Method for continuously authenticating the identity of an individual

Abstract

The invention relates to a method for continuously authenticating the identity of an individual, comprising a first step of authentication by recognizing at least one physiological signal of the individual transmitted by at least one sensor passively worn by the individual, and at least one step of iterative authentication of the identity, after the first authentication step, by recognizing at least one physiological signal of the individual, the recognition of the at least one physiological signal of the individual being performed during the first authentication step and during the step of iterative authentication from a set of authentication models specific to the individual, these models being unique for each individual to be authenticated and resulting from machine learning based on data of at least one physiological signal at least previously collected, at least in a previous training step, the authentication models specific to the individual being generated by processing a plurality of authentication test models designed by machine learning.

Description

[0001] This invention relates to the field of continuously authenticating the identity of individuals, specifically for authorized access.

[0002] Authentication involves proving a user's identity so that access can be authorized (access to locations, physical objects, software, etc.). To authenticate, a user must send their identity along with one or more authentication factors to the authentication system, which then verifies those factors.

[0003] The most common forms of authentication rely on passwords or access badges. However, this type of authentication is vulnerable to session hijacking. Furthermore, once a user has been authenticated, the system cannot verify that a user continuing to use the session after gaining access remains legitimate. In addition, biometric authentication (such as fingerprint recognition) has become more prevalent in recent years. However, fingerprint sensors can be spoofed by latex forgeries. Additionally, this authentication method is a one-time process; it authenticates a user once to log in, without requiring repeated authentication.

[0004] Therefore, the need to develop so-called continuous authentication has become apparent in recent years. Continuous authentication aims to re-authenticate a user several times during a session. For example, US Patent Application US2022012317 discloses a method and apparatus for implementing the method, which includes iterative biometric identification several times. The biometric identification described in this document involves voice recognition, fingerprint recognition, palm print recognition, or the recognition of facial features, hand shape, or vein patterns. However, this type of identification requires the user to actively interact with the biometric sensor for each identification request, which is undoubtedly inconvenient. In fact, it is already inconvenient for the user because the system notifies them that they must be identified again, and then they must spend time performing actions, such as interacting with the sensor by speaking or repositioning their fingers, hands, or face in an appropriate manner. Therefore, this continuous authentication method is too cumbersome for the user.

[0005] Patent application WO2012151680 also discloses a method for authenticating individuals, which can be continuously authenticated through iterative authentication over time. This document advantageously discloses authentication using physiological signals, such as ECG (electrocardiogram) and PPG (photoplethysmography), which are more difficult to forge. However, regardless of the number of individuals in the population, the method described in this document requires pre-recorded signals specific to each individual listed in the database to be input into the database beforehand. Authentication is performed by matching the measured physiological signals from the individual against the pre-recorded signals for that individual in the database. The pre-recorded signals are generated based on transformation rules derived from signals measured from a group of several individuals, using machine learning. However, this authentication method using template matching techniques has the following drawbacks: all individuals, along with their pre-recorded authentication signals derived from the transformation rules, need to be pre-stored in the database so that the measured signals can be compared with the pre-recorded signals. Another drawback is that although the pre-recorded signals are unique for each individual, they are generated using the same transformation rules for all individuals listed in the database. Ultimately, this method is not reliable enough.

[0006] Furthermore, the scientific publication "Opportunistic Multi-Modal User Authentication for Health-Tracking IoT Wearables" by ALEXA MURATYAN et al., dated September 28, 2021, describes research aimed at optimizing methods for authenticating individuals through training based on physiological signals. This disclosure does not disclose a continuous authentication method. It focuses solely on identifying the optimal classification-based training model for authenticating individuals. Several different training models using various classification methods (RF; KNN; NB; SVM-RBF; SVM-Poly) were tested on a set of individuals, and the results showed that the RF classifier model performed best. However, as previously explained, the same authentication model (RF model) was reused for each individual, ultimately lacking reliability. This prior art approach suffers from numerous biases, particularly demographic bias. In fact, it has been shown that two individuals with very similar physiological signals are indistinguishable.

[0007] Therefore, the object of the present invention is to provide a continuous authentication method that does not have the above-mentioned disadvantages, specifically a method that provides continuous authentication in a way that is transparent to the user and, most importantly, remains reliable and secure over time.

[0008] According to the present invention, the method for continuously authenticating the identity of an individual is as described in claim 1. Therefore, for each individual, performing each step of claim 1, and the final step involving selecting an authentication model for each individual, means that each individual thus has its own set of authentication models, i.e., each individual has a set of authentication models that differs from the set of authentication models of any other individual.

[0009] Specifically, according to the present invention, a method for continuously authenticating the identity of an individual (hereinafter sometimes referred to as a legitimate individual) comprises: a first authentication step (referred to as initial identity authentication) of identifying at least one physiological signal of the individual transmitted by at least one sensor passively worn by the individual; and, after the first authentication step, at least one identity authentication iteration step (to verify identity authentication) of identifying the at least one physiological signal of the individual. The authentication method is characterized by:

[0010] - Prior to the authentication and authentication iteration steps, the method includes: i) a so-called prior training step (relative to the individual), which includes processing data of at least one physiological signal specific to the individual (the data being based on the discriminative characteristics of the at least one physiological signal of the individual) through machine learning to generate multiple authentication test models; ii) using the individual's data to test the multiple authentication test models as data for the individual continues to be acquired; iii) ranking the authentication test models according to the probability that the authentication test model matches the data of the individual; and iv) retaining only a selected group (matching legitimate individuals) of the authentication test models ranked with the highest matching probability, which constitute the so-called authentication models retained for the authentication and authentication iteration steps (these authentication models are specific to each individual and therefore different from the authentication models of any other individual to be authenticated; for each individual, the preliminary step of generating several authentication models is performed);

[0011] The authentication and authentication iteration steps include: comparing the individual's acquired data with the authentication model specific to that individual, assigning a match probability to the result of the comparison with each authentication model in the authentication model, and processing the match probability to determine whether to accept or reject the authentication. An acceptance result indicates that the individual is indeed legitimate.

[0012] According to preferred features, in step i), which includes generating multiple authentication test models, data processing utilizes not only individual data but also data from at least one known imposter (training is performed using individual data and data from one or more known imposters), wherein the amount of data used by all known imposters considered when generating multiple test models must not exceed the ratio between the total amount of data from (legitimate) individuals and the number of known imposters. Using data from known imposters enables, for example, rejecting test models that might achieve a high matching probability, even though that probability should be low in reality, and also enables validating test models with high probabilities because these models can identify imposter data. This increases the reliability of the models reserved for authentication and authentication iteration steps. Additionally, considering a limited amount of data from known imposters relative to the amount of data from individuals avoids the risk of generating test models that consider data from known imposters more than individual data.

[0013] Therefore, several authentication models are used, and the comparison results generated by these models are analyzed to identify individuals; additionally, these authentication models are unique for a single individual because they are derived from models generated using only the individual's training data and optionally data from known imposters, rather than from transformation rules derived from a set of individuals whose data must be stored to list them. In fact, in [ Figure 4In the prior art, as illustrated in the illustration, a single model is created based on a single transformation rule (as in the aforementioned scientific publications), which has been developed from data of a given set of individuals (through machine learning, combining all data from all individuals in that set). Today, the entire scientific community still uses only a single model (single architecture) for all users. Although several instances of a model can coexist in the prior art (e.g., in the context of bagging), the implementation of several different models (different architectures) selected for each individual has not been proposed. The method of the present invention personalizes the authentication model for each user. Therefore, the method of the present invention not only generates a single model (which is more reliable), but it also does not only establish a single transformation rule. Thus, multiple models generated using individual data, and the fact that they have all been tested during prior training to retain only a small set of best models and are different from each other among individuals, provide a much more reliable authentication model for the authenticated individual and help reduce demographic bias associated with each biometric system (a problem that remains prevalent in biometric systems). Furthermore, the method of the present invention does not require data from all individuals in the group. Therefore, the method of the present invention uses data from a single individual and may additionally use data from at least one known imposter to generate several authentication test models, but does not require data from all individuals in the group to generate authentication test models. Thus, the method of the present invention does not require processing data from all individuals in the group to authenticate a single individual. Furthermore, each time a new individual not initially listed needs to be added to the database for authentication, it is not necessary to consider a new set of individuals and all their data (because the model reserved for authenticating that individual is independent of the authentication models reserved for other individuals, since there is no single transformation rule).

[0014] According to one feature, in step ii), at least one test model is tested using data from at least one other individual considered a known imposter, preferably using data from several known imposters. This helps in selecting the at least one authentication test model when the probability associated with the test model (as tested using data from known imposters) is appropriately low, since the imposter has been correctly detected. Preferably, all reserved authentication models are tested using data from known imposters (to confirm the effectiveness of the model when the associated probability is indeed low).

[0015] Based on one characteristic, it is known that the data of an imposter can belong to a real individual or can be digitally generated. When only the data of a legitimate individual is available, digital generation can be used to simulate one or more imposters.

[0016] In existing technologies, a given set of authorized individuals is considered to create transformation rules (through prior training). When a new individual needs to be added to the database of authorized individuals, the previous training process needs to be repeated for the entire new group to create new, more appropriate authentication rules because the group has changed. In contrast, the method of this invention does not require repeating the training process for all individuals every time a new individual needs to be added to the database of authorized individuals. In practice, the method of this invention generates a test model based on the data of the individual to be authenticated (legitimate user), and the data of another individual (a known imposter) is used only when necessary (though preferred) to generate and test that test model, and is assigned a very low probability value, or even a zero probability value, if the test model tested using the known imposter's data produces results too similar to those obtained using the data of the individual to be authenticated. This facilitates the generation of an even more reliable set of authentication models based on various test models; however, in any case, it is not necessary to use data from an exhaustive group of individuals corresponding to the group in which each individual is expected to be authorized after authentication. The method of this invention is implemented independently of the number of individuals to be authenticated.

[0017] The term “passive wearing” in relation to the use of sensors means that an individual does not need to perform any action on the sensor or even pay attention to it, but simply puts it on and wears it (the arrangement depends on the type of physiological signal being measured).

[0018] Physiological signals are understood to refer to any analog signal generated by the human body that can be measured and digitized.

[0019] Preferably, a decision tree is used to process the probabilities associated with the results of comparing the acquired data with the wear authentication model in order to determine whether to accept or reject the wear authentication.

[0020] According to one feature, the generation of authentication models and the authentication using these models do not require any active action from the individual.

[0021] Therefore, the authentication method of the present invention offers several advantages:

[0022] - The detection of signals that are necessarily physiological and the continuous repetition of the measurement of these physiological signals make the authentication request and its verification transparent to individuals who do not need to perform any action;

[0023] - The detection and processing of physiological signals reduces the risk of forgery because physiological signals have the advantage of being different for each individual;

[0024] - The additional authentication steps used to verify identity over time eliminate the risk of session hijacking. In fact, the fact that authentication is continuously verified during a session after the initial authentication by repeatedly measuring signals and analyzing them through the system ensures that the user can be continuously verified as the same person; if an authentication certificate has been issued, the session or access can remain open as long as authentication verification continues.

[0025] - In another intended use, the additional steps for authentication to verify identity over time enable detailed analysis of signals over a relatively long period of time before the issuance of authentication certificates;

[0026] - By generating several unique authentication models for each individual based on AI-based learning from the data of the individuals to be authenticated (which is more reliable than using a combination of data from a required group of individuals), and by assigning matching probabilities to each of these models during authentication, the reliability of the authentication process is enhanced. In fact, instead of providing the same data transformation rules based on a set of individuals and comparing data using only a single rule during authentication, the method of this invention enhances the reliability of the authentication assessment by testing several authentication models derived from data from legitimate users (and optionally additional data from one or more known imposters) and by evaluating their match relevance to said users. Furthermore, using data from known imposters to test the authentication models increases the reliability of the ranking of the best models to retain for the individuals to be authenticated.

[0027] - There is no need to store a large amount of data on individuals in order to build transformation rules based on all of this data, nor is there a need to perform a new training process to build new rules whenever one or more individuals are added to the list of individuals to be certified.

[0028] Based on a single characteristic, the authentication method builds a list of Y best known imposters for each legitimate user, and attempts to authenticate these Y best imposters each time a user authenticates (using various identification modules specific to each registered individual, relative to other individuals known in the authentication system). Specifically, this makes it possible to remove known individuals from the list of imposters.

[0029] Based on one feature, during the previous training steps, the individual to be certified must wear a sensor that detects at least one type of physiological signal for a specific period of time, during which the individual goes through rest, exercise and emotional phases (e.g., by listening to audio recordings and / or watching videos) in order to record the associated signals, and machine learning generates a relatively reliable certification model specific to that individual.

[0030] According to another feature, the authentication method includes the step of issuing an authentication certificate when authentication (true) is accepted, wherein an authentication iteration step is performed after the first authentication step and the issuance of the authentication certificate, or wherein the authentication iteration step is performed after the first authentication step and before the issuance of the authentication certificate. The execution of the authentication iteration step depends on the level of security associated with the authentication. Thus, the first implementation corresponds to high-frequency authentication on the order of one or a few seconds, for issuing an authentication certificate if the authentication is deemed true, and the authentication iteration step continues after the authentication certificate has been issued. The second implementation corresponds to authentication with what will be called a long identification duration, specifically, a duration of at least a few minutes or even tens of minutes, where the authentication certificate is issued only if the authentication is deemed true after the first authentication step and several authentication iterations within the selected identification duration; such analysis within the long measurement duration increases the legitimacy of the authentication result. This authentication implementation combining continuous authentication with the analysis of an individual's physiological signals provides highly reliable results; because forgery of physiological signals over long periods is relatively unlikely. This implementation can be very useful in high-security applications.

[0031] Advantageously, the authentication method implements machine learning during at least one authentication iteration step to update one or more authentication models specific to that individual. Based on a feature, a specific authentication model generated and retained during previous training steps constitutes a first authentication model, which is updated via machine learning during at least one iteration step of continuous authentication to constitute a new specific authentication model, which is preferably considered for the next authentication. Therefore, as long as the authenticated individual continues to wear the connected device, the authentication method continues to learn to recognize the individual's physiological signals. Thus, the learning module benefits from more data, which increases the reliability of the specific authentication model and, consequently, the reliability of the authentication method.

[0032] Preferably, the authentication method uses machine learning to update the authentication model and store the last N authentication models during the iterative authentication steps of an individual. These last N authentication models correspond to the most recently updated authentication models and those from one or more previous iterative steps. During new subsequent authentication periods (when the user is not wearing a sensor for a specific time period), the authentication method is executed using either the most recently stored authentication model or a combination of the N most recently stored authentication models. These stages of storing and retrieving the last authentication model or the combination of the last N models further increase the reliability of the authentication process and reduce the risk of rejecting authentications where the result should be true. In reality, an individual's mood may change over time, their health may change, and the obtained physiological signals may change accordingly; therefore, the method of the present invention enables the generation of authentication models that correspond as closely as possible to the individual's current physiological state.

[0033] According to another feature, the method considers signal quality when evaluating binary (true / false) authentication results, and preferably considers signal quality within a specific time period; specifically, signal quality is evaluated by processing signal noise.

[0034] According to another feature, the method measures and evaluates several types of physiological signals for a single authentication. According to one feature, the method implements a set of authentication models for each type of physiological signal, and the method includes an algorithm for evaluating the authentication based on a combination of recognition results generated by each set of authentication models for each of the multiple types of physiological signals.

[0035] According to another characteristic, one or more physiological signals are selected from PPG signals and / or ECG signals and / or the individual's physical activity and / or the individual's bioimpedance. Specifically, the authentication method also includes a step of assessing the individual's health status based on the acquired physiological signals. PPG signals use a plethysmography sensor, also known as a pulse oximeter, and involve measuring changes in blood volume by measuring the amount of light absorbed and reflected by blood vessels; PPG signals are associated with cardiac signals because fluctuations in blood volume occur with each heartbeat. Since each individual has their own unique PPG signal, this is a fairly reliable method for authenticating individuals. Additionally, PPG has the advantage of being a non-invasive technique.

[0036] According to another feature (in one embodiment), the method measures and evaluates data in addition to one or more physiological signals of the individual, said other data being processed such that the result is combined with the result of said one or more physiological signals to establish the identification, said other data being in particular biometric data used alone or in combination, such as fingerprints, facial features, irises, vein patterns, or passwords, said passwords may be one-time passwords, smart cards, badges, certificates, or encryption keys on removable media such as USB flash drives, or verification actions on a second device such as a telephone.

[0037] According to another feature, the method includes steps for detecting replay attacks and / or detecting spoofed signals. A replay attack occurs when a third party manages to capture an individual's physiological signals, record them, and then use them to impersonate that individual by transmitting them to a sensor, which then provides data that should normally be the individual's data to the device implementing the method of the invention.

[0038] According to another feature, the method includes the step of detecting spoofed signals (artificially generated signals that are therefore not derived from data from an individual wearing one or more sensors).

[0039] According to one feature, the method uses a secure communication protocol, such as HTTPS or a protocol chosen from other encryption protocols, specifically between one or more sensors worn by an individual and a server to which the device implementing the authentication method is connected.

[0040] The present invention also relates to a system for continuously authenticating the identity of individuals. This system includes an electronic processing component and an algorithm for implementing the authentication method described above. The processing component includes at least one identification module for each individual (and only for one type of physiological signal), an Identity and Access Management (IAM) module, and a data storage module. The individual identification module generates an authentication model specific to the individual and associated with the physiological signal through machine learning. The algorithm can be diverse, including, for example, support vector machines (also known as SVMs), neural networks, decision trees, principal component analysis, genetic algorithms, etc.

[0041] According to one feature, the processing components of the aforementioned authentication system include a replay detection module, a spoofing signal detection module, and a module for evaluating additional data specific to the individual (including the individual's health status).

[0042] Finally, the present invention relates to a computer program comprising code instructions that, when executed by a processor, perform the steps of the authentication method described above. Specifically, the program can be loaded via a network such as the Internet.

[0043] The invention will now be described by way of illustrative examples and with reference to the accompanying drawings, which are in no way intended to limit the scope of the invention. In the drawings:

[0044] -[ Figure 1 A flowchart is depicted that illustrates an authentication system capable of implementing the authentication method according to the present invention, which is used to authorize a user to access object X via a connection device worn by the user.

[0045] -[ Figure 2 The diagram illustrates a flowchart of an authentication system used to authenticate several individuals.

[0046] -[ Figure 3 The diagram schematically illustrates the steps of the previous training process that lead to the selection of a set of certification models specific to each individual.

[0047] -[ Figure 4 This illustration schematically depicts existing techniques related to certification by selecting a single model for all individuals.

[0048] The method for continuous authentication of an individual according to the present invention aims to first passively and without individual intervention (initial authentication), and then continue to passively and without individual intervention over time to verify that it is still the same individual (continuous authentication verification), the authentication being based on the individual's physiological signals.

[0049] Continuous authentication methods can be applied to a variety of purposes, such as accessing physical objects X, accessing software, accessing locations (via accessing objects such as doors), etc.

[0050] like[ Figure 1 As schematically illustrated in the diagram, the continuous authentication method is implemented by at least one connection device 1 worn by an individual and an authentication system 2 arranged remotely from the connection device 1 and receiving all data from said connection device 1. The authentication system 2 is capable of communicating with the object X. The authentication system 2 is capable of receiving and processing information from the connection device 1 to authenticate the individual wearing the connection device 1, so as to authorize them to access the object X, and continues to verify authentication as long as the connection device 1 is worn by the individual. The authentication system 2 is capable of receiving data from several connection devices 1-1, 1-2, 1-3, etc. Figure 2 The device receives information to authenticate individuals wearing the connected device so that, if they have been successfully authenticated, they are authorized to access objects (the same object or a different object for each individual).

[0051] Authentication System 2 (via the so-called IAM module) can issue an access authorization certificate after authentication, which is valid for a predetermined period of time, or must be updated with each authentication iteration to re-authenticate the correct identity and the validity of the access rights to be granted. Each certificate is signed using an asymmetric encryption system, allowing all parties to ensure the authenticity of each document and prevent document forgery.

[0052] The connecting device 1 may take the form of a smartwatch, a connecting garment worn directly on the skin, or a wristband connected to a telephone (which, if needed, can be used as an interface for controlling authentication and sending authorization certificates). The connecting device 1 must be worn by the individual and appropriately positioned on the individual depending on the nature of the physiological signals to be acquired.

[0053] The connection device 1 includes at least one sensor 3 that detects at least one physiological signal. The types of physiological signals detected include, for example, PPG signals (via a photoplethysmography sensor), ECG (via a sensor that measures heart activity), or signals indicating an individual's physical activity (via a three-dimensional accelerometer and gyroscope used as sensors), or bioimpedance signals (via an electrode-type sensor).

[0054] The authentication system 2 includes at least one authentication module 20 (also called an identification module), an identity and access management module 21 (also called an IAM module), and a data storage module 22. The identification module 20 is dedicated to a single individual. The authentication system 2 includes one identification module for each individual, or optionally several identification modules for each individual, each identification module relating to the measurement of a type of physiological signal. The identification module 20 has an algorithm implementing AI. The storage module 22 records a large amount of data, including all access requests.

[0055] The access request is made by the individual at object X. Object X issues an identifier ID to the connection device 1 worn by the individual; this identifier ID is unique to object X. From that moment on, the individual no longer needs to intervene but only needs to wait for access authorization. The main steps associated with the authentication method used to authorize the individual wearing connection device 1 to access object X are as follows:

[0056] - Connecting device 1 automatically connects to authentication system 2 and sends it the identifier ID of object X and the individual's physiological signals;

[0057] - If authentication system 2, specifically identification module 20, identifies a user, it issues an authentication certificate, which is transmitted to connection device 1 and stored in storage module 22 of authentication system 2;

[0058] - Connected device 1, having received the authentication certificate, requests access to the authorization certificate from authentication system 2, specifically from IAM module 21, by sending the object's identifier ID and the associated authentication certificate. A preferred encrypted copy of the access authorization certificate is stored in storage module 22;

[0059] -Authentication system 2, specifically, IAM module 21, first issues an identifier ID containing the individual requesting access to object X. user A copy of the access token is sent to connected device 1, and then a certificate authorizing access to object X is issued.

[0060] - Connect device 1 and then access the token and the individual's identifier ID. user Send it to object X, thereby unlocking object X and allowing the individual to access it.

[0061] The database of authentication system 2 includes one or more user identifiers (IDs). user The library associated with the identifier ID of object X.

[0062] Once an individual is authorized to access object X and within a given security period (e.g., one hour), the user does not need to interact with their connected device 1 as long as they wear it; continuous authentication continues transparently to the individual, and connected device 1 continues to measure the user's physiological signals and send them to authentication system 2, which issues an authentication certificate verifying that access is still authorized as long as identification is performed. Furthermore, it is anticipated that subsequent authentication verification during the user session can be performed by authentication system 2 even after the continuous authentication period has ended. Moreover, as will be seen later, the authentication method of the present invention advantageously continues to collect individual data during the session, even if continuous authentication has stopped (e.g., because the one-hour period has passed).

[0063] Therefore, authentication system 2 is able to:

[0064] - Identifies one or more connected devices 1;

[0065] -Process data received from one or more connected devices 1,

[0066] - In response to an access request, issue and send one or more authentication certificates, or reject the request due to authentication failure.

[0067] - Communicate with object X to issue a copy of the access token to that object.

[0068] -Issue access authorization certificates,

[0069] - Continuously issue and send authentication certificates.

[0070] - Store the data received and transmitted, specifically copies of authentication and authorization certificates.

[0071] In addition to the issuance of authentication and access authorization, authentication system 2 can detect, evaluate, and / or send other information, especially:

[0072] - Ensure long-term stability of recognition through continuous learning.

[0073] -Assess signal quality,

[0074] - Sending alarm,

[0075] - Detect replay attacks.

[0076] - Detect fake signal attacks.

[0077] - Detect brute-force attacks,

[0078] - Improve rejection performance relative to (unknown) imposters by using data from known imposters, where a known imposter is an individual whose authentication model has been tested with known individuals other than the individual being authenticated.

[0079] - Provides multiple pieces of information related to authentication / identification, such as, for example

[0080] • The health status of the individual wearing the connected device (sleep, stress, physical exertion leading to heart disease, etc.); for example, if the heart rate is high and the individual is not moving, the system infers a potential state of stress; or if the heart rate is low and the individual is not moving, the system infers that the individual is in a sleep stage or a state of reduced consciousness.

[0081] • The location or geographic location of an individual, or the detection of an individual's movement within a region.

[0082] Depending on the duration and number of iterations of the authentication process, authentication methods can provide several levels of security.

[0083] Furthermore, authentication system 2 has the advantage of being adaptable by adding multiple identification modules specific to each individual (every time a new user is added). Specifically ([ Figure 2 The authentication system 2 includes multiple identification modules 20-1, 20-2, 20-3, etc., each dedicated to an individual wearing a corresponding connection device 1-1, 1-2, 1-3, etc. This feature will be described in more detail below. Therefore, a unique authentication model for each individual (a set of authentication models for each type of physiological signal) is generated by an identification module dedicated to that individual, completely independent of the group of individuals. Thus, unlike existing technologies, it is not necessary to know and record data from multiple individuals belonging to the target group to establish data transformation rules dependent on all those individuals, nor is it necessary to perform a new training process based on data from previously registered individuals (which, incidentally, may have changed over time) and data from one or more individuals to be added to establish new transformation rules that consider both previously registered and added individuals.

[0084] More specifically, the authentication method includes the following steps:

[0085] - Step 1: Measure and record the signal output by the measuring device (such as sensor 3) within the acquisition period and preferably at the acquisition frequency;

[0086] Step 2: Send physiological signal data associated with the first item of identity information to authentication system 2, and optionally, simultaneously send the second item of identity information in the form of a unique key (encrypted or unencrypted) associated with sensor 3. If the measuring device includes a dedicated cryptographic processor, the unique key may be stored in the dedicated cryptographic processor;

[0087] Step 3: The authentication system 2 selects an identification module 20 dedicated to the user of the sensor 3; this is a personal identification module 20 for each individual to be authenticated, which stores a set of authentication models dedicated to the user in its memory. The process of generating this set of authentication models is described below. Authentication analysis is obtained by comparing the data obtained from the individual during the identification step with the authentication models specific to the individual stored in the identification module 20, then associating the matching probability with the comparison result of each authentication model, and finally processing the probabilities to determine whether the authentication is accepted (true) or rejected (false). Upon receiving an authentication request, each identification module 20 returns a binary result, true or false, which is associated with various probability values ​​or confidence scores generated from the comparison with each authentication model. Preferably, the true / false result is derived from a decision tree based on the probability values ​​obtained from the results of comparing the acquired data with the authentication models. In the algorithm implemented by the identification module 20, an algorithm provides initial authentication and several authentication iterations; these iterations are performed before and / or after the release of the true / false result, depending on the desired security level.

[0088] Step 4: Identification module 20 (20-1 for a given individual) transmits the true / false value to IAM module 21, which is then compared with the ID. user The probability value or confidence score of the comparison result with the authentication model is associated with the list of Y best identified imposters (these imposters are known imposters because they are considered other users and have their own identification modules 20-2, 20-3, etc.).

[0089] - Step 5: This step is optional but preferred; - IAM module 21 requests identification of Y known imposters by the identification modules of other individuals registered in authentication system 2. The identification modules of the other individuals independently return to IAM module 21 the true / false value of each of the Y known imposters, along with the associated probability or confidence score;

[0090] - Step 6: IAM module 21 issues a unique identity certificate for each authentication, mentioning several information items listed in the examples below.

[0091] In step 1, specifically, the acquisition duration for measuring the physiological signal for initial authentication is customized based on the nature of the physiological signal and the required level of security for the intended application. The acquisition for initial authentication is preferably periodic, for example, at a frequency of one second. Then, during the continuous verification phase of iterative authentication, the measurements are periodic and / or random.

[0092] Regarding steps 2 and 3, authentication system 2 advantageously includes several identification modules 20 (20-1, 20-2, 20-3, etc.), each dedicated to a single user. Identification modules 20 are personal modules because they are dedicated to a single individual based on a set of authentication models that are unique and depend only on the individual's data and are independent of the data of other individuals. Furthermore, identification modules 20 are dedicated to a specific type of signal, such as PPG, ECG, physical activity, or bioimpedance. Therefore, authentication system 2 may include several identification modules 20A, 20B, 20C, etc., for each individual, each dedicated to the measurement and authentication of a specific type of physiological signal (20A for PPG, 20B for ECG, etc.). Associated with each type of physiological signal is a set of authentication models, as described later, which have been pre-generated relative to the type of physiological signal being measured.

[0093] More specifically, for each individual, the identification module 20, which is personal (and independent of other individuals), is capable of:

[0094] - Prior to the authentication request (i.e., prior to step 1), a unique set of authentication models for each individual is generated by processing at least one type of physiological signal during a time period known as the prior training period; this step is called the prior training step.

[0095] - When an authentication request is received, the individual is authenticated by comparing one or more measured physiological signals against a set or more unique authentication models generated at the end of the previous training steps for each type of physiological signal (steps 1 to 6 mentioned above).

[0096] - During the authentication request process, learning continues using signals measured during each request to generate a set of authentication models that remain unique relative to the individual and are constantly updated (to account for any changes in the individual's physiological signals, which may vary over time, especially depending on the individual's health status). It should be noted that the prior art does not allow this step, which is referred to below as the step of updating the authentication model relative to the individual (because the prior art has a single transformation rule that remains unchanged over time).

[0097] For the purpose of initial authentication (step 1), prior training of user data is performed over a given time period according to the registration protocol. For example, prior training is performed throughout the day, during which the individual wears the connection device 1 and experiences different rest and activity periods, and even emotional states, at different times to obtain a wide range of the individual's heart rate range. Several variations of the registration protocol can be implemented. For example, the registration periods are different for sedentary individuals; specifically, they are shorter. During the prior training step corresponding to the first session, it is the operator of authentication system 2 who manually associates connection device 1 with the user. Connection device 1 contains a first token that can be used by the user as long as the user wears the connection device (e.g., a wrist-worn connection wristband). In parallel, authentication system 2 thus continuously records data from one or more sensors of connection device 1 to perform prior training, as indicated, for example, throughout the day. As the day progresses, authentication system 2 becomes increasingly reliable. Once the prior training phase is complete, the first token is replaced by an automatically issued authentication token from authentication system 2, and the continuous authentication method can be initiated immediately when needed.

[0098] During the previous training, the recognition module 20 performed the following steps (in […]) Figure 3 (Illustrated in the image) Implement one or more machine learning algorithms so that:

[0099] i) Generate multiple certification test models (e.g., approximately one hundred; this figure only illustrates a few models schematically represented by different geometric symbols) from the extraction of features related to a type of measured physiological signal of an individual, and then

[0100] ii) Use signals from individuals considered trustworthy to test the authentication test model.

[0101] iii) Rank the best test models (those with the highest probability of recognizing individuals), and

[0102] iv) Only one optimal set of test models (e.g., approximately ten) is retained, which is called the certification model (for simplicity, only a few models are depicted in the diagram, represented by different geometric symbols). Each individual has a personalized set of certification models. Each individual has a different set of several certification models (no two individuals have the same set of certification models).

[0103] Multiple authentication test models are obtained through machine learning based on the discriminative features of the acquired data. Multiple models generated using individual data, combined with the fact that they are all tested to retain only a small subset (i.e., the best models), provide the most reliable possible authentication model for authenticating individuals. This set of authentication models is linked to a unique key (ID) associated with the user's identity. user (related to)

[0104] Regarding step i) of extracting data related to physiological signals for generating multiple test models, identification module 20 (via an algorithm) implements known data processing steps or stages, namely, a preprocessing stage (specifically, windowing of a given type), filtering that can be performed before preprocessing (e.g., by using Fourier transform and / or digital filtering), extraction of (many) features, selection of those features (advantageously via machine learning, e.g., based on the PCA (Principal Component Analysis) function), and final classification (e.g., via SVM (Support Vector Machine) or via a neural network). Some neural networks are suitable for performing the extraction, selection, and classification steps. The algorithm used to perform these different stages can also be a genetic algorithm. For each generation of the test model, the steps or stages, optionally preprocessing, filtering, segmentation, normalization (optional), extraction, selection (which is the feature selection stage and can be optional), and classification, can each employ different types of methods (as illustrated above: different types of windowing, different filtering methods (Butterworth filter or others), different segmentation methods (e.g., based on duration or number of points), different extraction methods (Fourier transform, PCA, or DWT-db2 (discrete wavelet transform) etc.), and different classification methods (such as SVM, CNN that operates on the original signal, KNN). (db3, RF (Random Forest), DWT, and ADABOOST, etc.). Preferably, according to the invention, the algorithms used to perform these stages and subsequently generate multiple test models may optionally implement various combinations of different methods specific to each stage for certain test model results in a random manner. Therefore, the generated model examples are highly diverse, especially depending on different combinations of extraction and classification algorithms; for example, for user 1, this is at least one model that implements the extraction step via FFT (Fast Fourier Transform) followed by an RF classifier; for user 2, this is at least one model that implements DWT (Discrete Wavelet Transform), followed by SVM (Support Vector Machine), and CNN (Convolutional Neural Network) operating on the original signal; for user 3, this is at least one model that implements FFT with k-NN, FFT with RF, DWT, and ADABOOST, etc. Ultimately, each individual has its own authentication model.

[0105] Advantageously, in step i) of generating multiple authentication test models, data processing utilizes not only the data of the individual but also the data of at least one known imposter, wherein the amount of data used by all such known imposters considered during the generation of the multiple test models shall not exceed the ratio between the total amount of data of the individual and the number of known imposters.

[0106] Preferably, during step ii), the test model, which is at least retained as the authentication model, is also tested using data from at least one known imposter to verify that the confidence probability of identification using data from a known imposter is indeed extremely low, or even zero. This test using the imposter's data increases the reliability of the ultimately retained test model.

[0107] Regarding authentication, if an imposter is identified, the identification module 20 contains a list of the Y individuals with the highest imposter scores, and advantageously compares this list with other individuals registered and linked to its identification module. This makes it possible to remove known individuals from the imposter list if necessary.

[0108] Additionally, and very advantageously, the identification module 20 includes a machine learning algorithm that enables continued learning of the recognition of an individual's physiological signals while the individual continues to wear the connected device 1 after having already been initially authenticated; learning continues throughout the entire duration of continuous authentication verification, and even thereafter, while the connected device is worn (depending on a given number of times and / or a given time period). This continuous learning helps improve the legitimacy of the results of continuous authentication and subsequent initial authentication (upon a new access request), thereby providing enhanced security. Indeed, it has been found that initial authentication performance degrades by 15% to 30% when a relatively long period of time (such as a week) has passed before the same individual issues a new access request. However, the inventors have demonstrated that continuous learning during continuous authentication verification improves authentication performance not only during the continuous authentication process but also during subsequent authentication. Specifically, a portion of the signals used for continuous authentication throughout the day is retained to create a new dataset and train a new test model, thereby deriving a new, updated set of authentication models still specific to the user. This new, updated set of authentication models is stored on a chain (such as a blockchain), making it possible to track all training of the authentication models. During each new authentication, the most recently stored set of authentication models is used to determine the user's identity, thus contributing to higher reliability of the authentication results. Alternatively, the identification module 20 can use the last N authentication models (corresponding to the most recently stored set of authentication models and one or more other previously stored sets) to respond to authentication requests, thereby further enhancing security. Furthermore, storing successive versions of authentication models and associated authentication results in a chain facilitates investigations in the event of an attack on authentication system 2. If an attacker is able to modify the training dataset to insert their own personal information instead of the user's data, the system will be able to identify the attacker as an illegitimate user.

[0109] Regarding certification iterations, these can be carried out after the results have been determined so as to confirm the certification over time, or they can be carried out over a so-called long period of time so as to perform several verifications before the results are released.

[0110] In step 6, where the IAM module 21 generates the identity certificate, the data included in the certificate is, for example:

[0111] -Timestamp,

[0112] - Verify identity (true) or reject (false).

[0113] - Duration / expiration date of access authorization

[0114] - The identity being authenticated (the authenticated ID) user ),

[0115] - Identity token (which is unique and random).

[0116] Additional information, such as:

[0117] - A list of signals used for authentication.

[0118] - The decisions made by each identification module and their associated confidence scores,

[0119] -The wearer's health status,

[0120] -The activities detected

[0121] - The quantity and quality of the signal,

[0122] - The certificate's cryptographic signature.

[0123] This additional information is particularly useful during investigations or when an intrusion is detected. This additional information can be stored in a specific processing module linked to the intrusion.

[0124] To further enhance the effectiveness of authentication system 2 and contribute to its overall security (in order to deceive attackers), in addition to the data already listed above, various information items related to authentication and identification are preferably added, and these information items are recorded in individual records stored in storage module 22. This additional information specifically includes:

[0125] - Activities detected by accelerometers and gyroscopes (walking, running, resting, working on a computer, etc.).

[0126] - Emotional state (calm, tense, excited, afraid, etc.).

[0127] - Overall health status (good, poor, fair)

[0128] - Accident detection (falls, arrhythmia, heart attack, sudden changes in blood pressure, etc.)

[0129] - Internal geographical location (within a limited radius using proximity sensors, beacons, name tag systems, etc.)

[0130] -GPS geolocation

[0131] -Environmental information, such as light levels and ambient noise.

[0132] - Individual's body temperature (measured using appropriate sensors).

[0133] - Blood oxygen saturation (measured using an appropriate sensor).

[0134] Among the information listed above, the health status of a certified individual may be related to security or safety, depending on the intended application. For example, a rapid deterioration in the health status of a certified individual could lead to sudden health problems or an attack on the individual, which should be detected in highly sensitive environments such as military or nuclear facilities. This change in status is intended to be detected by the authentication system 2 to generate an alarm and / or trigger automated decisions. Specifically, the individual's health status is verified based on their heart rate, respiratory rate, sleep duration, body temperature, and oxygen saturation, parameters measured by means of a connected device 1 including appropriate sensors. The processing of this data produces a score related to the health status; depending on the score's value relative to a reference value, for example, if it falls below the reference value, the authentication system 2 sends an alarm, temporarily blocks access, or requests tiered approval.

[0135] In addition to the detection and processing of physiological signals, authentication methods can also use other biometric parameters to supplement authentication, such as through non-limiting examples: fingerprints, irises, faces, hand shapes, hand vein patterns, and specific gestures.

[0136] In addition to authentication via physiological signals, authentication methods can also utilize authentication protocols that do not use biometric data and rely on interaction with the object that the individual is requesting access to, such as passwords, one-time passwords, smart cards, badges, certificates, or encryption keys on removable media (USB flash drives, hard drives) or actions and / or verification on a second device such as a telephone.

[0137] Advantageously, the authentication method of the present invention takes into account the quality of the signal transmitted by the connecting device 1 and received by the authentication system 2 when evaluating the authentication result (true / false). Evaluating signal quality corresponds to evaluating at least the signal noise. Poor quality signals may be caused by events such as sensor movement or external conditions. Such events may include, for example, changes in how the wristband is worn on the wrist when the wristband is used as a connecting device, changes in the wearer, possible attacks on the wearer, falls, or loss of consciousness.

[0138] Therefore, the authentication method includes a step of evaluating signal quality; this step, combined with the step of processing the received physiological signals, leads to the rejection of unusable signals rather than the rejection of an individual's identity before generating an identification result. Furthermore, monitoring signal quality over time ensures that the detectable sensor is correctly worn over a long period (specifically, at least about ten minutes). Perfect signal quality over a long period can, for example, indicate a spoofing attack. To determine the proportion of poor-quality signals, Fourier transforms are used, for example. To evaluate signal quality, authentication system 2 includes a module for evaluating signal quality (over a given time period), which includes at least one algorithm for evaluating signal quality (noise), such as, for example, KNN (K-Nearest Neighbors) or SVM type. Thus, over a period of time, such as one hour, the signal quality evaluation module identifies portions of the signal that are too noisy to be usable and provides a probability about whether the user is actually wearing the connected device 1. This evaluation can also be performed at repeated time intervals, specifically from minutes to hours as needed, which increases authentication security. Additionally, a module dedicated to analyzing the user's movement (via accelerometers and gyroscopes) may be added to improve predictions and correlate noise with the user's typical movements. This makes it potentially easier to detect events such as device removal or the injection of spoofed signals.

[0139] Additionally, the authentication system 2 may include modules for improving signal quality using appropriate filters, such as Butterworth filters, FIR filters, automatic encoders, etc.

[0140] Advantageously, the authentication method of the present invention includes methods for detecting forged signals and for accelerating the recording of user data (thereby accelerating the prior training of the identification module 20). For this purpose, the authentication system 2 includes a module 23 for detecting forgery, which has one or more associated algorithms linked to the identification module 20. An example of methods for detecting forged signals and for accelerating data recording is the artificial generation of a signal similar to that of the currently recorded user (the generation of artificial signals, such as ECG signals, is known in itself). By artificially modifying the data, it is possible to detect potential attacks that aim only to reproduce artificial data rather than a combination of artificial and real data from an individual.

[0141] Advantageously, the authentication method of the present invention makes it possible to detect replay attacks (where a third party transmits data of an individual previously recorded without the individual's knowledge to sensor 3). Various methods for detecting replay attacks can be implemented. By way of example, one method involves using a hash function and storing a large number of hashes of the signal. Furthermore, by combining the method for detecting replay attacks with a "segmented hash" algorithm, authentication system 2 makes it possible to detect the reuse of pre-recorded signals.

[0142] Advantageously, as indicated above, the authentication method of the present invention is capable of identifying the best imposters (users attempting to impersonate genuine individuals). To this end, the step of identifying the best imposters includes calculating a success score for each user, and calculating a success score or imposter score for each imposter when matched with that user. Assuming that the distributions of user scores and imposter scores each follow a normal distribution, their intersection and the size of that intersection are considered to determine the probability that an imposter will be successfully authenticated by the authentication system during an attack. The smaller the intersection, the more reliably the authentication system 2 authenticates the user. Therefore, to further improve authentication and identification, the authentication method establishes a list of Y best imposters for each user, and each time a user is authenticated, attempts are made to authenticate Y best imposters (relative to other individuals known to the authentication system via various identification modules specific to each registered individual); this makes it possible to reduce the probability and score of each of these ultimately identified imposters relative to the user, in order to obtain verification of the user and rejection of each imposter. However, since identifying the best imposter is a resource-intensive task, this step is only advantageous when the identity of the person is suspected or authentication is required for highly sensitive actions.

[0143] The following describes examples of connection devices associated with exemplary embodiments of the authentication method of the present invention.

[0144] In the first example, a smartwatch is used as the connecting device 1. Most smartwatches are equipped with the following sensors: PPG (for heart rate, SpO2 for oxygen saturation), a 3D accelerometer (for detecting body activity), a gyroscope, and a bioimpedance sensor for detecting whether the object is being worn. Therefore, the authentication method uses the PPG sensor, the 3D accelerometer, and the gyroscope as sensors 3. Two identification modules are implemented: an identification module 20A for PPG and an identification module 20B for movement, which use signals from the accelerometer and gyroscope. Data from the bioimpedance sensor is used to determine whether the watch is being worn. The authentication method aims to determine whether the person wearing the watch is indeed the correct user, and if so, to issue an identity certificate to the user, allowing the user to use it to be authorized to access, for example, software or physical spaces. If the state of the bioimpedance sensor changes (the watch is no longer being worn), the watch's token and memory are reset.

[0145] In this watch example, an authentication frequency of, for example, once per minute is used, with a measurement time of 30 seconds. An authorized user wearing the watch can gain access to a secure room; in the case of a secure room, continuous authentication for a relatively long duration, such as verification within a ten-minute period, may be required. The watch continuously measures signals from all sensors and stores them in memory. Every minute, the watch sends the signal from the most recent minute to authentication system 2, which analyzes the last 30 seconds to verify the user's identity (authentication). The authentication analysis includes steps 1 through 6 of the authentication method described above. The watch has been worn for a certain period beforehand, allowing authentication system 2 to generate a unique and specific set of authentication models for each individual based on selection from multiple authentication test models implemented through machine learning using only the individual's personal data. Subsequently, upon authentication request, if identity has already been verified by authentication system 2, the authentication system generates an authentication certificate, which may contain the following information: date, verification or rejection of identity, confidence level, signal quality, emotional state, health status, SpO2 saturation, verification that the state of the bioimpedance sensor has not changed (the watch has not been removed), detected activity, and an identity token. In return, an identity token is sent to the user's watch 1, which the user can then use to authenticate with the software service or access their authorized physical space. When an authorized user wishes to access a secure room using a ten-minute continuous authentication process, authentication system 2 analyzes physiological signals recorded via the smartwatch over the previous ten minutes to verify the user's identity and the continuity of the authentication status during the ten minutes of wearing the watch. If successful, a new identity certificate and token are issued, granting the user access to the room. As long as the user continues to wear the watch while in the room, the continuous authentication method continues, again transparent to the user, allowing for continued learning of the user's data to update the authentication model for that user.

[0146] In another example, a user wearing a smartwatch implements the continuous authentication method of the present invention to protect their phone and watch, as well as their personal data. When the smartwatch is first worn, authentication system 2 (server-based) collects a set of PPG signals sent by the smartwatch to generate a user-specific authentication model through machine learning. Once this set of authentication models is generated, it is stored on the user's phone. Subsequently, whenever authentication of the user relative to the phone is required, the watch transmits PPG signals to the phone, which performs authentication analysis (via a dedicated software application) based on the (set of) dedicated authentication models stored on the phone to identify the user. At regular intervals, a portion of the day's authentication data is sent from the phone to server-based identification system 2 to continuously train and update the user-specific authentication model.

Claims

1. A method for continuously authenticating the identity of an individual, the method comprising a first step of authenticating by identifying at least one physiological signal of the individual transmitted by at least one sensor passively worn by the individual, and, after the first authentication step, iteratively authenticating the identity by identifying the at least one physiological signal of the individual, characterized in that... - Prior to the authentication and authentication iteration steps, the method includes, for each individual: i) a so-called prior training step, which includes processing data of at least one physiological signal specific to the individual through machine learning to generate multiple authentication test models; ii) using the data to test the multiple authentication test models as data for the individual continues to be acquired; iii) ranking the authentication test models according to the probability that the authentication test models match the data of the individual; and iv) retaining only a selected set of authentication test models ranked with the highest matching probability, the selected authentication test models being so-called authentication models reserved for the authentication and authentication iteration steps, each individual having its own set of authentication models; The authentication and authentication iteration steps include: comparing the acquired data of the individual with the authentication model specific to the individual, assigning a matching probability to the result of the comparison with each of the authentication models, and processing the matching probability to determine whether to accept or reject the authentication.

2. The method according to claim 1, characterized in that, In step i), which includes generating multiple authentication test models, data processing utilizes not only the data of the individual but also the data of at least one known imposter, wherein the amount of data used by all the known imposters considered during the generation of the multiple test models shall not exceed the ratio between the total amount of data of the individual and the number of known imposters.

3. The method according to claim 1 or 2, characterized in that, In step ii), at least one test model is tested using data from at least one other individual who is considered a known imposter; preferably, in step ii), a list of Y best imposters is established for each user, and each time the user is authenticated, the test model attempts to authenticate the Y best imposters relative to the other individuals known to the authentication system.

4. The method according to any one of the preceding claims, characterized in that, A decision tree is used to process the probabilities associated with the results of comparing the acquired data with the authentication model in order to determine whether to accept or reject the authentication.

5. The method according to any one of the preceding claims, characterized in that, The method includes the following steps: issuing an authentication certificate when identification is accepted, wherein the authentication iteration step is performed after the first authentication step and the issuance of the authentication certificate, or wherein the authentication iteration step is performed after the first authentication step and before the issuance of the authentication certificate.

6. The method according to any one of the preceding claims, characterized in that, During at least one authentication iteration step, the method implements machine learning to update one or more authentication models specific to the individual.

7. The method according to the preceding claim, characterized in that, During the iterative authentication steps for an individual, the method updates the authentication model using machine learning and stores the last N authentication models, which correspond to the most recently updated authentication model and authentication models from one or more previous iteration steps. During new and subsequent authentications, the authentication method is executed using the most recently stored authentication model or a combination of the N most recently stored authentication models.

8. The method according to any one of the preceding claims, characterized in that, When evaluating the authentication result of binary (true / false), the method takes into account signal quality, and preferably considers signal quality over a specific time period, specifically by processing signal noise to evaluate the signal quality.

9. The method according to any one of the preceding claims, characterized in that, The one or more physiological signals are selected from PPG and / or ECG signals and / or the individual's physical activity and / or the individual's bioimpedance. Specifically, the authentication method further includes a step of assessing the individual's health status based on the acquired physiological signals.

10. The method according to any one of the preceding claims, characterized in that, The method implements a set of authentication models for each type of physiological signal, and the method includes an algorithm for evaluating the authentication based on a combination of recognition results generated by each set of authentication models for each of the multiple types of physiological signals.

11. The method according to any one of the preceding claims, characterized in that, The method measures and evaluates data in addition to one or more physiological signals of the individual, wherein the other data is processed such that the results are combined with the results of the one or more physiological signals to establish the identification, wherein the other data is in particular biometric data used alone or in combination, such as fingerprints, facial features, irises, vein patterns, or passwords, wherein the passwords may be one-time passwords, smart cards, badges, certificates, or encryption keys on removable media such as USB flash drives, or verification actions on a second device such as a telephone.

12. The method according to any one of the preceding claims, characterized in that, The method includes steps for detecting replay attacks and / or detecting spoofed signals.

13. A system (2) for continuously authenticating the identity of an individual, the system comprising an electronic processing component and an algorithm for implementing the authentication method according to any one of the preceding claims, the processing component comprising at least one individual identification module (20), an identity and access management (IAM) module (21), and a data storage module (22), wherein the individual identification module generates an authentication model specific to the individual and associated with physiological signals through machine learning.

14. The authentication system according to the preceding claim, characterized in that, The processing unit includes a replay detection module, a spoofing signal detection module, and a module for evaluating additional data specific to the individual, including the individual's health status.

15. A computer program comprising code instructions that, when executed by a processor, perform the steps of the authentication method according to any one of claims 1 to 12.

Images