nose,
The current state simultaneously utilizing nose updates object module.If nose lost during following the tracks of, utilize testing result weight
The original state that new definition is followed the tracks of.On the basis of the state determining each moment nose, by computer automatically to thermal infrared figure
As carrying out cutting, obtain qualified breath data frame.
In one implementation, above-noted persons' identifying system includes a breathing modeling device and an evaluator.System is from exhaling
Inhale several sample breath data of recording equipment reception staff, wherein each sample breath data comprises a number of exhaling
Inhale Frame, cover the temperature changing process of nose areas in continuously multiple breathing cycles for the personnel.Each is such
Sample breath data all will be provided to breathing modeling device, by breathing modeling device, it be processed.As previously mentioned, for each
Sample breath data, breathing modeling device will be sequentially completed:Respiration parameter vector extracts;Breathing morpheme cluster and son breathing morpheme are gathered
Class;The dynamic step such as respiratory characteristic parameter and respiration indices characteristic parameter extraction, finally gives one group of respiratory characteristic parameter.Corresponding
The respiratory characteristic parameter of all sample breath data is supplied to evaluator the most at last.Evaluator mainly has two functions:In registration
In the stage, train a grader using above-mentioned some groups of respiratory characteristic parameters, so that this grader is possessed to registered personnel's
Good discrimination ability;In cognitive phase, carry out personal identification using this grader.
In one implementation, system comprises processor.Processor is probably one or more microprocessors, microcomputer, micro-control
Device processed, digital signal processor, CPU, state machine, logic circuit and/or any operated based on operational order
The equipment of signal.Processor can obtain storage software instruction in memory, and executes these instructions.
System includes interface.Here interface can be the various interfaces based on software instruction or hardware based interface.
Using these interfaces, system can be with other equipment(Server, data source and external data warehouse etc.)Communicated.Additionally,
System can be by a communication network and other communication apparatus communications.
System comprises a memorizer.Memorizer may be coupled to processor.Memorizer can be any computer-readable
Medium, including volatile memory and/or nonvolatile memory:The former such as static RAM (SRAM) and dynamically
Random access memory (DRAM);The latter's such as read only memory (ROM), erasable read-only memory (EROM), erasable programmable are only
Read memorizer, flash memory, hard disk, CD, tape etc..
System includes module data.Module data, may be connected to processor.Potentially include for holding in module
Row particular task or realize the various routines of specific data type, program, object, assembly, data structure.Module is potentially based on
Realizing, these hardware potentially include various hardware:Signal processor, state machine, logic circuit, and any referred to according to operation
Make equipment or the assembly of peration data.Data is used for the data storing acquired in module, receiving or generate.
Module can be hardware or one group of software instruction, or the combination of hardware and software.For module it is
The situation of hardware, module is implemented as the equipment such as some computers, processor, state machine, logic array or device.In module it is
In the case of software instruction, module is one group and is instructed by one group that certain processing unit executes, when this processing unit executes these
During instruction, any required function can be executed.Above-mentioned software instruction is potentially stored in various storage devices.According further to need
Will, software instruction can be downloaded by network.
In one implementation, module includes feature extractor, breathing modeling device, evaluator and other modules.These modules
By system execution.Data module includes breath data, extract characteristic, cluster data, classifier data, Yi Jiqi
His data.These data modules are all used for storing the data from module.
The identification process of system on human person is described below, the core of this identification process is to breathing morpheme and son breathing
The cluster analyses of morpheme, and and then extract dynamic respiratory characteristic parameter and respiration indices characteristic parameter.
As previously mentioned, for the personnel in the visual field of breathing recording equipment, breathing recording equipment will catch its some number
The sample breath data of amount.Each sample standard deviation such comprises a number of multiple breath data frame, covers a people's
Continuously multiple breathing cycles.These sample breath data will be sent to system.In one example, breathing recording equipment is to people
Member carries out data capture and generates a sample breath data and be supplied to system, comprises 3000 breath data in this sample
Frame.These breath data frames can be expressed as f1, f2, f3 ..., f3000.These breath data frames will be made further by system
Process.
After system receives this sample breath data, feature extractor 212 will be to breath data frame all of in this sample
Processed, extracted a respiration indices parameter vector and a dynamic respiration parameter vector from each breath data frame.Often
Respiration indices parameter vector in one frame comprises multiple respiration indices parameters, and dynamic respiration parameter vector comprises multiple dynamic breathings
Parameter.Nose temperature data, respiration indices parameter vector, dynamic respiration parameter vector, are stored in nose temperature respectively and exhale
Inhale in supplemental characteristic 216.
In one example, a respiration indices parameter vector comprises a respiration indices parameter, is designated as [ds1 ...
ds12];One dynamic respiration parameter vector comprises 14 dynamic respiration parameters, is designated as [df1 ... df14].So, for upper
3000 breath data frames in the sample stated(From f1 to f3000), have 3000 respiration indices parameter vectors and 3000 dynamic
State respiration parameter vector.All these respiration indices parameter vector and dynamic respiration parameter vector representation can be one two
Dimension matrix, its order is [3000 x (14+12)].
Breathing morpheme and the modeling process of son breathing morpheme are described below.By breathing modeling device to by sample breath data institute
The all dynamic respiration parameter vector obtaining carries out cluster analyses, forms multiple breathing morpheme clusters, and breathing morpheme therein is gathered
The possible range of the quantity of class is from 2 to 10.
In one example, the quantity needing the breathing morpheme of identification is 3.Breathing modeling device, for f1 to f3000 frame
All dynamic respiration parameter vectors, execute K mean cluster algorithm, create 3 clusters.For all of dynamic in hyperspace
Respiration parameter vector, breathing modeling device 110 randomly chooses three dynamic respiration parameter vectors(VA、VB、VC)As initial clustering
Center, and calculate Euclidean distance D (Vi, VA) of other dynamic respiration parameters vectors and VA, VB, VC, D (Vi, VB), D (Vi,
VC), its computational methods be D (Vi, V ' k)=(Σ (Vij-V ' kj) ^2) ^0.5, wherein Vi represents that to be sorted certain dynamically breathes
Parameter vector (i=1 ... 3000), V ' k is certain initial cluster center (k=1,2,3), and Vij represents dividing of j-th dimension of Vi
Amount (j=1 ... 14), V ' kj represents the component (j=1 ... 14) of j-th dimension of V ' k.All of Euclidean distance calculates after finishing,
Create cluster around three initial cluster centers, that is, for each dynamic respiration parameter vector Vi to be sorted, find with its away from
From nearest initial cluster center, and Vi is divided into belongs to the corresponding class of this initial cluster center.Complete in initial division
Afterwards, all of dynamic respiration parameter vector is divided into three clusters.Hereafter, breathing modeling device 110 calculates all of each cluster
The average center Mk of dynamic respiratory characteristic vector, its computational methods is Mk=(Σ Vi)/Nk, and wherein Vi is contained in k-th cluster
Certain dynamic respiration parameter vector (i=1 ... Nk) having, Nk is the sum (k=of the dynamic respiration parameter vector in k-th cluster
1,2,3).Three initial cluster center V ' k, are replaced by three averages calculating.After have modified cluster centre, breathing
Modeling device 110 creates cluster again around three amended cluster centres.In the manner described above, breathing modeling device 110 is continuous
Three cluster centres of modification, and create amended cluster around amended cluster centre, until cluster centre no longer changes
Till.Three clusters will be obtained in this approach, the corresponding breathing morpheme that can identify of each cluster.Meanwhile, each
All several points are contained, these points all represent separated dynamic respiration parameter vector in cluster.
For the breathing situation of finer description personnel 104, by breathing modeling device, each breathing morpheme is gathered further
Class carries out cluster analyses again, finally forms many height breathing morpheme clusters in each breathing morpheme cluster.Realize at one
In, the possible range of the quantity of sub- breathing morpheme is from 2 to 10.
In one example, the quantity of son breathing morpheme to be identified is 8.For each breathing morpheme cluster, breathing is built
Mould device all to dynamic respiration parameter vector execution K mean cluster algorithm therein, is clustered with creating 8 sons.So, for 3
Breathing morpheme cluster, is obtained 24 son breathing morpheme clusters.Obtained every height cluster, both corresponding to one can be at certain
The son breathing morpheme identifying in individual breathing morpheme.Meanwhile, all contain a number of dynamic breathing in every height cluster
Parameter vector.
Above-mentioned breathing morpheme cluster and son breathing morpheme cluster process can use any clustering method, including but do not limit
In:K mean cluster method, fuzzy C-means clustering method.Number with above-mentioned breathing morpheme cluster and son breathing morpheme cluster correlation
According to being stored in cluster data.
One group of respiratory characteristic ginseng of personnel to be identified after creating cluster and clustering with son, is determined by breathing modeling device
Number.Its concrete grammar is:For above-mentioned every height breathing morpheme cluster, calculate all dynamic respiration parameter vector that it comprises
Center.One son breathing morpheme cluster center, can be expressed as one 14 dimension vector, be set to VDij, wherein i be this to
The numbering of the breathing morpheme cluster belonging to amount, j is son breathing morpheme cluster belonging to this vector in the numbering of this breathing morpheme cluster.
Therefore, if there being 8 son breathing morpheme clusters in a breathing morpheme cluster, the center of this little breathing morpheme cluster, can
To be expressed as the two-dimensional matrix of [8 x 14], it is set to WDi, wherein i is the numbering of this breathing morpheme cluster.WDi from 8 to
Amount composition, i.e. [VDi1 VDi2 ... VDi8], WDi is the dynamic respiratory characteristic parameter of corresponding i-th breathing morpheme.In addition,
Calculate the average of the respiration indices parameter vector in each breathing morpheme cluster by breathing modeling device.Each breathing morpheme is gathered
Class, its average can be expressed as the vector of one 12 dimension, is set to VSi, and wherein i is the numbering of this breathing morpheme cluster.VSi is
The respiration indices characteristic parameter of corresponding i-th breathing morpheme.Breathe this average of morpheme cluster calculation for each, then that it is attached
It is added in the dynamic respiration parameter vector VDi of this cluster.This provides the data set corresponding to each breathing morpheme cluster, and it is permissible
It is expressed as a two-dimensional matrix, is set to Wi, wherein i is the numbering of this breathing morpheme cluster.Understand Wi order be [8x (14+
12)].Further it is assumed that personnel have 3 breathing morpheme clusters, then the complete breath data collection of this personnel can represent
Become a two-dimensional matrix, its order is [3 x 8 x (14+12)], i.e. [24 x 26].This data set contains to should exhale
Inhale dynamic respiratory characteristic parameter and the respiration indices characteristic parameter of data sample, referred to as respiratory characteristic parameter.In general, one group
Complete respiratory characteristic parameter can be expressed as a two-dimensional matrix, and its order is [(quantity of breathing morpheme to be identified) x
(son breathing morpheme quantity to be identified in a breathing morpheme) x (parameter sum in respiration indices parameter vector+dynamic
Parameter sum in respiration parameter vector)].The data related to respiratory characteristic data, is stored in feature classifiers data.
In sum, the method for breathing modeling device 110 acquisition respiratory characteristic parameter includes:There is provided for by feature extractor
Dynamic respiration parameter vector carry out cluster analyses, create the breathing morpheme cluster of predetermined quantity;Gather for each breathing morpheme
Class, carries out cluster analyses again for all dynamic respiration parameter vectors therein, and the son breathing morpheme forming predetermined quantity is gathered
Class;Be calculated as follows index for this sample breath data, to determine to should sample respiratory characteristic parameter:Every height breathing morpheme
The center of all dynamic respiration parameter vector of cluster(For generating dynamic respiratory characteristic parameter), and each breathing morpheme is poly-
The average of the respiration indices parameter vector of apoplexy due to endogenous wind(For generating respiration indices characteristic parameter).As it was previously stated, complete exhaling
Inhaling characteristic parameter is a two-dimensional matrix, and its order is that [(quantity of breathing morpheme to be identified) x is (in each breathing breathing shape
Son breathing morpheme quantity to be identified in position) x (the parameter sum in respiration indices parameter vector+dynamic respiration parameter vector
In parameter sum)].
Above-mentioned realization breathes the method for morpheme cluster and son breathing morpheme cluster it may be possible to K mean cluster method, fuzzy
C means clustering method, or any other clustering method.
Gathered by the collection sample breath data extraction respiration parameter vector that personnel to be identified are repeated with above-mentioned
The process of respiratory characteristic parameter is extracted in alanysis, it is possible to obtain multigroup respiratory characteristic parameter of this personnel, they all reflect
The individual character respiratory characteristic of this personnel.The meaning extracting multigroup respiratory characteristic parameter is:Any data acquisition all cannot be complete
Entirely avoid effect of noise, the process of the acquisition respiratory characteristic parameter of the system is no exception.If relying only on the list of a people
Only one group of respiratory characteristic parameter generating the model for characterizing certain personnel, then in respiratory characteristic parameter generating process in model
Effect of noise may be obvious, and rely on multigroup respiratory characteristic parameter of a people to generate the model of this personnel
Words, then the model generating can preferably eliminate noise and accidentalia, have preferably representative and stability, and then lift knowledge
The performance of other system.
It is after personnel to be identified determine one or more groups of respiratory characteristic parameters in breathing modeling device, by identifying
Device 112 be based on one grader of these respiratory characteristic parameter trainings so that this grader registered personnel are possessed good
Separating capacity.Complete to train and mean to complete the registration to this personnel, hereafter evaluator can be come in fact using this grader
When this personnel of identification.This grader potentially includes a supervised learning algorithm, such as support vector machine (SVM) grader, shellfish
Ye Si estimation, decision tree, neutral net etc..For the registration of multiple personnel, carry out successively according to above-mentioned flow process.
Once completing the registration to personnel, system can be identified to it any time afterwards.The system
The live Real time identification of middle personnel is by evaluator 112(Respiratory characteristic parameter according to personnel to be identified)Carry out.Extraction is treated
The process of identification personnel's respiratory characteristic parameter comprises the steps:A breathing by breathing recording device records personnel to be identified
Data sample;The respiration indices parameter of this sample breath data and dynamic respiration parameter are extracted by feature extractor 212;By breathing
Modeling device 110 obtains one group of respiratory characteristic parameter of this sample breath data;Finally by this group respiratory characteristic parameter input identification
Device 112, to be calculated the matching degree of the respiratory characteristic parameter extracted and each existing model in data base by it.
One application scenarios of the present invention are the automatic identification to vehicle drivers.Being embodied as method is:Certain bits in automobile
Put(The lower section of such as windshield)Several thermal infrared imaging equipment are installed, as breathing recording equipment.Adjusted by prior
Whole said one(Or it is multiple)The position of thermal infrared imaging equipment and focal length are so that this thermal infrared imaging device just can collect
Stand on the nose areas of the people of front side ad-hoc location.There is an initialization routine in processor in identifying system, should
Initialization routine is based on above-mentioned thermal infrared imaging equipment work, and its principle is the Temperature Distribution in monitoring of environmental, when
Temperature Distribution meets certain condition(For example there is the region of large-area higher temperature, the temperature of such as face)When, initialization
Routine thinks and detects a people, and now thermal infrared imaging equipment starts target(Nose)Detection, and carry out follow-up identification
Work.For example, when a people wants car is unlocked, go to predetermined position first, towards above-mentioned thermal infrared imaging equipment station
Vertical.Initialization routine can detect the presence of personnel, and starts thermal infrared imaging equipment, obtains the multiple continuous of a period of time
Thermal infrared images, wherein contains the nasal area of people to be identified.Then, system automatically to these thermal infrared images at
Reason, extracts respiratory characteristic, and then multiple respiratory characteristic models of the feature extracting and preservation in data base is compared.Look for
Go out one model of matching degree highest of comparison it may be determined that the identity of personnel on current driver seat.By this way,
Can avoid carrying out during recognition of face, being subject to the big problem of illumination effect with visible images.Furthermore it is possible to effectively evade with visible
Light mode carries out the potential risk of recognition of face.After personal identification completes, some intelligentized operations can be carried out, for example
Autonomous driving vehicle unblock or automatic chair regulation etc..
Another application scenarios of the present embodiment are cellphone subscriber's identification.Specific implementation method is:Screen one in mobile phone
The upper end of side or lower end, install a low profile thermal infrared imaging device.By the prior position adjusting above-mentioned thermal infrared imaging equipment
Put with focal length so that this thermal infrared imaging device can collect the nose areas of the personnel holding mobile phone.When a people wants
When mobile phone is unlocked, need to gather the thermal infrared images of the breathing situation comprising a period of time.Then, system is automatically to this
A little thermal infrared images are processed, and extract respiratory characteristic, and then the respiratory characteristic extracting is carried out with the model in data base
Compare.The matching degree highest model comparing, may be confirmed as the identity of current persons.In this way it is possible to avoid
Carry out during recognition of face, being subject to the big problem of ray images with visible images, effectively to evade, pedestrian can be entered in visible ray mode
The potential risk of face identification.
It is noted that the description to personal identification system and correlation technique above, simply related ultimate principle is concrete
Realization, feature described herein or method, do not produce to the present invention and limit.