Specific embodiment
Below in conjunction with attached drawing in the embodiment of the present invention, technical solution in the embodiment of the present invention carries out clear, complete
Ground description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Usually exist
The component of the embodiment of the present invention described and illustrated in attached drawing can be arranged and be designed with a variety of different configurations herein.Cause
This, is not intended to limit claimed invention to the detailed description of the embodiment of the present invention provided in the accompanying drawings below
Range, but it is merely representative of selected embodiment of the invention.Based on the embodiment of the present invention, those skilled in the art are not doing
Every other embodiment obtained under the premise of creative work out, shall fall within the protection scope of the present invention.
Referring to Fig. 1, the contactless physiologic information detection method provided in the present embodiment, comprising the following steps:
Step 1, facial video image is acquired.For the ease of the processing of subsequent step, it is preferred to use the higher camera shooting of pixel
Machine carries out facial video image acquisition.
Step 2, region of interest ROI is chosen from the facial video image of acquisition.
Step 3, it is based on the region of interest ROI, obtains tri- channel time sequences of RGB, i.e., is obtained respectively red logical
Road time series data, green channel time series data, blue channel time series data.Each frame video image be all by
RGB three primary colors composition, three primary colors constitute the data of triple channel, as soon as such as second have 30 frame images, that minute has 3 × 1800
A data, so that it may be known as triple channel time series.
Step 4, tri- channel time sequence datas of the RGB of acquisition are pre-processed, reject the pixel of doubtful noise,
And rgb signal is transformed into the space CHROM, obtain the PPG signal of CHROMization.
Step 5, the PPG signal of CHROMization is decomposed using EEMD, generates intrinsic mode function IMFS, abandon and belong to noise
IMF, and by remaining IMFSDimensionality reduction is carried out through principal component analysis PCA technology.
Step 6, heart rate HR and respiratory rate RR is calculated using AR model.
More specifically, referring to Fig. 2, choosing area-of-interest from the facial video image of acquisition in above-mentioned steps 2
The specific operation process of ROI is as follows:
Step 21, face is detected from collected facial video image.Herein preferably using by Shenzhen technical research
MTCNN provided by advanced technology research institute (Multi-task convolutional neural networks) algorithm is examined
Face is surveyed, because MTCNN has higher accuracy and robustness compared with other currently a popular method for detecting human face.
Step 22, using CE-CLM (Convolutional Experts Constrained Local Model) method
Form 68 human face characteristic points, this method carries out the calculating of response diagram and the update of form parameter using convolution network of experts.
The region near the wing of nose is selected to can be obtained by the estimation of hrv parameter again, but wing of nose region is easy to be influenced by expression shape change
To generate noise, so to reject the region comprising the corners of the mouth and eyes, and selects and make comprising the region of nose and forehead together
For facial ROI, as shown in Figure 3.In Fig. 3, small circle indicates that the Partial Feature point chosen, irregular frame area domain representation are chosen
Rough ROI, rectangle wire region indicates finally selected facial ROI.
It should be noted that human face characteristic point is formed using CE-CLM algorithm in this step, so what is formed is
68 characteristic points are then not necessarily to form 68 characteristic points according to other methods.The seat of characteristic point in the facial ROI of selection
It is denoted as the coordinate for ROI.In addition, CE-CLM algorithm is the prior art, so not just being unfolded to do careful explain to save space herein
It states.In addition, the selection for area-of-interest, in the application for a patent for invention that also may refer to Publication No. CN109589101A
Associated description.
More specifically, as shown in figure 4, being located in advance in above-mentioned steps 4 to the RGB triple channel time series data of acquisition
Reason, is converted to CHROM spacing wave for each Color Channel time series data, specific operation process is as follows:
Step 41, smooth to RGB triple channel time series data 5 points of overlapping sliding windows of progress of acquisition, more put down
Sliding new data set.As shown in figure 5 a and 5b, Fig. 5 a is raw data set, and Fig. 5 b is the smooth new data set obtained later.
When 5 points of overlapping sliding windows are smooth herein, be by the 1st, 2,3,4,5 totally 5 points take the average value as third point;By the 2,3rd,
4,5,6 totally 5 points take it is average be used as the 4th point;And so on;It can obtain new 2,3,4..n-2 points, as the new the 1st,
2 and (n-1)th and n-th point of value take initial value.
It is readily comprehensible, it is only that use 5 points of overlapping sliding windows smooth herein, it of course, can also be using 3 points, 7 points etc.
Multiple spot overlapping sliding window is smooth, has only carried out 5 points of overlapping sliding window smoothing tests herein, and test effect is fine, what is obtained is new
Data set is smooth enough.
Step 42, it by smoothed out PPG signal bi-directional scaling, is allowed to be mapped to [0,1] section, removes the unit of data
Limitation, is converted into nondimensional pure values.New data set shown in Fig. 5 b is as shown in Figure 5 c after mapping.
Step 43, linear fit is carried out to the signal after mapping using linear least square, then from after mapping in signal
Resulting linear deflection amount after being fitted is subtracted, that realizes signal goes trending to handle.Data set shown in Fig. 5 c is through going at trend
After reason as fig 5d.
Step 44, in general, noise corresponds to the part that pixel value is excessive or too small in grey level histogram, has in image
Row in pixel value be significantly greater or less than typical values, this correspond to noise section.In order to will likely be noise (doubtful noise)
Pixel reject, need by pixel carry out rayleigh distributed matching, i.e.,
Wherein σ is the scale parameter of distribution.Choose the pixel between [0.5 σ, 1.5 σ].Data set shown in Fig. 5 d
After rayleigh distributed is screened as depicted in fig. 5e.The pixel that tri- channels RGB are chosen finally is subjected to space average as just again
Beginning signal.Pass through the method for pixel median sampling (rayleigh distributed screening), the initial signal of available very big conservative estimation, phase
The accuracy of subsequent processing is improved over the ground.
Step 45, the original one-dimensional signal in three channels, including danger signal y are obtained after have passed through intermediate value samplingR(t)、
Green channel yG(t), blue channel yB(t), it reuses CHROM method and carries out data reconstruction processing, it may be assumed that
Xraw(t)=1.5yR(t)+yG(t)-1.5yB(t) (2)
G.de Haan andV.Jeanne, " Robust can be referred to about CHROM method more detailed description
pulse rate from chrominance-based rPPG,”IEEE Trans.Biomed.Eng.,vol.60,no.10,
Pp.2878-2886, Oct.2013. as shown in figure 5f, choose the PPG signal of one section of video acquisition of subject, it can be seen that
The PPG signal of CHROM linear combination method building is more regular, and the frequency for meeting cardiopulmonary section on frequency domain is more prominent.So
In this step, RGB is transformed into the space CHROM, obtains the PPG signal X of CHROMizationraw(t)。
More specifically, as shown in fig. 6, decomposing CHROMization using EEMD (set empirical mode decomposition) in the step 5
PPG signal, generate intrinsic mode function IMFs operating process it is as follows:
Step 51, EEMD principle is that the true intrinsic mode function IMFs of data is seen as to the average value of test set,
White noise of each signal by original signal plus finite amplitude forms.I.e. an input signal x (t) can pass through EEMD
Technology is decomposed into N number of IMFs, and it is as follows that all IMFs and residual error can reconstruct input signal:
Therefore according to EEMD principle, original PPG signal x (t) can be decomposed are as follows:
1) x is generatedj=x+ α ωj, the white noise of the zero mean unit variance of j ∈ (1,2 ..., n), the amplitude of white noise is big
Small is α=0.2.
2) x is countedjThe EEMD of (j=1 ... n) obtains intrinsic modeWherein k=1 ... K indicates mode.
3) willAs the kth rank mode of x (t), by will be correspondingIt is average to obtain:
All IMF signals after can obtaining original PPG signal decomposition.Original PPG signal in this step refers to
The PPG signal X of CHROMization obtained in above-mentioned steps 45raw(t)。
Step 52, it after obtaining all intrinsic mode function IMFs, needs to abandon the IMFs for belonging to noise.PPG signal master
Will be by the modulation of cardiopulmonary frequency, and frequency range can be set between (0.1Hz~0.7Hz, 0.7Hz~3Hz).In order to identify
Artifact determines basic frequency to each IMFs application Fast Fourier Transform (FFT) (FFT), that is, obtains the frequency of amplitude peak.Once
The basic frequency of each IMFs is obtained, certain IMFs for belonging to noise can be abandoned according to the frequency range of setting, it then again will symbol
The IMFs for closing the frequency range (0.7Hz~3Hz) of heart rate is included into HR- group, and meets respiratory rate range (0.1Hz~0.7Hz)
IMFs be included into RR- group.As shown in figure 4, IMF4 and IMF5 are included into HR- group, and IMF6 and IMF7 are included into RR- group.
Step 53, the IMFs application principal component analysis PCA technology of HR- group and RR- group is handled respectively, i.e., by just
The change commanders IMFs of existing multiple groups correlation of alternation is converted to some linear incoherent variables, this group of variable after conversion, which is named, to be led
Ingredient PCs.PCs is ranked up, first PCs maintains largely to be changed present in selected IMFs.
Therefore, the main activities of heart frequency are represent using first PCs that PCA is obtained on the IMFs in HR- group,
Equally, first PCs of RR- group represents the main activities of respiratory rate, as shown in Figure 7.
More specifically, as shown in figure 8, calculating the specific of heart rate HR and respiratory rate RR using AR model in above-mentioned steps 6
It operates as follows:
Step 61, after obtaining the respective principal component PCs of breathing and heart rate, it is converted into P rank AR model.
Assuming that the value of current sample x (n) be p of x (n) before value and q of Disturbance e (n) before value
Linear combination, wherein e (n) be white Gaussian noise distribution, then be AR model.Here x (n) is the sampled value of PPG time-domain signal
(the respective principal component PCs of breathing and heart rate), wherein n is hits.I.e. are as follows:
In formula, x (n) is the linear regression to itself preceding value, i.e. k from 1 to p (model order) summation, ak(k=1,
2 ..., p) be auto-regressive equation model undetermined coefficient.εnIt is the error returned, if P x (n) of estimation, can be write as
Matrix form, it is more convenient in this way to calculate least square solution, i.e., are as follows:
In formula, a aptIt is obtained most for orthogonality principle is applied to least square method
Excellent predictive coefficient, makes column vectorεIt is orthogonal toXEach column vectorx i, i=1,2 ..., P, and make mean square errorεIt minimizes.εWithXIt is
It is independent, multiply together in both membersXTranspositionX T, it may be assumed that
In order to acquire optimum prediction coefficienta opt, continue to be multiplied by both sidesX T XIt is inverse (X T X)-1, it may be assumed that
(X T X)-1(X T X)a opt=Ia opt=a opt=(X T X)-1 X T x (12)
It is this directly to be asked with least square solutiona optMethod be called covariance method.And this new matrixX T XWithX T xIt is by having
It is made of the sum of the auto-correlation function of Unequal time lag, it is possible to be approximately:
It in other words, can be by calculating the auto-correlation between sample come approximate generation when the time series of given certain length
The auto-correlation of table entirety.Likewise,X T xIt can also be indicated with auto-correlation vector:
It is available in conjunction with (10) and (11):
a opt=R -1 r (15)
And this equation be referred to as ' Yule-Walker ' equation, can by ' Levinson-Durbin ' recursive operator come
It solves.P rank AR model can be obtained finally by (9) or (12),Wherein: c is constant
?;εtIt is assumed to be the random error value that average is equal to σ equal to 0, standard deviation;It is assumed to be all constant for any t.
Step 62, frequency domain information is converted by time series.
It, can be by transform by analysis time sequence in AR (p) model conversation to complex plane after obtaining AR (p) model
Frequency domain information.The variation of description frequency is gone with pole, i.e., estimates the different frequencies of time series by calculating the position of pole
Rate component.In the domain Z, the pole on axis corresponds to the spectral peak of time series signal.The frequency f at each peak and corresponding pole
The relationship of angular frequency θ:
The π f Δ t of θ=2 (16)
In formula, Δ t is the sampling interval, and θ is the angular frequency indicated with radian.
Step 63, after PPG time series signal being converted into spectrum analysis, the corresponding pole of corresponding order has been obtained.Root
It was found that heart rate is corresponding to amplitude response maximum one in frequency range, i.e., it is corresponding near that pole of unit circle
Frequency, this is because the pole outside unit circle is unstable, and only in unit circle corresponding to the maximum pole of amplitude response
Frequency be only the frequency component that performance is most strong within the scope of this, that is, correspond to heart rate.And respiratory rate corresponds to frequency model
Interior angular frequency the smallest one and same reason are enclosed, has only carried out down-sampled step again.So for heart rate pole
The selection of point selects phase angle to correspond to the maximum pole of amplitude response in [0, π].And for the selection of poles of respiratory rate,
Amplitude is that the smallest pole of angular frequency is chosen in 90% or more pole of maximum amplitude.The AR mould of heart rate and respiratory rate estimation
Shape parameter is as shown in table 1 below:
Table 1
Heart rate frequency domain value corresponding with respiratory rate brings following formula into, then heart rate and respiratory rate can be estimated respectively are as follows:
HR=fhr·60 (17)
RR=fbr·60 (18)
F represent be the corresponding frequency of current frequency domain section peak-peak, due to calculate frequency be 1 second, heartbeat HR with exhale
Inhaling RR was calculated by minute, 1 second * 60 times=1min, it is therefore desirable to which frequency f is obtained into conventional heartbeat/breathing multiplied by 60
Metering method.
In the above method, the filtering of environment light is simplified, the accurate selection and application CHROM method for focusing on ROI can be with
Motion artifacts are played with certain inhibiting effect, the intrinsic mode function for obviously belonging to noise is then rejected using EEMD, and
PCA has then further refined the selection of cardiopulmonary frequency, and then AR method can be selected in the noise frequency close with cardiopulmonary frequency
Select optimal estimation frequency, to reduce illumination effect, therefore accuracy rate compared to currently used FFT, ICA, it is simple from
It is all high to return AR etc..The mark structure of distinct methods is as shown in table 2 below:
Table 2
In table 2, " Our " indicates the present embodiment the method, and * shows that correlation has statistics meaning in p=0.05 level
Justice.
In the above method, after having chosen region of interest ROI, the method for having used CHROM makes to be that signal more tends to advise
The waveform of rule, makes frequency distribution more have discrimination, convenient for prominent respiratory rate and palmic rate.EEMD method is reused
Signal is decomposed, makes signal decomposition into the waveform of multiple and different frequencies, is more conducive to be chosen at respiratory rate and signal frequency
The waveform of surrounding, conducive to the calculating of HR and BR, therefore, by the physiologic information that the above method of the present invention detects, accuracy is more
It is high.
Referring to Fig. 9, being based on identical inventive concept, a kind of contactless physiologic information is provided in the present embodiment simultaneously
Detection system, comprising: ROI selection module, time series acquisition module, preprocessing module, EEMD decomposing module, physiologic information are true
Cover half block.
Wherein, ROI chooses module for choosing region of interest ROI from collected facial video image.
Time series obtains module and is used to be based on the facial video image, obtains RGB triple channel time series.
Preprocessing module rejects doubtful noise for pre-processing to the RGB triple channel time series data of acquisition
Pixel, and rgb signal is transformed into the space CHROM, obtain the PPG signal of CHROMization.
More specifically, as shown in Figure 10, preprocessing module includes: sliding submodule, and mapping submodule goes trend submodule
Block, Rayleigh screen submodule, data reconstruction submodule.Wherein, sliding submodule is used for tri- channel time sequences of RGB to acquisition
It is smooth that column data carries out sliding window respectively, obtains more smooth new data set;Mapping submodule is used for PPG signal in proportion
Scaling, is allowed to be mapped to [0,1] section;Go trend submodule for using linear least square to after mapping signal progress
Linear fit, then the resulting linear deflection amount after subtracting fitting after mapping in signal, that realizes signal goes trending to handle;It is auspicious
Treated for will remove trending that pixel carries out rayleigh distributed matching for benefit screening submodule, and screening obtains the original in three channels
Beginning one-dimensional signal;Data reconstruction submodule is for counting the original one-dimensional signal in three channels using CHROM method
According to reconstruct, the PPG signal of CHROMization is obtained.
EEMD decomposing module is used to decompose the PPG signal of CHROMization using EEMD, generates intrinsic mode function IMFS, lose
Abandon the IMF for belonging to noiseS, and by remaining IMFSThrough principal component analysis PCA method carry out dimension-reduction treatment, respectively obtain breathing and
The principal component of heart rate.
More specifically, as shown in figure 11, EEMD decomposing module includes: decomposition submodule, for decomposing CHROM using EEMD
The PPG signal of change obtains several and generates intrinsic mode function IMFS;Noise remove submodule, for being applied to each IMFs
Fast Fourier Transform (FFT) FFT determines basic frequency, and according to the range of preset heart rate and respiratory rate, discarding belongs to noise
IMFS;Principal component analysis submodule is used for remaining IMFSDimension-reduction treatment is carried out through principal component analysis PCA method, is respectively obtained
The principal component of respiratory rate and heart rate.
Physiologic information determining module, for heart rate HR and respiratory rate RR to be calculated using AR model.
Since contactless physiologic information detection system right and wrong contact physiologic information detection method is based on identical hair
Bright design, therefore, place not described herein may refer to the associated description in preceding method embodiment.
As shown in figure 12, the present embodiment provides a kind of electronic equipment simultaneously, which may include processor 51
With memory 52, wherein memory 52 is coupled to processor 51.It is worth noting that, the figure is exemplary, can also use
The structure is supplemented or substituted to other kinds of structure, realizes data extraction, report generation, communication or other function.
As shown in figure 12, which can also include: input unit 53, display unit 54 and power supply 55.It is worth note
Meaning, the electronic equipment are also not necessary to include all components shown in Figure 12.In addition, electronic equipment can also wrap
The component being not shown in Figure 12 is included, the prior art can be referred to.
Processor 51 is sometimes referred to as controller or operational controls, may include microprocessor or other processor devices and/
Or logic device, the processor 51 receive the operation of all parts of input and controlling electronic devices.
Wherein, memory 52 for example can be buffer, flash memory, hard disk driver, removable medium, volatile memory, it is non-easily
The property lost one of memory or other appropriate devices or a variety of, can store configuration information, the processor 51 of above-mentioned processor 51
The instruction of execution, record the information such as list data.Processor 51 can execute the program of the storage of memory 52, to realize information
Storage or processing etc..It in one embodiment, further include buffer storage in memory 52, i.e. buffer, with the intermediate letter of storage
Breath.
Input unit 53 is for example for providing each text report to processor 51.Display unit 54 is processed for showing
It is various as a result, the display unit can be for example LCD display in journey, but the present invention is not limited thereto.Power supply 55 is for being
Electronic equipment provides electric power.
The embodiment of the present invention also provides a kind of computer-readable instruction, wherein when executing described instruction in the electronic device
When, described program makes electronic equipment execute the operating procedure that the method for the present invention is included.
The embodiment of the present invention also provides a kind of storage medium for being stored with computer-readable instruction, wherein the computer can
Reading instruction makes electronic equipment execute the operating procedure that the method for the present invention is included.
Those of ordinary skill in the art may be aware that list described in conjunction with the examples disclosed in the embodiments of the present disclosure
Member and algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware
With the interchangeability of software, each exemplary composition and step are generally described according to function in the above description.This
A little functions are implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Specially
Industry technical staff can use different methods to achieve the described function each specific application, but this realization is not
It is considered as beyond the scope of this invention.
If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product
When, it can store in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially
The all or part of the part that contributes to existing technology or the technical solution can be in the form of software products in other words
It embodies, which is stored in a storage medium, including some instructions are used so that a computer
Equipment (can be personal computer, server or the network equipment etc.) executes the complete of each embodiment the method for the present invention
Portion or part steps.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only
Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. are various can store journey
The medium of sequence code.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.