WO2018152711A1 - 一种基于心电认证的门禁系统及其认证方法 - Google Patents

一种基于心电认证的门禁系统及其认证方法 Download PDF

Info

Publication number
WO2018152711A1
WO2018152711A1 PCT/CN2017/074437 CN2017074437W WO2018152711A1 WO 2018152711 A1 WO2018152711 A1 WO 2018152711A1 CN 2017074437 W CN2017074437 W CN 2017074437W WO 2018152711 A1 WO2018152711 A1 WO 2018152711A1
Authority
WO
WIPO (PCT)
Prior art keywords
ecg
control system
access control
feature
authentication
Prior art date
Application number
PCT/CN2017/074437
Other languages
English (en)
French (fr)
Inventor
张跃
肖志博
雷夏飞
张拓
Original Assignee
清华大学深圳研究生院
深圳市岩尚科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 清华大学深圳研究生院, 深圳市岩尚科技有限公司 filed Critical 清华大学深圳研究生院
Priority to PCT/CN2017/074437 priority Critical patent/WO2018152711A1/zh
Priority to CN201780002092.9A priority patent/CN107980151B/zh
Publication of WO2018152711A1 publication Critical patent/WO2018152711A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G07—CHECKING-DEVICES
    • G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
    • G07C9/00—Individual registration on entry or exit
    • G07C9/00174—Electronically operated locks; Circuits therefor; Nonmechanical keys therefor, e.g. passive or active electrical keys or other data carriers without mechanical keys
    • G07C9/00563—Electronically operated locks; Circuits therefor; Nonmechanical keys therefor, e.g. passive or active electrical keys or other data carriers without mechanical keys using personal physical data of the operator, e.g. finger prints, retinal images, voicepatterns
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2136—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on sparsity criteria, e.g. with an overcomplete basis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Definitions

  • the present invention relates to the field of identity authentication technologies, and in particular, to an access control system based on ECG authentication and an authentication method thereof.
  • biometric identification methods applied to the access control system mainly include fingerprints, faces and languages. These methods have greatly improved the security compared with the traditional access control systems, but they still exist to be imitated, copied, stolen or stolen. danger.
  • the present invention provides an access control system based on ECG authentication and an authentication method thereof.
  • An authentication method for an access control system based on ECG authentication comprising an ECG registration and an ECG authentication step
  • the ECG registration step includes: S11, adjusting the access control system to the ECG registration state; S12, collecting and registering through the ECG collection device
  • the ECG signal is preprocessed to detect the R wave position and intercept the QT band; S13, the intercepted QT band is extracted by the autocorrelation transform algorithm to obtain the ECG autocorrelation sequence; S14, the ECG obtained will be obtained
  • the correlation sequence is subjected to dimensionality reduction by orthogonal polynomial fitting regression to generate a feature template; S15, selecting and evaluating the optimal template of the electrocardiogram from the generated feature template; S16, from the ECG
  • the optimal threshold is obtained in the optimal feature template;
  • the ECG authentication step includes: S21: adjusting the access control system to the ECG registration state; S22, collecting the tester's ECG signal through the ECG acquisition device and performing preprocessing to detect the R wave Position
  • the present invention proposes an access control system based on ECG authentication and an authentication method thereof.
  • the method uses the autocorrelation transform algorithm to extract the features of the pre-processed ECG signals, obtains the ECG autocorrelation sequence, and then reduces the dimension of the ECG autocorrelation sequence by orthogonal polynomial fitting regression to generate feature templates.
  • the optimal template of the ECG is evaluated, and the optimal threshold is obtained.
  • the user performs identity authentication based on the optimal template of the ECG and the optimal threshold.
  • the authentication method has high security, high recognition accuracy, and small authentication storage information.
  • Figure 1 is a flow chart of an ECG user registering.
  • FIG. 2 is a flow chart of an ECG user performing authentication.
  • FIG. 3 is a structural block diagram of an access control system in an embodiment of the present invention.
  • An authentication method for an access control system based on ECG authentication includes an ECG registration and an ECG authentication step.
  • the ECG registration step includes:
  • Step 201 adjusting the access control system to the ECG registration state
  • Step 202 Collect an ECG signal of the registrant through the ECG acquisition device and perform pre-processing to detect the R wave position and intercept the QT band;
  • Step 203 Perform feature extraction on the intercepted QT band by using an autocorrelation transform algorithm to obtain an ECG autocorrelation sequence
  • Step 204 the acquired ECG autocorrelation sequence is subjected to dimensionality reduction by orthogonal polynomial fitting regression.
  • Step 205 Select and evaluate an ECG optimal feature template from the generated feature templates.
  • Step 206 Obtain an optimal threshold from the ECG optimal feature template, complete user registration, and re-register if the user fails to register.
  • Both the error acceptance rate (FAR) and the error rejection rate (FRR) are functions of the threshold.
  • FAR error acceptance rate
  • FRR error rejection rate
  • the cost of error acceptance and error rejection is different. It is assumed that the cost of error acceptance is cost 1 and error rejection occurs.
  • the cost is cost 2 (cost 1 > cost 2 ).
  • the ROC curve is made according to FAR and FRR, and then the cost curve is made according to the ROC curve.
  • the threshold corresponding to the minimum cost of the overall selection is selected as the optimal threshold best_thd.
  • the ECG authentication step includes:
  • Step 301 adjusting the access control system to the ECG authentication state
  • Step 302 collecting the ECG signal of the tester through the ECG acquisition device and performing preprocessing, detecting the R wave position, and intercepting the QT band;
  • Step 303 extracting the QT band by using an autocorrelation transform algorithm to obtain an ECG autocorrelation sequence
  • Step 304 Perform dimension reduction on the acquired ECG autocorrelation sequence by orthogonal polynomial fitting regression to generate a feature template.
  • Step 305 Perform feature matching on the generated feature template and the ECG optimal feature template, and complete the authentication according to the optimal threshold. Then the certification is passed, otherwise the certification will not pass.
  • the switch of the access control system is turned on. If the authentication fails, the switch of the access control system remains off. Preferably, if the authentication fails, an alert is issued.
  • the autocorrelation sequence R xx [m] between different individuals has obvious differences and can be used as an intrinsic feature of individual ECG signals. Since the QRS complex is the smallest and most stable component of the ECG signal in different kinds of test environments, the value of m is close to the length of the QRS wave, and the autocorrelation sequence R xx of the ECG sequence x[i] after autocorrelation processing. [m] is still a high-dimensional signal and needs to be reduced in dimensionality.
  • the dimension reduction is performed by orthogonal polynomial fitting regression described in step 204 or step 304, and the feature template is generated by approximating the ECG autocorrelation sequence by using a polynomial, and the feature template is used to represent the ECG autocorrelation sequence.
  • A (a 0 , a 1 , a 2 , ..., a k ) T
  • F i (1, f 1 (x i ), f 2 (x i ),...,f k (x i )) T
  • f is the sampling frequency of the ECG signal.
  • the ECG optimal feature template described in step 205 is obtained by using the leave-one method, and the discriminant is Where D(A i , A j ) represents a distance metric between the feature vector A i and the feature vector A j ; Indicates that 1 is taken when the distance between the feature A i and the feature A j is less than the preset threshold THD, otherwise 0 is taken; the value of THD is the average value of the distance between the n feature vectors, and i, j is 1 to n. I ⁇ j.
  • the i-th feature template is selected as the high-quality feature template; when not satisfied, the i-th template is an outlier and is rejected. Finally, the optimal feature template is selected, A 1 , A 2 , ..., A nl , where nl ⁇ n.
  • the intercept QT band described in step 202 or 302 is a Q wave point within a minimum of 90 milliseconds of the left side of the R wave, and a maximum point within 300 milliseconds of the right side of the R wave.
  • the first-order difference from the right-order first-order difference of the T-wave peak point is the T-waveform cut-off point for the first time, and then the fixed-length QT band is generated by the waveform correction.
  • the pre-processing in step 202 or 302 includes: filtering the ECG signal, collecting the ECG signal of the user for a certain period of time, and filtering the power frequency interference, the baseline drift, and the muscle by using a suitable filter.
  • Noise such as electrical interference.
  • the frequency point of the power frequency 50 Hz is notched to remove 50 Hz power frequency interference in the waveform; the Butterworth low-pass filter with a cutoff frequency of 40 Hz is used to filter the myoelectric interference; and the high-pass filter greater than 1 Hz is used to eliminate the baseline drift. .
  • the extracting the feature in step 203 or step 303 further comprises acquiring using a differentiated dictionary learning algorithm for sparse representation, specifically,
  • J (D, C) is the solved dictionary D and the sparse feature C
  • Verif (X i , X j , D, C i , C j ) is the feature distinguishing attribute
  • ⁇ is the sparsity degree coefficient
  • ⁇ is the regularization.
  • the coefficients, ⁇ and ⁇ range from 0 to 1.
  • X i and X j represent the i-th and j-th QT waves, respectively, and C i and C j represent the sparse features corresponding to X i and X j , respectively. Among them, i ⁇ j.
  • label (X i) X i represents the category number.
  • X (X 1 , X 2 , . . . , X n ) represents n QT waves;
  • D (d 1 , d 2 , . . . , d l ) represents the dimension of the dictionary, and l is greater than 1 Any value; Represents sparse features.
  • the feature extraction in step 203 or step 303 includes the following steps:
  • C1 Determine the length of the window including at least one ECG information on the ECG signal.
  • the length of the window is greater than the length of a heart beat, ensuring that each window contains at least one complete information of the heart beat.
  • the normal person's heartbeat is 60-100 beats/min, and the special crowd is usually more than 40 beats/min. Therefore, the window length is selected for 1-2 seconds or more to ensure that the window contains at least one complete beat of the heart beat.
  • Complete information not limited to the same heart beat, also contains the complete information of the different parts of the two heart beats that can be combined into one heart beat. After the window d is fixed, the window length does not change during training and testing.
  • the sliding window intercepts the corresponding length of the ECG window from any position of the ECG signal. During the process of intercepting the ECG window, there is no restriction on the starting point position of the window, which is especially important in the real-time testing phase. .
  • C3 dividing the ECG window into a plurality of fixed length segments to obtain a plurality of the feature segments, each ECG window being divided into n arbitrary fixed length segments, wherein n is greater than or equal to 1, assuming by window d
  • the intercepted ECG window is x
  • the arbitrary fixed length is divided into n feature segments ⁇ x 1 , x 2 , . . . , x n ⁇ , and any fixed length segment means that the segment length is smaller than the length of the ECG window and is fixed.
  • the feature segment includes two phases through the fully automatic feature extraction layer: a convolution layer and a maximum pool layer; wherein: the following steps are included:
  • A21 Parallel convolution of the feature segments by a plurality of the convolution layers to obtain a plurality of vector values, here a plurality of 1 ⁇ m-dimensional vector values.
  • the number of convolutional layers is n
  • the number of layers of each convolution layer is greater than 1
  • the convolution kernel K is a one-dimensional convolution kernel; n feature segments ⁇ x 1 , x 2 , . . . , x n ⁇
  • i is in the range [1, n]
  • l is the number of convolution layers
  • b is the offset
  • the initial value can be set to zero.
  • A22 A plurality of the vector values generate a deep fusion feature through the maximum pool layer.
  • the maximum pool core acts on matrix A:
  • A23 The depth fusion feature performs training classification through the fully connected layer, outputs a classification determination result, and extracts the fully automatic feature extraction layer as the feature extractor according to the classification determination result.
  • the method for feature extraction in the authentication method of the ECG-based access control system is different, including: first detecting each reference point in the ECG signal to extract a quasi-periodic heart beat as the original ECG feature.
  • the ECG signal is a quasi-periodic signal, but not all components in the heart cycle are specific.
  • the P wave, QRS group and T wave in each heart cycle contain most of the ECG. Specific information.
  • the band in each heart cycle is cut out from the continuous electrocardiographic signal as the original electrocardiographic feature. To do this, locate the reference point for your heartbeat. In addition, in the subsequent waveform correction, it is necessary to further process the P wave and the T wave.
  • the reference points for each heartbeat detection in the embodiments of the present invention include: P wave start point (Ps) and P wave end point (Pe), R wave peak (R), J wave start point (J), T wave peak (Tp) and T wave. End point (Te), a total of 6 types of benchmark points.
  • the overall ECG signal is moderate, and the R wave is the sharpest part.
  • the R wave is located at the minimum value of the second order difference of the signal, and the first order difference is zero.
  • the coarse position of the R wave is determined by the minimum value of the second-order differential signal of the original signal. After locating the rough position of the R wave, according to the characteristic that the amplitude of the R wave is at the maximum value, the first derivative is 0, and in the discrete case, that is, the first difference signal is closest to zero. This locates the exact R peak position.
  • a P wave start point Ps is preferably in a range of 160-180 milliseconds on the left side of each R wave, such as 170 milliseconds; and a P wave is preferably in a range of 80-100 milliseconds on the left side of each R wave, such as 90 milliseconds.
  • End point Pe a range of 80-100 milliseconds on the right side of each R wave peak, preferably at a time of 90 milliseconds, is the J wave start point (J);
  • the maximum value in the region to the right of each R peak (R) is the T peak (Tp) from the beginning of the J wave to 2/3 current RR intervals (ie, between the adjacent two R peaks)
  • the time limit is cut off;
  • the first-order differential signal on the right side of the T-peak (Tp) is the end of the T-wave (Te) for the first time from negative to positive.
  • the embodiment of the present invention proposes a method of segmentation waveform correction to eliminate the influence of heart rate variability.
  • the basic method of correction is to divide the original heart beat signal.
  • Segment resampling specifically, upsampling the P-band, extending the P-band duration after upsampling, unified to 460-500 milliseconds, preferably 480 milliseconds; for the QRS band remains unchanged, for example 180 milliseconds long; for the T-band
  • the J-Tp segment and the Tp-Tp segment are respectively downsampled, so that the durations of the two segments after resampling are unified to 10-20 milliseconds, preferably 15 milliseconds.
  • the total length of the corrected heart beats is basically the same, for example, 690 milliseconds. Because people have different heart rates at different times and after different sports, this difference in heart rate should not be a measure of human identity.
  • the invention uses the QRS band as a reference to generate a signal for convenient detection, and the heartbeat cycle length is uniform, thereby eliminating the difference caused by heart rate variability.
  • PAC(X) is a PCA dimensionality reduction for the signal X after the waveform
  • LDA is a linear discriminant analysis for the signal X after the waveform
  • DCT is a discrete cosine transform of the signal X after the waveform.
  • the present invention also provides an ECG-based access control system using any of the above authentication methods.
  • the power supply device 101 is connected to the main controller 106.
  • the ECG acquisition device 102 is used to collect the user's heart.
  • the electrical signal is transmitted to the main controller 106; the main controller 106 performs preprocessing, feature extraction, dimensionality reduction, generation of feature templates, acquisition of optimal thresholds, and acquisition of ECG signals or authentication of the collected ECG signals.
  • the button 103 is connected to the main controller 106 for setting and determining the working state of the access control system;
  • the door locker 104 is connected to the main controller 106 for opening the access control system. Switch or maintain the access control system off state.
  • the ECG acquisition device 102 performs ECG signal acquisition through two electrodes.
  • a transmission device 107 is further included for transmitting the ECG signal collected by the smart wearable device to the main controller 106.
  • the transmission device 107 comprises Bluetooth or WIFI.
  • the ECG-based access control system further includes an alarm 108, and the alarm 108 is connected The master will issue an alert if the user authentication fails.

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

一种基于心电认证的门禁系统的认证方法，包括对预处理过的心电信号采用自相关变换算法进行特征提取，获取心电自相关序列，再通过正交多项式拟合回归对心电自相关序列进行降维，生成特征模板，之后选择和评估出心电最优特征模板，获取最佳阈值，用户基于心电最优特征模板，最佳阈值进行身份认证，该认证方法安全性高，同时识别准确度高、认证存储信息小。

Description

一种基于心电认证的门禁系统及其认证方法 技术领域
本发明涉及身份认证技术领域，具体涉及一种基于心电认证的门禁系统及其认证方法。
背景技术
本项研究工作得到了中国国家自然科学基金资助(项目批准号：61571268)。
随着电子技术的发展，日常生活用品的品质也在不断提升，门禁系统在向着高科技化、智能化方向发展。传统的门禁系统有机械锁式的、密码式的、卡片式的等等。然而这些门禁系统存在一定的缺点：机械锁式的门禁系统由钥匙来控制，但是钥匙容易携带，数量多了携带也不方便；密码式的门禁系统密码容易泄露、安全性差；卡片式的门禁系统由于卡片容易被复制、窃取，安全性也不高。近年来，生物特征识别技术被应用到了门禁系统上，其主要利用人体的生理或行为特征，自动实现对个人身份的识别和认证，进行控制门禁系统的开关。
目前应用到门禁系统上的生物特征识别方法主要有指纹、人脸、语言，这些方式相比传统的门禁系统在安全性上有了很大的提高，但还是存在被模仿、复制、窃取或盗用的危险。
发明内容
为了解决现有技术识别安全性低的问题，本发明提出一种基于心电认证的门禁系统及其认证方法。
本发明的技术问题通过以下的技术方案予以解决：
一种基于心电认证的门禁系统的认证方法，包括心电注册和心电认证步骤，心电注册步骤包括：S11、将门禁系统调至心电注册状态；S12、通过心电采集设备采集注册者的心电信号并进行预处理，检测R波位置，截取QT波段；S13、将截取的QT波段采用自相关变换算法进行特征提取，获取心电自相关序列；S14、将获取的心电自相关序列通过正交多项式拟合回归进行降维，生成特征模板；S15、从生成的特征模板中选择和评估出心电最优特征模板；S16、从心电 最优特征模板中获取最佳阈值；心电认证步骤包括：S21、将门禁系统调至心电注册状态；S22、通过心电采集设备采集测试者的心电信号并进行预处理，检测R波位置，截取QT波段；S23、将截取的QT波段采用自相关变换算法进行特征提取，获取心电自相关序列；S24、将获取的心电自相关序列通过正交多项式拟合回归进行降维，生成特征模板；S25、将生成的特征模板与心电最优特征模板进行特征比对，根据最佳阈值完成认证，若认证通过，则门锁器打开，若认证失败，则门锁器维持关闭状态。
本发明与现有技术对比的有益效果包括：本发明提出了一种基于心电认证的门禁系统及其认证方法。该方法对预处理过的心电信号采用自相关变换算法进行特征提取，获取心电自相关序列，再通过正交多项式拟合回归对心电自相关序列进行降维，生成特征模板，之后选择和评估出心电最优特征模板，获取最佳阈值，用户基于心电最优特征模板，最佳阈值进行身份认证，该认证方法安全性高，同时识别准确度高、认证存储信息小。
附图说明
图1是心电用户进行注册的流程图。
图2是心电用户进行认证的流程图。
图3是本发明具体实施方式中的一种门禁系统的结构框图。
具体实施方式
下面对照附图并结合优选的实施方式对本发明作进一步说明。
一种基于心电认证的门禁系统的认证方法，包括心电注册和心电认证步骤，如图1所示，所述心电注册步骤包括：
步骤201，将门禁系统调至心电注册状态；
步骤202，通过心电采集设备采集注册者的心电信号并进行预处理，检测R波位置，截取QT波段；
步骤203，将截取的QT波段采用自相关变换算法进行特征提取，获取心电自相关序列；
步骤204，将获取的心电自相关序列通过正交多项式拟合回归进行降维，生 成特征模板；
步骤205，从生成的特征模板中选择和评估出心电最优特征模板；
步骤206，从心电最优特征模板中获取最佳阈值，完成用户注册，若用户注册失败则重新注册。
在本具体实施方式中，取注册时特征模板向量两两之间的最小距离为(thd_down)，最大距离为(thd_up)，那么阈值的取值范围为(thd_down，thd_up)，迭代次数为iternum，则变化的步长为
Figure PCTCN2017074437-appb-000001
阈值的取值为
Figure PCTCN2017074437-appb-000002
其中i＝1，2,...,iternum。
错误接受率(FAR)与错误拒绝率(FRR)都是阈值的函数，在门禁系统中，发生错误接受和错误拒接的代价是不同的，假设发生错误接受的代价为cost1,发生错误拒绝的代价为cost2(cost1>cost2)，首先根据FAR和FRR做出ROC曲线，再根据ROC曲线做出代价曲线，选择使得总体代价最小所对应的阈值为最佳阈值best_thd。
如图2所示，所述心电认证步骤包括：
步骤301，将门禁系统调至心电认证状态；
步骤302，通过心电采集设备采集测试者的心电信号并进行预处理，检测R波位置，截取QT波段；
步骤303，将截取的QT波段采用自相关变换算法进行特征提取，获取心电自相关序列；
步骤304，将获取的心电自相关序列通过正交多项式拟合回归进行降维，生成特征模板；
步骤305，将生成的特征模板与心电最优特征模板进行特征比对，根据最佳阈值完成认证，当
Figure PCTCN2017074437-appb-000003
则认证通过，否则认证不通过。
如果用户通过认证，则门禁系统的开关打开，若认证失败，则门禁系统的开关维持关闭状态。优选地，若认证失败，发出警报。
在本具体实施方式中，步骤203或步骤303中所述的自相关变换算法的公式为
Figure PCTCN2017074437-appb-000004
其中，x[i]表示心电序列，N表示心电序列的长度，x[i+m]表示对心电序列平移m个序列后的心电序列，m＝0,1，2,...,M-1，M<<N，Rxx[m]表示心电信号自相关序列，Rxx[0]表示心电序列的能量。
需要说明的是，不同个体之间自相关序列Rxx[m]具有明显的差异性，可作为个体心电信号的固有特征。由于QRS波群是心电信号在不同种测试环境下变化最小最稳定的成分，所以m的取值接近QRS波的长度，心电序列x[i]经过自相关处理后的自相关序列Rxx[m]仍是高维信号，需要进行降维处理。
在本具体实施方式中，步骤204或步骤304中所述的通过正交多项式拟合回归进行降维，生成特征模板是通过用多项式近似表示心电自相关序列，得出用特征模板表示所述心电自相关序列。
具体地，所述多项式为a0+a1f1(xi)+a2f2(xi)+...+akfk(xi)＝ATFi≈Rxx[i]。其中A表示特征模板，且上述多项式中，A＝(a0,a1,a2,...,ak)T,Fi＝(1,f1(xi),f2(xi),...,fk(xi))T,i＝0,1,2,3,...,M-1,其中1,f1(xi),f2(xi),...,fk(xi)分别是x的0次，1次，2次,...，k次正交多项式，其中
Figure PCTCN2017074437-appb-000005
即
Figure PCTCN2017074437-appb-000006
f为心电信号的采样频率。
特征模板的计算公式为：
Figure PCTCN2017074437-appb-000007
其中,λ,α∈(0,1)，λ，α为正则化系数，
Figure PCTCN2017074437-appb-000008
为求解后的向量A和Fi,得到的特征模板为A＝(a0,a1,a2,...,ak)T，k<<M，采取n个模板生成n个特征模板A1,A2,...,An，10≤n≤20。
在本具体实施方式中，步骤205中所述的心电最优特征模板是采用留一法获得的，判别式为
Figure PCTCN2017074437-appb-000009
其中D(Ai,Aj)表示特征向量Ai和特征向量Aj之间的距离度量；
Figure PCTCN2017074437-appb-000010
表示当特征Ai与特征Aj的间距小于预设阈值THD时取1，否则取0；THD的取值为n个特征向量之间距离的平均值，i，j取值为1到n，i≠j。
当上式条件满足时，第i个特征模板被选为优质特征模板；不满足时，第i个模板即为异常值，被剔除。最终选出最优特征模板，A1,A2,...,Anl，其中nl≤n。
在本具体实施方式中，步骤202或302中所述的截取QT波段是以所述R波左侧90毫秒内的最小值点为Q波点，所述R波右侧300毫秒以内最大值点为T波峰值点，以所述T波峰值点右侧一阶差分首次由负到正的位置为所述T波形截止点，再通过波形矫正生成固定长度的所述QT波段。
在本具体实施方式中，步骤202或302中所述预处理包括：对心电信号进行滤波处理，采集用户一定时长的心电信号，采用合适的滤波器滤出工频干扰、基线漂移及肌电干扰等噪声。优选地，对工频50Hz的频率点进行陷波，除去波形中50Hz工频干扰；使用截止频率40Hz的巴特沃斯低通滤波器滤除肌电干扰；使用大于1Hz的高通滤波器消除基线漂移。
在一些实施例中，在步骤203或步骤303中所述特征提取还包括采用用于稀疏表示的区分字典学习算法获取，具体地，
Figure PCTCN2017074437-appb-000011
其中，J(D,C)是求解后的字典D和稀疏特征C，Verif(Xi,Xj,D,Ci,Cj)是特征区分属性，λ为稀疏程度系数，α为正则化系数，λ和α取值范围都为0到1之间。
Xi与Xj分别表示第i个和第j个QT波，Ci和Cj分别表示与Xi和Xj相对应的稀疏特征。其中,i≠j。
Figure PCTCN2017074437-appb-000012
其中，dm是设定的不同类之间的最小距离，label(Xi)表示Xi的类别编号。
s.t.||dj||＝1,1≤j≤l
其中，X＝(X1,X2,...,Xn)表示n个QT波；D＝(d1,d2,...,dl)表示字典的维数，l为大于1的任意数值；
Figure PCTCN2017074437-appb-000013
表示稀疏特征。
在另一些实施例中，在步骤203或步骤303中所述特征提取包括以下步骤：
C1：确定截取心电信号上至少包括一个心电信息的窗口长度，窗口长度大于一个心拍长度，确保每个窗口至少包含一个心拍的完整信息。正常人的心跳在60-100次/分钟，特殊人群一般也在40次/分钟以上，因此窗口长度选择1-2秒以上，即可确保窗口中至少包含一个心拍的完整信息，这里一个心拍的完整信息，不局限于同一个心拍，也包含两个心拍的不同部分能够组合成一个心拍的完整信息。窗口d固定之后，训练和测试时窗口长度不再变化。
C2：根据确定好的窗口长度，滑动窗口，从心电信号的任意位置截取对应长度的心电窗口，截取心电窗口过程中，对窗口的起始点位置没有任何限制，在实时测试阶段尤为重要。
C3：将所述心电窗口划分为多个的定长片段，得到多个所述特征片段，每个心电窗口划分为n个任意定长片段，其中n大于或等于1，假设由窗口d截取的心电窗口为x，任意定长划分为n个特征片段{x1，x2,...,xn}，任意定长片段指片段长度小于心电窗口长度且固定。
特征片段通过全自动特征提取层包括两个阶段：卷积层和最大池层；其中：包括以下步骤：
A21：通过多个所述卷积层对所述特征片段进行并行卷积，得到多个向量值，此处为多个1×m维向量值。其中卷积层个数为n，每个卷积层层数大于1，卷积核K为一维卷积核；n个特征片段{x1，x2,...,xn}通过卷积层后生成n个向量 {c1，c2,...,cn}，其中：
Figure PCTCN2017074437-appb-000014
Figure PCTCN2017074437-appb-000015
其中i取值范围为[1,n]，l为卷积层数，b为偏置，初始值可置零。
n个向量{c1，c2,...,cn}直接生成矩阵A＝[c1，c2,...,cn]m×n。
A22：多个所述向量值通过所述最大池层生成深度融合特征。最大池核尺寸为1×n，最大池核作用于矩阵A后生成深度融合特征DeepFusionFeature＝[f1,f2,...fm]T。最大池核作用于矩阵A：
fi＝max(ci1，ci2,...,cin)；
A23：所述深度融合特征通过所述全连接层进行训练分类，输出分类判断结果，根据分类判断结果抽取所述全自动特征提取层作为所述特征提取器。
在某些实施例中，基于心电认证的门禁系统的认证方法中特征提取的方法不同，包括：首先检测心电信号中各个基准点以提取出准周期性的心搏作为原始心电特征。心电信号是一种准周期的信号，但并不是整个心搏周期中的成分都具有特异性，其中每个心搏周期中的P波、QRS波群和T波包含了大部分的心电特异性信息。本发明实施例从连续的心电信号中截出各个心搏周期中的波段作为原始的心电特征。为此，要定位出心搏的基准点。此外，在后续的波形矫正环节，还需要进一步对P波和T波进行处理。因此，需要定位出这些波形的关键位置，将这些点统称为基准点。本发明实施例针对每个心搏检测的基准点包括：P波起点(Ps)和P波终点(Pe)，R波峰(R)，J波起点(J)，T波峰(Tp)和T波终点(Te)，共计6类基准点。
其中，心电信号总体比较缓和，R波为最尖锐的部分。R波位于信号二阶差分的极小值位置，并且一阶差分为0。本发明实施例用原信号的二阶差分信号的极小值确定R波的粗略位置。定位出R波的粗略位置后，再根据R波幅值处于极大值位置这一特点，其一阶导数为0，在离散情况下，即一阶差分信号最接近于零的那一个，据此定位精确的R波峰位置。
进一步地，以各R波左侧160-180毫秒范围内一处优选如170毫秒处为P波起点Ps；以各R波左侧80-100毫秒范围内一处优选如90毫秒处为P波终点Pe；以各R波峰右侧80-100毫秒范围内一处优选如90毫秒处为J波起点(J)； 以各R波峰(R)右侧一段区域内的最大值为T波峰(Tp)，该段区域从J波起点开始到2/3个当前RR间期(即相邻两个R波峰之间的时长)处截止；以T波峰(Tp)右侧一阶差分信号首次由负到正的位置为T波终点(Te)。
由于心率的变化，各个准周期内的心搏并不相同，因此本发明实施例提出了一种分段波形矫正的方法来消除心率变异的影响，矫正的基本方法是对原心搏信号进行分段重采样，具体的，对P波段进行上采样，经过上采样后延长P波段时长，统一为460-500毫秒，优选如480毫秒；对于QRS波段保持不变，例如长180毫秒；对于T波段，分别对其中J-Tp段和Tp-Tp段进行下采样，使得重采样后两小段各时长统一为10-20毫秒，优选如15毫秒。最终，矫正后的心搏总长基本一致，例如为690毫秒。由于人在不同时间和经过不同运动后心率是不一样的，而这心率的差异不应该成为衡量人身份特征的标准。本发明以QRS波段为基准，生成一种方便检测的信号，心搏周期长度一致，从而消除心率变异带来的差异。
波形矫正后的信号X，对X进行特征提取和特征重组，重组后特征F：
F＝λ1×PCA(X)+λ2×LDA(X)+λ3×DCT(X)
其中，λ1,λ2,λ3是特征重组系数，取值范围均为[0,1]，且满足λ1+λ2+λ3＝1。PAC(X)是对波形后的信号X进行PCA降维，LDA是对波形后的信号X进行线性判别分析降维，DCT是对波形后的信号X进行离散余弦变换。
同时本发明还提供了采用上述任一认证方法的基于心电认证的门禁系统，如图3所示，包括电源供给装置101，连接主控制器106；心电采集设备102用来采集用户的心电信号，并传输到所述主控制器106；所述主控制器106对采集的心电信号进行预处理、特征提取、降维、生成特征模板、获取最佳阈值，进行心电注册或认证，并将所有数据储存在存储器105；按键103连接所述主控制器106，用来设定、更定门禁系统的工作状态；门锁器104连接所述主控制器106，用来打开门禁系统的开关或维持门禁系统关闭状态。
进一步地，所述心电采集设备102是通过双电极进行心电信号采集。
更进一步地，还包括传输装置107，用于将智能可穿戴设备采集的心电信号传输给主控制器106。优选地，所述传输装置107包括蓝牙或WIFI。
优选地，基于心电认证的门禁系统还包括报警器108，所述报警器108连接 所述主控器，如果用户认证失败，则会发出警报。
以上内容是结合具体的优选实施方式对本发明所作的进一步详细说明，不能认定本发明的具体实施只局限于这些说明。对于本发明所属技术领域的技术人员来说，在不脱离本发明构思的前提下，还可以做出若干等同替代或明显变型，而且性能或用途相同，都应当视为属于本发明的保护范围。

Claims (10)

  1. 一种基于心电认证的门禁系统的认证方法，包括心电注册和心电认证步骤，其特征在于，所述心电注册步骤包括：
    S11、将门禁系统调至心电注册状态；
    S12、通过心电采集设备采集注册者的心电信号并进行预处理，检测R波位置，截取QT波段；
    S13、将截取的QT波段采用自相关变换算法进行特征提取，获取心电自相关序列；
    S14、将获取的心电自相关序列通过正交多项式拟合回归进行降维，生成特征模板；
    S15、从生成的特征模板中选择和评估出心电最优特征模板；
    S16、从心电最优特征模板中获取最佳阈值；
    所述心电认证步骤包括：
    S21、将门禁系统调至心电认证状态；
    S22、通过心电采集设备采集测试者的心电信号并进行预处理，检测R波位置，截取QT波段；
    S23、将截取的QT波段采用自相关变换算法进行特征提取，获取心电自相关序列；
    S24、将获取的心电自相关序列通过正交多项式拟合回归进行降维，生成特征模板；
    S25、将生成的特征模板与心电最优特征模板进行特征比对，根据最佳阈值完成认证的步骤，若认证通过，则门禁系统的开关打开，若认证失败，则门禁系统的开关维持关闭状态。
  2. 如权利要求1所述的基于心电认证的门禁系统的认证方法，其特征在于，在步骤S13或步骤S23中所述的自相关变换算法的公式为
    Figure PCTCN2017074437-appb-100001
    其中，x[i]表示心电序列，N表示心电序列的长度，x[i+m]表示对心电序列平移m个序列后的心电序列，m＝0,1，2，...，M-1，M<<N，Rxx[m]表示心电信号自相关序列，Rxx[0]表示心电序列的能量。
  3. 如权利要求1所述的基于心电认证的门禁系统的认证方法，其特征在于，在步骤S14或步骤S24中所述的通过正交多项式拟合回归进行降维生成特征模板是通过用多项式近似表示心电自相关序列，得出用特征模板表示的所述心电自相关序列。
  4. 如权利要求3所述的基于心电认证的门禁系统的认证方法，其特征在于，所述多项式为a0+a1f1(xi)+a2f2(xi)+...+akfk(xi)＝ATFi≈Rxx[i]，其中A表示特征模板，所述多项式中A＝(a0,a1,a2,...,ak)T,Fi＝(1,f1(xi),f2(xi),...,fk(xi))T,i＝0,1,2,3,...,M-1,其中1,f1(xi),f2(xi),...,fk(xi)分别是x的0次，1次，2次,...，k次正交多项式，其中即
    Figure PCTCN2017074437-appb-100003
    f为心电信号的采样频率。
  5. 如权利要求1所述的基于心电认证的门禁系统的认证方法，其特征在于，在步骤S15中所述的心电最优特征模板是采用留一法获得的，判别式为
    Figure PCTCN2017074437-appb-100004
    其中D(Ai,Aj)表示特征向量Ai和特征向量Aj之间的距离度量；
    Figure PCTCN2017074437-appb-100005
    表示当特征Ai与特征Aj的间距小于预设阈值THD时取1，否则取0；THD的取值为n个特征向量之间距离的平均值，i，j取值为1到n，i≠j。
  6. 如权利要求1-5任一所述的基于心电认证的门禁系统的认证方法，其特征在于，在步骤S12或步骤S22中所述心电信号是通过有线和/或无线方式进行采集。
  7. 一种采用权利要求1-6任一所述认证方法的基于心电认证的门禁系统，其特征在于，包括电源供给装置，连接主控制器；心电采集设备用来采集用户的心电信号，并传输到所述主控制器；所述主控制器对采集的心电信号进行预处理、特征提取、降维、生成特征模板、获取最佳阈值，进行心电注册或认证，并将所有数据储存在存储器；按键连接所述主控制器，用来设定、更定门禁系统的工作 状态；门锁器连接所述主控制器，用来打开门禁系统的开关或维持门禁系统关闭状态。
  8. 如权利要求7所述的基于心电认证的门禁系统，其特征在于，还包括传输装置，用于将智能可穿戴设备采集的心电信号传输给所述主控制器。
  9. 如权利要求8所述的基于心电认证的门禁系统，其特征在于，所述传输装置包括蓝牙或WIFI。
  10. 如权利要求7-9任一所述的基于心电认证的门禁系统，其特征在于，所述心电采集设备是通过双电极进行心电信号采集。
PCT/CN2017/074437 2017-02-22 2017-02-22 一种基于心电认证的门禁系统及其认证方法 WO2018152711A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2017/074437 WO2018152711A1 (zh) 2017-02-22 2017-02-22 一种基于心电认证的门禁系统及其认证方法
CN201780002092.9A CN107980151B (zh) 2017-02-22 2017-02-22 一种基于心电认证的门禁系统及其认证方法

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2017/074437 WO2018152711A1 (zh) 2017-02-22 2017-02-22 一种基于心电认证的门禁系统及其认证方法

Publications (1)

Publication Number Publication Date
WO2018152711A1 true WO2018152711A1 (zh) 2018-08-30

Family

ID=62006106

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/074437 WO2018152711A1 (zh) 2017-02-22 2017-02-22 一种基于心电认证的门禁系统及其认证方法

Country Status (2)

Country Link
CN (1) CN107980151B (zh)
WO (1) WO2018152711A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110403599A (zh) * 2019-08-19 2019-11-05 深圳旭宏医疗科技有限公司 心电监测方法、装置、计算机设备和存储介质
WO2020053610A1 (en) 2018-09-14 2020-03-19 SOLOVEV, Sergei, Vladimirovich A method for automatic contactless authentication
RU2719023C1 (ru) * 2019-07-10 2020-04-16 Общество с ограниченной ответственностью «Компоненты и технологии 3Д» Способ регистрации электрокардиограммы водителя транспортного средства и устройство для его осуществления
CN113520403A (zh) * 2021-06-21 2021-10-22 浙江好络维医疗技术有限公司 一种基于峰谷特征的心电伪差识别方法
CN117338309A (zh) * 2023-08-21 2024-01-05 合肥心之声健康科技有限公司 一种心电信号近似度阈值计算方法、身份识别方法及存储介质

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109875570B (zh) * 2019-01-30 2020-08-18 华南理工大学 一种运动前后心电信号身份识别的有效方法
CN110269625B (zh) * 2019-05-31 2022-02-11 杭州电子科技大学 一种新型的多特征融合的心电认证方法及系统
CN112016405B (zh) * 2020-08-07 2022-07-15 浙江大学 一种基于可穿戴设备的心电图身份认证方法
CN114469123B (zh) * 2022-01-30 2024-04-09 北京理工大学 运动过程中的心电数据分类与健康特征识别方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102231213A (zh) * 2011-06-29 2011-11-02 哈尔滨工业大学深圳研究生院 Ecg门禁卡身份识别方法及系统
CN102929487A (zh) * 2012-10-31 2013-02-13 广东欧珀移动通信有限公司 一种移动终端的解锁方法及装置
CN104055522A (zh) * 2014-07-01 2014-09-24 清华大学深圳研究生院 一种心律失常情况下心电信号身份识别方法
CN104090722A (zh) * 2014-07-08 2014-10-08 惠州Tcl移动通信有限公司 一种基于心电图的对电子设备进行解锁的方法及系统
US20160042219A1 (en) * 2014-08-07 2016-02-11 Samsung Electronics Co., Ltd. User authentication method and apparatus based on fingerprint and electrocardiogram (ecg) signal

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1638463A4 (en) * 2003-07-01 2007-11-21 Cardiomag Imaging Inc USE OF MACHINE LEARNING FOR CLASSIFICATION OF MAGNETOCARDIOGRAMS
PL3462449T3 (pl) * 2014-01-24 2021-06-28 Nippon Telegraph And Telephone Corporation Urządzenie, sposób i program do analizy liniowo-predykcyjnej oraz nośnik zapisu
CN104102915B (zh) * 2014-07-01 2019-02-22 清华大学深圳研究生院 一种心电异常状态下基于ecg多模板匹配的身份识别方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102231213A (zh) * 2011-06-29 2011-11-02 哈尔滨工业大学深圳研究生院 Ecg门禁卡身份识别方法及系统
CN102929487A (zh) * 2012-10-31 2013-02-13 广东欧珀移动通信有限公司 一种移动终端的解锁方法及装置
CN104055522A (zh) * 2014-07-01 2014-09-24 清华大学深圳研究生院 一种心律失常情况下心电信号身份识别方法
CN104090722A (zh) * 2014-07-08 2014-10-08 惠州Tcl移动通信有限公司 一种基于心电图的对电子设备进行解锁的方法及系统
US20160042219A1 (en) * 2014-08-07 2016-02-11 Samsung Electronics Co., Ltd. User authentication method and apparatus based on fingerprint and electrocardiogram (ecg) signal

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020053610A1 (en) 2018-09-14 2020-03-19 SOLOVEV, Sergei, Vladimirovich A method for automatic contactless authentication
RU2719023C1 (ru) * 2019-07-10 2020-04-16 Общество с ограниченной ответственностью «Компоненты и технологии 3Д» Способ регистрации электрокардиограммы водителя транспортного средства и устройство для его осуществления
CN110403599A (zh) * 2019-08-19 2019-11-05 深圳旭宏医疗科技有限公司 心电监测方法、装置、计算机设备和存储介质
CN113520403A (zh) * 2021-06-21 2021-10-22 浙江好络维医疗技术有限公司 一种基于峰谷特征的心电伪差识别方法
CN113520403B (zh) * 2021-06-21 2024-04-02 浙江好络维医疗技术有限公司 一种基于峰谷特征的心电伪差识别方法
CN117338309A (zh) * 2023-08-21 2024-01-05 合肥心之声健康科技有限公司 一种心电信号近似度阈值计算方法、身份识别方法及存储介质
CN117338309B (zh) * 2023-08-21 2024-03-15 合肥心之声健康科技有限公司 一种身份识别方法及存储介质

Also Published As

Publication number Publication date
CN107980151B (zh) 2020-03-17
CN107980151A (zh) 2018-05-01

Similar Documents

Publication Publication Date Title
WO2018152711A1 (zh) 一种基于心电认证的门禁系统及其认证方法
CN105787420B (zh) 用于生物认证的方法、装置以及生物认证系统
CA2494491C (en) Method and apparatus for electro-biometric identity recognition
Labati et al. ECG biometric recognition: Permanence analysis of QRS signals for 24 hours continuous authentication
CN108460318B (zh) 基于心电信号的身份认证/识别方法及设备
Chamatidis et al. Using deep learning neural networks for ECG based authentication
Boumbarov et al. ECG personal identification in subspaces using radial basis neural networks
Shen et al. The PLR-DTW method for ECG based biometric identification
Wu et al. ECG biometric recognition: unlinkability, irreversibility, and security
El_Rahman Biometric human recognition system based on ECG
Nait-Ali Hidden biometrics: Towards using biosignals and biomedical images for security applications
KR101578167B1 (ko) 심전도 생체정보를 이용한 실시간 개인 인증방법
Rehman et al. ECG based authentication for remote patient monitoring in IoT by wavelets and template matching
CN112804937A (zh) 在生物识别中的高频qrs
Lee et al. Wearable Bio-Signal (PPG)-Based Personal Authentication Method Using Random Forest and Period Setting Considering the Feature of PPG Signals.
Labati et al. Adaptive ECG biometric recognition: a study on re-enrollment methods for QRS signals
CN109840451A (zh) 一种基于心电身份识别的智能支付可穿戴环及其支付方法
Safie et al. ECG biometric authentication using Pulse Active Width (PAW)
Hwang et al. Variation-stable fusion for PPG-based biometric system
Chandrashekhar et al. Pulse ID: the case for robustness of ECG as a biometric identifier
Canento et al. Review and comparison of real time electrocardiogram segmentation algorithms for biometric applications
Huang et al. Multi-view discriminant analysis with sample diversity for ECG biometric recognition
Matos et al. Biometric recognition system using low bandwidth ECG signals
Singh Individual identification using linear projection of heartbeat features
Komeili et al. Feature selection from multisession electrocardiogram signals for identity verification

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17898288

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 24.01.2020)

122 Ep: pct application non-entry in european phase

Ref document number: 17898288

Country of ref document: EP

Kind code of ref document: A1