A dual biometric method based on millimeter wave radar

Abstract

The application discloses a kind of double biological identification methods based on millimeter wave radar, belong to biological characteristic identification technical field.The method described in the application, including by setting millimeter wave radar signal parameter and collecting the biological activity signal of human body;Then the phase of all pulse periods of frequency-modulated continuous wave unwrapping after processing reflection signal and transmission signal;Then the phase after unwrapping is segmented to obtain biological activity signal segment;Further extract biological characteristic in biological activity signal segment;According to the above characteristic information identity registration and identification.The application collects biological characteristic by millimeter wave radar, avoids equipment and direct contact with body, simultaneously, through double biological characteristic identification, enhances identification accuracy, in addition, signal processing method and feature extraction technology make identity characteristic identification method have higher reliability.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of biometric recognition, and particularly relates to a double biometric recognition method based on millimeter wave radar. BACKGROUND

[0002] With the progress of society, people's awareness of privacy protection is also getting stronger, and through identity recognition, the protection of legal users and the resistance of intruders can be realized. The identity recognition method is mainly divided into two categories: one is the traditional token-based identity recognition method. The other is the biometric-based identity recognition method. The traditional token-based identity recognition method refers to the identity recognition through physical tokens (such as keys, access cards) or virtual tokens (such as account passwords). The second category is to use the biometric features of the human body for identity recognition. With the rapid development of the Internet of Things technology, biometric-based identity recognition technology has also been more widely applied. Various biometric identity recognition technologies such as face, voice, and fingerprint have been more and more widely applied.

[0003] However, the existing biometric identity recognition technology is vulnerable to impersonation attacks, such as: through printed photos and screen shots, the face data of the registered user can be easily obtained, leading to face impersonation intrusion. Using pre-recorded audio of the registered user or using artificial intelligence technology to synthesize language for impersonation intrusion. The fingerprint pattern is accurately carved on the latex, gelatin, wood glue and other materials to copy the registered user's finger model for fingerprint impersonation intrusion. For the above intrusion, researchers have also made many studies to overcome it.

[0004] For face impersonation attacks, Albakri and Alghowinem developed a system that uses the 3D sensor of the Microsoft Kinect device to obtain depth information. The system calculates the depth value on the image to determine the real face and the 2D photo, and the judgment reaches 100% accuracy, but it cannot face the 3D mask impersonation intrusion.

[0005] In order to resist voice impersonation attacks, researchers have proposed a series of solutions based on the differences between humans and loudspeakers, but these methods may be subject to site restrictions, require high user cooperation, and are not effective when invaded by high-fidelity recording devices. Recently, Meng et al. used a ring-shaped microphone array to determine the identity corresponding to the collected audio, and tested a dataset of 32780 audio samples and 14 intrusion devices, with an accuracy of 99.84%. The above research has broken away from the method of resisting intrusion by adding additional sensors in the past, but it is through the microphone to collect the sound wave signal, so the recognition accuracy is easily affected by external noise.

[0006] For fingerprint impersonation attack, researchers have added additional sensor modules to distinguish between human fingers and fake finger models from pulse oximetry, smell, blood flow, etc. However, such methods have low reliability and require expensive hardware devices.

[0007] In addition to the above-mentioned facial, voice, and fingerprint-based biometric identity recognition methods, there is also a breath-based biometric identity recognition. Respiratory rate is an indicator of human health and is commonly used in the medical field to monitor patient sleep and health status. In recent studies, many scholars have used it as a biometric feature to explore identity verification methods. C. Pham et al. designed a wearable Internet of Things device to capture breathing sounds from the nose through acoustic sensors, and used accelerometer and gyroscope sensors to collect TF to verify user identity, but the device needs to be in direct contact with the body and is inconvenient to wear. J. Liu et al. used CSI to extract signals related to breathing from WiFi-enabled devices to achieve non-contact chest fluctuation signal detection, and proposed a user verification method based on deep learning, but the detection distance is limited and the accuracy is affected by the message transmission rate and WiFi signal. Leem et al. used an ultra-wideband radio frequency transmitter to capture TF for identity authentication, but the verification method proposed in the article is based on the pre-set TF speed rule as the basis for determining whether to pass the authentication, rather than the difference in TF between different users as the basis for verification.

[0008] In summary, the existing biometric-based identity recognition technology has the following shortcomings:

[0009] (1) Lack of reliability. First, it is vulnerable to impersonation attacks; in addition, voice identity recognition technology is also susceptible to acoustic noise in the environment, and when the noise in the environment is large, the accuracy of the recognition will be affected.

[0010] (2) High cost and inconvenience. In order to resist impersonation intrusion, many technologies have added additional sensors, which also increase the cost. Some also need to be worn on the body or in contact with the body, which increases the inconvenience in the use process.

[0011] (3) Single feature. Most biometric recognition technologies can only recognize one biometric feature, and single biometric feature cannot achieve higher recognition accuracy and is easily copied by intruders. SUMMARY

[0012] In view of the above prior art defects, the present application proposes a dual biometric method based on millimeter wave radar. The millimeter wave radar is used to collect biometric features, avoiding direct contact between the device and the body, and the dual biometric recognition is used to enhance the recognition accuracy. In addition, the signal processing method and feature extraction technology are proposed to make the identity feature recognition method have high reliability.

[0013] A dual biometric method based on millimeter wave radar, comprising the following steps:

[0014] (1) Set the millimeter wave radar signal parameters, and collect the biological activity signals of the human body by the millimeter wave radar transmitting frequency-modulated continuous wave; the biological activity signals include the fluctuation signals of the thorax when the human body breathes and the vibration signals of the skin in the vocal cord area when the human body speaks;

[0015] (2) Mix the reflected signal with the transmitted signal to obtain a beat signal, and perform Fourier transform on the beat signal to obtain distance unit information between the radar and the human body; the transmitted signal is the frequency-modulated continuous wave transmitted by the millimeter wave radar to the human body; the reflected signal is the signal reflected by the human body after the frequency-modulated continuous wave reaches the human body;

[0016] (3) Extract the phase of the frequency-modulated continuous wave in all pulse periods and perform unwrapping processing to obtain unwrapped phases;

[0017] (4) Separate the two biological activity signals carried by the unwrapped phases and segment them into biological activity signal segments;

[0018] (5) Extract the biological features in the biological activity signal segments;

[0019] (6) Identity authentication and recognition, specifically including:

[0020] S6-1: In the registration stage, the biological features of the legal user are sent into the classifier to generate a user profile;

[0021] S6-2: In the authentication stage, the biological features of the access personnel are compared with those of the legal user, and the identity recognition result is output.

[0022] As a preferred embodiment of the present application, the millimeter wave radar signal parameters include: the number of slow time sampling points , the slow time sampling rate , the frame period , the number of fast time sampling points , the fast time sampling rate , the pulse period , and the number of frames ;

[0023] The setting method of the millimeter wave radar signal parameters includes the following steps: first, determine to be set determining set to ; the required slow-time sampling rate is determined as then at most ; at most ; at least ; at this point, all radar parameters are calculated, wherein F is set as the value of the slow-time sampling rate, and the value of F needs to be greater than 2 times the required target signal bandwidth.

[0024] As a preferred embodiment of the present application, in the step (2), the transmitted signal is a frequency-modulated continuous wave transmitted by the millimeter wave radar to the human body; and the reflected signal is a signal received by the millimeter wave radar after the frequency-modulated continuous wave reaches the human body and is reflected.

[0025] As a preferred embodiment of the present application, in the step (2), the method for obtaining the distance unit information of the human body specifically comprises:

[0026] S2-1: assuming that the signal amplitude is 1 and no attenuation occurs in the propagation process, then the signal expression in one pulse period of the frequency-modulated continuous wave transmitted by the millimeter wave radar is: ; wherein, is the initial frequency of the signal, is the frequency modulation slope, is the observation time, and j is the imaginary symbol.

[0027] S2-2: the expression of the reflected signal is: ; wherein, is the delay between the transmitted signal and the reflected signal, and satisfies , is the speed of light, and R is the distance unit information between the radar and the human body.

[0028] S2-3: after mixing the transmitted signal and the reflected signal, the beat signal is obtained: ; wherein, is the reflected signal, is the transmitted signal, and since , it is approximately 1, so the beat signal at this time can be written as: .

[0029] S2-4: the time-domain signal beat signal is subjected to Fourier transform to obtain the frequency-domain signal :

[0030] ; wherein, ​is the frequency of the beat signal, f is the frequency argument.

[0031] S2-5: the range spectrum of the beat signal is When , the value of the range spectrum is maximum, and the range unit information of the human body is obtained as , R is the range unit information between the radar and the human body.

[0032] As a preferred embodiment of the present application, in the step (3), the phases of the frequency-modulated continuous waves in all pulse periods are extracted and unwrapped to obtain unwrapped phases, and specifically comprising:

[0033] S3-1: assuming that the number of frequency-modulated continuous waves in each frame signal is , and there are frame signals in total, then for the frequency-modulated continuous waves in pulse periods, the beat signal is expressed as after Fourier transform;

[0034] S3-2: the phases at the position of the human target are extracted and unwrapped, and when , the phase of the beat signal is , wherein , x=1,2..., ;

[0035] S3-3: the phases at the position of the human target in all frequency-modulated continuous wave periods are expressed as / λ;R x is the range unit information between the radar and the target at each pulse moment, wherein x=1,2..., is the pulse period; T is a transpose symbol, λ is the wavelength of the beat signal; L is an omitted part, and T is a transpose symbol.

[0036] S3-4: the formula for phase unwrapping is expressed as: ; wherein is the unwrapped phase, is the phase before unwrapping, is the period in which is located.

[0037] As a preferred embodiment of the present application, in the step (4), first, the separated unwrapped phase signals are respectively input into a high-pass filter with a cutoff frequency of 0.1 Hz and a band-pass filter with a passband range of 80-500 Hz, the biological activity signal A output by the high-pass filter corresponds to the fluctuation signal of the thorax when the human body breathes, and the biological activity signal B output by the band-pass filter corresponds to the vibration signal of the skin in the vocal cord region when the human body speaks.

[0038] Then the bioactivity signal A is divided into expiration-inhalation signal segments and inhalation-expiration signal segments according to a periodic rule; the segments of the talking part of the bioactivity signal B are retained, and the segments of the silent part are discarded, and the retained segments are vocal cord vibration signal segments.

[0039] As a preferred embodiment of the present application, in the step (5), the biological features in the bioactivity signal segments are extracted, specifically including:

[0040] S5-1: respectively extracting waveform features of the expiration-inhalation signal segments and waveform features of the inhalation-expiration signal segments;

[0041] S5-2: sequentially extracting statistical feature components and time-frequency features of the expiration-inhalation signal segments and the inhalation-expiration signal segments;

[0042] S5-3: extracting filter bank features of the vocal cord vibration signal segments.

[0043] The biological features include waveform features of the expiration-inhalation signal segments, waveform features of the inhalation-expiration signal segments, statistical feature components and time-frequency features of the expiration-inhalation signal segments and the inhalation-expiration signal segments, and filter bank features of the vocal cord vibration signal segments.

[0044] As a preferred embodiment of the present application, the expiration-inhalation signal segment, i.e. the peak segment, is composed of an expiration process and an inhalation process; the waveform feature extraction method thereof includes: first finding the maximum value of the peak segment, i.e. the peak height , and the maximum width of the peak ; then selecting a group of sampling points on the peak segment, each group of sampling points including a peak rising segment sampling point and a peak falling segment sampling point , and the amplitude ratio corresponding to each group of sampling points is 1:1; the time interval between each group of sampling points is ; the peak height is defined as ; therefore, the waveform feature of the peak is , wherein , , , are all corresponding coordinate values, ; the inhalation-expiration signal segment, i.e. the valley segment, is composed of an inhalation process and an expiration process; the waveform feature extraction method thereof includes: first finding the minimum value of the valley segment, i.e. the valley depth , and the maximum width of the valley ; then selecting a group of sampling points on the valley segment, each group of sampling points including a valley rising segment sampling point and a peak falling segment sampling point , and the amplitude ratio of each group of sampling points is 1:1; the time interval between each group of sampling points is ; the valley depth is defined as , and the waveform characteristics of the valley are , wherein , , , , and each is a corresponding coordinate value .

[0045] As a preferred embodiment of the present application, in S5-2, the extraction of the statistical feature component is: using K-level wavelet packet decomposition, the peak segment and the valley segment signal of the biological activity signal A are decomposed into sub-signals, a total of 2xJ sub-signals are obtained; the mean, variance and skewness of each sub-signal are calculated to obtain Jx3 statistical features, and the value of J is required to be as close as possible to the value of a, and the value of K can be determined accordingly;

[0046] The extraction of the time-frequency feature component is: selecting an arbitrary inhale-breath-inhale signal segment to calculate its short-time Fourier transform as its time-frequency feature.

[0047] As a preferred embodiment of the present application, in S5-3, the filter bank feature of the vocal cord vibration signal segment is extracted, specifically including: each vocal cord vibration signal segment is framed, the Fourier transform of each frame is calculated to obtain a time-frequency spectrum, X overlapping triangular filters are used to map the time-frequency spectrum to the Mel frequency scale, and finally the logarithm of the triangular filter output time-frequency spectrum is calculated to obtain the filter bank feature of the vocal cord vibration signal. The size of the feature vector obtained for each vocal cord vibration signal segment is a x X, wherein the larger X is, the higher the accuracy of identity recognition is, but the longer the recognition time is, so the selection can be balanced according to the demand.

[0048] As a preferred embodiment of the present application, in step (6), identity authentication and recognition specifically include:

[0049] S6-1: When in the registration phase, the user collects the features of the algorithm user through steps (1)-(5), and enters them into the database to form a corresponding legal user profile; when a new user registers successfully, a new legal user profile is formed, and when there are N legal users, the label of the Nth user profile is represented as , wherein each element in the above formula is a positive number not greater than 1;

[0050] S6-2: Collect the features of the access personnel features through steps (1)-(5), and use ECAPA-TDNN as the identity recognition classifier, assuming that the access personnel feature label obtained by the identity recognition classifier is wherein is xxx, is an element in , the Euclidean distance vector of the legal user profile label and the access personnel feature label is ; if there is any element in , so that the identity of the access personnel is the identity label corresponding to the element index; otherwise, the access personnel is determined as an illegal user and will be refused access.

[0051] Compared with the prior art, the present application has the following beneficial effects:

[0052] (1) The non-contact biological characteristic signal acquisition mode is adopted, so that the user can acquire the signal containing the biological characteristics without direct contact with the device, the user is not bound by the device on the body, and the use is more convenient.

[0053] (2) The present application solves the problem that the traditional face, voice, fingerprint and other identity recognition modes are vulnerable to intrusion attacks. The millimeter wave sensor acquires subtle body movements caused by normal physiological activities, rather than sound signals. These subtle body movements are difficult to replicate and vary from person to person, so the intruder cannot intrude into the user's system by stealing the user's data. Even if the intruder steals the user's legal data using other millimeter wave sensor devices, the intruder cannot intrude into the user's system because the intruder cannot obtain the same radar parameters as the user's system. Therefore, the present application has excellent attack resistance, and the user's privacy is better protected.

[0054] (3) Compared with one biological characteristic, the present application adopts two biological characteristics, which brings higher recognition accuracy and is more difficult to be replicated by intruders. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 is a schematic diagram of the millimeter wave sensor acquisition device of the present application.

[0056] Figure 2 is a frequency and time relationship diagram of the frequency-modulated continuous wave transmission signal and the reflected signal of the present application, wherein is the modulation bandwidth of the frequency-modulated continuous wave, is the modulation period, is the time delay of the transmission signal and the reception signal, is the frequency modulation slope.

[0057] Figure 3 is a flowchart of the dual-biometric identification method based on millimeter wave radar of the present application.

[0058] Figure 4Signal diagram of the biological activity signal A described in the present application.

[0059] Figure 5 Signal diagram of the biological activity signal B described in the present application.

[0060] Figure 6 Accuracy comparison diagram of waveform features, statistical features, time-frequency features, filter bank features, and combined features of the above features. DETAILED DESCRIPTION

[0061] For the purpose, technical solutions and advantages of the present application, the present application will be further described below in combination with specific embodiments.

[0062] Related parameter name:

[0063] a. Slow time sampling point number : The number of frequency-modulated continuous waves (pulses) in each frame period.

[0064] b. Slow time sampling rate : The ratio of the number of frequency-modulated continuous waves to the unit time.

[0065] c. Fast time sampling point number : The number of sampling points in each frequency-modulated continuous wave period.

[0066] d. Fast time sampling rate : The ratio of the fast time sampling point number to the unit time.

[0067] e. Pulse period : That is, the frequency-modulated continuous wave period.

[0068] f. Frame period : Composed of multiple pulse periods.

[0069] g. Frame number .

[0070] Example 1

[0071] A dual biometric method based on a millimeter wave radar, comprising the following steps:

[0072] Step one, using Figure 1The millimeter wave sensor acquisition device is used for millimeter wave radar signal parameter calculation and biological activity signal acquisition. The millimeter wave sensor acquisition device has a millimeter wave sensor with three transmitting antennas and four receiving antennas, and includes a signal generator, a mixer, and a microprocessor. The microprocessor includes a filter, an AD conversion module, and a signal processing module. The signal generator transmits a signal through the three transmitting antennas. The receiving antennas receive the transmitted signal, and the transmitted signal is mixed in the mixer and then sent to the filter of the microprocessor for filtering. Then, the AD conversion module is used to obtain a beat signal, and the signal processing module is used to process the beat signal.

[0073] S1-1: Set millimeter wave radar signal parameters: 2000 Hz; 128; 256; thus at most ; at most ; at least ; at this time, the radar parameters are all calculated.

[0074] S1-2: Collect biological activity signals of the human body by using the millimeter wave radar to transmit a frequency-modulated continuous wave. The chest fluctuation signal (signal A) caused by breathing and the skin vibration signal (signal B) caused by the vocal cords during speaking; the frequency range of signal B is 90-200 Hz, which is much higher than the highest frequency of signal A, so the slow-time sampling rate needs to be greater than 400 Hz.

[0075] Step two, signal processing:

[0076] (1) Mix the reflected signal with the transmitted signal to obtain a beat signal, and perform Fourier transform on the beat signal to obtain distance unit information of the human target:

[0077] S2-1: The transmitted signal is a frequency-modulated continuous wave transmitted by the millimeter wave radar to the human body; the reflected signal is a frequency-modulated continuous wave that reaches the human body and is reflected, and then is received by the millimeter wave radar. In this example, the distance between the millimeter wave sensor and the human target is ; wherein is the initial displacement between the millimeter wave sensor and the human target, is the displacement of the human target relative to the millimeter wave sensor, which is caused by the human target speaking, breathing, and random body movement. is the time delay of the transmitted signal and the received signal, , wherein is the speed of light.

[0078] S2-2: According toFigure 2 The transmitted signal is a frequency-modulated continuous wave emitted towards the human body via millimeter-wave radar. The transmitted signal of millimeter-wave radar is a periodic signal, composed of multiple pulse periods. Assuming the signal amplitude is 1 and there is no attenuation during propagation, the signal expression within a single pulse period of the transmitted signal is: ;in, The starting frequency of the signal. For frequency modulation slope, The observation time is j, where j is the imaginary number.

[0079] S2-3: The reflected signal is a frequency-modulated continuous wave that is reflected when it reaches the human body. ;in, Let be the delay between the transmitted signal and the reflected signal, and satisfy . , R represents the speed of light, and R represents the distance unit information between the radar and the target being measured.

[0080] S2-4: After the receiving antenna receives the echo of the transmitted signal, it mixes with the transmitted signal and sends the mixture to the microprocessor for filtering. The A / D conversion yields the beat signal. .because Therefore If the value is 1, then the beat signal at this time can be written as: .

[0081] S2-5: Beat signal Perform Fourier transform: ,when hour, The value reaches its maximum, at which point the distance unit information between the radar and the human body, i.e., the distance between the radar and the human body, is... , The frequency is the independent variable.

[0082] (2) Extract the phase of the frequency-modulated continuous wave of all pulse periods and perform unwinding processing to obtain the unwound phase signal;

[0083] S3-1: The phase extraction method is as follows: Because the target remains relatively stationary with respect to the radar, the amplitude of the biological activity signal generated by the target is extremely small. Therefore, changes in its biological activity signal can be reflected in the phase of the beat signal. So, for a series of... By extracting the phase corresponding to each pulse period from a frequency-modulated continuous wave with several pulse cycles, the following can be obtained:

[0084] / λ;R x This represents the range cell information between the radar and the target at each pulse moment, where x=1,2,..., , is the pulse period; T is the transpose symbol, and l is the wavelength of the beat signal; L is the omitted part, and T is the transpose symbol.

[0085] S3-2: Since the phase value extracted in S3-1 is in the range of , Z is an integer, and a 180° jump occurs when , in order to obtain the intuitive phase change trend, phase unwrapping needs to be performed: , where is the unwrapped phase, is the phase before unwrapping, is the period in which

[0086] Step three, feature extraction:

[0087] (1) Separate the two biological activity signals carried by the unwrapped phase and segment them into biological activity signal segments:

[0088] S4-1: Design a high-pass filter (filter A) with a cutoff frequency of 0.1 Hz; design a band-pass filter (filter B) with a passband range of 80-500 Hz.

[0089] S4-2: Input the unwrapped phase into filter A and filter B respectively, the biological activity signal A output by filter A corresponds to the fluctuation signal of the thoracic cavity during human respiration, and the biological activity signal B output by filter B corresponds to the vibration signal of the skin in the vocal cord region during human speech.

[0090] S4-3: As shown in Figure 4 , the biological activity signal A presents a periodicity similar to a sine function, which needs to be segmented into individual periodic segments according to the periodic law. A period can be regarded as a wave peak segment (A-B-C) or a wave trough segment (B-C-D). Therefore, each period can be segmented into an expiration-inhalation signal segment (wave peak segment) and an inhalation-expiration signal segment (wave trough segment). The vibration of the vocal cord region is produced along with the speech behavior, and when silent, the vocal cord region does not vibrate, so the signal during silence is invalid. Therefore, for the biological activity signal B, the invalid segment needs to be discarded, and the effective vocal cord vibration signal segment needs to be extracted.

[0091] (2) Extract biological features from the biological activity signal segment:

[0092] S5-1: Extract the waveform features of the expiration-inhalation signal segment and the waveform features of the inhalation-expiration signal segment respectively;

[0093] S5-2: Extract the statistical feature components and time-frequency features of the expiration-inhalation signal segment and the inhalation-expiration signal segment in turn;

[0094] S5-3: Extract the filter bank feature of the effective vocal cord vibration signal segment.

[0095] S5-1: The exhalation-inhalation signal segment, i.e. the peak segment, is composed of an exhalation process and an inhalation process; the waveform feature extraction method thereof includes: first finding the maximum value of the peak segment: peak height and the maximum width of the peak ; then selecting 126 groups of sampling points on the peak segment, each group of sampling points including a peak rising segment sampling point and a peak falling segment sampling point , and the amplitude ratio corresponding to each group of sampling points is 1:1; the time interval between each group of sampling points is ; the peak height is defined as ; therefore the waveform feature of the peak is ; and the waveform feature of the exhalation-inhalation signal segment is a 126x1 vector.

[0096] S5-2: The inhalation-exhalation signal segment, i.e. the valley segment, is composed of an inhalation process and an exhalation process; the waveform feature extraction method thereof includes: first finding the minimum value of the valley segment, i.e. the valley depth , the maximum width of the valley ; then selecting 126 groups of sampling points on the valley segment, each group of sampling points including a valley rising segment sampling point and a peak falling segment sampling point , and the amplitude ratio corresponding to each group of sampling points is 1:1; the time interval between each group of sampling points is ; the valley depth is defined as ; therefore the waveform feature of the valley is , and the waveform feature of the inhalation-exhalation signal segment is a 126x1 vector.

[0097] S5-3: Perform 6-level wavelet packet decomposition on the exhalation-inhalation signal segment and the inhalation-exhalation signal segment, and a total of 126 signal subspaces can be obtained; calculate the mean, variance and skewness of the signal in each subspace, and the statistical feature is a 126x3 vector.

[0098] S5-4: Calculate the short-time Fourier transform of the exhalation-inhalation signal segment and the inhalation-exhalation signal segment to obtain the time-frequency feature. Since the biological activity signal is a non-stationary signal, the length of the time-frequency spectrum calculated from different signal segments is also inconsistent, and the length of the time axis of the time-frequency spectrum needs to be increased to 126 by zero padding, so the size of the time-frequency feature is 126xL, and L is the length of the frequency axis.

[0099] S5-5: Each vocal cord vibration signal segment is framed, and the Fourier transform of each frame is calculated to obtain the time spectrum. The time spectrum is mapped to the Mel frequency scale using 40 overlapping triangular filters. Finally, the logarithm of the time spectrum output by the triangular filters is calculated to obtain the filter bank characteristics of the vocal cord vibration signal. The feature vector size of each vocal cord vibration signal segment is 126×40.

[0100] Step four, identity authentication and recognition, specifically includes:

[0101] S6-1: During the registration phase, the user collects the characteristics of legitimate users through steps (1)-(5) and enters them into the database to form the corresponding legitimate user profile; when a new user successfully registers, a new legitimate user profile will be formed. When there are N legitimate users, the profile tag of the Nth user is represented as ,in All elements in the array are positive numbers with a value not exceeding 1;

[0102] S6-2: The characteristics of visitors are collected through steps (1)-(5). ECAPA-TDNN is used as the identity recognition classifier. Assume that the visitor feature labels output by the identity recognition classifier are... ,in It is xxx. for The elements in The Euclidean distance vector between the legitimate user profile tags and the visitor feature tags is: ;if There exists any element in the array such that If the access is valid, the accessor's identity is the identity tag corresponding to the element index; otherwise, the accessor is determined to be an illegal user and will be denied access.

[0103] Table 1. Comparison of Noise Resistance Performance (Accuracy)

[0104]

[0105] While the millimeter-wave radar collects human biological activity signals, a recording device is used to collect the actual speech signals of the test subject. At the same time, a speaker continuously plays white noise to simulate noise interference. The accuracy of identity recognition under the presence of interference is compared with the actual speech signals and the dual biometric recognition method described in this invention, as shown in Table 1.

[0106] According to Table 1, by adjusting the level of noise, four signal-to-noise ratio levels of 0 dB, 10 dB, 20 dB and 30 dB are obtained. It can be observed that as the signal-to-noise ratio decreases, the accuracy of identity recognition using real language signals continuously decreases, while the recognition accuracy of the present application is basically not disturbed by noise. This is because the millimeter wave radar collects the body movement caused by normal physiological activities, and the external noise cannot be captured by the millimeter wave radar, so it has excellent anti-noise performance.

[0107] Table 2 Anti-attack performance

[0108]

[0109] A-F represent 6 different legal users respectively, and two identity recognition networks are pre-trained using real language signals and biological activity signals of the double biological recognition method of the present application respectively, and then the language signals of the 6 legal users are recorded by another recording device, 600 samples for each user, a total of 2400 samples. The biological activity signals of the 6 legal users are collected again by another millimeter wave radar device, 600 samples for each user, a total of 2400 samples. These samples are input into the corresponding identity recognition network to simulate intrusion estimation. Assuming that 600 estimations are launched against user A and 200 estimations are successful, it is recorded as 200 / 600. According to Table 2, it can be found that it is difficult to resist intrusion estimation using real language signals, and the number of successful attacks is 3445 times in a total of 3600 attacks. While the present application has only 39 successful attacks in 3600 attacks.

[0110] From Figure 6 it can be observed that the identity recognition accuracy of using waveform features, statistical features, time-frequency features and filter bank features respectively is 75.14%, 84.52%, 81.98% and 92.56% respectively. If the several feature components are combined to obtain combined features, the correct rate can reach 96.66%. Combining multiple features can more completely represent the identity of the user and achieve higher recognition accuracy.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit the protection scope of the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the essence and scope of the technical solutions of the present application.

Claims

1. A dual biometric method based on millimeter wave radar, characterized by, The method comprises the following steps: (1) setting millimeter wave radar signal parameters, collecting biological activity signals of a human body by transmitting a frequency-modulated continuous wave through a millimeter wave radar; the biological activity signals include chest fluctuation signals during human respiration and vibration signals of skin in a vocal cord region during human speech; (2) mixing the reflected signals and the transmitted signals to obtain beat signals, performing Fourier transform on the beat signals to obtain distance unit information between the radar and the human body; the transmitted signals are frequency-modulated continuous waves transmitted to the human body through the millimeter wave radar; the reflected signals are signals reflected by the human body after the frequency-modulated continuous waves arrive at the human body; (3) extracting phases of the frequency-modulated continuous waves in all pulse cycles and performing unwrapping processing to obtain unwrapped phases; (4) separating two biological activity signals carried by the unwrapped phases and dividing the biological activity signals into biological activity signal segments; (5) extracting biological features in the biological activity signal segments, specifically comprising: S5-1: respectively extracting waveform features of exhalation-inhalation signal segments and inhalation-exhalation signal segments; S5-2: sequentially extracting statistical feature components and time-frequency features of the exhalation-inhalation signal segments and the inhalation-exhalation signal segments; S5-3: extracting filter bank features of the vocal cord vibration signal segments; The waveform feature extraction method of the expiratory-inspiratory signal segment comprises the following steps: first, finding the maximum value of the peak segment, i.e. the peak height , the maximum width of the peak ; then selecting a group of sampling points on the peak segment, each group of sampling points comprising a peak rising segment sampling point and a peak falling segment sampling point , wherein , and the amplitude ratio corresponding to each group of sampling points is 1:1; the time interval between each group of sampling points is ; the peak height is defined as ; therefore, the waveform feature of the peak is ; wherein , , , are the corresponding coordinate values; The waveform feature extraction method of the inhalation-exhalation signal segment comprises the following steps: first, finding the minimum value of the trough segment, i.e. the trough depth , the maximum width of the trough ; then selecting a group of sampling points on the trough segment, each group of sampling points comprising a trough rising segment sampling point and a peak descending segment sampling point , wherein , and the amplitude ratio corresponding to each group of sampling points is 1:1; the time interval between each group of sampling points is ; the trough depth is defined as , and thus the waveform feature of the trough is ; wherein , , , are the corresponding coordinate values. (6) identity authentication and recognition, specifically comprising: S6-1: in a registration stage, sending biological features of a legal user into a classifier to generate a user profile; S6-2: in an authentication stage, comparing biological features of an access person with those of the legal user and outputting an identity recognition result.

2. The dual biometric method based on millimeter wave radar of claim 1, wherein, The millimeter wave radar signal parameters include: slow time sampling point number , slow time sampling rate , frame period , fast time sampling point number , fast time sampling rate , pulse period , frame number ; The setting method of the millimeter wave radar signal parameter comprises the following steps: firstly determining is set to , determining is set to ; determining the required slow-time sampling rate as , then is at most ; is at most ; is at least ; at this time, the radar parameters are all calculated.

3. The dual biometric method based on millimeter wave radar of claim 1, wherein, In the step (2), the method for obtaining the distance unit information between the radar and the human body specifically comprises: S2-1: The transmitting signal of the millimeter wave radar is a periodic signal composed of multiple pulse periods. Assuming that the signal amplitude is 1 and no attenuation occurs in the propagation process, the complex signal expression in a single pulse period in the transmitting signal is: ; wherein, is the starting frequency of the signal, is the frequency modulation slope, is the observation time, and j is the imaginary symbol; S2-2: The expression of the reflected signal is: ; wherein, is the delay between the transmitted signal and the reflected signal, and satisfies , is the speed of light, and R is the distance unit information between the radar and the human body; S2-3: The beat signal is obtained after mixing the transmitted signal and the reflected signal. ;in, For reflected signals, In order to transmit a signal, due to Therefore If the value is 1, then the current time difference signal is... It can be written as: ; S2-4: performing Fourier transform on the beat signals: ; wherein, is the frequency of the beat signal, f is the frequency argument; S2-5: the distance spectrum of the beat signal is ; when , the value of the distance spectrum is maximum, and the distance unit information of the human body is obtained as , and R is the distance unit information between the radar and the human body.

4. The dual biometric method based on millimeter wave radar as claimed in claim 1, wherein, In the step (3), the method for extracting the phases of the frequency-modulated continuous waves in all pulse cycles and performing unwrapping processing to obtain the unwrapped phases specifically comprises: S3-1: Assuming the number of frequency-modulated continuous waves per frame signal is , there are frame signals in total, then for the frequency-modulated continuous wave of pulse periods, the beat signal is expressed as after Fourier transform; S3-2: Extract the phase at the human target position and unwrap, in case of a beat signal with where , x = 1,2... ; S3-3: The phase at the range cell of the human target within all frequency-modulated continuous wave cycles is represented as / λ ; R x is the range unit information between the radar and the target at each pulse moment, where x = 1, 2,..., is the pulse period; T is the transpose symbol, and λ is the wavelength of the beat signal. S3-4: The formula of phase unwrapping is expressed as: ; wherein is the unwrapped phase, is the phase before unwrapping, is the period in which is located.

5. The dual biometric method based on millimeter wave radar as claimed in claim 4, wherein, In the step (4), first, the unwrapped phases are respectively input into a high-pass filter with a cutoff frequency of 0.1 Hz and a band-pass filter with a passband range of 80-500 Hz, the biological activity signal A output by the high-pass filter corresponds to the chest fluctuation signals during human respiration, and the biological activity signal B output by the band-pass filter corresponds to the vibration signals of the skin in the vocal cord region during human speech; Then, the biological activity signal A is divided into exhalation-inhalation signal segments and inhalation-exhalation signal segments according to a periodic law, and the segments of the biological activity signal B corresponding to the speech are retained, and the segments corresponding to the silence are discarded, and the retained segments are vocal cord vibration signal segments.

6. The dual biometric method based on millimeter wave radar as claimed in claim 1, wherein, In the S5-2, the extraction of the statistical characteristic component is as follows: the peak segment and the valley segment signal of the biological activity signal A are respectively decomposed into 2xJ sub-signals by using K-level wavelet packet decomposition; the mean value, the variance and the skewness of each sub-signal are calculated to obtain Jx3 statistical characteristics; and the statistical characteristics of the peak segment and the valley segment are combined to obtain the final statistical characteristic vector. In the S5-2, the extraction of the statistical characteristic component is as follows: the peak segment and the valley segment signal of the biological activity signal A are respectively decomposed into 2xJ sub-signals by using K-level wavelet packet decomposition; the mean value, the variance and the skewness of each sub-signal are calculated to obtain Jx3 statistical characteristics; and the statistical characteristics of The extraction of the time-frequency feature components is that in the biological activity signal A, a short-time Fourier transform of an arbitrary inhalation-exhalation-inhalation signal segment is calculated as the time-frequency feature of the signal segment.

7. The dual biometric method based on millimeter wave radar as claimed in claim 1, wherein, In the S5-3, the filter bank features of the vocal cord vibration signal segments are extracted, specifically comprising: each vocal cord vibration signal segment is framed, the Fourier transform of each frame is calculated to obtain a time-frequency spectrum, the time-frequency spectrum is mapped to a mel frequency scale by using X overlapping triangular filters, and finally the logarithm of the time-frequency spectrum output by the triangular filters is calculated to obtain the filter bank features of the vocal cord vibration signal, and the size of a feature vector obtained for each vocal cord vibration signal segment is a×X.

8. The dual biometric method based on millimeter wave radar as claimed in claim 1, wherein, The step (6) specifically includes: S6-1: When in the registration stage, the user combines the biological characteristics of the user through steps (1)-(5), combines these biological characteristic vectors, and then enters them into the database to form a corresponding legal user profile; when a new user registers successfully, a new legal user profile is formed, and when there are N legal users, the label of the Nth user profile is represented as wherein the elements in the above formula are positive numbers with a size not greater than 1. S6-2: Collect the biometric features of the access person through steps (1)-(5), and use ECAPA-TDNN as the identity recognition classifier. Assume that the access person feature label obtained by the identity recognition classifier output is , wherein is , an element in , the Euclidean distance vector between the legal user profile label and the access person feature label is ; if there is any element in such that , then the identity of the access person is the identity label corresponding to the element index; otherwise, the access person is determined to be an illegal user and is denied access.

Images