Method for continuously authenticating the identity of an individual
Patent Information
- Application Number
- AE202602717
- Authority / Receiving Office
- AE · AE
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2024-02-13
- Filing Date
- 2025-02-12
Smart Images
Figure ABST_ABST
Abstract
Description
Full specification Method for continuously authenticating the identity of an individual The invention relates to the field of continuously authenticating the identity of an individual, in particular for authorizing access. Authentication consists in proving the identity of a user so that access authorization can be granted (access to a location, a physical object, software, etc.). To authenticate, the user must transmit, to an authentication system, their identity together with one or more authentication factors, and the system verifies the authentication factor. The most common form of authentication relies on a password or an access badge. However, this type of authentication is susceptible to session hijacking. Moreover, once a user has been authenticated, the authentication system has no means of verifying that the user who continues to use the session after gaining access to it is still legitimate. In addition, biometric authentication, for example fingerprint recognition, has become more widespread in recent years. However, fingerprint sensors can be deceived by a latex forgery. Additionally, this authentication method is a one-time process; it authenticates the user once to log in without requiring repeated authentication. Thus, in recent years, the need to develop so-called continuous authentication has become apparent. Continuous authentication aims to re-authenticate the user several times during a session. For example, U.S. patent application US2022012317 discloses a method and a device for implementing same that consists in reiterating biometric recognition several times. The biometric recognition described in this document relates to voice recognition, fingerprint recognition, palmprint recognition, or recognition of facial features, hand shape, or vein patterns. Nevertheless, this type of recognition requires the user to actively interact with the biometric sensors for each recognition request, which is undoubtedly an inconvenience. Indeed, the user is already inconvenienced because the system informs them that they must be recognized again, and then they must take the time to perform an action, such as interacting with a sensor, by speaking or once again positioning their finger, hand, or face in an appropriate manner. Such a continuous authentication method is therefore too burdensome for the user. Patent application WO2012151680 also discloses a method for authenticating an individual that may be continuous by reiterating the authentication over time. This document advantageously discloses the use of physiological signals for authentication, such as ECG (electrocardiogram) and PPG (photoplethysmography), which are signals that are more difficult to forge. However, regardless of the number of individuals in the population, the method described in this document requires that a pre-recorded signal specific to each individual listed in the database be entered into the database beforehand. Authentication is performed by matching a measured physiological signal from an individual with the pre-recorded signal for that individual in the database. The pre-recorded signal is generated by a transformation rule resulting from machine learning based on signals measured from a group of several individuals. However, such an authentication method, which uses the template matching technique, has the shortcoming of requiring all individuals, along with their pre-recorded authentication signals derived from the transformation rule, to be stored in advance in a database so as to compare the measured signal with the pre-recorded signal. Another shortcoming is that, although the pre-recorded signal is unique to each individual, it is generated using the same transformation rule for all individuals listed in the database. Ultimately, this method is not reliable enough. Furthermore, the scientific publication by ALEXA MURATYAN ET AL: “Opportunistic Multi-Modal User Authentication for Health-Tracking IoT Wearables”, dated September 28, 2021, describes research directed to optimizing a method for authenticating an individual by training on the basis of physiological signals. This publication does not disclose a continuous authentication method. This publication focuses solely on identifying the best classification-based training model for authenticating an individual. Several different training models using various classification methods (RF; KNN; NB; SVM-RBF; SVM-Poly) were tested on a group of individuals, and the results show that the RF classifier model performs best. However, as explained previously, the same authentication model (RF model) is used again for each individual, which ultimately lacks reliability. This method of the prior art has numerous biases, especially demographic bias. In fact, it has been shown that two individuals with very similar physiological signals are indistinguishable. The purpose of the invention is therefore to provide a continuous authentication method that does not have the aforementioned shortcomings, in particular one that provides continuous authentication in a manner that is transparent to the user and, above all, remains reliable and secure over time. According to the invention, the method for continuously authenticating the identity of an individual is as set forth in claim 1. Consequently, for each individual, each of the steps of claim 1 is performed, and the final step, which involves selecting the authentication models for each individual, means that each individual therefore has their own group of authentication models, that is, each individual has a group of authentication models that is different from the group of authentication models of any other individual. In particular, according to the invention, the method for continuously authenticating the identity of an individual (hereinafter sometimes referred to as the legitimate individual) includes a first authentication step (referred to as initial identity authentication) of recognizing 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 the authentication of the identity) by recognizing said at least one physiological signal of the individual, the authentication method being characterized in that:- prior to the authentication and authentication iteration steps, the method includes: i) a so-called prior training step (with respect to said individual), which consists in processing, by machine learning, data of at least one physiological signal specific to the individual (the data being based on discriminating characteristics of said at least one physiological signal of the individual) in order to generate a plurality of authentication test models; ii) testing this plurality of authentication test models using data of the individual as such data continues to be acquired; iii) ranking the authentication test models according to their probability of matching with the data of the individual; and iv) retaining only a selected group of (several) authentication test models ranked with the highest probability of matching (matching with the legitimate individual), the selected authentication test models constituting the so-called authentication models that are retained for the authentication and authentication iteration steps (these authentication models are specific to each individual and therefore differ from those of any other individual to be authenticated; for each individual, a preliminary step of generating several authentication models is performed);- the authentication and authentication iteration steps consist in comparing the acquired data of the individual with the authentication models specific to the individual, assigning a probability of matching to the result of the comparison with each of the authentication models, and processing the matching probabilities in order to determine whether to accept or reject the authentication. An accepted result indicates that the individual is indeed legitimate. According to a preferred feature, in step i), which consists in generating a plurality of authentication test models, the data processing utilizes not only the data of the individual but also the data of at least one known impostor (training is performed using the data of the individual and the data of one or more known impostors), wherein the amount of data used of all the known impostors considered when generating the plurality of test models must not exceed the ratio between the total amount of data of the (legitimate) individual and the number of known impostors. The use of data of known impostors makes it possible, for example, to reject test models that might achieve a high probability of matching even though, in reality, the probability should be low, and also makes it possible to validate test models that have high probability because they were able to recognize the impostor data. This increases the reliability of the models retained for the authentication and authentication iteration steps. Additionally, considering a limited amount of data of known impostors with respect to the amount of data of the individual avoids the risk of generating test models that take greater account of the data of known impostors than of the data of the individual. Thus, several authentication models are used, and the comparison results produced by these models are analyzed to identify an individual; additionally, these authentication models are unique to a single individual because they are derived from models generated using the training data of that individual alone and, optionally, data of known impostors, rather than from a transformation rule derived from data of a group of individuals, all of whom had to be listed by storing their respective data. Indeed, in the prior art, schematically illustrated in [Fig. 4], a single model (as in the aforementioned scientific publication) is created based on a single transformation rule, this transformation rule having been developed from the data of a given group of individuals (by combining, through machine learning, all the data of all the individuals in the group). Today, the entire scientific community still uses only a single model (a single architecture) for all users. Although, in the prior art, several instances of the model may coexist (for example, in the context of bagging), it has not been proposed to implement several different models (different architectures) selected for each individual. The method of the invention personalizes the authentication models for each user. Consequently, the method of the invention does not generate only a single model (which is more reliable), nor does it establish only a single transformation rule. Thus, the plurality of models generated using the data of an individual, and the fact that they have all been tested during prior training to retain only a small group of the best models, and which differ from one individual to another, provide authentication models that are far more reliable for authenticating individuals and help reduce the demographic biases associated with each biometric system (a problem that is still widespread in biometric systems). Moreover, the method of the invention does not require data of all the individuals in a group. The method of the invention therefore uses the data of a single individual and may additionally use the data of at least one known impostor to generate several authentication test models, but does not require data of all the individuals in a group in order to generate an authentication test model. The method of the invention therefore does not need to process data of all the individuals in a group in order to authenticate a single individual. Nor is it necessary to consider a new group of individuals, together with all their data every time a new individual who was not initially listed needs to be added to a database for authentication (since the models retained for authenticating that individual are independent of the authentication models retained for other individuals, as there is no single transformation rule). According to one feature, in step ii), at least one test model is tested using data of at least one other individual considered to be a known impostor, preferably using data of several known impostors, which aids in the selection of said at least one authentication test model when the probability associated with that test model, as tested with the data of the known impostor, is appropriately low because the impostor has been correctly detected. Preferably, all the retained authentication models are tested using the data of the known impostor (to confirm the validity of the model when the associated probability is indeed low). According to one feature, the data of a known impostor may belong to a real individual or may have been generated digitally. Digital generation may be useful for simulating one or more impostors when only the data of the legitimate individual are available. In the prior art, a given group of authorized individuals is considered in order to create (through prior training) a transformation rule. When a new individual needs to be added to the database of authorized individuals, because the group has changed, it is necessary to repeat the prior training process for the entire new group in order to create a new, more appropriate authentication rule. In contrast, the method of the invention does not require the training process to be repeated for all the individuals each time a new individual needs to be added to the database of authorized persons. Indeed, the method of the invention generates test models based on the data of the individual to be authenticated (legitimate user), the data of another individual (known impostor) being used only if necessary (although preferentially used) solely to generate and test the test models and to establish that, if the test model tested with the data of the known impostor yields a result that is too similar to the result obtained with the data of the individual to be authenticated, that test model is assigned a very low, or even zero, probability value. This contributes to generating, based on the various test models, a group of authentication models that are even more reliable; however, under no circumstances is it necessary to use the data of an entire exhaustive group of individuals corresponding to a group in which each individual is expected to obtain authorization after authentication. The method of the invention is implemented independently of the number of individuals to be authenticated. The expression “passively worn”, used in relation to the sensor, means that the individual does not need to perform any action on the sensor or even pay attention to it, but merely to put it on and wear it (the arrangement depends on the type of physiological signal being measured). A physiological signal is understood to mean any analog signal generated by the human body that can be measured and digitized. Preferably, the processing of the probabilities associated with the results of comparing the acquired data with the authentication models, in order to determine whether to accept or reject the authentication, is implemented by means of a decision tree. According to one feature, the generation of authentication models and the authentication using these authentication models do not require any active action to be carried out by the individual. Consequently, the authentication method of the invention provides several advantages:- the detection of signals that are necessarily physiological and the continuous repetition of the measurement of these physiological signals make the authentication request and the verification thereof transparent to the individual, who does not need to perform any action;- the detection and processing of physiological signals reduce the risk of counterfeiting because physiological signals have the advantage of being different for every individual;- the additional steps for verifying the authentication of the identity over time eliminate the risk of session hijacking. Indeed, during a session, the fact that, after the initial authentication, the authentication is continuously verified by repeated measurement of the signal and systematic analysis thereof, makes it possible to continuously verify that the user is still the same person; if an authentication certificate has been issued, the session or access may remain open for as long as authentication verification continues;- in another intended use, the additional steps for verifying the authentication of the identity over time enable detailed analysis of a signal continuously over a relatively long period of time before issuing an authentication certificate;- generating several authentication models that are unique to each individual from artificial intelligence-based learning of the data of the individual to be authenticated (which is more reliable than using combined data of a required group of individuals), and implementing these models during authentication, by assigning each of them a probability of matching reinforces the reliability of the authentication process. Indeed, instead of providing an identical data transformation rule based on the data of a group of individuals and comparing the data during authentication using only this single rule, the method of the invention, by testing several authentication models derived from the data of the legitimate user (and optionally from additional data of one or more known impostors) and by evaluating their matching relevance for said user, this reinforces the reliability of the authentication evaluation. Moreover, testing the authentication models using data of a known impostor increases the reliability of the ranking of the best models to be retained for the individual to be authenticated;- there is no need to store the data of a multitude of individuals in order to establish a transformation rule from all these data, nor is there a need to carry out a new training process in order to establish a new rule whenever one or more individuals are added to the list of individuals to be authenticated. According to one feature, the authentication method establishes a list of the Y best known impostors for each legitimate user and, every time the user authenticates, attempts to authenticate these Y best impostors (with respect to the other individuals known in the authentication system, using the various recognition modules dedicated to each registered individual). In particular, this makes it possible to remove known individuals from the list of impostors. According to one feature, during the prior training step, the individual who is to be subsequently authenticated must wear, for a certain period of time, a sensor that detects at least one type of physiological signal, the individual passing through phases of rest, exercise, and emotions (for example, by listening to audio recordings and / or watching videos), in order to record the associated signals, and the machine learning results in authentication models dedicated to that individual that are relatively reliable. According to another feature, the authentication method includes a step of issuing an authentication certificate when the authentication is accepted (TRUE), with the authentication iteration steps being performed after the first authentication step and the issuing of the authentication certificate, or with the authentication iteration steps being performed after the first authentication step and before the issuing of the authentication certificate. The execution of the authentication iteration steps depends on the degree of security associated with the authentication. Thus, a first embodiment corresponds to a high-frequency authentication, on the order of one second or a few seconds, for issuing an authentication certificate if the authentication is deemed TRUE, and the authentication iteration steps continue after the authentication certificate has been issued. A second embodiment corresponds to an authentication having what will be referred to as a long recognition duration, in particular a duration of at least several minutes, or even several tens of minutes, the authentication certificate being issued only if the authentication is considered to be TRUE, after the first authentication step and several authentication iterations over the selected recognition duration; this analysis over a long measurement duration increases the legitimacy of the authentication result. This authentication embodiment, which combines continuous authentication with the analysis of physiological signals of an individual, provides highly reliable results; since the forgery of physiological signals over a long period of time is relatively unlikely. This embodiment may be very useful in high-security applications. Advantageously, the authentication method implements machine learning during at least one identity authentication iteration step in order to renew one or more authentication models dedicated to that individual. According to one feature, the dedicated authentication models that were generated and retained during the prior training step constitute first authentication models that are updated via machine learning during at least one iteration step of the continuous authentication in order to constitute new dedicated authentication models, that are preferably to be taken into account for the next authentication. Thus, the authentication method continues to learn to recognize the physiological signals of the individual for as long as the authenticated individual continues to wear the connected device. The learning module therefore benefits from more data, which increases the reliability of the dedicated authentication models and, consequently, the reliability of the authentication method. Preferably, the authentication method uses machine learning to renew, during the steps of iterating the authentication of an individual, the authentication models and stores the last N authentication models, which correspond to those that have been most recently renewed and to those from one or more previous iteration steps; and the execution of the authentication method during a new, subsequent authentication (when the user has not been wearing the sensor for a certain period of time) is carried out using the most recently stored authentication models or a combination of the N most recently stored authentication models. These phases of storing and retrieving the last authentication models or a combination of the last N models, further increase the reliability of the authentication process and reduce the risk of rejecting an authentication for which the result should have been TRUE. Indeed, the mood of an individual may change over a time period, their health may change, and the resulting physiological signals change accordingly; the method of the invention therefore makes it possible to generate authentication models that correspond as closely as possible to the current physiological state of the individual. According to another feature, the method takes into account, when evaluating the authentication result, which is binary (TRUE / FALSE), the signal quality and, preferably, the signal quality over a certain period of time; the signal quality being evaluated in particular by processing the signal noise. 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 group of authentication models for each type of physiological signal, and the method includes an algorithm for evaluating the authentication based on the combination of the recognition results produced by each group of authentication models for each of the types of physiological signals. According to another feature, the physiological signal or signals are selected from PPG signals and / or ECG signals and / or the physical activity of the individual and / or the bio-impedance of the individual. In particular, the authentication method further includes a step of evaluating the health status of the individual on the basis of the acquired physiological signals. PPG signals use plethysmography sensors, also referred to as pulse oximetry sensors, and relate to the measurement of changes in blood volume by measuring the amount of light absorbed and reflected by the blood vessels; the PPG signal is associated with a cardiac signal because the 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 an individual. Additionally, PPG has the advantage of being a non-invasive technique. According to another feature (in one embodiment), the method measures and evaluates data other than one or more physiological signals of the individual, said other data being processed such that the results are combined with the results of the physiological signal or signals in order to establish the recognition, said other data being, especially, taken alone or in combination, biometric data such as a fingerprint, face, iris, vein pattern, or a password that may be a one-time password, a smart card, a badge, a certificate, or an encryption key on a removable medium such as a USB flash drive, or a validation action on a second device such as a phone. According to another feature, the method includes a step of detecting replay attacks and / or a step of detecting forged signals. A replay attack occurs when a third party manages to capture the physiological signal of an individual, record it, and then use it to impersonate that individual, by sending it to the sensor, which then supplies, to the device implementing the method of the invention, data that should normally be the data of the individual. According to another feature, the method includes a step of detecting forged signals (signals that are generated artificially and therefore do not originate from the data of the individual wearing the sensor or sensors). According to one feature, the method uses a secure communications protocol such as HTTPS or a protocol selected from other cryptographic protocols, in particular between the sensor or sensors worn by the individual and a server to which the device implementing the authentication method is connected. The invention also relates to a system for continuously authenticating the identity of an individual, including electronic processing means and algorithms for implementing the aforementioned authentication method, the processing means comprising at least one recognition module for each individual (and for only one type of physiological signal), an identity and access management (IAM) module, and a data storage module, wherein the individual recognition module generates, by machine learning, the authentication models dedicated to the individual and associated with a physiological signal. The algorithms may be diverse, including, for example, support vector machines (also referred to as SVMs), neural networks, decision trees, principal component analysis, genetic algorithms, etc. According to one feature, the processing means of the aforementioned authentication system comprise a replay detection module, a forged signal detection module, and a module for evaluating additional data specific to the individual, including the health status of the individual. Finally, the invention relates to a computer program including code instructions for executing the steps of the aforementioned authentication method, when said program is executed by a processor. In particular, the program may be loaded over a network such as the Internet. The present invention will now be described by way of illustrative examples, which are in no way limiting of the scope of the invention, and with reference to the accompanying drawings, in which:- [Fig. 1] depicts a flowchart of the authentication system capable of implementing the authentication method according to the invention for authorizing a user to access an object X via a connected device worn by the user.- [Fig. 2] illustrates a flowchart of the authentication system used to authenticate several individuals.- [Fig. 3] schematically shows the steps of the prior training process leading to the selection of a group of authentication models specific to each individual.- [Fig. 4] schematically shows the prior art relating to authentication by selecting a single model for all individuals. The method for continuously authenticating the identity of an individual according to the invention is intended to verify, a first time, the identity of the individual (initial authentication), passively and without intervention by the individual, and then to continue verifying over time that it is still the same individual (continuous authentication verification), again passively and without intervention by the individual, the authentication being carried out based on the physiological signals of the individual. The continuous authentication method may be applied to various uses, such as access to a physical object X, access to software, access to a location (via access to an object such as a door), etc. As schematically illustrated in [Fig. 1], the continuous authentication method is implemented by at least one connected device 1 worn by the individual and by an authentication system 2 arranged remotely from the connected device 1 and receiving all the data of said connected device 1. The authentication system 2 is capable of communicating with the object X. The authentication system 2 is capable of receiving information from the connected device 1 and of processing it to authenticate the individual wearing the connected device 1 in order to authorize them to access the object X and to continue verifying the authentication for as long as the connected device 1 is being worn by the individual. The authentication system 2 is capable of receiving information from several connected devices 1-1, 1-2, 1-3, etc. ([Fig. 2]) in order to authenticate the individuals wearing said connected devices so as to authorize them to access an object (either the same object or a different object for each individual) if they have been successfully authenticated. The authentication system 2 (via a so-called IAM module) is capable of issuing, after authentication, an access authorization certificate that is valid for a predetermined period of time or that must be renewed upon each request for authentication reiteration in order to once again validate the correct identity and the validity of the access rights to be granted. Each certificate is signed by means of an asymmetric encryption system so that all parties can ensure the authenticity of each document and prevent document forgery. The connected device 1 is, for example, in the form of a smartwatch, a connected garment worn directly against the skin, or a wristband connected to a phone (the phone being able, if necessary, to serve as an interface for controlling the authentication and transmitting the authorization certificate). The connected device 1 must be worn by the individual and is arranged on the individual appropriately, depending on the nature of the physiological signals to be acquired. The connected device 1 includes at least one sensor 3 that detects at least one physiological signal. The types of physiological signals detected include, for example, a PPG signal (via a photoplethysmography sensor), an ECG (via a sensor measuring cardiac activity), or a signal indicating the physical activity of the individual (via a three-dimensional accelerometer and a gyroscope serving as sensors) or a bio-impedance signal (via electrode-type sensors). The authentication system 2 includes at least one authentication module 20, also referred to as recognition module, an identity and access management module 21, also referred to as IAM module, and a data storage module 22. A recognition module 20 is dedicated to a single individual. The authentication system 2 includes one recognition module for each individual, or optionally several recognition modules for each individual, each relating to the measurement of one type of physiological signal. The recognition module 20 features algorithms that implement AI. The storage module 22 records a wide range of data, including all the access requests. The access request is made by the individual at the object X. The object X issues an identifier ID to the connected device 1 worn by the individual; this identifier ID is unique to the object X. From that moment on, the individual no longer needs to intervene and simply has to wait for the access authorization. The main steps associated with the authentication method for authorizing access by an individual wearing connected device 1 to the object X are as follows:- the connected device 1 automatically connects to the authentication system 2 and transmits to it the identifier ID of the object X and a physiological signal of the individual;- if the authentication system 2, in particular the recognition module 20, recognizes the user, it issues an authentication certificate, which is sent to the connected device 1 and stored in the storage module 22 of the authentication system 2;- the connected device 1, having received the authentication certificate, requests an access authorization certificate from the authentication system 2, in particular from the IAM module 21, by transmitting the identifier ID of the object and the associated authentication certificate. A preferably encrypted copy of the access authorization certificate is stored in the storage module 22;- the authentication system 2, in particular the IAM module 21, issues, firstly, to the object X a copy of an access token containing an identifier IDuser of the individual requesting access, and, secondly, to the connected device 1, the certificate authorizing access to the object X;- the connected device 1 then transmits the access token and the identifier IDuser of the individual to the object X, thereby unlocking the object X, allowing the individual to access it. The database of the authentication system 2 comprises a library that associates one or more user identifiers (IDuser) with the identifier ID of the object X. Once the individual has been granted access to the object X and for a given security period (for example, one hour), the user likewise does not need to interact with their connected device 1 as long as they are wearing it; the continuous authentication continues transparently for the individual, the connected device 1 continues to measure the physiological signals of the user and to transmit them to the authentication system 2 which, for as long as recognition is performed, issues authentication certificates validating that access is still authorized. Furthermore, it is possible to anticipate that, even after the continuous authentication period has ended, a subsequent authentication verification during the user session may be implemented by the authentication system 2. Moreover, as will be seen later, the authentication method of the invention advantageously continues to collect data of the individual during the session, even if the continuous authentication has stopped (for example, because the one-hour period has elapsed). The authentication system 2 is therefore capable of:- identifying the one or more connected devices 1;- processing the data received from the one or more connected devices 1,- issuing and transmitting the one or more authentication certificates in response to the access request, or rejecting the request due to failed authentication,- communicating with the object X to issue it with copies of access tokens,- issuing access authorization certificates,- continuously issuing and transmitting authentication certificates,- storing the data received and sent, in particular a copy of the authentication certificates and the authorization certificates. Moreover, in addition to the authentication and the issuing of access authorization, the authentication system 2 may detect, evaluate, and / or transmit other information, especially:- ensuring long-term stability of the recognition through continuous learning,- evaluating signal quality,- sending alerts,- detecting replay attacks,- detecting forged signal attacks,- detecting brute-force attacks,- improving the rejection performance with respect to (unknown) impostors by using data of known impostors, that is to say, individuals for whom it is known that the results of the authentication models have been tested using known individuals other than the individual to be authenticated,- providing multiple items of information relating to authentication / identification, such as for example• the health status of the individual wearing the connected device (sleep, stress, physical exertion leading to a heart condition, 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 phase or in a state of reduced consciousness;• the position or geolocation of the individual, or the detection of the movement of the individual within an area. Depending on the duration of the authentication and the number of iterations, the authentication method makes it possible to provide several levels of security. Moreover, the authentication system 2 has the advantage of being adaptable at any time by adding multiple recognition modules (each time a new user is to be added) that are specific to each individual. In particular ([Fig. 2]), the authentication system 2 includes a plurality of recognition modules 20-1, 20-2, 20-3, etc., each of which is dedicated to an individual wearing a respective connected device 1-1, 1-2, 1-3, etc. This feature will be described in greater detail hereinafter. Thus, authentication models that are unique to each individual (one group of authentication models for each type of physiological signal) are generated by a recognition module dedicated to one individual, entirely independently of the population of individuals. Therefore, unlike in the prior art, there is no need to know and record data of a plurality of individuals belonging to a targeted group in order to establish a data transformation rule that is dependent on all those individuals, nor is it necessary to perform a new training process based on the data of the previously registered individuals (which, incidentally, may have changed over time) and the data of one or more individuals to be added in order to establish a new transformation rule that takes into account both the previously registered individuals and the added individuals. More particularly, the authentication method includes the following steps:- Step 1: measuring and recording the signals output by the measurement device, such as the sensor 3, over an acquisition period and preferably at an acquisition frequency;- Step 2: transmitting, to the authentication system 2, the physiological signal data which relate to a first item of identity information, and, optionally, simultaneously transmitting a second item of identity information in the form of a unique key (encrypted or unencrypted) associated with the sensor 3. This unique key may be stored in a dedicated cryptographic processor, if the measurement device includes one;- Step 3: selecting, by the authentication system 2, the recognition module 20 dedicated to the user of the sensor 3; this is a personal recognition module 20 for each individual to be authenticated, which stores in memory a group of authentication models dedicated to the user. The process of generating the group of authentication models is described hereinafter. Authentication analysis is obtained by comparing the data acquired from the individual during the recognition step with the authentication models specific to the individual, which are stored in the recognition module 20, then associating a probability of matching with the comparison result for 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 recognition module 20 returns a binary result — TRUE or FALSE — which is associated with the various probability values or confidence scores resulting 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. Among the algorithms implemented by the recognition module 20, one algorithm provides initial authentication and several authentication iterations; these iterations are performed before and / or after the TRUE / FALSE result is issued, depending on the desired security level.- Step 4: the recognition module 20 (20-1 for a given individual) sends the TRUE / FALSE value to the IAM module 21, associated with the IDuser and the probability values or confidence scores of the results of comparison with the authentication models, and preferably a list of the Y best identified impostors (which are known impostors because they are considered to be other users and have their own recognition module 20-2, 20-3, etc.);- Step 5: this step is optional but preferred; the IAM module 21 requests recognition of the Y known impostors by the recognition modules of the other individuals registered in the authentication system 2. The recognition modules of the other individuals independently return a TRUE / FALSE value to the IAM module 21 for each of the Y known impostors, together with the associated probability or confidence score;- Step 6: the IAM module 21 issues a unique identity certificate for each authentication, mentioning several items of information that are listed hereinafter by way of example. In Step 1, the acquisition duration for measuring physiological signals for the purpose of the initial authentication is tailored, in particular, to the nature of the physiological signal and to the degree of security required for the intended application. The acquisition for the initial authentication is preferably periodic, for example, at a frequency of one second. Then, during the phase of iterating the continuous verification of the authentication, the measurement is periodic and / or random. With regard to steps 2 and 3, the authentication system 2 advantageously includes several recognition modules 20 (20-1, 20-2, 20-3, etc.), each dedicated to one user. A recognition module 20 is a personal module because it is dedicated to a single individual, on the basis of a group of authentication models that are unique and depend solely on the data of the individual and are independent of the data of other individuals. Moreover, the recognition module 20 is dedicated to one type of signal, for example PPG, ECG, physical activity, or bio-impedance. The authentication system 2 may therefore comprise several recognition modules 20A, 20B, 20C, etc. for each individual, each dedicated to measurement and authentication with respect to one type of physiological signal (20A for PPG, 20B for ECG, etc.). Associated with each type of physiological signal is a group of authentication models, said authentication models of a group having been generated beforehand, as will be described later, with respect to the type of physiological signal being measured. More particularly, the recognition module 20, which is personal to each individual (and independent of the other individuals), is capable of:- prior to an authentication request (that is, prior to step 1), generating a group of authentication models that are unique to the individual by processing at least one type of physiological signal over a period of time referred to as the prior training period, this step being referred to as the prior training step;- authenticating the individual when an authentication request is received (steps 1 to 6 mentioned hereinbefore), by comparing the one or more measured physiological signals with respect to the one or more groups of unique authentication models generated at the end of the prior training step for each type of physiological signal;- continuing the learning during authentication requests, using the signals measured during each request, in order to generate a group of authentication models that remain unique with respect to the individual and are updated (to take into account any changes in the physiological signals of the individual, which may vary over time, especially depending on the health status of the individual). It should be noted that the prior art does not permit this step, which is referred to hereinafter as the step of updating authentication models with respect to the individual (since the prior art has a single transformation rule that remains unchanged over time). For the purpose of the initial authentication (step 1), prior training of the user’s data is therefore carried out over a given period of time according to a registration protocol. The prior training is carried out, for example, over the course of one day, during which the individual wears the connected device 1 and passes through various periods of rest and activity, and even emotional states, at various times, in order to obtain a wide range of heart rate ranges for the individual. Several variants of the registration protocol may be implemented. For example, the registration periods are different for a sedentary individual, in particular they are shorter. During the prior training step, which corresponds to a first session, it is the operator of the authentication system 2 who manually associates the connected device 1 with a user. The connected device 1 contains a first token usable by the user for as long as the user wears the connected device (such as, for example, a wrist-worn connected wristband). In parallel, the authentication system 2 therefore continuously records the data from the one or more sensors of the connected device 1 in order to perform the prior training, as indicated for example over the course of one day. As the day progresses, the authentication system 2 becomes increasingly reliable. Once the prior training phase has been completed, the first token is replaced with an automated authentication token issued by the authentication system 2, and the continuous authentication method can begin as soon as needed. During the prior training, the recognition module 20 implements one or more machine learning algorithms, according to several steps (schematically shown in [Fig. 3]) in order to:i) generate, from the extraction of features relating to one type of measured physiological signal of the individual, a plurality of authentication test models (for example, about a hundred; the figure only illustrates a few models that are schematically shown by different geometric symbols), thenii) test said authentication test models using the signals of the individual considered to be authentic,iii) rank the best test models (those having the highest probability of recognizing the individual), andiv) retain only a group of the best (for example, about ten) tested models, which are referred to as authentication models (for simplicity, only a few models have been depicted in the figure, represented by different geometric symbols). Each individual has a personalized group of authentication models. Each individual has a distinct group of several authentication models (each individual does not have the same group of authentication models). The plurality of authentication test models is obtained by machine learning based on discriminating features of the acquired data. The plurality of models generated using the data of the individual, combined with the fact that they are all tested to retain only a small group thereof, namely the best models, provides authentication models that are as reliable as possible for authenticating the individual. The group of authentication models is associated with a unique key linked to the identity of the user (IDuser). With regard to the extraction of data relating to the physiological signals in order to carry out step i) of generating the plurality of test models, the recognition module 20 implements data processing steps or phases (via algorithms) that are known per se, namely preprocessing phases (in particular according to a given type of windowing), filtering (for example, by using a Fourier transform and / or digital filtering), which may be carried out before the preprocessing, extraction of (numerous) features, selection of those features (advantageously by machine learning, for example according to a PCA (Principal Component Analysis) function) and, finally, classification (for example, by SVM (Support Vector Machine) or by a neural network). Some neural networks are suitable for carrying out the extraction, selection, and classification steps. The algorithms used to carry out these various phases may also be genetic algorithms. The steps or phases, optionally of preprocessing, filtering, segmentation, normalization (optional), extraction, selection (which is a feature selection phase and may be optional), and classification, for each generation of a test model, may each employ different types of methods (as exemplified hereinbefore: different types of windowing, different filtering methods (Butterworth filter or others), different segmentation methods (for example, according to 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 operating on the raw signal, KNN db3, RF (Random Forest), DWT and ADABOOST, etc.). Preferably, according to the invention, the algorithms used to perform these phases and subsequently generate a multiplicity of test models implement, optionally in a random manner for certain test model results, various combinations of the different methods specific to each phase. Thus, the generated model examples are highly diversified, especially according to different combinations of extraction and classification algorithms; for example, for user No. 1, this is at least one model implementing an extraction step by FFT (Fast Fourier Transform) followed by an RF classifier; for user No. 2, this is at least one model implementing a DWT (discrete wavelet transform) followed by an SVM (support vector machine) as well as a CNN (convolutional neural network) operating on the raw signal; for user No. 3, this is at least one model implementing FFT with k-NN, FFT with RF, DWT and ADABOOST, etc. Ultimately, each individual has their own authentication models. Advantageously, in step i) of generating a plurality of authentication test models, the data processing utilizes not only the data of the individual but also the data of at least one known impostor, wherein the amount of data used from all the known impostors considered during the generation of the plurality of test models must not exceed the ratio between the total amount of data of the individual and the number of known impostors. Preferably, at least the test models retained as authentication models are also tested during step ii) using the data of at least one known impostor in order to verify that the confidence probability for recognition using the data of a known impostor is indeed extremely low, or even zero. This test, which uses data of an impostor, increases the reliability of the test models ultimately retained. With regard to authentication, if impostors are identified, the recognition module 20 contains the list of Y individuals having the highest impostor scores, and advantageously has compared this list with the other individuals registered and linked with their recognition module. Where necessary, this makes it possible to remove known individuals from the list of impostors. Additionally, and very advantageously, the recognition module 20 comprises a machine learning algorithm that enables continued learning of the recognition of the physiological signals of the individual for as long as the individual continues to wear the connected device 1 after having been authenticated a first time; the learning continues throughout the entire duration of the continuous verification of the authentication and even thereafter, for as long as the connected device is worn (according to a given number of times and / or given periods). This continuous learning contributes to increasing the legitimacy of the result of the continuous authentication and of a subsequent initial authentication (upon a new access request), thereby providing increased security. Indeed, it has been found that initial authentication performance decreases by 15% to 30% when a relatively long period of time, such as one week, has elapsed before the same individual makes a new access request. However, the inventors have demonstrated that continuous learning during the continuous verification of authentication increases authentication performance not only during the continuous verification process but also during a subsequent authentication. In particular, a portion of the signals used for the continuous authentication over one day are retained, used to create a new dataset, and to train new test models, thereby deriving a new updated group of authentication models that remain dedicated to the user. This new updated group of authentication models is stored in a chain, such as a blockchain, which makes it possible to trace all the training of the authentication models. Upon each new authentication, the most recently stored group of authentication models is used to determine the identity of the user, thereby contributing to greater reliability of the authentication result. Alternatively, the recognition module 20 may respond to an authentication request using the last N authentication models (corresponding to the most recently stored group of authentication models and one or more other previously stored groups), thereby further increasing security. Moreover, storing the successive versions of the authentication models and the associated authentication results in the form of chains facilitates investigations in the event of an attack on the authentication system 2. If an attacker is able to modify the training dataset in order to insert their own personal information in place of the data of the user, the system will be able to recognize the attacker as a non-legitimate user. With regard to the authentication iterations, these may be carried out after a result has been determined in order to confirm the authentication over time, or may be carried out over a so-called long period of time in order to perform several verifications before issuing the result. In step 6 of generating the identity certificate by the IAM module 21, the data included in this certificate are, for example:- the timestamp,- the validation (TRUE) or rejection (FALSE) of the identity,- the duration / expiry date of the access authorization,- the certified identity (certified IDuser),- the identity token (which is unique and random).Additional information may be incorporated, such as:- the list of signals used for the authentication,- the decision made by each recognition module and the associated confidence score,- the health status of the wearer,- the detected activity,- the number of signals and the signal quality,- the cryptographic signature of the certificate.This additional information is especially useful in the event of an investigation or when an intrusion is detected. This additional information may be stored in a particular processing module linked with the intrusion. In order to further improve the effectiveness of the authentication system 2 and contribute to its overall security (so as to deceive attackers), it is preferable to add various items of information related to authentication and identification, in addition to the data already listed hereinbefore, which are recorded in the record of the individual stored in the storage module 22. This additional information especially includes:- detected activity (walking, running, resting, working on a computer, etc.) via an accelerometer and a gyroscope,- emotional state (calm, stressed, excited, afraid, etc.),- overall health status (good, poor, fair),- accident detection (falls, arrhythmias, heart attacks, sudden changes in blood pressure, etc.),- internal geolocation (an area within a defined radius by means of proximity sensors, beacons, badge systems, etc.),- GPS geolocation,- environmental information such as light levels, ambient noise,- the body temperature of the individual (measured using an appropriate sensor),- blood oxygen saturation (measured using an appropriate sensor). Among the information listed hereinbefore, the health status of the individual after authentication may be relevant to security or safety, depending on the intended application. For example, a rapid deterioration in the health status of the authenticated individual may result in a sudden health problem or an attack against the individual, which should be detected in ultra-sensitive environments such as military or nuclear facilities. This status change is intended to be detected by the authentication system 2 in order to generate an alert and / or lead to automated decisions. The health status of an individual is verified, especially, from the heart rate, respiratory rate, sleep duration, body temperature, and oxygen saturation of the individual, these parameters being measured by means of the connected device 1, which comprises the appropriate sensors. Processing of these data yields a score relating to the health status; depending on the value of the score relative to a reference value, for example, below the reference value, the authentication system 2 sends an alert, temporarily blocks access, or requests hierarchical approval. In addition to the detection and processing of physiological signals, the authentication method may use other biometric parameters to supplement the authentication, such as, by way of non-limiting example: fingerprints, iris, face, hand geometry, hand vein patterns, and dedicated gestures. In addition to authentication by physiological signals, the authentication method may make use of authentication protocols that do not use biometric data and that rely on an interaction with the object to which the individual is requesting access, such as a password, a one-time password, a smart card, a badge, a certificate, or an encryption key on a removable medium (USB flash drive, hard drive), or an action and / or validation on a second device such as a phone. Advantageously, the authentication method of the invention takes into account, in evaluating the authentication result (TRUE / FALSE), the quality of the signal sent by the connected device 1 and received by the authentication system 2. Evaluating the signal quality corresponds to evaluating at least the signal noise. A poor-quality signal may result from events such as movements of the sensor or external conditions. Such events may include, for example, in the case of a wristband serving as the connected device, a change in how the wristband is worn on the wrist, a change of wearer, a possible attack on the wearer, a fall, or a loss of consciousness. Thus, the authentication method includes a step of evaluating the signal quality; this step, combined with the step of processing the received physiological signals, leads, prior to generating the recognition result, to a step of rejecting unusable signals rather than rejecting the identity of the individual. Moreover, monitoring the signal quality over time also makes it possible to detect that the sensor is being worn properly over a long period (in particular, for at least about ten minutes). A perfect signal quality over a long period of time may, for example, indicate a forging attack. In order to determine the proportion of poor-quality signals, a Fourier transform is, for example, used. In order to evaluate the signal quality, the authentication system 2 includes a module for evaluating signal quality (over a given period of time) comprising at least one algorithm for evaluating signal quality (noise), such as, for example, of the KNN (K-nearest neighbors) or SVM type. Thus, over a period of time, for example of one hour, the signal quality evaluation module determines the portions of the signal that are too noisy to be usable and provides probabilities regarding whether the user is actually wearing the connected device 1. This evaluation may also be carried out over repeated time intervals, in particular from several minutes to several hours, as needed, which adds to the authentication security. Moreover, it is possible to add modules dedicated to analyzing the movements of the user (via an accelerometer and a gyroscope) in order to improve the predictions and associate the noise with movements typical of the user. This makes it possible, for example, to more readily detect incidents such as removal of the device or injection of forged signals. Additionally, the authentication system 2 may include a module for improving signal quality using appropriate filters, such as Butterworth filters, FIR filters, autoencoders, etc. Advantageously, the authentication method of the invention includes a method for detecting forged signals and for speeding up the recording of the user’s data (thereby speeding up the prior training by the recognition module 20). For this purpose, the authentication system 2 includes a module 23 for detecting forging having one or more associated algorithms, which is linked to the recognition module 20. One example of a method for detecting forged signals and for speeding up the recording of data is to artificially generate signals that resemble those of the users currently being recorded (the generation of artificial signals, such as ECG signals, is known per se). By artificially modifying the data, it becomes possible to detect a potential attack that aims solely to reproduce artificial data, rather than a combination of artificial and authentic data of an individual. Advantageously, the authentication method of the invention makes it possible to detect replay attacks (where a third party sends, to the sensor 3, data of the individual that were previously recorded without the individual’s knowledge). Various methods for detecting replay attacks may be implemented. By way of example, one method consists in using a hash function and storing a large number of hashes of the signals. Moreover, by combining the method for detecting replay attacks with a “piecewise hashing” algorithm, the authentication system 2 makes it possible to detect the reuse of pre-recorded signals. Advantageously, as already indicated hereinbefore, the authentication method of the invention is capable of identifying the best impostors (users who attempt to impersonate the authentic individual). For this purpose, the step of identifying the best impostors includes calculating a success score for each user and calculating a success score or an impostor score for each impostor when matched against that user. Assuming that the distributions of user scores and impostor scores each follow normal distributions, their intersection, and the size of this intersection, are considered in order to determine the probability that an impostor will successfully be authenticated by the authentication system during an attack. The smaller the intersection, the more reliably the authentication system 2 authenticates the user. Thus, in order to further improve the authentication and identification, the authentication method establishes the list of the Y best impostors for each user and, upon each authentication of the user, attempts to authenticate the Y best impostors (with respect to the other individuals known to the authentication system via the various recognition modules dedicated to each registered individual); this makes it possible to reduce the probabilities and the scores of each of these ultimately identified impostors with respect to the user, so as to obtain validation for the user and rejection for each impostor. Nevertheless, since this step of identifying the best impostors is a resource-intensive task, it is advantageously used only when there is doubt as to the identity of the person or where authentication is needed for an ultra-sensitive action. An example of a connected device associated with an exemplary implementation of the authentication method of the invention is described next. In the first example, a smartwatch is used as the connected device 1. Most smartwatches are equipped with the following sensors: PPG (for heart rate and SpO2 for oxygen saturation), a three-dimensional accelerometer (for detecting physical activity), a gyroscope, and a bio-impedance sensor for detecting whether or not the object is being worn. The authentication method therefore uses, as sensors 3: the PPG sensors, the three-dimensional accelerometer, and the gyroscope. Two recognition modules are implemented: a recognition module 20A for PPG and a recognition module 20B for movement, which uses the signals from the accelerometer and the gyroscope. The data from the bio-impedance sensor are used to determine whether or not the watch is being worn. The authentication method is intended to determine whether the person wearing the watch is indeed the correct user and, if so, to issue the user with an identity certificate so that the user can use it to be authorized to access, for example, software or a physical space. If the state of the bio-impedance sensor changes (the watch is no longer being worn), the tokens and the memory of the watch are reset. In this watch example, an authentication frequency of, for example, once per minute is used, with a measurement time of 30 seconds. Wearing the watch by an authorized user may provide access to a secure room; in the case of a secure room, it may be required for the continuous authentication to be of relatively long duration, for example with verification over a ten-minute period. The watch continuously measures the signals from all of the sensors and stores them in memory. Every minute, the watch transmits the most recent minute of signal to the authentication system 2, which analyzes the last 30 seconds in order to validate the identity of the user (authentication). The authentication analysis consists in implementing the aforementioned steps 1 to 6 of the authentication method. Beforehand, the watch has been worn for a certain period of time so that the authentication system 2 generates a group of authentication models that are unique and specific to the individual, based on selection from the plurality of authentication test models that were implemented by machine learning using only the personal data of the individual. Subsequently, upon an authentication request, if the identity has been validated by the authentication system 2, the authentication system produces an authentication certificate, which may contain the following information: date, validation or rejection of the identity, confidence level, signal quality, emotional state, health status, SpO2 saturation, validation that there has been no change in the state of the bio-impedance sensor (the watch has not been removed), detected activity, and the identity token. In return, the identity token is transmitted to the user’s watch 1, which the user may then use to authenticate with software services or to access a physical space for which they are authorized. When the accredited user wishes to access a secure room using ten minutes of continuous authentication, the authentication system 2 analyzes the physiological signals recorded over the previous 10 minutes via the smartwatch in order to validate, a second time, the identity of the user and the continuity of the authentication states over the previous ten minutes during which the watch was worn. If successful, a new identity certificate and a token are issued so that the user can access the room. For as long as the watch continues to be worn by the user while the user is in the room, the continuous authentication method continues, again transparently to the user, thereby enabling continued learning of the user’s data in order to update the group of authentication models for that user. In another example, the user wearing the smartwatch implements the continuous authentication method of the invention in order to protect their phone and watch, as well as their personal data. When the smartwatch is worn for the first time, the authentication system 2 (server-based) collects a certain number of PPG signals transmitted by the smartwatch in order to generate, by machine learning, the authentication models dedicated to the user. Once the group of authentication models has been generated, it is stored on the user's phone. Subsequently, whenever authentication of the user with respect to the phone is required, the watch sends PPG signals to the phone, which implements (via a dedicated software application) the authentication analysis based on the dedicated authentication models (of the group) stored on the phone in order to recognize the user. At regular intervals, a portion of the authentication data for the day is transmitted from the phone to the server-based recognition system 2 in order to continue training and updating the authentication models dedicated to the user.
Claims
1. A method for continuously authenticating the identity of an individual, including 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, after the first authentication step, at least one step of iterating the authentication of the identity by recognizing said 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 consists in processing, by machine learning, data of at least one physiological signal specific to the individual in order to generate a plurality of authentication test models; ii) testing this plurality of authentication test models using data of the individual as such data continues to be acquired; iii) ranking the authentication test models according to their probability of matching with the data of the individual; and iv) retaining only a selected group of authentication test models ranked with the highest probability of matching, the selected authentication test models being the so-called authentication models that are retained for the authentication and authentication iteration steps, each individual having their own group of authentication models;- the authentication and authentication iteration steps consist in comparing the acquired data of the individual with the authentication models specific to the individual, assigning a probability of matching to the result of the comparison with each of the authentication models, and processing the matching probabilities in order to determine whether to accept or reject the authentication. 2. The method according to claim 1, characterized in that, in step i) which consists in generating a plurality of authentication test models, the data processing utilizes not only the data of the individual but also the data of at least one known impostor, wherein the amount of data used from all the known impostors considered during the generation of the plurality of test models must not exceed the ratio between the total amount of data of the individual and the number of known impostors. 3. The method according to claim 1 or 2, characterized in that, in step ii), at least one test model is tested using the data of at least one other individual considered to be a known impostor; preferably, the list of the Y best impostors for each user is established in step ii), and, upon each authentication of the user, the test models attempt to authenticate the Y best impostors with respect to the other individuals known to the authentication system. 4. The method according to any one of the preceding claims, characterized in that the processing of the probabilities associated with the results of comparing the acquired data with the authentication models, in order to determine whether to accept or reject the authentication, is implemented by means of a decision tree. 5. The method according to any one of the preceding claims, characterized in that it includes a step of issuing an authentication certificate when the recognition is accepted, with the authentication iteration steps being performed after the first authentication step and the issuing of the authentication certificate, or with the authentication iteration steps being performed after the first authentication step and before the issuing 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, it implements machine learning to renew one or more authentication models dedicated to said individual. 7. The method according to the preceding claim, characterized in that, during the steps of iterating the authentication of an individual, it renews the authentication models by machine learning and stores the last N authentication models, which correspond to those that have been most recently updated and to those from one or more previous iteration steps, and in that the execution of the authentication method during a new and subsequent authentication is carried out using the most recently stored authentication models 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 it takes into account, when evaluating the authentication result, which is binary (TRUE / FALSE), the signal quality and, preferably, the signal quality over a certain period of time, in particular the signal quality being evaluated by processing the signal noise. 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 physical activity of the individual and / or the bio-impedance of the individual, in particular, the authentication method further includes a step of evaluating the health status of the individual on the basis of the acquired physiological signals. 10. The method according to any one of the preceding claims, characterized in that it implements a group of authentication models for each type of physiological signal, and the method includes an algorithm for evaluating the authentication based on the combination of the recognition results produced by each group of authentication models for each of the types of physiological signals. 11. The method according to any one of the preceding claims, characterized in that it measures and evaluates data other than one or more physiological signals of the individual, said other data being processed such that the results are combined with the results of the physiological signal or signals in order to establish the recognition, said other data being, especially, taken alone or in combination, biometric data such as a fingerprint, face, iris, vein pattern, or a password that may be a one-time password, a smart card, a badge, a certificate, or an encryption key on a removable medium such as a USB flash drive, or a validation action on a second device such as a phone. 12. The method according to any one of the preceding claims, characterized in that it includes a step of detecting replay attacks and / or a step of detecting forged signals. 13. A system (2) for continuously authenticating the identity of an individual, including electronic processing means and algorithms for implementing the authentication method according to any one of the preceding claims, the processing means comprising at least one individual recognition module (20), an identity and access management (IAM) module (21), and a data storage module (22), wherein the individual recognition module generates, by machine learning, the authentication models dedicated to the individual and associated with a physiological signal. 14. The authentication system according to the preceding claim, characterized in that the processing means comprise a replay detection module, a forged signal detection module, and a module for evaluating additional data specific to the individual, including the health status of the individual. 15. A computer program including code instructions for executing the steps of the authentication method according to any one of claims 1 to 12, when said program is executed by a processor.