User gesture fingerprint authentication method based on heart rate sensor

Abstract

The present invention discloses a user gesture fingerprint authentication method based on a heart rate sensor, which extracts data during gesture actions through the heart rate sensor to determine the identity authentication fingerprint, and finally constructs an identity authentication model, thereby the user identity authentication fingerprint is completely based on the heart rate sensor of the device, and such user identity fingerprint is not portable. In addition, the user identity fingerprint based on the heart rate sensor in the wrist wearable device is based on the physiological movement characteristics of a person, and the identity authentication fingerprint extracted by the heart rate sensor cannot be modified at the hardware level without changing the heart rate sensor and the user's gesture style and the user's wrist wearable device wearing habits; the advantages are low cost, not easy to crack, high security, and no security risk of user personal information being stolen.

Description

Technical Field

[0001] The present invention relates to a user gesture fingerprint authentication method, and in particular to a user gesture fingerprint authentication method based on a heart rate sensor. Background Art

[0002] The current continuous identity authentication method mainly achieves continuous identity authentication of users by authenticating the user once every certain period of time. The current continuous identity authentication mainly uses face and fingerprint authentication. Face authentication requires the use of a high-precision camera to capture the face image, while fingerprint authentication requires the use of a fingerprint sensor. The prices of high-precision cameras and fingerprint sensors are both high, which leads to high costs for continuous identity authentication. In addition, face data and fingerprint data are both sensitive data of users, and there is a risk of abuse and theft of faces and fingerprints by certification agencies. At the same time, with the maturity of 3D printing technology, hardware imitation of faces and fingerprints has posed a security threat to face and fingerprint authentication, and simulated attacks by virtual software also pose a threat to face recognition. Therefore, the existing continuous identity authentication method is not only costly, easy to crack, and not very secure, but also user personal information may be stolen, posing a major security risk.

[0003] With the development of wearable devices, smart watches and smart bracelets are becoming more popular. Relying on the development of embedded IoT technology and the rapid improvement of chip processing capabilities of mobile devices, the functions of wearable IoT devices such as smart watches and smart bracelets are becoming more and more perfect. At the same time, due to people's pursuit of healthy life, health monitoring modules such as heart rate sensors are rapidly popularized on wearable devices such as smart watches and smart bracelets. The readings of heart rate sensors do not pose a great threat to people's information security, and they are dependent on wearable devices. Once the user changes the wearable device, the fingerprint of the heart rate sensor on it will change immediately; at the same time, everyone's heart rate sensor readings cannot be easily imitated, making the continuous identity authentication based on the heart rate sensor safer; in addition, the cost of heart rate sensors is much cheaper than high-precision cameras and fingerprint sensors. In view of this, it is of great significance to use heart rate sensors for continuous identity authentication and design a user gesture fingerprint authentication method based on heart rate sensors that is low in cost, not easy to crack, high in security, and does not have the security risk of user personal information being stolen. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a user gesture fingerprint authentication method based on a heart rate sensor, which has low cost, is not easy to crack, has high security, and does not have the potential safety hazard of user personal information being stolen.

[0005] The technical solution adopted by the present invention to solve the above technical problem is: a user gesture fingerprint authentication method based on a heart rate sensor, comprising the following steps:

[0006] Step 1: Construct the total data set. The specific process is as follows: record the total number of users as I, let I users be I testers, let I testers wear the same wearable device with a heart rate sensor, and set the sampling frequency of the heart rate sensor to 100Hz, let each tester make 9 gestures representing the 9 numbers 1-9, where each gesture is performed 40 times in a row, and the 9 gestures are numbered in advance from 1 to 9. At this time, the heart rate sensor worn by each tester samples the tester and outputs the readings corresponding to each sampling point; the authentication server obtains the readings output by the heart rate sensor carried by each tester during the 9 gestures, and arranges the obtained readings of the heart rate sensor carried by each tester in the order of output to form a data set for each tester. I testers are randomly numbered from 1 to I, and the data set of the i-th tester is recorded as A i , i=1,2,…,I; the data sets of I testers are used to form the total data set, and the total data set is recorded as A, A={A1,A2,A3,...,A I}; The data set of the jth gesture of the i-th tester is recorded as A i，j , j = 1, 2, ..., 9, A i，j The number of data contained in is recorded as M i，j , each data corresponds to a sampling point of the heart rate sensor, that is, A i，j Contains M i，j sampling points;

[0007] Step 2: Perform data preprocessing on the total data set A obtained in step 1 to obtain the preprocessed total data set B, perform data analysis on the preprocessed total data set, and extract data segments related to gestures. The specific process is as follows:

[0008] S2-1, use low-pass filter to perform low-pass filtering on all data sets in the total data set A to make the data smoother, and convert the data set A of the i-th tester i The data set obtained after filtering is recorded as B i , the dataset A of the jth gesture of the i-th tester i，j The data set obtained after filtering is recorded as B i，j , Dataset B i，j There is also M i，j sampling points, the filtered data sets of I test persons are used to form the total filtered data set B, where B = {B1, B2, B3, ..., B I};

[0009] S2-2. When the heart rate sensor is idle, the output reading of the heart rate sensor will fluctuate slightly around a fixed value. The fixed value is obtained by directly reading the reading of the heart rate sensor when it is idle, and the fixed value is recorded as

[0010] S2-3, 0.75 seconds is used as the duration of each gesture action. Under the sampling frequency of 100 Hz, each gesture action has 75 sampling points, which is data set B. i，j Define a length (M i，j -75) i，j , Y i，j The nth element in is denoted as Y i，j [n], n = 1, 2, ..., (M i，j -75), n represents Y i，j [n] in Y i，j The position in the image is obtained by the following formula (1) i，j :

[0011]

[0012] In formula (1), B i，j [g] is data set B i，j The g-th sampling point of , | | is the absolute value symbol, g represents B i，j [g] In dataset B i，j Position in

[0013] S2-4. Determine Y i，j The maximum point in Y i，j The number of maximum points in is denoted as m i，j , get each Y i，j The maximum point in Y i，j The position in Y i，j The set of positions of all maximum points in is denoted as a i，j , is to i，j The position of the u-th maximum point in is recorded as a i，j [u],u=1,2,…,m i，j , B i，j [a i，j [u]] as B i，j The starting point of the u-th gesture slice in B i，j [a i，j The 74th sampling point after [u]] is taken as B i，j The end point of the u-th gesture slice in B i，j From the starting point of the u-th gesture slice to B i，j The end point of the u-th gesture slice in B is 75 sampling points in total.i，j The u-th gesture slice in B i，j Medium i，j gesture slices;

[0014] S2-5. Using Dataset B i，j Medium i，j The gesture slices constitute the slice dataset C of the jth gesture of the i-th tester. i，j , Dataset B i，j The mth gesture slice in is the slice dataset C i，j The mth data in is denoted as C i，j，m , C i，j，m represents the mth gesture slice of the jth gesture of the i-th tester, where m=1, 2, ..., m i，j ;

[0015] Step 3. Extract gesture motion fingerprints. The specific process is as follows:

[0016] S3-1, C i，j，m The kth sampling point in is denoted by C i，j，m [k],k=1,2,…,75,C i，j，m is the time domain data, and C is calculated using formula (2) to formula (7) i，j，m The time domain eigenvalue of i，j，m The time domain characteristic values ​​include the average value x i，j，m , standard deviation σ i，j，m , mean difference D i，j，m , skewness γ i，j，m , Kurtosis β i，j，m and the RMS amplitude A i，j，m :

[0017]

[0018]

[0019] S3-2, through the fast Fourier transform, the time domain data C i，j，m Convert to C i，j，m Frequency domain related data, C i，j，m Frequency domain related data include frequency magnitude coefficient data and frequency interval data, which will be shown in C i，j，m The frequency magnitude coefficient data after Fourier transformation is recorded as p i，j，m , C i，j，m The frequency interval data after Fourier transformation is recorded as f i，j，m , f i，j，m and p i，j，m The number of data in is the same, and this number of data is recorded as N i，j，m ; i,j,mThe k′th data in is denoted as p i，j，m [k′], f i，j，m The k′th data is recorded as f i，j，m [k′], the value range of k′ is = 1, 2, ..., N i，j，m ; Use formula (8) to formula (13) to calculate C i，j，m The frequency domain eigenvalue, C i，j，m The frequency domain eigenvalues ​​include the standard deviation σs i,j,m , Spectral centroid Cs i,j,m , spectral peak CRs i，j,m , Smoothness Fs i,j,m , skewness γs i,j，m and kurtosis βs i，j，m :

[0020]

[0021] In the above formula, the symbol * is the multiplication symbol, the symbol / is the division symbol, Π is the continuous multiplication symbol, ∑ is the continuous addition symbol, Max is the maximum value symbol, Max(p i,j，m [k′])| k′＝1 to H Indicates taking p i，j，m [1] to p i,j，m The maximum value in [H];

[0022] S3-3, through discrete wavelet transform in the "haar" wavelet mode to C i，j，m Perform a discrete wavelet transform and get C i，j，m The time-frequency domain data, C i，j，m The time-frequency domain data includes C i，j，m The low-frequency data after discrete wavelet transformation and C i，j，m After the high-frequency data is transformed by discrete wavelet, C i，j，m The low-frequency data after discrete wavelet transform is recorded as q i,j,m , C i，j，m The high-frequency data after discrete wavelet transform is recorded as t i，j,m , q i,j，m The hth low-frequency value in is denoted as q i,j，m [h],t i，j，m The hth high frequency value in is denoted as t i，j，m [h], h = 1, 2, ..., 37; use formula (14) to formula (23) to calculate C i，j，m The time-frequency domain eigenvalue, C i，j，m The time-frequency domain eigenvalues ​​include the average value xz i，j,m , mean difference Dz i,j，m , skewness γz i,j,m , Kurtosis βz i,j,m、Maximum value Maxz i,j，m and minimum value Minz i,j,m ;

[0023]

[0024] Among them, Max(q i,j,m ) means taking q i,j，m The maximum value in, Max(t i，j，m ) means taking t i，j,m The maximum value in the i,j，m ) means taking q i,j,m The minimum value in i，j，m ) means taking t i，j,m The minimum value in ;

[0025] S3-4, using C i，j，m The 6 time-domain eigenvalues, 6 frequency-domain eigenvalues, and 6 time-frequency-domain eigenvalues ​​of the eigenvalues ​​form an 18-dimensional vector, which is the m-th gesture motion fingerprint of the j-th gesture of the i-th tester, denoted by D i，j,m , the set of all gesture motion fingerprints of the jth gesture of the i-th tester is recorded as D i，j , the set of all gesture motion fingerprints of all gestures of all testers is recorded as D;

[0026] Step 4: Use set D to train the identity authentication model. Specifically, randomly select 100 gesture fingerprints of other testers except the i-th tester from set D and D i，j All gesture motion fingerprints in D are transformed using random forest, SVM and gradient boosting tree methods. i，j All gesture motion fingerprints in and the selected 100 gesture motion fingerprints are classified and learned together to construct D i，j The authentication model is saved in the authentication server. i，j The identity authentication model is the identity authentication model of the jth gesture of the i-th user;

[0027] Step 5: When a user needs to be authenticated, the user makes a gesture from 1 to 9 gestures. At this time, the heart rate sensor worn by the user collects the data of the user making the gesture in real time and sends it to the authentication server. The authentication server uses the same method as steps 2 and 3 to obtain an 18-dimensional vector as the user's gesture motion fingerprint P. Then the authentication server identifies the gesture motion fingerprint P through the 9 gesture authentication models of the user stored therein. If one of the authentication models determines that the gesture is in compliance, the authentication is successful and the user is authenticated. Otherwise, the authentication fails and the user is not authenticated.

[0028] All the identity authentication models in the authentication server are updated every Q days, where the value of Q is greater than or equal to 7. Specifically, the user's latest gesture motion fingerprint is obtained every Q days according to the same method as steps 2 and 3, and the user's latest gesture motion fingerprint is added to set D. Then, gesture motion fingerprints older than 30 days in set D are randomly deleted to update set D. Then, based on the method in step 4, the updated D is used to train again to obtain the identity authentication model, so as to update the identity authentication model to adapt to the user's daily physiological changes and ensure the accuracy of identity authentication.

[0029] Compared with the prior art, the advantage of the present invention is that the data of gesture actions are extracted by the heart rate sensor to determine the identity authentication fingerprint, and finally an identity authentication model is constructed, so that the user identity authentication fingerprint is completely based on the heart rate sensor of the device, and such user identity fingerprint is not portable. In addition, the user identity fingerprint based on the heart rate sensor in the wrist wearable device is based on the physiological movement characteristics of a person. Under the premise of not changing the heart rate sensor and the user's gesture style and the user's wrist wearable device wearing habits, the identity authentication fingerprint extracted by the heart rate sensor cannot be modified at the hardware level. Therefore, the present invention has low cost, is not easy to crack, has high security, and does not have the security risk of user personal information being stolen. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 Schematic diagram of 9 gestures for users;

[0031] Figure 2 The heart rate sensor readings and the corresponding Y when there is a gesture and when there is no gesture i，j result. DETAILED DESCRIPTION

[0032] The present invention is further described in detail below with reference to the accompanying drawings.

[0033] Embodiment: A user gesture fingerprint authentication method based on a heart rate sensor comprises the following steps:

[0034] Step 1: Construct the total data set. The specific process is as follows: record the total number of users as I, let I users be I testers, let I testers wear the same wearable device with a heart rate sensor, and set the sampling frequency of the heart rate sensor to 100Hz. Let each tester make 9 gestures representing the 9 numbers 1-9. The 9 gestures are as follows: Figure 1As shown, each gesture is performed 40 times continuously, the interval between each gesture is 2 seconds, and the 9 gestures are pre-numbered from 1 to 9. At this time, the heart rate sensor worn by each tester samples the tester and outputs the readings corresponding to each sampling point; the authentication server obtains the readings output by the heart rate sensor carried by each tester during the 9 gestures, and arranges the obtained readings of the heart rate sensor carried by each tester in the order of output to form a data set for each tester. I testers are randomly numbered from 1 to I, and the data set of the i-th tester is recorded as A i , i=1,2,…,I; the data sets of I testers are used to form the total data set, and the total data set is recorded as A, A={A1,A2,A3,...,A I}; The data set of the jth gesture of the i-th tester is recorded as A i，j , j = 1, 2, ..., 9, A i，j The number of data contained in is recorded as M i，j , each data corresponds to a sampling point of the heart rate sensor, that is, A i，j Contains M i，j sampling points;

[0035] Step 2: Perform data preprocessing on the total data set A obtained in step 1 to obtain the preprocessed total data set B, perform data analysis on the preprocessed total data set, and extract data segments related to gestures. The specific process is as follows:

[0036] S2-1, use low-pass filter to perform low-pass filtering on all data sets in the total data set A to make the data smoother, and convert the data set A of the i-th tester i The data set obtained after filtering is recorded as B i , the dataset A of the jth gesture of the i-th tester i，j The data set obtained after filtering is recorded as B i，j , Dataset B i，j There is also M i，j sampling points, the filtered data sets of I test persons are used to form the total filtered data set B, where B = {B1, B2, B3, ..., B I};

[0037] S2-2. When the heart rate sensor is idle, the output reading of the heart rate sensor will fluctuate slightly around a fixed value. The fixed value is obtained by directly reading the reading of the heart rate sensor when it is idle, and the fixed value is recorded as

[0038] S2-3, 0.75 seconds is used as the duration of each gesture. Then, at a sampling frequency of 100 Hz, each gesture has 75 sampling points, which is data set B.i，j Define a length (M i，j -75) i，j , Y i，j The nth element in is denoted as Y i，j [n], n = 1, 2, ..., (M i，j -75), n represents Y i，j [n] in Y i，j The position in the image is obtained by the following formula (1) i，j :

[0039]

[0040] In formula (1), B i，j [g] is data set B i，j The g-th sampling point of , | | is the absolute value symbol, g represents B i，j [g] In dataset B i，j Position in

[0041] S2-4. Determine Y i，j The maximum point in Y i，j The number of maximum points in is denoted as m i，j , get each Y i，j The maximum point in Y i，j The position in Y i，j The set of positions of all maximum points in is denoted as a i，j , is to i，j The position of the u-th maximum point in is recorded as a i，j [u],u=1,2,…,m i，j , B i，j [a i，j [u]] as B i，j The starting point of the u-th gesture slice in B i，j [a i，j The 74th sampling point after [u]] is taken as B i，j The end point of the u-th gesture slice in B i，j From the starting point of the u-th gesture slice to B i，j The end point of the u-th gesture slice in B is 75 sampling points in total. i，j The u-th gesture slice in B i，j Medium i，j gesture slices;

[0042] S2-5. Using Dataset B i，j Medium i，j The gesture slices constitute the slice dataset C of the jth gesture of the i-th tester. i，j , Dataset B i，jThe mth gesture slice in is the slice dataset C i，j The mth data in is denoted as C i，j，m , C i，j，m represents the mth gesture slice of the jth gesture of the i-th tester, where m=1, 2, ..., m i，j ;

[0043] Step 3. Extract gesture motion fingerprints. The specific process is as follows:

[0044] S3-1, C i，j，m The kth sampling point in is denoted by C i，j，m [k],k=1,2,…,75,C i，j，m is the time domain data, and C is calculated using formula (2) to formula (7) i，j，m The time domain eigenvalue of i，j，m The time domain characteristic values ​​include the average value x i，j，m , standard deviation σ i，j，m , mean difference D i，j，m , skewness γ i，j，m , Kurtosis β i，j，m and the RMS amplitude A i，j，m :

[0045]

[0046]

[0047] S3-2, through the fast Fourier transform, the time domain data C i，j，m Convert to C i，j，m Frequency domain related data, C i，j，m Frequency domain related data include frequency magnitude coefficient data and frequency interval data, which will be shown in C i，j，m The frequency magnitude coefficient data after Fourier transformation is recorded as p i,j，m , C i，j，m The frequency interval data after Fourier transformation is recorded as f i，j，m , f i，j，m and p i,j，m The number of data in is the same, and this number of data is recorded as N i，j，m ; i,j,m The k′th data in is denoted as p i,j，m [k′], f i，j，m The k′th data is recorded as f i，j，m [k′], the value range of k′ is = 1, 2, ..., N i，j，m ; Use formula (8) to formula (13) to calculate C i，j，m The frequency domain eigenvalue, C i，j，m The frequency domain eigenvalues ​​include the standard deviation σs i,j,m , Spectral centroid Csi,j,m , spectral peak CRs i,j，m , Smoothness Fs i，j,m , skewness γs i,j，m and kurtosis βs i,j，m :

[0048]

[0049] In the above formula, the symbol * is the multiplication symbol, the symbol / is the division symbol, Π is the continuous multiplication symbol, ∑ is the continuous addition symbol, Max is the maximum value symbol, Max(p i，j,m [k′])| k′＝1 to H Indicates taking p i,j,m [1] to p i,j，m The maximum value in [H];

[0050] S3-3, through discrete wavelet transform in the "haar" wavelet mode to C i，j，m Perform a discrete wavelet transform and get C i，j，m The time-frequency domain data, C i，j，m The time-frequency domain data includes C i，j，m The low-frequency data after discrete wavelet transformation and C i，j，m After the high-frequency data is transformed by discrete wavelet, C i，j，m The low-frequency data after discrete wavelet transform is recorded as q i，j,m , C i，j，m The high-frequency data after discrete wavelet transform is recorded as t i，j,m ,q i,j，m The hth low-frequency value in is denoted as q i,j，m [h],t i，j,m The hth high frequency value in is denoted as t i，j，m [h], h = 1, 2, ..., 37; use formula (14) to formula (23) to calculate C i，j，m The time-frequency domain eigenvalue, C i，j，m The time-frequency domain eigenvalues ​​include the average value xz i,j，m , mean difference Dz i,j,m , skewness γz i,j,m , Kurtosis βz i，j，m 、Maximum value Maxz i，j，m and minimum value Minz i，j，m ;

[0051]

[0052] Among them, Max(q i,j，m ) means taking q i,j，m The maximum value in, Max(t i，j，m ) means taking t i，j，mThe maximum value in the i,j，m ) means taking q i,j，m The minimum value in i，j，m ) means taking t i，j，m The minimum value in ;

[0053] S3-4, using C i，j，m The 6 time-domain eigenvalues, 6 frequency-domain eigenvalues, and 6 time-frequency-domain eigenvalues ​​of the eigenvalues ​​form an 18-dimensional vector, which is the m-th gesture motion fingerprint of the j-th gesture of the i-th tester, denoted by D i，j,m , the set of all gesture motion fingerprints of the jth gesture of the i-th tester is recorded as D i，j , the set of all gesture motion fingerprints of all gestures of all testers is recorded as D;

[0054] Step 4: Use set D to train the identity authentication model. Specifically, randomly select 100 gesture fingerprints of other testers except the i-th tester from set D and D i，j All gesture motion fingerprints in D are transformed using random forest, SVM and gradient boosting tree methods. i，j All gesture motion fingerprints in and the selected 100 gesture motion fingerprints are classified and learned together to construct D i，j The authentication model is saved in the authentication server. i，j The identity authentication model is the identity authentication model of the ith gesture of the ith user;

[0055] Step 5: When a user needs to be authenticated, the user makes one of the gestures 1-9. At this time, the heart rate sensor worn by the user collects the data of the user making the gesture in real time and sends it to the authentication server. The authentication server uses the same method as steps 2 and 3 to obtain an 18-dimensional vector as the user's gesture motion fingerprint P. Then the authentication server identifies the gesture motion fingerprint P through the 9 gesture authentication models of the user stored therein. If one of the authentication models determines that the gesture is in compliance, the authentication is determined to be successful and the user is authenticated. Otherwise, the authentication is determined to have failed and the user is not authenticated.

[0056] In this embodiment, all identity authentication models in the authentication server are updated every Q days, and the value of Q is greater than or equal to 7. Specifically, the user's latest gesture motion fingerprint is obtained every Q days according to the same method as step 2 and step 3, and the user's latest gesture motion fingerprint is added to set D, and then the gesture motion fingerprints older than 30 days in set D are randomly deleted to update set D, and then based on the method of step 4, the updated D is used to train again to obtain the identity authentication model, so as to update the identity authentication model to adapt to the user's daily physiological changes and ensure the accuracy of identity authentication.

[0057] The heart rate sensor readings and corresponding Y values ​​when the user is making gestures and when the user is not making gestures i，j The results are as follows Figure 2 As shown, analysis Figure 2 It can be seen that if there is no gesture action, the reading of the heart rate sensor fluctuates very little, and the corresponding result Y i，j If there is a gesture, the reading will fluctuate greatly. i，j will also be very large, thus through Y i，j The determination of the maximum value can accurately determine the gesture slice, and then extract the gesture motion fingerprint based on the gesture slice to build an identity authentication model.

Claims

1. A user gesture fingerprint authentication method based on a heart rate sensor, characterized in that The steps include: Step 1: Construct the total data set. The specific process is as follows: record the total number of users as I, let I users be I testers, let I testers wear the same wearable device with a heart rate sensor, and set the sampling frequency of the heart rate sensor to 100Hz, let each tester make 9 gestures representing the 9 numbers 1-9, where each gesture is performed 40 times in a row, and the 9 gestures are numbered in advance from 1 to 9. At this time, the heart rate sensor worn by each tester samples the tester and outputs the readings corresponding to each sampling point; the authentication server obtains the readings output by the heart rate sensor carried by each tester during the 9 gestures, and arranges the obtained readings of the heart rate sensor carried by each tester in the order of output to form a data set for each tester. I testers are randomly numbered from 1 to I, and the data set of the i-th tester is recorded as A i , i=1,2,…,I; The data sets of I testers are used to form a total data set, which is denoted as A, where A = {A1, A2, A3, ..., A I }; The data set of the jth gesture of the i-th tester is recorded as A i，j , j = 1, 2, ..., 9, A i，j The number of data contained in is recorded as M i，j , each data corresponds to a sampling point of the heart rate sensor, that is, A i，j Contains M i，j sampling points; Step 2: Perform data preprocessing on the total data set A obtained in step 1 to obtain the preprocessed total data set B. Perform data analysis on the preprocessed total data set B to extract data segments related to gestures. The specific process is as follows: S2-1, use low-pass filter to perform low-pass filtering on all data sets in the total data set A to make the data smoother, and convert the data set A of the i-th tester i The data set obtained after filtering is recorded as B i , the dataset A of the jth gesture of the i-th tester i，j The data set obtained after filtering is recorded as B i，j , Dataset B i，j There is also M i，j sampling points, the filtered data sets of I test persons are used to form the total filtered data set B, where B = {B1, B2, B3, ..., B I }; S2-2. When the heart rate sensor is idle, the output reading of the heart rate sensor will fluctuate slightly around a fixed value. The fixed value is obtained by directly reading the reading of the heart rate sensor when it is idle, and the fixed value is recorded as S2-3, 0.75 seconds is used as the duration of each gesture action. Under the sampling frequency of 100 Hz, each gesture action has 75 sampling points, which is data set B. i，j Define a length (M i，j -75) i，j , Y i，j The nth element in is denoted as Y i，j [n], n = 1, 2, ..., (M i，j -75), n represents Y i，j [n] in Y i，j The position in the equation (1) is used to calculate Y i，j : In formula (1), B i，j [g] is data set B i，j The g-th sampling point of , || is the absolute value symbol, g represents B i，j [g] In dataset B i，j Position in S2-4. Determine Y i，j The maximum point in Y i，j The number of maximum points in is denoted as m i，j , get each Y i，j The maximum point in Y i，j The position in Y i，j The set of positions of all maximum points in is denoted as a i，j , is to i，j The position of the u-th maximum point in is recorded as a i，j [u],u=1,2,…,m i，j , B i，j [a i，j [u]] as B i，j The starting point of the u-th gesture slice in B i，j [a i，j The 74th sampling point after [u]] is taken as B i，j The end point of the u-th gesture slice in B i，j From the starting point of the uth gesture slice to B i，j The end point of the u-th gesture slice in B is 75 sampling points in total. i，j The u-th gesture slice in B i，j Medium i，j gesture slices; S2-5. Using Dataset B i，j Medium i，j The gesture slices constitute the slice dataset C of the jth gesture of the i-th tester. i，j , Dataset B i，j The mth gesture slice in is the slice dataset C i，j The mth data in is denoted as C i，j，m , C i，j，m represents the mth gesture slice of the jth gesture of the i-th tester, where m=1, 2, ..., m i，j ; Step 3. Extract gesture motion fingerprints. The specific process is as follows: S3-1, C i，j，m The kth sampling point in is denoted by C i，j，m [k],k=1,2,…,75,C i，j，m is the time domain data, and C is calculated using formula (2) to formula (7) i，j，m The time domain eigenvalue of i，j，m The time domain characteristic values ​​include the average value x i，j，m , standard deviation σ i，j，m , mean difference D i，j，m , skewness γ i，j，m , Kurtosis β i，j，m and the RMS amplitude A i，j，m : S3-2, through the fast Fourier transform, the time domain data C i，j，m Convert to C i，j，m Frequency domain related data, C i，j，m Frequency domain related data include frequency magnitude coefficient data and frequency interval data, which will be shown in C i，j，m The frequency magnitude coefficient data after Fourier transformation is recorded as p i，j，m , C i，j，m The frequency interval data after Fourier transformation is recorded as f i，j，m , f i，j，m and p i，j，m The number of data in is the same, and this number of data is recorded as N i，j，m ; i，j，m The k′th data in is denoted as p i，j，m [k′], f i，j，m The k′th data is recorded as f i，j，m [k′], k′=1, 2,…,N i，j，m ; Use formula (8) to formula (13) to calculate C i，j，m The frequency domain eigenvalue, C i，j，m The frequency domain eigenvalues ​​include the standard deviation σs i，j，m , Spectral centroid Cs i，j，m , spectral peak CRs i，j，m , smoothness Fs i，j，m , skewness γs i，j，m and kurtosis βs i，j，m : CRs i，j，m =(Max(p i，j，m [k′])| k′＝1 to H ) / Cs i，j，m ,H=N i，j，m (10) In the above formula, the symbol * is the multiplication symbol, the symbol / is the division symbol, Π is the continuous multiplication symbol, ∑ is the continuous addition symbol, Max is the maximum value symbol, Max(p i，j，m [k′])| k′＝1 to H Indicates taking p i，j，m [1] to p i，j，m The maximum value in [H]; S3-3, through the discrete wavelet transform "haar" wavelet mode to C i，j，m Perform a discrete wavelet transform and get C i，j，m The time-frequency domain data, C i，j，m The time-frequency domain data includes C i，j，m The low-frequency data after discrete wavelet transformation and C i，j，m After the high-frequency data is transformed by discrete wavelet, C i，j，m The low-frequency data after discrete wavelet transform is recorded as q i，j，m , C i，j，m The high-frequency data after discrete wavelet transform is recorded as t i，j，m , q i，j，m The hth low-frequency value in is denoted as q i，j，m [h],t i，j，m The hth high frequency value in is denoted as t i，j，m [h], h = 1, 2, ..., 37; Use formula (14) to formula (23) to calculate C i，j，m The time-frequency domain eigenvalue, C i，j，m The time-frequency domain eigenvalues ​​include the average value xz i，j，m , mean difference Dz i，j，m , skewness γz i，j，m , Kurtosis βz i，j，m 、Maximum value Maxz i，j，m and minimum value Minz i，j，m ; Among them, Max(q i，j，m ) means taking q i，j，m The maximum value of the data, Max(t i，j，m ) means taking t i，j，m The maximum value of the data, Min(q i，j，m ) means taking q i，j，m The minimum value of the data, Min(t i，j，m ) means taking t i，j，m The minimum value of the data in; S3-4, using C i，j，m The 6 time-domain eigenvalues, 6 frequency-domain eigenvalues, and 6 time-frequency-domain eigenvalues ​​of the eigenvalues ​​form an 18-dimensional vector, which is the m-th gesture motion fingerprint of the j-th gesture of the i-th tester, denoted by D i，j，m , the set of all gesture motion fingerprints of the jth gesture of the i-th tester is recorded as D i，j , the set of all gesture motion fingerprints of all gestures of all testers is recorded as D; Step 4: Use set D to train the identity authentication model. Specifically, randomly select 100 gesture fingerprints of other testers except the i-th tester from set D and D i，j All gesture motion fingerprints in D are transformed using random forest, SVM and gradient boosting tree methods. i，j All gesture motion fingerprints in and the selected 100 gesture motion fingerprints are classified and learned together to construct D i，j The authentication model is saved in the authentication server. i，j The identity authentication model is the identity authentication model of the jth gesture of the i-th user; Step 5: When a user needs to be authenticated, the user makes one of 1-9 gestures. At this time, the heart rate sensor worn by the person collects the data of the user making the gesture in real time and sends it to the authentication server. The authentication server uses the same method as steps 2 and 3 to obtain an 18-dimensional vector as the user's gesture motion fingerprint P. Then the authentication server identifies the gesture motion fingerprint P through the 9 gesture authentication models of the user stored therein. If one of the authentication models determines that the gesture is in compliance, the authentication is determined to be successful and the user is authenticated. Otherwise, the authentication is determined to fail and the user is not authenticated.

2. According to claim 1, a user gesture fingerprint authentication method based on a heart rate sensor is characterized in that All the identity authentication models in the authentication server are updated every Q days, where the value of Q is greater than or equal to 7. Specifically, the user's latest gesture motion fingerprint is obtained every Q days according to the same method as steps 2 and 3, and the user's latest gesture motion fingerprint is added to set D. Then, gesture motion fingerprints older than 30 days in set D are randomly deleted to update set D. Then, based on the method in step 4, the updated D is used to train again to obtain the identity authentication model, so as to update the identity authentication model to adapt to the user's daily physiological changes and ensure the accuracy of identity authentication.

Images