CN106580301A

CN106580301A - Physiological parameter monitoring method, device and hand-held device

Info

Publication number: CN106580301A
Application number: CN201611189222.8A
Authority: CN
Inventors: 刘锦; 邹煜晖
Original assignee: Guangzhou Heart And Tide Mdt Infotech Ltd
Current assignee: Guangzhou Heart And Tide Mdt Infotech Ltd
Priority date: 2016-12-21
Filing date: 2016-12-21
Publication date: 2017-04-26
Anticipated expiration: 2036-12-21
Also published as: CN106580301B

Abstract

The invention discloses a physiological parameter monitoring method, a physiological parameter monitoring device and a hand-held device. The method includes the following steps that: a current face image is acquired; a cheek part image in the face image is extracted; the mean values of the green channels of the cheek part image are calculated, so that a mean value sequence can be obtained; sliding mean value filtering is performed on the mean value sequence, a physiological change curve is obtained by adjusting the filtering interval of the sliding mean value filtering; and feature extraction is performed on the physiological change curve, so that physiological parameters can be obtained. According to the physiological parameter monitoring method, different physiological parameter curves can be obtained through different filtering intervals; and the curves can be correspondingly processed, so that a plurality of physiological parameter values can be obtained. The invention also discloses a handheld device using the method. The method, device and the hand-held device of the invention have the advantages of low computation amount, high accuracy and low hardware requirements.

Description

A kind of monitoring method of physiological parameter, device and handheld device

Technical field

The present invention relates to physiological compensation effects research field, more particularly to a kind of monitoring method of physiological parameter, device and Handheld device.

Background technology

Human body physiological parameter monitoring is included in electrocardio, blood pressure, heart rate, blood oxygen saturation, respiratory rate, body temperature, breathing The isoparametric monitoring of gas concentration lwevel, be now widely used for the fields such as clinical medicine, wearable device, mobile terminal.

For example, heart rate non-invasive measurement means ripe at present have sphygmodynamometry, electrocardiosignal method, photoplethysmographic Graphical method (PhotoPlethysmoGraphy, PPG).Wherein：

First, sphygmodynamometry：This is in fact most ancient method, is exactly feeling the pulse for the traditional Chinese medical science.In wrist or neck both sides, Percutaneously the pressure regularly fluctuation of artery can be touched by skin.This signal can be become by heart rate by pressure sensor. This scheme is also commercial at present most jejune, and reason one is that pressure sensor needs for a long time to press the artery half of wearer Compel, there is sense of discomfort.Two is that pressure sensor is difficult to be fixed on skin surface in an appropriate manner：Fixed can tightly cause very much blood flow It is not smooth, it is fixed too loose and cannot realize measurement.This problem shows especially obvious in sport wrist-watch design.So the party Method is general only to be used postoperative tranquillization patient in operation within the hospital.Under normal circumstances, pulse is consistent with heart rate, is had One time heart contraction will produce a pulse, but in the case where blood pressure is extremely low, the hematopenia that heart contraction is pushed is with energy It is enough that beating is perceived on blood vessel, then there is situation of the pulse less than heart rate.

2nd, electrocardiosignal method：Antrum controls heart contraction diastole so as to trunk pump blood with having the rhythm and pace of moving things.This control letter Number it is an electric signal (human nerve signal all shows as electric signal on nerve), body surface can be gradually diffused into, can be in skin Skin passes through electrode measurement.The electrocardiogram equipment that Hospitals at Present is used is exactly using this principle.This rhythm is exactly heart rate, except this it Outward, electrocardiosignal can also provide many reference informations for diagnosis.Current wearable heart rate measurement most accurate on the market Instrument, heart rate band, are also using this method.But due to the wavelength of electrocardiosignal it is very long, in order to measure enough accuracy Signal, signal electrode and reference electrode just must be on torso spaces every enough to remote.Distant 2 points usually on chest, or Person's left hand and the right hand, or hand and pin etc..Watch just compares difficult using this scheme, unless someone is ready two tables of band simultaneously.

3rd, photoplethysmographic graphical method：This is a kind of simple lossless measurement heartbeat component and evaluation body surface circulation Method.According in organism optical to the description of light propagation in biological tissues, distribution and light and the interaction of tissue and Research, when during the light of certain wavelength is by (injecting) bio-tissue, bio-tissue (skin, fat, blood, muscle etc.) Scattering will be produced to light and will be absorbed, make the light intensity decays for detecting.Wherein its hetero-organization such as skin, muscle and bone is to light Absorb almost constant, but when transmission region exist arteries beating or vein blood vessel it is full when, with CBF Increase and decrease, main extinction composition content of hemoglobin also accordingly increases and decreases in blood, and blood will change therewith to the uptake of light.Blood When pipe is full, absorbing amount is maximum, and the luminous intensity for detecting is minimum, is believed the intensity variation for detecting using photoelectric sensor Number it is converted into electric signal detection, you can the change for tracing out intravascular volume obtains pulse wave signal.

At present, the implementation of PPG can be divided into contact and contactless.

Contact PPG is realized needing photoelectric sensor to carry out data acquisition, such as that common is clipped in forefinger tip/ear Vertical Cardiotachometer, various wearable devices, sport wrist-watch etc..Some mobile phones such as the S5/S6 of Samsung, the ZukZ2Pro of association Also it is integrated with infrared heart rate sensor at mobile phone back, it is only necessary to finger by can just collect heart rate on a sensor.

Due to needing transmitting light source and receiving terminal, institute can also realize in this way on mobile phone.Also have many at present App utilizes this principle, it is only necessary to finger is pressed on camera, flash lamp is opened, by the image light and shade of acquisition camera Change to obtain the numerical value (a kind of form of photoelectric signal transformation can be regarded as) of heart rate.But current scheme generally existing Following defects：1st, can wear and tear camera lens, affect cam lens imaging effect；2nd, flash lamp is opened, finger may be scalded；3rd, day When gas is colder or in the case of body abnormality, blood possibly cannot flow through finger, cause to can't detect heart rate；4th, finger Permeability must get well, and if spot or have covering, can affect measurement result；5th, without flash lamp or flash lamp distance The distant equipment of camera cannot carry out heart rate collection (wherein 3 and 4 is also one of defect of infrared sensor).

On wearable device (referring mainly to bracelet, wrist-watch here), typically at least there are a green glow Led light sources (general feelings Under condition, more accurate data can be obtained by measuring the Absorption of green glow, be dripped with sweat in hot environment, such as gymnasium When, skin surface moisture increases, because more green glows have been predominantly absorbed, so need to be switched to other light sources, such as in vain Light or infrared light supply), some bracelets employ two kinds of light sources of green glow and white light, and AppleWatch is then using green glow and infrared Light.Except light source, in addition it is also necessary to there is individual photoelectric sensor to sense reflected light.Although this method can only obtain heart rate signal, no The noise immunity for comparatively bringing to motion is crossed stronger, so being well suited for current sport wrist-watch.Therefore each manufacturer is only (to remove integrated optic-electronic sensor on respective SOC (System-on-a-Chip integrating chips, typically bluetooth SOC) AppleWatch is oneself research and development, and it is third-party that major part is all, such as Philips, the offer of the manufacturer such as Kionix ), by the algorithm of heart rate of respective research and development, go the impact of compensation campaign noise to calculate heart rate using accelerograph.Due to being Contact PPG, is ensureing steady, and under the premise of laminating skin is close, the interference being subject to is less, thus numerical value should all it is poor not It is many.

But even under ideal conditions, AppleWatch and other bracelet/wrist-watches cannot guarantee that all users are every It is secondary to obtain accurate heart rate data.The reason for causing destabilizing factor be also mainly：1st, such as, under cold environment, hand DBF on wrist may be too low, causes heart rate sensor to read data；2nd, there is interference in the region that sensor is covered, Such as tattoo, birthmark etc.；3rd, the impact that causes is moved, such as arm bends and stretches the change that can all cause blood flow.At this moment, then Need to consider to get accurate heart rate by heart rate detectors such as third-party bluetooth pectoral girdles.

Contactless PPG measuring methods in actual applications, can bring great convenience to the life of user, its Advantage is the blood flow information that can monitor larger skin area so that result of calculation is more accurate.Additionally, relative to contact For PPG technologies, contactless PPG technologies have more be widely applied field, for example, for some skin injury/injuries Crowd, it has not been convenient to have crowd etc. that equipment is contacted on body.Current implementation method is probably as follows：1st, photoelectric transfer is utilized Sensor (typically camera) to human body somewhere blood vessel concentrated area (any region all can, wrist or face are covered without clothes Lid is more convenient) carry out video data acquiring；Secondary light source or special light source emitter are generally needed in gatherer process；2、 ROI (interested, i.e., to the obvious region of light absorption reacting condition) region is isolated from each two field picture of video data, RGB primary channel separation is carried out to the area image, and average is taken to each passage all pixels, as the two field picture at this The characteristic value of Color Channel, generates three new signal Xr, Xg, Xb, and these three signals are standardized, and obtains standardized Signal Yr, Yg, Yb；3rd, blind source separating (ICAIndependent Component are carried out to the signal after standardization Correlation Algorithm, independent component analysis).Because the independent signal source after separation is unordered, it is impossible to directly select The number of winning the confidence, is at this moment accomplished by carrying out signal screening, and by carrying out correlation analysis with green passage Xb, selection obtains final signal Zi；4th, the signal transactings such as band filtering are carried out to selected signal, cycle analysis (frequency spectrum point is finally done to filtered signal Analysis), that is, DFT transform is carried out, spectrogram is obtained, peak-peak frequency is chosen as palmic rate, you can obtain heart rate.

But such scheme there are problems that following：1st, need first to shoot one section of video, then video could be carried out Analysis, the general video length for shooting needs 30 seconds or so, while shooting process has higher requirements to ambient light；2nd, it is right to need The each frame of video carries out ROI region separation, while tested personnel should avoid displacement as far as possible, in order to avoid affect result of calculation；3rd, need Red/Green/Blue3 channel data average is calculated simultaneously；4th, need to carry out blind source separating mixed signal using ICA, need Want more operation time；5th, need to carry out final signal spectrum analysis (DFT calculating), it is equally time-consuming more, and after changing The spectrogram for arriving is higher to data source requirement, if there have more interference to directly result in evaluation to be incorrect.

The maximum defect of above scheme is exactly that hardware requirement is high, and operand is big, then there has been proposed many improved methods Such as：1st, replace ICA with fastICA or jade algorithms, accelerate arithmetic speed；2nd, replace DFT with FFT, reduce computing Amount, accelerates arithmetic speed；3rd, the detection number of times of ROI region is reduced, it is to avoid excessive computing consumption；4th, high and low frequency filtering is increased Device makes result of calculation tend to accurate reducing noise.But effect is still not satisfactory.

Therefore, seek that a kind of accuracy of measurement is high, low to System Hardware Requirement, physiological parameter that is can in real time obtaining result There is the monitoring method and device of (particularly heart rate, respiratory rate) important practical to be worth.

The content of the invention

Present invention is primarily targeted at overcoming the shortcoming and deficiency of prior art, there is provided a kind of monitoring side of physiological parameter Method, the method has the advantages that amount of calculation is low, the degree of accuracy is high, low to hardware requirement.

Another object of the present invention is to provide a kind of monitoring device of physiological parameter, the device has that amount of calculation is low, standard Exactness is high, the advantage low to hardware requirement.

Another object of the present invention is to a kind of handheld device of the monitoring device including above-mentioned physiological parameter is provided, here Handheld device can be one it is independent, be only used for carrying out the equipment of rhythm of the heart, or one not only carrying out the heart Rate is monitored, and the physiological parameter monitor of other specification monitoring is also carried out, it is of course also possible to be the prison for being integrated with above-mentioned physiological parameter Survey the mobile terminal of device, such as mobile phone, removable computer etc..

The purpose of the present invention is realized by following technical scheme：A kind of monitoring method of physiological parameter, including step：

Collection current face's image；

Extract the cheek parts of images in facial image；

The average of Green passages is calculated to cheek parts of images, equal value sequence is obtained；

Moving average value filtering is carried out to equal value sequence, by adjusting its filtering interval physiological change curve is obtained；

Feature extraction is carried out to physiology change curve and obtains physiological parameter.

Preferably, for the facial image for gathering, diminution process is first carried out, then OpenCV is passed through to the image after diminution The Face datection algorithm of offer after making face position, then carries out the amplification of equal proportion detecting face, obtains face area Domain.Such that it is able to improve the speed of method.

Preferably, the location determining method of cheek parts of images is：It is highly h if human face region width is w, start bit It is set to (x, y), the width range of two cheek regions is respectively 1/8w～3/8w and 5/8w～7/8w, altitude range 1/ 2h～3/4h.The region for probably omitting nose is limited by above-mentioned zone, only retains two cheek regions, reduced data and do Disturb.

Preferably, the average of Green passage is calculated cheek parts of images, that is, the Green for adding up in each pixel leads to Road component, then obtain divided by the pixel quantity of image, the mode of interval sampling is taken in calculating process, such that it is able to improve Processing speed, and ensure to get overall situation of change.

Preferably, the frame per second of current gathered data is set as m, then be 3m/2 by arranging the filtering of equal value sequence interval ～2m, can obtain change of respiratory rate curve；The filtering interval for setting equal value sequence is m/3～2m/3, can obtain heart rate Change curve.Here filter interval setting range and the minimum of computation cycle it is similar by obtain corresponding change curve.

Further, if physiological change curve is changes in heart rate curve, extracted using following fisrt feature and calculated Method carries out feature extraction：

Preferentially find the position of first crest；

From the beginning of the position of previous crest, search in the range of R thereafter and be worth highest primary peak；

Then from the beginning of the position of primary peak, search in the range of R thereafter and be worth highest secondary peak；

Whether primary peak peak value is judged more than or equal to secondary peak peak value, if it is, determining that primary peak is to work as prewave Peak, then starts to continue search for from primary peak；If it is not, then determine that secondary peak is current crest, then from the second ripple Peak starts to continue search for；

After complete changes in heart rate curve of search, a crest sequence is obtained, described crest sequence includes each crest Peak value and position, position i.e. eartbeat interval；

By the time interval of 2 crests of head and the tail divided by the number of crest, heart rate value is obtained.

Further, heart rate 30～220 is divided into into n interval, correspondence n R value of setting, each R value is respectively adopted Ask for crest sequence；For each crest sequence, using following formula f values are calculated：

f_i=sd_i*sd_i/avg_i；

Wherein, i=1 ... ..., n；sd_iFor the standard deviation of eartbeat interval, avg_iTo calculate the equispaced of eartbeat interval, take Min f_iCorresponding i calculates heart rate value as the current sequence for selecting using the sequence data.

Further, if physiological change curve is change of respiratory rate curve, carried using following second feature Taking algorithm carries out feature extraction：

Step 1, the data sequence in change of respiratory rate curve, find first wave crest point O point；

Step 2, from O points, find next waveform flex point upwards, obtain first trough, mark 1；

Step 3, from the point of mark 1, constantly find the downward flex point of waveform, and be labeled as crest, then draw ripple Peak sequence O, A, B, C, D, E ..., and trough sequence 1,2,3,4,5 ..., draw the wave character of signal；

Step 4, according to wave character, it is peak-to-peak every the time difference and peak separation quantity to obtain head and the tail, by formula：

Respiratory rate br=head and the tail are peak-to-peak every the time difference/peak separation quantity；

Draw current respiratory rate.Respiratory rate has hysteresis quality, when lag time is the half of front respiration duration Between.Under normal circumstances, once complete waveform can judge the general frequency for breathing.

Further, because the oxygen whole content in blood changes a necessarily relatively slow process, so The wave crest point for searching out is judged by choosing the r of a real-time change.Using as follows during crest is found Denoising method：

If peak to peak separation twice recently is respectively d and d1, r=(d+d1)/2 is calculated；

Judge whether the time that the current crest found is located is more than last time peak time+r, if it is, being judged as new Crest, if it is not, then being judged as noise spot.

Certainly, the computing interval certain situation also occurs, such as the corresponding sequence number of the minimum of a value got every time constantly becomes, when Front numerical value exists abnormal, and the numerical value of heart rate is unstable etc., in order to improve the accuracy of data, to physiology change curve data elder generation Following screening operation is carried out, method is as follows：

Because experiment before has been proven that the length of calculating time will not have an impact to the concrete numerical value of heart rate, here The time span for taking the L seconds is interval as rate calculation, and the HR values increased in a last L1 second are referred to as a comparison, its The numerical value of middle heart rate is denoted as hbr, and the HR values that the last L1 seconds calculate are denoted as hbr L1, and hbr L1 can also regard that heart rate becomes as The trend of change, if the unexpected fluctuation of appearance or ectocine, hbr L1 and hbr have obvious numerical value difference.

Step 0, last heart rate waveform sequence number lastchoice for selecting of initialization, get the cumulative number of heart rate Count, used as the cumulative number e of heart rate, (e is that, when long-time rhythm of the heart is done, opposing is extraneous dry to Current heart rate waveform sequence number The parameter disturbed, for avoiding because interference causes to choose other heart rate sequences), recalculate number of times retrytime, assignment For 0；Here count is the number of times for getting heart rate for adding up, and is all to use Minfi before p1 heart rate is not got Corresponding sequence as heart rate sequence undetermined, behind just with the conduct heart rate sequence undetermined that average fluctuation is minimum；

Step 1, according to the n different filtering interval R for arranging, calculate heart rate hbr, hbr L1 and situation f that fluctuates (if External condition in the case of the excessive explanation of fluctuation is the currently monitored is undesirable, needs to abandon current sampling), and record storage, together When cumulative count；

If Step 2, count>=p1, then calculateAs average fluctuation；

Step 3, take g=Min g_i(i=1～n), and corresponding i is recorded as the alternative of current optimal heart rate

If lastchoice=0 (is not recorded for the first time), lastchoice=i, e=1 are made；

If lastchoice=i, e=e+1, e is made to be not more than p2；

If lastchoice is not equal to i, e=e-1, e is made to be not less than 0；

Step 4, difference d for calculating hbr and hbr L1, during heart rate stabilization, the numerical value of d is not over p4；

If Step 5, Current heart rate f>P3 (empirical value that many experiments draw), then illustrate currently to be subject to external interference Than larger, or the heart rate of collection has abnormal conditions；

At this moment judge：

If f>G and f>P3, illustrates to affect increasing, needs to abandon some abnormal datas and re-starts calculating, holds Row Step 6；If f<=g or f<=p3, performs Step 7；

Step 6, the data for abandoning current sequence of calculation the first half, make lastchoice, count, e be 0, retrytime =retrytime+1；If retrytime=2, total data is abandoned, make lastchoice, count, e be 0；

If retrytime=3, prompting is provided, preset test environment is undesirable, monitoring failure performs Step 8；

If Step 7, difference d of continuous L2 calculating<P4, illustrates that rate calculation numerical value tends towards stability, and at this moment judges to exhale The numerical value of suction whether calculated (numerical value of heart rate can tend towards stability for most fast 7 seconds, the frequency of breathing then due to the cycle compared with Long, typically at least needing the time of a respiratory cycle can just find out that waveform changes), and tend towards stability and (due to heart rate and exhale Suction is to calculate simultaneously, the consideration so the two is put together), if the numerical value of breathing is also stable, export heart rate and respiratory rate Numerical value, performs Step 8；

Step 8, stopping data acquisition, terminate monitoring process.

It is also possible to pass through this method to carry out vivo identification (such as picture or image etc.), because the light in the external world Line reflection is substantially irregular, if so excessive (f of fluctuation situation for drawing>P3, under normal circumstances f<=p3/2), then can be with Judge that current light environment is undesirable, or tested can not provide normal rule heart rate.

A kind of monitoring device of physiological parameter, including：

Image capture module, for gathering current face's image；

Image interception module, for extracting facial image in cheek parts of images；

Value sequence computing module, for calculating cheek parts of images the average of Green passages, obtains equal value sequence；

Filtration module, for carrying out moving average value filtering to equal value sequence, by adjusting its filtering interval physiology is obtained Change curve；

Analysis and processing module, for carrying out feature extraction to physiology change curve physiological parameter is obtained.

Preferably, described image acquisition module includes Zoom module and face detection module, and Zoom module is used for will collection Facial image carry out diminution process, then the image after diminution is sent to into face detection module, and for examining in face Survey module to orient behind face position, then carry out the amplification of equal proportion, obtain human face region；Face detection module is used for fixed Position goes out face position.

Preferably, image interception module is specially：It is highly h if human face region width is w, original position is (x, y), The width range of two cheek regions is respectively 1/8w～3/8w and 5/8w～7/8w, altitude range 1/2h～3/4h.

As one kind preferably, the filtration module adjustment filtering interval is m/3～2m/3, and m is current gathered data Frame per second, obtains changes in heart rate curve；The analysis and processing module includes fisrt feature extraction module, fisrt feature extraction module bag Include：

First data read module, for finding the position of first crest；

Primary peak search module, for from the beginning of the position of previous crest, search is worth thereafter highest the in the range of R One crest；

Secondary peak search module, for from the beginning of the position of primary peak, searching in the range of R thereafter highest second is worth Crest；

First judge module, for whether judging primary peak more than or equal to secondary peak, if it is, determining primary peak For current crest, then start to continue search for from primary peak；If it is not, then determine that secondary peak is current crest, then Start to continue search for from secondary peak；

Primary peak sequence preserving module, for searching for complete changes in heart rate curve after, obtain a crest sequence, it is described Crest sequence including each crest peak value and position, position i.e. eartbeat interval；

Heart rate value computing module, divided by the number of crest, heart rate value is obtained for by the time interval of 2 crests of head and the tail.

Further, crest sequence preserving module includes several sub- preserving modules of crest sequence, by heart rate 30～220 N interval, correspondence n R value of setting is divided into, each R value is respectively adopted and is asked for crest sequence, a crest sequence is stored in respectively In arranging sub- preserving module；

For each crest sequence, using following formula f values are calculated：

f_i=sd_i*sd_i/avg_i；

As another kind preferably, the filtration module adjustment filtering interval is 3m/2～2m, and m is current gathered data Frame per second, obtains change of respiratory rate curve；The analysis and processing module includes second feature extraction module, and second feature extracts mould Block includes：

First crest seeking module, for the data sequence in change of respiratory rate curve, finds first ripple Peak dot O points；

First trough finds module, for from O points, finding next waveform flex point upwards, obtains first Trough, mark 1；

Wave character determining module, for from the point of mark 1, constantly finding the downward flex point of waveform, and is labeled as Crest, then draw crest sequence O, A, B, C, D, E ..., and trough sequence 1,2,3,4,5 ..., draw the waveform of signal Feature；

Respiratory rate computing module, it is peak-to-peak every the time difference and peak separation quantity for according to wave character, obtaining head and the tail, lead to Cross formula：Respiratory rate br=head and the tail are peak-to-peak every the time difference/peak separation quantity；Draw current respiratory rate.

Further, the second feature extraction module also includes denoising module, and denoising module includes：

Step-size in search setting module, if peak to peak separation twice recently is respectively d and d1, calculates step-size in search r=(d+ d1)/2；

Judge module, for whether judging the time that the crest of current searching is located more than last time peak time+r, if It is then to be judged as new crest, if it is not, then being judged as noise spot.

Further, the monitoring device includes data screening module, and the data screening module includes：

Initialization module, for initializing heart rate waveform sequence number lastchoice of last selection, gets the tired of heart rate Metering number count, Current heart rate waveform sequence number recalculates number of times retrytime as the cumulative number e of heart rate, is entered as 0；According to the n different filtering interval R for arranging, heart rate hbr, hbr L1 and fluctuation situation f, and record storage are calculated, while tired Plus count；

Average fluctuation computing module, for count>During=p1, calculateAs average fluctuation；

Alternate data module, takes g=Min g_i, and corresponding i is recorded as the alternative of current optimal heart rate；If Lastchoice=0, then make lastchoice=i, e=1；If lastchoice=i, e=e+1, e is made to be not more than p2； If lastchoice is not equal to i, e=e-1, e is made to be not less than 0；

Difference calculating module, for calculating difference d of hbr and hbr L1；

Second judge module, for judging f<=g or f<Whether=p3 sets up, if set up, performs the 3rd and judges Module, otherwise, performs data removing module；

Data removing module, for abandoning the data of current sequence of calculation the first half, initialization lastchoice, count, E is 0, makes retrytime=retrytime+1；If retrytime=2, total data is abandoned, initialized Lastchoice, count, e are 0；If retrytime==3, prompting is provided, preset test environment is undesirable, monitoring is lost Lose, perform terminate module；

3rd judge module, for whether judging difference d of continuous L2 calculating less than p4, if explanation rate calculation Numerical value tends towards stability, then continue to judge whether the numerical value for breathing has calculated and tended towards stability, if it is, the output heart Rate and respiratory rate numerical value, perform terminate module；

Terminate module, for stopping data acquisition, terminates monitoring process.

The present invention compared with prior art, has the advantage that and beneficial effect：

1st, computation complexity of the present invention is low, can in real time obtain the physiology such as heart rate, the respiratory rate of current picker ginseng Number, the degree of accuracy is high.By comparing with existing Medical Devices acquired results, it is found that algorithm stability is very strong, and be not susceptible to To the interference of outside various conditions, with extremely strong applicability.

2nd, the present invention is by extracting cheek parts of images, denoising etc., it is possible to reduce the environmental disturbances factor in image, enters one Step improves the degree of accuracy of detection.

Description of the drawings

Fig. 1 is the flow chart of the method for the present embodiment 1.

Fig. 2 is the flow chart of the fisrt feature extraction algorithm of the present embodiment 1.

Fig. 3 is the flow chart of the second feature extraction algorithm of the present embodiment 1.

Fig. 4 is an original series curve of actual acquisition in embodiment 1.

Fig. 5 is that the heart rate curve that obtains after smothing filtering is carried out to Fig. 3.

Fig. 6 is the schematic diagram that heart rate is calculated by time domain method.

Fig. 7 is the heart rate curve of perfect condition.

Fig. 8 is that the respiratory rate curve that obtains after smothing filtering is carried out to Fig. 3.

Fig. 9 is the schematic diagram that respiratory rate is calculated by time domain method.

Specific embodiment

With reference to embodiment and accompanying drawing, the present invention is described in further detail, but embodiments of the present invention are not limited In this.

Embodiment 1

As shown in figure 1, a kind of monitoring method of physiological parameter of the present embodiment, including step：

First, current face's image is gathered

Because face is most often exposure skin area outside and with high identification, it is easy to position, so Non-contact capture system generally chooses human face region as data acquisition region.Collection face is also due to confession of the heart to human body Blood is preferentially to ensure brain and important internal organs, followed by four limbs, so collection face can obtain than gathering four limbs More reliable and accurate numerical value is obtained, because four limbs are in body tip, especially in winter sluggish metabolism, low temperature makes blood vessel Shrink, blood reflux capability weakens so that the position such as trick, especially finger tip, tiptoe poor blood circulation, likewise, some People can be presented pale after strenuous exercise, the ice-cold situation of four limbs, and this is also due to heart blood supply scarce capacity causes blood Liquid circulation is not smooth.And it, for detecting that oxygen content of blood changes, is cannot to detect accurately if poor blood circulation that PPG is Numerical value.

The present embodiment is capable of achieving the collection demand of PPG only by natural light or only by stable ambient light, due to It is to calculate in real time, it is only necessary to the G passages average of ROI region is preserved as PPG signals, without the need for calculating R passages and channel B Numerical value separating noise.

2nd, the cheek parts of images in facial image is extracted

It is the part of the skin area that blood flows through due to need acquisition, so needs reduce other regions as far as possible causing Interference, such as eyes, nose, face, because brow portion may be covered by bang, so the present embodiment propose only take face Compare and round a face and can get more accurate change numerical value in the region of cheek point.

It is not the emphasis of present invention discussion with regard to face recognition module, therefore can adopts a variety of existing Face recognition module is realizing.In mobile platform, IOS and Android has respective face recognition module, naturally it is also possible to adopt With any third party library, such as OpenCV, Face++ etc..The examples herein is realized at PC ends, is used and is increased income OpenCV storehouses, version is 2.48.OpenCV storehouses are the method for detecting human face by comprehensive AdaBoost and Cascade algorithms (be otherwise known as Harr-Cascade face detection devices) calls the cascade for training by OpenCV carrying out Face datection (grader) carries out pattern match to realize recognition of face.OpenCV supports at present HAAR, LBP (Local Binary Patterns) and 3 kinds of training aids of HOG, compared with Haar features, LBP features are integer features, therefore are trained and detection process Will be faster several times than Haar feature.In actual test, the time-consuming difference of detection face of different feature databases is called, but detect knot Fruit is without difference, difference such as following table.That what is chosen herein is the haarcascade_frontalface_default of acquiescence.

The present embodiment experimental situation machine is configured：CPU：I3-4010u 1.7G, internal memory：8GB, system：Win7 Ultimates 64 Position, development environment：Java8.0, OpenCV：2.48, developing instrument：Eclipse, resolution ratio of camera head：640×480.

The present embodiment can be carried out first to improve detection efficiency in face recognition process is carried out to image to be detected Diminution is processed, and carries out Face datection to the image after diminution, it is assumed here that the image size of initial acquisition is 640*480, is carried out Face datection is time-consuming to need 0.12 second, and the image size reduced after 1/4 is 320*240, at this moment carries out Face datection and only needs to It is 0.03 second, per second calculating by the frame of video 30, also can substantially can ensure that each two field picture all gets accurate face institute in place Put, (0.03 second × 30 frame=0.9 second, the operand needed for the scaling of image is 320 × 240 × 30=2304000, during computing Between be negligible substantially) only need to carry out equal proportion scaling, it is possible to obtain the region that face is located in original image, Bu Guoshi In the operation of border, all ROI region separation is carried out to each two field picture it is not necessary that, the operation of the Face datection of 1～5 time per second is The demand of real-time monitoring can be met, it is per second in this example to carry out 2 Face datections.

In the present embodiment, the location determining method of cheek parts of images is：If the width of face region is w, height For h, then the width range of cheek region is 1/8w～3/8w, 5/8w～7/8w, altitude range 1/2h～3/4h, the region The region of nose is probably omitted, only retains two cheek regions, reduce data interference.

3rd, the average of Green passages is calculated cheek parts of images, equal value sequence, the datagram ginseng of equal value sequence is obtained See Fig. 4.

Calculate the average of Green channel components in cheek region image, that is, the Green passages for adding up in each pixel point Amount, then be obtained divided by the pixel quantity of image.In order to improve processing speed, the mode of interval sampling can be taken, i.e., Odd-numbered line takes even number column data, and even number line takes the data of odd column, it is ensured that while covering, gets overall situation of change.

When processing data, in order to reduce the impact that baseline drift brings, the data to being currently needed for processing all are entered Row normalized, takes the processing method of zero-mean, i.e., all of data all deduct data mean value as pending data.

Because different filtering intervals can respectively obtain changes in heart rate curve and change of respiratory rate curve, below to two The detailed process of individual change curve is described in detail respectively.

4th, moving average value filtering is carried out to equal value sequence, is 4～6 by adjusting its filtering interval, obtain changes in heart rate Curve.As shown in Figure 5.

With regard to digital filtering, there are a variety of methods, such as arithmetic equal value filtering, weighted mean filter, medium filtering etc.. Sliding average filter method is adopted herein, because comparing other Mean Filtering Algorithms, often calculating an efficiently sampling value must be even Continuous sampling n times, for sample rate is relatively slow or requires the higher real-time system of data computation rate, these methods are to make .And sliding average filtering method only needs to consider continuous N number of data, it is possible to obtain new filter value.Slide flat Mean filter method has good inhibitory action to PERIODIC INTERFERENCE, and smoothness is high, and sensitivity is low；But to the accidental arteries and veins for occurring The inhibitory action of punching property interference is poor, is difficult the deviation for eliminating the sampled value that impulse disturbances cause.Therefore impulse disturbances are not suitable for it Than more serious occasion.

5th, feature extraction is carried out to changes in heart rate curve and obtains heart rate value.

In common noncontact PPG methods, after PPG signals are collected, need to enter PPG signals by amplifying circuit Row amplifies, and then using low-pass filter circuit and high-pass filtering circuit operation is filtered successively, then using the trap of 50hz Circuit for eliminating Hz noise, then the signal just to finally giving carry out power spectral analysis computing (method be shown in spectrum analysis side Method).Just include in the PPG signals of script and compare clearly heart rate, breath signal, why common methods are not but using most straight The temporal analysis (i.e. direct detection maximum, the method for calculating the peak separation time) for connecing, and frequency-domain calculations method is adopted, It is less susceptible to recognize noise and real heart rate signal mainly due to time domain approach, so just needing to enter again by repeatedly filtering Line frequency analysis of spectrum.And the present embodiment methods described can realize judging noise by temporal analysis, thus no longer need into The multiple filtering of row and complicated frequency spectrum conversion, so as to lift the degree of accuracy and the operational efficiency of rate calculation.

Referring to Fig. 2, and Fig. 6 is combined, the present embodiment carries out feature extraction using following fisrt feature extraction algorithm：

Because the signal for collecting is exactly data Y with time shaft as coordinate_i, therefore it is special first to find out the waveform of signal Levy, such as crest/trough/peak separation etc..Step is as follows：

Step 1, according to sequence to be calculated, from first point, find first downward flex point of waveform as setting out Point, i.e. Y_i>Y_i+1(Y is sequential digit values, i=1,2,3 ...), if Y₁>Y₂(first point may not be peak), then first find ripple Shape flex point upwards, i.e. Y_i<Y_i+1(i=1,2,3 ...), is further continued for looking for afterwards the downward flex point of first waveform as starting point, That is the O points in Fig. 6, mark O points are first crest.

Step 2, from O points, find next waveform flex point upwards, i.e. Y_i<Y_i+1(i=1,2,3 ...), i.e. Fig. 6 In 1 point, 1 point of mark is first trough.

Step 3, from 1 point, constantly find the downward flex point of waveform, and be labeled as crest, then draw crest sequence Row O, A, B, C, D, E ..., and trough sequence 1,2,3,4,5 ... (as shown in Figure 6), have at this moment shown that the waveform of signal is special Levy, feature calculation can be carried out according to formula：

Heart rate hbr=head and the tail are peak-to-peak every the time difference/peak separation quantity.

Can run into some noise points during crest sequence is found, such as between two crests of C and D, J and K two Between crest, can all there is the noise of doubtful crest, need to filter out.At this moment can by find scope R in peak come Spurious peaks are filtered out, when a crest is found, it is assumed that the spurious peaks between C and D points are C1, are found backward within R from C1 Maximum, if peak value not high than C1 in the range of R, C1 continues the search of next crest as real crest；Such as Fruit have found the peak value D higher than C1, then D continues the search of next crest as real crest.Here set between J and K Spurious peaks are J1, and when crest J is found, continuation has searched J1 in the range of R backward, so J1 is considered as spurious peaks, are needed Continually look for real crest K.

Noise data is filtered out in order to obtain optimal R values, accurate heart rate is obtained.Enter here by Medical Devices Row contrast, through mass data analysis, by cycle calculations different R values is arranged, and has been obtained and R during Medical Devices same heart rate Scope, concrete numerical value such as table 1：

The relation of the heart rate range of table 1 and R intervals

Heart rate range	30～60	60～80	80～110	110～150	150～220
						R intervals	R1	R1～R2	R2～R3	R3～R4	R4

In by table 1,4 numerical value R1-R4 are obtained, by arranging the interval of R from R1～R4, human normal can be covered substantially The range intervals of heart rate.The method can also be used in the denoising of other any waveforms.With it, directly filter much making an uproar Sound data, reduce calculation error.

As waveform such as Fig. 7, due to basic without noise, the change of R does not interfere with result of calculation, but if the number of monitoring Value contains more noise (spurious peaks), will obtain different 4 heart rate data, it is therefore desirable to determine by additive method Accurate heart rate.Because the heart rate of normal person is hole rule heart rate, therefore every time the amplitude of variation of eartbeat interval is not in Too big change, and if adding or miss out spurious peaks, this amplitude of variation exception is inevitably resulted in, for this purpose, we draw Enter function：

F=sd*sd/avg；

To calculate the fluctuation situation of Current heart rate, wherein sd is the mark of computation interval eartbeat interval (corrugation pitch) each time It is accurate poor, avg for computation interval heartbeat equispaced (corrugation pitch average), emotionally condition is minimum is then closest to very for average wave Real HR values.

Certainly, the computing interval certain situation also occurs, such as the corresponding sequence number of the minimum of a value got every time is continually changing, How to judge whether current numerical value is abnormal, and the numerical value for how judging heart rate has been stablized, and concrete processing method is as follows：

Take that the time span of 10 seconds is interval as rate calculation, and increase the HR values in last 5 seconds as right Than reference, the numerical value of wherein heart rate is denoted as hbr, and the HR values for calculating for last 5 seconds are denoted as hbr5, and hbr5 can also regard the heart as The trend of rate change, if the unexpected fluctuation of appearance or ectocine, hbr5 and hbr has obvious numerical value difference.

Step 0, initialization lastchoice (the heart rate waveform sequence number that the last time selects), count (gets the tired of heart rate Metering number), e (cumulative number of the Current heart rate waveform sequence number as heart rate), retrytime (recalculating number of times) they are 0；

Step 1, according to 4 different filtering intervals R (R1～R4) for arranging, calculate heart rate hbr, hrv5 and fluctuation situation F, and record storage, while cumulative count is (in count<All it is to take the corresponding sequence datas of Min fi to calculate heart rate value when 5)；

If Step 2, count>=5, then calculateAs average fluctuation；

Step 3, g=Min gi (i=1～n) are taken, and record corresponding i as the alternative of current optimal heart rate

If lastchoice=i, e=e+1, e is made to be not more than 5；

If lastchoice is not equal to i, e=e-1, e is made to be not less than 0；

Step 4, difference d for calculating hbr and hbr5, during heart rate stabilization, the numerical value of d is not over 3.

If the f of Step 5, Current heart rate>10 (empirical values that many experiments draw), then illustrate current by extraneous dry Disturb than larger, or the heart rate of collection has abnormal conditions.

At this moment judge：If f<=g or f<=10, then perform Step 7；Otherwise perform Step 6；；

Step 6, the data for abandoning current sequence of calculation the first half, make lastchoice, count, e be 0, order Retrytime=retrytime+1；

If retrytime=2, total data is abandoned, make lastchoice, count, e be 0；

If Step 7, difference d of continuous 5 times<3, illustrate that rate calculation numerical value tends towards stability, at this moment judge the number for breathing Value whether calculated (numerical value of heart rate can tend towards stability for most fast 7 seconds, the frequency of breathing then because the cycle is longer, lead to Often at least needing the time of a respiratory cycle can just find out that waveform changes), and tend towards stability (because heart rate and breathing are same When calculate, considerations so the two is put together), if the numerical value of breathing is also stable, export heart rate and respiratory rate numerical value, hold Row Step 8.

Step 8, stopping data acquisition, terminate monitoring process.

It is also possible to pass through this method to carry out vivo identification (such as picture or image etc.), because the light in the external world Line reflection is substantially irregular, if so excessive (f of fluctuation situation for drawing>10, f under normal circumstances<=5), then can sentence Disconnected current light environment is undesirable, or tested can not provide normal rule heart rate.

In laboratory environments, the HR values that the method is calculated are basic identical with the equipment that Medical Devices draw.

Respiratory rate calculating process is illustrated below.

6th, moving average value filtering is carried out to equal value sequence, is 28～32 by adjusting its filtering interval, obtain breathing frequency Rate change curve.As shown in Figure 8.

5th, feature extraction is carried out to change of respiratory rate curve and obtains respiratory rate value.

Referring to Fig. 3, and Fig. 9 is combined, the method that the present embodiment calculates respiratory rate is identical with the method for calculating heart rate, but Due to respiratory rate it is different with heart rate feature, so here the setting of r is real-time change.The step of second feature extraction algorithm It is as follows：Step 1, according to sequence to be calculated, from first point, find first downward flex point of waveform as starting point, That is Y_i>Y_i+1(Y is sequential digit values, i=1,2,3 ...), if Y₁>Y₂(first point may not be peak), then first find waveform Flex point upwards, i.e. Y_i<Y_i+1(i=1,2,3 ...), is further continued for looking for afterwards the downward flex point of first waveform as starting point, i.e., O points in Fig. 9, mark O points are first crest.

Step 2, from O points, find next waveform flex point upwards, i.e. Y_i<Y_i+1(i=1,2,3 ...), i.e. Fig. 9 In 1 point, 1 point of mark is first trough.

Step 3, from 1 point, constantly find the downward flex point of waveform, and be labeled as crest, then draw crest sequence Row O, A, B, C, D, E ..., and trough sequence 1,2,3,4,5 ... (as shown in Figure 9), have at this moment shown that the waveform of signal is special Levy, feature calculation, such as peak to peak separation etc. can be carried out.

Step 4, find crest during can run into some noise points, such as at this moment spurious peaks A1 between A and B need R is set according to time interval d between its respiratory intervals for drawing before, i.e. O and A, the scope of r is d/2.Due in blood An oxygen whole content change necessarily relatively slow process, so the value of r should be according between front breathing twice Real-time adjustment is done every d and d1.

R=(d+d1)/2

Step 5, due to the particularity of respiratory rate, after noise data is filtered out, can directly pass through formula：

Respiratory rate br=head and the tail are peak-to-peak every the time difference/peak separation quantity

Current respiratory rate is calculated, respiratory rate has hysteresis quality, and lag time is the one of front respiration duration Half time.Under normal circumstances, once complete waveform can judge the general frequency for breathing.In laboratory environments, the party Method calculate breathing numerical value and actual numerical value error within positive and negative 2, substantially conform to tested actual conditions.

The present embodiment algorithm stability is strong, by different illumination intensity, face's pickup area, collection distance test, Compared with existing medical equipment in hospital result of calculation, it is not easy to disturbed by outside various conditions, with extremely strong applicability.

Statistical analysis is carried out by gathering substantial amounts of data sample, following result is drawn：

(1) accuracy of inventive algorithm --- contrast (correlation analysis and regression analysis) with Medical Devices

[data are Medical Devices test and the present invention under different time (15s, 60s) difference light intensity (high, medium and low) All data of algorithm]

The present invention of table 2 is contrasted with the correlation of Medical Devices

Pearson product moment correlation result：Correlation is notable in 0.01 bilateral level.Coefficient correlation is 0.906.

With the present invention as dependent variable, constant, medical treatment are the model that predictive variable is set up to table 3

The Regression Analysis Result of table 4

Regression Analysis Result：Return significantly, regression coefficient is 0.906.The error estimated by two above result standard 2.92393 and regression coefficient 0.906 understand that inventive algorithm result is close with authoritative Medical Devices test result, it is seen that its standard True property.

(2) impact (two analysis of variance) that the time determines with light intensity to the present invention

The inspection of effect between the main body of table 5

Dependent variable:The present invention

A.R side=.019 (adjustment R side=- .033)

The result of multiple comparisons of table 6

Dependent variable:The present invention

Based on the average for observing.

Error term is average side (mistake)=43.948.

By above-mentioned variance analysis, main effect analysis shows with the result of multiple comparisons, as a result very not significantly, illustrates difference Selecting time and different light intensities do not affect on measurement result.

Embodiment 2

The monitoring device of the physiological parameter described in the present embodiment, corresponding to each method and step described in embodiment 1, each The specific works of module are referring to described in embodiment 1.Here repeat no more.

A kind of monitoring device of physiological parameter, including

Image capture module, for gathering current face's image；

It is m/3～2m/3 that the filtration module adjustment filtering is interval, and m is the frame per second of current gathered data, obtains heart rate Change curve；The analysis and processing module includes fisrt feature extraction module, and fisrt feature extraction module includes：

First data read module, for finding the position of first crest；

Crest sequence preserving module includes several sub- preserving modules of crest sequence, and heart rate 30～220 is divided into into n area Between, correspondence n R value of setting is respectively adopted each R value and asks for crest sequence, crest sequence is stored in respectively and preserves mould In block；For each crest sequence, using following formula f values are calculated：

f_i=sd_i*sd_i/avg_i；

Wherein, i=1 ... ..., n；sd_iFor the standard deviation of eartbeat interval, avg_iTo calculate the equispaced of eartbeat interval, take The corresponding i of Min fi calculate heart rate value as the current sequence for selecting using the sequence data.

The monitoring device includes data screening module, and the data screening module includes：

Initialization module, for initializing heart rate waveform sequence number lastchoice of last selection, gets the tired of heart rate Metering number count, Current heart rate waveform sequence number recalculates number of times retrytime as the cumulative number e of heart rate, is entered as 0；According to n different filtering interval R (R1～Rn) for arranging, heart rate hbr, hbr L1 and fluctuation situation f are calculated, and record is deposited Storage, while cumulative count is (in count<When 5, take the corresponding sequence datas of Min fi and calculate heart rate value)；

Average fluctuation computing module, for count>=5, calculateAs average fluctuation；

Alternate data module, takes g=Mingi (i=1～n), and records corresponding i as current optimal heart rate Alternatively.If lastchoice=0 (is not recorded for the first time), lastchoice=i, e=1 are made；If lastchoice =i, then make e=e+1, e be not more than 5；If lastchoice is not equal to i, e=e-1, e is made to be not less than 0；

Difference calculating module, for calculating difference d of hbr and hbr L1, during heart rate stabilization, the numerical value of d is not over 3；

Second judge module, for judging f>G and f>Whether 10 set up, if set up, performs data removing module, Otherwise, the second judge module is performed；Also correspond to, judge f<=g or f<Whether=10 set up, if set up, performs the Three judge modules, otherwise, perform data removing module；

Data removing module, for abandoning the data of current sequence of calculation the first half, makes lastchoice, count, e be 0, make retrytime=retrytime+1；If retrytime=2, abandon total data, make lastchoice, Count, e are 0；If retrytime=3, prompting is provided, preset test environment is undesirable, monitoring failure is performed and terminates mould Block；

3rd judge module, for whether judging difference d of continuous L2 calculating less than 3, if it is, continue judgement exhaling Whether the numerical value of suction has calculated and has tended towards stability, if it is, exporting heart rate and respiratory rate numerical value, execution terminates Module；

Terminate module, for stopping data acquisition, terminates monitoring process.

It is 3m/2～2m that the filtration module adjustment filtering is interval, and m is the frame per second of current gathered data, obtains breathing frequency Rate change curve；The analysis and processing module includes second feature extraction module, and second feature extraction module includes：

The technology of present invention description can be implemented by various means.For example, these technologies may be implemented in hardware, consolidate In part, software or its combination.For hardware embodiments, processing module may be implemented in one or more special ICs (ASIC), digital signal processor (DSP), programmable logic device (PLD), field-programmable logic gate array (FPGA), place Reason device, controller, microcontroller, electronic installation, other be designed to perform the electronic unit of function described in the invention or In its combination.

For firmware and/or Software implementations, can be with module (for example, process, the step for performing functions described herein Suddenly, flow process etc.) implementing the technology.Firmware and/or software code are storable in memory and by computing device.Storage Device may be implemented in processor or outside processor.

One of ordinary skill in the art will appreciate that：Realizing all or part of step of said method embodiment can pass through Completing, aforesaid program can be stored in a computer read/write memory medium the related hardware of programmed instruction, the program Upon execution, the step of including said method embodiment is performed；And aforesaid storage medium includes：ROM, RAM, magnetic disc or light Disk etc. is various can be with the medium of store program codes.

Above-described embodiment is the present invention preferably embodiment, but embodiments of the present invention not by above-described embodiment Limit, other any Spirit Essences without departing from the present invention and the change, modification, replacement made under principle, combine, simplification, Equivalent substitute mode is should be, is included within protection scope of the present invention.

Claims

1. a kind of monitoring method of physiological parameter, it is characterised in that including step：

Collection current face's image；

Extract the cheek parts of images in facial image；

2. the monitoring method of physiological parameter according to claim 1, it is characterised in that the frame of the current gathered data of setting Rate is m, and the filtering interval for setting equal value sequence is m/3～2m/3, obtains changes in heart rate curve, is carried using following fisrt feature Taking algorithm carries out feature extraction：

Preferentially find the position of first crest；

Whether primary peak peak value is judged more than or equal to secondary peak peak value, if it is, determine that primary peak is current crest, so Start afterwards to continue search for from primary peak；If it is not, then determine that secondary peak is current crest, then from the beginning of secondary peak Continue search for；

After complete changes in heart rate curve of search, a crest sequence is obtained, described crest sequence includes the peak value of each crest And position, position i.e. eartbeat interval；

3. the monitoring method of physiological parameter according to claim 2, it is characterised in that heart rate 30～220 is divided into into n Interval, correspondence n R value of setting, is respectively adopted each R value and asks for crest sequence；For each crest sequence, using following public affairs Formula calculates f values：

f_i=sd_i*sd_i/avg_i；

Wherein, i=1 ... ..., n；sd_iFor the standard deviation of eartbeat interval, avg_iTo calculate the equispaced of eartbeat interval, Min is taken f_iCorresponding i calculates heart rate value as the current sequence for selecting using the sequence data.

4. the monitoring method of physiological parameter according to claim 1, it is characterised in that the frame of the current gathered data of setting Rate is m, set equal value sequence filtering it is interval be 3m/2～2m, obtain change of respiratory rate curve, it is special using following second Levying extraction algorithm carries out feature extraction：

Step 3, from the point of mark 1, constantly find the downward flex point of waveform, and be labeled as crest, then draw crest sequence Row O, A, B, C, D, E ..., and trough sequence 1,2,3,4,5 ..., draw the wave character of signal；

Draw current respiratory rate；

Following denoising method is adopted during crest is found：

Judge whether the time that the current crest found is located is more than last time peak time+r, if it is, being judged as new ripple Peak, if it is not, then being judged as noise spot.

5. the monitoring method of physiological parameter according to claim 1, it is characterised in that while to physiology change curve data It is filtered, respectively obtains changes in heart rate curve and change of respiratory rate curve, for above-mentioned data following screening is first carried out Operation, method is as follows：

The time span for taking the L seconds is interval as rate calculation, and the HR values increased in a last L1 second are joined as a comparison Examine, the numerical value of wherein heart rate is denoted as hbr, the HR values that the last L1 seconds calculate are denoted as hbr L1, calculate for continuous L2 time Changes in heart rate represents that the HR values for obtaining tend towards stability less than certain limit, p1, p2, p3, and p4 is empirical parameter value；

Step 0, last heart rate waveform sequence number lastchoice for selecting of initialization, get the cumulative number count of heart rate, Current heart rate waveform sequence number recalculates number of times retrytime as the cumulative number e of heart rate, is entered as 0；

Step 1, according to the n different filtering interval R for arranging, calculate heart rate hbr, hbr L1 and fluctuation situation f, and record and deposit Storage, while cumulative count, in count<During p1, take the corresponding sequence datas of Minfi and calculate heart rate value；

If Step 2, count>=p1, then calculateAs average fluctuation；

Step 3, take g=Min g_i, wherein i=1～n, and corresponding i is recorded as the alternative of current optimal heart rate：

If lastchoice=0, lastchoice=i, e=1 are made；

If lastchoice=i, e=e+1, e is made to be not more than p2；

If lastchoice is not equal to i, e=e-1, e is made to be not less than 0；

Step 4, difference d for calculating hbr and hbr L1；

If Step 5, f<=g or f<=p3, then perform Step 7；Otherwise perform Step 6；

Step 6, the data for abandoning current sequence of calculation the first half, make lastchoice, count, e be 0, make retrytime= retrytime+1；

If retrytime=2, total data is abandoned, make lastchoice, count, e be 0；

If Step 7, difference d of continuous L2 calculating<P4, at this moment judge breathe numerical value whether calculated and Tend towards stability, if the numerical value of breathing is also stable, export heart rate and respiratory rate numerical value, perform Step 8；

Step 8, stopping data acquisition, terminate monitoring process.

6. a kind of monitoring device of physiological parameter, it is characterised in that include：

Image capture module, for gathering current face's image；

Filtration module, for carrying out moving average value filtering to equal value sequence, by adjusting its filtering interval physiological change is obtained Curve；

7. the monitoring device of physiological parameter according to claim 6, it is characterised in that the filtration module adjustment filtering area Between be m/3～2m/3, m is the frame per second of current gathered data, obtains changes in heart rate curve；The analysis and processing module includes the One characteristic extracting module, fisrt feature extraction module includes：

First data read module, for finding the position of first crest；

Primary peak search module, for from the beginning of the position of previous crest, searching in the range of R thereafter highest first wave is worth Peak；

Secondary peak search module, for from the beginning of the position of primary peak, searching in the range of R thereafter the ripple of highest second is worth Peak；

First judge module, for whether judging primary peak more than or equal to secondary peak, if it is, determining that primary peak is to work as Front crest, then starts to continue search for from primary peak；If it is not, then determine that secondary peak is current crest, then from the Two crests start to continue search for；

Primary peak sequence preserving module, for searching for complete changes in heart rate curve after, obtain a crest sequence, described ripple Peak sequence includes the peak value and position, position i.e. eartbeat interval of each crest；

8. the monitoring device of physiological parameter according to claim 6, it is characterised in that the filtration module adjustment filtering area Between be 3m/2～2m, m is the frame per second of current gathered data, obtains change of respiratory rate curve；The analysis and processing module bag Second feature extraction module is included, second feature extraction module includes：

First crest seeking module, for the data sequence in change of respiratory rate curve, finds first wave crest point O Point；

Wave character determining module, for from the point of mark 1, constantly finding the downward flex point of waveform, and is labeled as crest, Then draw crest sequence O, A, B, C, D, E ..., and trough sequence 1,2,3,4,5 ..., draw the wave character of signal；

Respiratory rate computing module, it is peak-to-peak every the time difference and peak separation quantity for according to wave character, obtaining head and the tail, by public affairs Formula：Respiratory rate br=head and the tail are peak-to-peak every the time difference/peak separation quantity；Draw current respiratory rate；

The second feature extraction module also includes denoising module, and denoising module includes：

Step-size in search setting module, if peak to peak separation twice recently is respectively d and d1, calculates step-size in search r=(d+d1)/2；

Judge module, for whether judging the time that the crest of current searching is located more than last time peak time+r, if it is, It is judged as new crest, if it is not, then being judged as noise spot.

9. the monitoring device of physiological parameter according to claim 6, it is characterised in that the monitoring device includes data sieve Modeling block, the data screening module includes：

Initialization module, for initializing heart rate waveform sequence number lastchoice of last selection, get heart rate accumulative time Number count, Current heart rate waveform sequence number recalculates number of times retrytime, is entered as 0 as the cumulative number e of heart rate；Root According to n arranged different filtering interval R, heart rate hbr, hbr L1 and fluctuation situation f, and record storage are calculated, while cumulative count；

Second judge module, for judging f<=g or f<Whether=p3 sets up, if set up, performs the 3rd judge module, Otherwise, data removing module is performed；

Data removing module, for abandoning the data of current sequence of calculation the first half, initialization lastchoice, count, e are 0, make retrytime=retrytime+1；If retrytime=2, total data is abandoned, initialization lastchoice, Count, e are 0；If retrytime==3, prompting is provided, preset test environment is undesirable, monitoring failure, execution terminates Module；

3rd judge module, for judging that difference d of continuous L2 calculating, whether less than p4, then shows what is obtained if less than p4 HR values tend towards stability, then continuation judges whether the numerical value for breathing has calculated and tended towards stability, if it is, Output heart rate and respiratory rate numerical value, perform terminate module；

Terminate module, for stopping data acquisition, terminates monitoring process.

10. handheld device, it is characterised in that including the monitoring device of the physiological parameter described in any one of claim 6-9.