A method and device for identity-recognizable respiratory monitoring
By using bandpass filtering and generating an adversarial network to identify impulse responses generated by breathing in a dynamic environment, the problem of inability to identify breathing signals and voice privacy leakage in a dynamic environment is solved in the prior art, and high-precision breathing monitoring and identity are achieved Identify.
Patent Information
- Application Number
- CN202210072386.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-01-21
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2042-01-21
AI Technical Summary
Existing non-contact breath monitoring methods based on sound waves cannot effectively identify target breath signals in dynamic environments, and there is a risk of user voice privacy exposure, so it is impossible to achieve non-private violation identity-identifiable breath monitoring.
By sending transmission signals and receiving feedback signals, the voice signal and chest reflection signals are extracted, the voice audio band is filtered out using bandpass filtering, and the frequency and power characteristics are extracted through short-time Fourier transform, and combined with the Generative Adversarial Network (GAN) to identify breathing generation The impulse response is to realize identity identification and breathing rate monitoring, protect user privacy.
High-precision breathing rate monitoring is achieved in a dynamic environment, solving the problem of voice privacy leakage, improving the recognition accuracy in multiple users, and realizing contactless daily breathing monitoring.
Smart Images

Figure CN114617543B_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of respiratory monitoring technology, and specifically relates to an identity-identifiable respiratory monitoring method and device. Background Art
[0002] Respiratory rate is an important indicator of a person's health. Existing respiratory monitoring methods are mainly divided into contact and non-contact methods. Contact respiratory monitoring methods usually require users to wear specific measuring instruments on their body parts. The measuring equipment is usually expensive and inconvenient for users to use for daily monitoring. Non-contact respiratory monitoring methods mainly rely on wireless sensing technology or infrared image processing technology. Among them, using electromagnetic waves as transmission signals to sense changes in chest amplitude caused by breathing requires the setting of specific transceiver equipment, and the detection range of such wireless signals is wide and is easily interfered by the surrounding environment or other actions. Other methods use the camera of the mobile terminal to collect the user's image information and evaluate the user's respiratory parameters by analyzing the user's facial image or chest image. Such methods are easily restricted by environmental factors and involve user privacy issues.
[0003] In recent years, there have been many studies on non-contact respiratory monitoring methods using acoustic wave technology, and the implementation of acoustic wave technology often relies on the transmission and processing of two types of signals. One is based on the transmission and processing of frequency modulated continuous wave signals, extracting respiratory signals from the signal components in the distance range of the target user, but the distance resolution of frequency modulated continuous wave is low. If the distance resolution cannot be improved, high-precision respiratory monitoring cannot be achieved. The other is based on the transmission and processing of continuous wave signals, estimating the respiratory rate by analyzing the phase change of the signal. In the past, respiratory monitoring methods based on continuous wave signals were all aimed at static environments, that is, when the target user is stationary, the phase change of the signal is caused by breathing, which can achieve high-precision estimation of the respiratory rate. When the target is in motion, that is, in daily walking, the phase change of the transmitted signal is not only induced by breathing, but also by the large-scale movement of various parts of the body. The induction of multiple dynamic paths makes the phase change of the received signal no longer a trend of periodic breathing, but a trend of superposition of periodic and non-periodic signals. Therefore, compared with the previous respiratory monitoring method based on continuous wave signals, respiratory monitoring technology in non-static environments faces the challenge of being unable to effectively identify the target respiratory signal.
[0004] Sound waves are often implemented on smart devices, and smart devices have a variety of application scenarios. Applying identity-recognizable respiratory monitoring methods on smart devices can not only verify the true identity of the target user and prevent the forgery of health codes, but also obtain records of the respiratory rate under that identity. Existing identity recognition methods are mostly based on voice recognition, and voice recognition requires recording the user's voice signal, which poses a risk of exposing the user's voice privacy. Therefore, it is impossible to implement a non-privacy-invading identity-recognizable respiratory monitoring method. Summary of the invention
[0005] The purpose of the present application is to provide an identity-recognizable breathing monitoring method and device to avoid the shortcomings of existing sound wave-based non-contact breathing monitoring methods that cannot achieve identity-recognizable breathing monitoring and need to be achieved in a quiet environment but cannot be achieved when the user is moving.
[0006] In order to achieve the above purpose, the technical solution of this application is as follows:
[0007] An identity-recognizable breathing monitoring method, comprising:
[0008] Sending a transmission signal to a target object and receiving a feedback signal, wherein the feedback signal includes a speech signal and a chest reflection signal of the target object, extracting a speech segment from the feedback signal and then performing a bandpass filter to remove the speech segment to obtain a received signal;
[0009] Perform short-time Fourier transform on the received signal, extract frequency and power features, compare them with samples in the identity database, and identify the identity of the target object;
[0010] The correlation between the transmission signal and the receiving signal is calculated using a cross-correlation function, and the time sampling point with the maximum correlation is used as the time delay between the transmission signal and the receiving signal to synchronize the receiving signal and the transmission signal;
[0011] Performing fast Fourier transform on the received signal and the transmitted signal after synchronization processing, calculating the frequency response function, and performing inverse fast Fourier transform on the frequency response function to obtain the impulse response of the received signal;
[0012] Input the impulse response of the received signal into the trained generative adversarial network to extract the impulse response generated by breathing;
[0013] The impulse response generated by breathing is smoothed, and then the number of peak points is detected to obtain the breathing rate.
[0014] Furthermore, the frequency of the transmission signal is 18 kHz, and the passband frequency of the bandpass filter is 17.5 kHz-18.5 kHz.
[0015] Furthermore, when performing short-time Fourier transform on the received signal, the length of the input received signal ranges from 512 to 4800 sampling points.
[0016] Further, the extracting of frequency features and power features includes:
[0017] The received signal power characteristics are divided into four power intervals, and the centroid frequencies of each power interval are calculated as four characteristics of the power domain;
[0018] The frequency characteristics of the received signal are divided into five frequency intervals, and the average power of each frequency range is calculated as the five characteristics of the frequency domain.
[0019] Furthermore, the correlation between the transmission signal and the received signal is calculated using a cross-correlation function, and the cross-correlation function used is as follows:
[0020]
[0021] Among them, x * (n) represents the conjugate matrix of the transmission signal x(n), l represents the length of the transmission signal, n represents the sampling point of the transmission signal, m represents the number of offset sampling points of the received signal relative to the transmission signal, and R(m) represents the similarity between the transmission signal and the received signal when the received signal y(n) is shifted to the left by m sampling points.
[0022] The present application also proposes an identity-recognizable respiratory monitoring device, comprising a processor and a memory storing a plurality of computer instructions, wherein the computer instructions implement the steps of the identity-recognizable respiratory monitoring method when executed by the processor.
[0023] The present application proposes an identity-identifiable breathing monitoring method and device, which analyzes the Doppler frequency shift caused by the chest changes of the target user when speaking as a feature for identity recognition, and uses a generative adversarial network (GAN) to reduce the noise of the original pulse sequence and identify the pulse response generated by breathing, thereby realizing privacy-protected identity-identifiable breathing monitoring based on acoustic Doppler technology. The present application can filter out the signal in the voice frequency band, and only needs to use the Doppler frequency shift of the target chest activity to achieve identity recognition, solving the problem of user voice privacy infringement that is common in existing identity recognition methods. Breathing monitoring is mainly aimed at non-static environments, that is, the target user is in a daily walking state, and the existing audio equipment is used to monitor the breathing rate.
[0024] The beneficial effects of the present application are mainly reflected in: (1) It solves the potential risk of privacy infringement in existing identity recognition based on voice information, adopts acoustic Doppler technology to realize identity recognition, and has high recognition accuracy in the case of multiple users; (2) It solves the problem that previous respiratory monitoring methods are not suitable for dynamic environments, and is a non-contact daily respiratory monitoring method; (3) It can realize the identity recognition of the target user and the corresponding recording and tracking of the respiratory rate. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 A flow chart of an identity-recognizable respiratory monitoring method for this application;
[0026] Figure 2 A schematic diagram of transmitting and receiving signals according to an embodiment of the present application;
[0027] Figure 3 This is a schematic diagram of the identity identification process for this application;
[0028] Figure 4 Schematic diagram of the respiratory monitoring process for this application. DETAILED DESCRIPTION
[0029] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0030] The overall idea of this application is that the reflected signal of the sound wave signal contains a lot of available information, among which the breathing characteristics of the target user and the chest change characteristics during the speaking process can be obtained from the reflected signal at the chest. Compared with the traditional method, this application can protect the voice privacy of the target user by filtering out the signal in the voice band and extracting the high-frequency signal. The Doppler information contained in the high-frequency signal represents the frequency offset characteristics and power distribution characteristics of the signal, which can be used as a characteristic parameter to distinguish different targets. In addition, this application is also suitable for respiratory monitoring in a dynamic environment. The transmission signal reaches the receiving end via a variety of dynamic reflection paths in a dynamic environment. The reflected signals of other dynamic paths have a great interference with the reflected signals of the respiratory path, and the GAN network can eliminate the noise signal in the received signal. Therefore, the respiratory monitoring method based on the GAN network to eliminate noise can realize the estimation of respiratory rate in a dynamic environment. Combined with identity authentication and respiratory monitoring, this application can construct a pattern that matches identity and breathing, and can realize the tracking of respiratory change trends of the corresponding target.
[0031] In one embodiment, Figure 1 As shown, a method for identity-recognizable breathing monitoring is provided, comprising:
[0032] Step S1, sending a transmission signal to a target object and receiving a feedback signal, wherein the feedback signal includes a speech signal and a chest reflection signal of the target object, extracting a speech segment from the feedback signal and then performing a bandpass filter to remove the speech segment to obtain a received signal.
[0033] This application is implemented using a handheld smart terminal, using the speaker of the smart terminal to send a transmission signal (18kHz), and using the microphone of the smart terminal to receive a feedback signal. When sending a transmission signal, the speaker sends 48,000 transmission signal samples per second, and when receiving, the microphone also receives 48,000 samples per second. This application refers to the number of transmission signal samples sent per second as the sampling frequency. Sampling is the process of converting a signal into a numerical sequence, with 48,000 sampling points per second forming a digital signal. The digital-to-analog converter will convert the digital signal into an analog signal, which is sent out by the speaker and then received by the microphone. The digital-to-analog converter will convert the received analog signal into a digital signal in the form of 48,000 sampling points per second, and then perform subsequent processing. The above is explained using a sampling frequency of 48kHz as an example. The sampling frequency of this application is at least twice or more than the transmission signal frequency.
[0034] When in use, point the smart terminal toward the chest of the target object. Figure 2 As shown, when the speaker sends a transmission signal, the microphone is activated and a feedback signal reaching the microphone is received. The feedback signal includes the target object's voice signal and the chest cavity reflection signal. In this process, there is no contact with the target object. The voice signal is the low-frequency sound emitted by the target object when speaking, and the chest cavity reflection signal is the signal reflected by the chest cavity when the transmission signal is spoken by the target object.
[0035] After receiving the feedback signal, the starting point and the ending point of the speech in the feedback signal are detected, and the speech segment in the feedback signal is extracted.
[0036] Specifically, for speech signals, the short-time average energy and short-time average zero-crossing rate in the time domain are calculated. Considering that the collected speech signals have a medium or small signal-to-noise ratio, empirical thresholds are set for the short-time average energy and short-time average zero-crossing rate to distinguish between speech segments and silence segments of the speech signal. Then, each speech segment is extracted based on the detected starting and ending points.
[0037] Then use a bandpass filter to filter the speech segment. The passband frequency of the bandpass filter is 17.5kHz-18.5kHz, and the frequency range of the speech band is generally below 10kHz. After passing through the bandpass filter, the speech band is filtered out, and the high-frequency signal of the filtered speech band is output as the receiving signal for subsequent use.
[0038] It should be noted that by filtering out the speech band of the target object through bandpass filtering and processing the filtered object, the speech information of the target object can be prevented from being leaked, thereby effectively protecting the privacy of the target object.
[0039] Step S2: Perform short-time Fourier transform on the received signal, extract frequency characteristics and power characteristics, compare them with samples in the identity database, and identify the identity of the target object.
[0040] STFT is a method for calculating Doppler shift, which includes the change of signal frequency over time, that is, Doppler shift. Figure 3 , the present application calculates the spectrum of the received signal by using a short-time Fourier transform algorithm (STFT), and the spectrum of the received signal describes the relationship between the signal time, frequency and power.
[0041] It should be noted that the length of the signal L for Fourier transform has a certain influence on the signal spectrum calculation result, and an appropriate length should be selected. If the length of L is too small, the signal for Fourier transform is too short, resulting in low frequency resolution. On the contrary, if L is too large, it results in low time resolution. According to experimental conditions and specific implementation methods, the range of L can be set to [512, 4800] sampling points.
[0042] The frequency features and power features are extracted from the spectrum of the received signal to represent the frequency shift features and power distribution features of the target object during speaking, respectively.
[0043] Specifically, the power characteristics of the received signal are divided into 4 power intervals, and the centroid frequency of each power interval is calculated as the four characteristics of the power domain; the frequency characteristics of the received signal are divided into 5 frequency intervals, and the average power of each frequency range is calculated as the five characteristics of the frequency domain.
[0044] The obtained centroid frequency sequence and average power sequence are matched with the sample sequence in the identity database, that is, the dynamic time planning algorithm is used to calculate the similarity between the sample in the identity database and the centroid frequency sequence or the average power sequence. Since each person's speaking speed and pronunciation are different, the detected voice segments are sequences of different lengths, so the similarity algorithm based on dynamic time planning is used to determine the similarity between the centroid frequency sequence, the average power sequence and the sample sequence in the identity database. The sample sequence with the highest similarity is taken as the matching result. If the two matched sample sequences represent the same target user, the identity of the target object is identified and the identity authentication is successful.
[0045] Step S3: using a cross-correlation function to calculate the correlation between the transmission signal and the reception signal, taking the time sampling point with the maximum correlation as the time delay between the transmission signal and the reception signal, and synchronizing the reception signal and the transmission signal.
[0046] Reference Figure 4 In this step, the transmission signal sent by the speaker is represented as x(n), and the received signal is represented as y(n). The correlation between x(n) and y(n) is calculated based on the cross-correlation function. The cross-correlation function describes the relationship between the signal correlation and the time sampling point, which can be expressed as:
[0047]
[0048] Where n represents the sampling point of the transmission signal, m represents the number of offset sampling points of the received signal relative to the transmission signal, and R(m) represents the similarity between the transmission signal and the received signal when the received signal y(n) is moved to the left by m sampling points. When R(m) reaches the maximum value, the transmission signal and the received signal reach the maximum similarity. * (n) represents the conjugate matrix of the transmission signal x(n), and l represents the length of the transmission signal.
[0049] The time sampling point with the largest correlation is used as the time delay between the transmission signal and the received signal to synchronize the received signal and the transmission signal. That is, m is taken as the offset to be synchronized. If m ≥ 0, the transmission signal is shifted to the left by m samples to align it with the received signal. If m < 0, the received signal is shifted to the left by |m| samples to align it with the transmission signal.
[0050] Step S4: Perform fast Fourier transform on the received signal and the transmitted signal after synchronization processing, calculate and obtain a frequency response function, perform inverse fast Fourier transform on the frequency response function, and obtain an impulse response of the received signal.
[0051] This step analyzes the impulse response of the received signal. The impulse response is the time response characteristic of the signal transmission process, that is, the impulse response of the received signal is a set of response characteristics of the transmission signal arriving at the microphone at different time points via multiple paths.
[0052] First, the received signal y(n) is expressed in the form of an impulse response function, which is represented in the frequency domain as Y(f):
[0053]
[0054] Y(f)=H(f)X(f)
[0055] Here, k is the sampling point of the impulse response.
[0056] The frequency response function H(f) is estimated by the fast Fourier transform function (FFT):
[0057]
[0058] Perform an inverse fast Fourier transform on the frequency response function and convert the frequency domain function to the time domain function to express the impulse response h(n):
[0059] h(n)=iFFT(H(f)).
[0060] Step S5: input the impulse response of the received signal into the trained generative adversarial network to extract the impulse response generated by breathing.
[0061] This application uses a generative adversarial network (GAN) for noise reduction. During training, the impulse response of the received signal and the impulse response generated by real breathing are input into the GAN network, and the parameters of the GAN network are trained. The GAN network includes a generating network and a discriminating network, wherein the input of the generating network is the impulse response of the received signal, and the output is close to the impulse response generated by real breathing. The discriminating network discriminates whether the output of the generating network is true or false from the impulse response generated by breathing and the impulse response generated by the generating network. The training of the GAN network is a relatively mature technology in this field and will not be repeated here.
[0062] After the GAN network is trained, the impulse response of the received signal corresponding to the target object is input into the generation network of the GAN network, and the impulse response after noise reduction, that is, the impulse response generated by breathing, is output.
[0063] Since the impulse response of the received signal contains other interference responses in the environment, the GAN network is used to eliminate other interference responses from the impulse response of the received signal, thereby extracting the impulse response generated by breathing.
[0064] Step S6: Smooth the impulse response generated by breathing, and then detect the number of peak points to obtain the breathing rate.
[0065] In this step, the impulse response generated by breathing is firstly smoothed by using a smoothing function, and then the number of peak points is detected by using a peak detection algorithm.
[0066] Specifically, the moving average method is used to set a fixed-length window, and the amplitude average of the signal in the window is calculated as the new amplitude, and the sequence composed of the smoothed amplitudes is used as the new signal sequence. According to the experimental conditions, the window length is generally in the range of [5,50]. Then the peak detection algorithm is used to detect the number of peak points. Specifically, the average value μ and standard deviation σ of all local peaks of the signal are first calculated, and μ-2σ is used as the peak threshold, and the time interval of 2s is used as the time threshold. Combine the two thresholds to identify the correct peak point and calculate its number.
[0067] BR=Fs*(N-1) / (L N -L1) calculate the respiratory rate, Fs represents the sampling frequency, N represents the number of peak points, L NL1 and L2 represent the time points of the Nth peak point and the 1st peak point respectively, and the respiratory rate is calculated by the above formula.
[0068] This enables respiratory monitoring based on identity recognition. After successful identity recognition, the target's respiratory rate is estimated, and the successfully identified target and the corresponding respiratory monitoring results are recorded under the corresponding target user, thereby achieving respiratory tracking and recording of the authenticated target.
[0069] In one embodiment, the present application further provides an identity-identifiable respiratory monitoring device, comprising a processor and a memory storing a plurality of computer instructions, wherein the computer instructions implement the steps of the identity-identifiable respiratory monitoring method when executed by the processor.
[0070] For the specific definition of the identity-identifiable respiratory monitoring device, please refer to the definition of the identity-identifiable respiratory monitoring method above, which will not be repeated here. The above-mentioned identity-identifiable respiratory monitoring device can be implemented in whole or in part by software, hardware and a combination thereof. It can be embedded in or independent of the processor in the computer device in the form of hardware, or it can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the above corresponding operations.
[0071] The memory and the processor are electrically connected directly or indirectly to achieve data transmission or interaction. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines. The memory stores a computer program that can be run on the processor, and the processor implements the network topology layout method in the embodiment of the present invention by running the computer program stored in the memory.
[0072] The memory may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc. The memory is used to store a program, and the processor executes the program after receiving an execution instruction.
[0073] The processor may be an integrated circuit chip with data processing capabilities. The above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0074] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.
Claims
1. An identity-recognizable breathing monitoring method, characterized in that: The identity-identifiable breathing monitoring method comprises: Sending a transmission signal to a target object and receiving a feedback signal, wherein the feedback signal includes a speech signal and a chest reflection signal of the target object, extracting a speech segment from the feedback signal and then performing a bandpass filter to remove the speech segment to obtain a received signal; Perform short-time Fourier transform on the received signal, extract frequency and power features, compare them with samples in the identity database, and identify the identity of the target object; The correlation between the transmission signal and the receiving signal is calculated using a cross-correlation function, and the time sampling point with the maximum correlation is used as the time delay between the transmission signal and the receiving signal to synchronize the receiving signal and the transmission signal; Performing fast Fourier transform on the received signal and the transmitted signal after synchronization processing, calculating the frequency response function, and performing inverse fast Fourier transform on the frequency response function to obtain the impulse response of the received signal; Input the impulse response of the received signal into the trained generative adversarial network to extract the impulse response generated by breathing; Smooth the impulse response generated by breathing, then detect the number of peak points to get the breathing rate; The extracting of frequency characteristics and power characteristics comprises: The received signal power characteristics are divided into four power intervals, and the centroid frequencies of each power interval are calculated as four characteristics of the power domain; The received signal frequency characteristics are divided into five frequency intervals, and the average power of each frequency range is calculated as five characteristics in the frequency domain; The cross-correlation function is used to calculate the correlation between the transmission signal and the received signal. The cross-correlation function used is as follows: ; in, Indicates transmission signal The conjugate matrix of Indicates the transmission signal length, represents the sampling point of the transmission signal, Indicates the number of offset sampling points of the received signal relative to the transmitted signal. It means that the signal will be received Move Left The similarity between the transmitted signal and the received signal at the sampling point.
2. The identity-recognizable breathing monitoring method according to claim 1, characterized in that: The frequency of the transmission signal is 18 kHz, and the passband frequency of the bandpass filter is 17.5 kHz-18.5 kHz.
3. The identity-recognizable breathing monitoring method according to claim 1, characterized in that: The short-time Fourier transform is performed on the received signal. When the short-time Fourier transform is performed, the length of the input received signal ranges from 512 to 4800 sampling points.
4. An identity-recognizable respiratory monitoring device, comprising a processor and a memory storing a plurality of computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the method described in any one of claims 1 to 3 are implemented.
