System and method for analyzing human behavior

Abstract

The invention relates to a system (2) for analyzing human behavior, the system (2) comprising a plurality of sensors (4) for input data relating to human behavior, a plurality of monomodal behavioral analysis engines (6), and a cyber-cognitive analysis platform (8), each monomodal behavioral analysis engine (6) being capable of acquiring the input data relating to human behavior; the cyber-cognitive analysis platform (8) comprising a meta-engine (10) and at least two modules (12a-12f) configured to reproduce cognitive functions of a human brain, the meta-engine (10) being configured to aggregate the input data, to consolidate said input data, and to transmit said consolidated input data to said at least two modules (12a-12f);the meta-engine (10) being provided with a self-learning statistical analysis module (14) configured to analyze and correlate the input data from the plurality of single-modal behavioral analysis engines (6), and to provide an analysis of human behavior from said analyzed and correlated input data and data processed by said at least two modules (12a-12f). Abstract figure: Fig. 1;

Description

Title of the invention: System and method for analyzing human behavior

[0001] The invention relates to a system and method for analyzing human behavior. The human behavior analysis system is provided with a cyber-cognitive analysis platform. The cyber-cognitive analysis platform is intended to analyze and statistically represent the complexity of human behavior, the latter being modeled and captured by different monomodal behavioral analysis engines connected to sensors. State of the art

[0002] The use of behavioral analysis engines is known in the field of human behavioral interaction. However, such behavioral analysis engines only allow single-modal interactions, without correlating the results of several engines with each other. The different modes of behavioral interaction are: analysis of facial expression (including eye movements and facial micro-expressions), analysis of vocal intonation, semantic analysis of expressed language (vocal or gestural) and analysis of gestures (excluding sign language or sign language more generally).

[0003] From the point of view of cognitive science (and in particular in psychology, neuroscience and linguistics), the absence of correlation between different behavioral analysis engines prevents the scientific veracity of the analyses produced by these same monomodal engines. We can cite for example that the behavior of a subject rolling his eyes while blowing and saying "but yes of course" will be interpreted differently by engines taken individually. No engine can detect that this is a typical case of irony. The absence of multimodal correlation therefore degrades the performance of these behavioral analysis engines.

[0004] There is therefore a need to have a common language between such single-modal behavioral analysis engines, in order to enable efficient and valid multimodal communication, and thus improve the accuracy and reliability of human behavior analysis engines. Description of the invention

[0005] The present invention aims to resolve all or part of the drawbacks of the state of the art cited above.

[0006] Thus, according to a first aspect, the invention relates to a system for analyzing human behavior, the system comprising a plurality of input data sensors relating to human behavior, a plurality of analysis engines com single-modal behavioral analysis platforms connected to data sensors, and a cyber-cognitive analysis platform connected to single-modal behavioral analysis engines, each single-modal behavioral analysis engine being capable of acquiring input data relating to human behavior.

[0007] According to the invention, the cyber-cognitive analysis platform comprises a meta-engine and at least two modules configured to reproduce cognitive functions of a human brain, the meta-engine being connected to said at least two modules and being configured to aggregate the input data from the plurality of monomodal behavioral analysis engines, to consolidate said input data, and to transmit said consolidated input data to said at least two modules; the meta-engine being provided with a self-learning statistical analysis module, said self-learning statistical analysis module being configured to analyze and correlate the input data from the plurality of monomodal behavioral analysis engines, and to provide an analysis of human behavior from said analyzed and correlated input data and data processed by said at least two modules.

[0008] In the context of the present invention, "cyber-cognitive" is understood to mean that the analysis platform according to the invention deals with cybernetics by adding methods and principles from cognitive sciences, as opposed to artificial intelligence which uses algorithms and statistics, without employing any particular cognitive science. In the context of the present invention, "cybernetics" means the study of the information mechanisms of complex systems, explored with a view to being standardized.

[0009] Thanks to the cognitive analysis platform thus configured, the human behavior analysis system allows the meta-engine to analyze the data produced from different separate and external monomodal analysis engines, to correlate them and to produce a scientifically valid result from the point of view of cognitive science. This makes it possible to significantly improve the accuracy and reliability of the monomodal human behavior analysis engines used.

[0010] Typical use cases of the analysis system according to the invention concern, for example, the field of public health (cases of mental illness, cognitive deficits, depression), that of communication and media (for example, real-time feedback of the impact of a conference on its audience to the speaker, or even prediction of marketing or advertising campaigns), as well as that of industry (a priori effectiveness of corporate social responsibility policies, measurement of quantitative but above all qualitative risk in terms of cybersecurity, etc.). The analysis system according to the invention finds applications in particular in devices such as smart boxes embedded in vehicles (in particular in cars), “magic mirror” type devices, or even in digital desks intended for human assistance.

[0011] Preferably, the self-learning statistical analysis module is a multi-factorial and temporal statistical analysis module. This makes it possible to further improve the accuracy and reliability of the single-modal human behavior analysis engines used.

[0012] According to a particular technical characteristic of the invention, the cyber-cognitive analysis platform comprises six modules configured to reproduce cognitive functions of a human brain; a first module reproducing a perception function; a second module reproducing an attention function; a third module reproducing a memory function; a fourth module reproducing a motor function; a fifth module reproducing a language function; a sixth module reproducing executive functions.

[0013] According to one embodiment of the invention, the meta-engine is configured to aggregate the input data via a pivot model for analyzing and interpreting human behavior data. Such a pivot model makes it possible to provide a common language in order to standardize the data acquired by the single-modal behavioral analysis engines before they are transmitted to the meta-engine.

[0014] According to one embodiment of the invention, the self-learning statistical analysis module is configured to statistically synthesize the understanding of the cyber-cognitive analysis platform into a comprehensible and probable statistical model.

[0015] According to one embodiment of the invention, the self-learning statistical analysis module is configured to provide data explaining the analysis. This makes it possible to transcribe in near real time the precise analysis of the human behavior studied.

[0016] According to one embodiment of the invention, the self-learning statistical analysis module is configured to aggregate historical data to evolve its statistical model. This makes it possible to further improve the accuracy and reliability of the single-modal human behavior analysis engines used.

[0017] According to one embodiment of the invention, the self-learning statistical analysis module is configured to implement a hierarchical hidden Markov chain structure. This makes it possible to provide explainability and multimodality to the underlying problem of multivariate analysis and integration of multiple sensors, all within a comprehensible structure.

[0018] According to a particular technical characteristic of the invention, the cyber-cognitive analysis platform is an integrated platform.

[0019] According to a second aspect, the invention relates to a method for analyzing the com human behavior, implemented by a system comprising a plurality of input data sensors relating to human behavior, a plurality of single-modal behavioral analysis engines connected to the data sensors, and a cyber-cognitive analysis platform connected to the single-modal behavioral analysis engines, each single-modal behavioral analysis engine being capable of acquiring the input data relating to human behavior; the cyber-cognitive analysis platform comprising a meta-engine and at least two modules configured to reproduce cognitive functions of a human brain, the meta-engine being connected to said at least two modules and being configured to aggregate the input data from the plurality of single-modal behavioral analysis engines, to consolidate said input data, and to transmit said consolidated input data to said at least two modules;the meta-engine being provided with a self-learning statistical analysis module, said self-learning statistical analysis module being configured to analyze and correlate the input data from the plurality of single-modal behavioral analysis engines, and to provide an analysis of human behavior from said analyzed and correlated input data and data processed by said at least two modules, the method comprising the following steps: ; - a step of capturing input data relating to human behavior, by at least one of the data sensors; - a step of acquisition, by at least one of the single-modal behavioral analysis engines, of input data relating to human behavior captured by said at least one data sensor; - a step of transmitting to the cyber-cognitive analysis platform, by said at least one monomodal behavioral analysis engine, the acquired input data; - an aggregation step, by the meta-engine, of the transmitted input data; - a consolidation step, by the meta-engine, of the input data aggregated; - a step of transmission to said at least two modules, by the meta-engine, of the consolidated input data; - a step of analysis and correlation, by the self-learning statistical analysis module, of the input data transmitted to the cyber-cognitive analysis platform; and - a step of providing, by the self-learning statistical analysis module, an analysis of human behavior from said analyzed and correlated input data and data processed by said at least two modules. Figures

[0020] [Fig.l] is a schematic representation of a human behavior analysis system according to one embodiment of the invention; and

[0021] [Fig.2] is a flowchart representing a method of analyzing human behavior, implemented by the analysis system of [Fig.l], according to the present invention. Detailed description of the invention

[0022] In [Fig.l] is shown a system 2 for analyzing human behavior according to an embodiment of the invention. The system 2 for analyzing human behavior comprises several sensors 4 of input data relating to human behavior, several monomodal behavioral analysis engines 6 connected to the data sensors 4, and a cyber-cognitive analysis platform 8 connected to the monomodal behavioral analysis engines 6.

[0023] The sensors 4 of input data relating to human behavior are for example made up of microphones, video cameras, etc.

[0024] Each single-modal behavioral analysis engine 6 is capable of acquiring input data relating to human behavior. Each single-modal behavioral analysis engine 6 is for example implemented in the form of software installed on a dedicated server (not shown in [Fig.l]). The cyber-cognitive analysis platform 8 comprises a meta-engine 10 and six modules 12a-12f configured to reproduce cognitive functions of a human brain. Preferably, the cyber-cognitive analysis platform 8 is an integrated platform.

[0025] The meta-engine 10 is connected to the modules 12a-12f and is configured to aggregate the input data from the single-modal behavioral analysis engines 6, to consolidate the input data, and to transmit the consolidated input data to the modules 12a-12f. In other words, the meta-engine 10 receives as input behavioral data, over time, in a specific context, of a given individual (for whom a maximum of information has been recorded in order to even allow correlations between individuals according to multifactorial comparisons), such data being derived from several of the single-modal engines 6; in order to determine which analysis and which data are the most relevant to reflect the multiple reality of the human behavior in question, transmitted by the sum of these single-modal meta-engines 6.

[0026] A first module 12a reproduces a perception function, a second module 12b reproduces an attention function, a third module 12c reproduces a memory function, a fourth module 12d reproduces a motor function, a fifth module 12e reproduces a language function, and a sixth module 12f reproduces executive functions.

[0027] The first module 12a constitutes the means by which system 2 acquires the data from reality, which it is responsible for interpreting. Each of the monomodal behavioral analysis engines 6 can be seen as a sense of the human being. System 2 therefore has N senses, as opposed to the five senses of the human being. N senses, because the existing technologies (interpretations of facial expression, vocal intonation, semantic expression and gestures) must be multiplied by the two cognitive science methods used. For example, the analysis of semantics can be treated differently in linguistics and cognitive psychology. Even within psychological theories, a Rogerian approach will give a different result from an Eriksonian approach.The possibilities are multiple and system 2 can therefore benefit from these N sources of interpretation of captured data, taken from reality, but seen through the prisms of different, not to say differentiating, cognitive theories.

[0028] The second module 12b is a module which focuses temporally on an exogenous stimulus (for example an exploding firecracker) or on an endogenous stimulus (for example a feeling of frustration) and which focuses a salient analytical axis (the cyber-cognitive analysis platform 8 decides, like the brain, to analyze a situation or an event through this salient analytical axis). The identification of this analytical axis is identified upon acquisition of the data from reality via the monomodal behavioral analysis engines 6.

[0029] Attention is subject to what is called in cognitive sociology parasitic variables. Cognitive sociology identifies three types of variables for analyzing the same event: independent variables (the real causes of the phenomenon), dependent variables (the causes that can modulate the intensity of the phenomenon) and parasitic variables (which seem statistically to correlate with the interpretation of the phenomenon, but which in fact represent either statistical chance or a cognitive bias, and in particular confirmation bias or anchoring heuristic bias).

[0030] The third module 12c is a module that stores and organizes information, in accordance with the regulations in force. In the third module 12c, a unique individual is identified from a non-significant identifier. The data sources studied by the monomodal behavioral analysis engines 6 (e.g., audio or video recordings) are never stored, but only the results data transmitted by these same engines, as well as personal data only when their users concerned have given their consent and for which these same users are given the possibility of accessing, modifying or deleting them upon simple request.

[0031] The human brain distinguishes five major types of memory, for which the third module 12c reproduces cyber cognitive functions. These five major types of memory are as follows: - working memory, which stores the analyzed information. This is what is called "buffer memory" in IT. This memory is volatile and is therefore only active in the event of statistical reconciliations until the end of the production of the final statistical report data; - procedural memory (automatic and often conscious), which represents all of an individual's know-how and automatism. Linked to the cerebellum, it deals with ordinary motor actions subject to automatism. Due to their nature, it is often transcribed into non-verbal action. This is often a source of parasitic variable (because the intention behind it is difficult to evaluate, since it is by nature an automatism). The third module 12c therefore identifies the probabilities for a non-verbal action to be of a procedural type, in order to treat this action in the particular case of the various related psychological theories. It is important to note that a gestural slip of the tongue, a form of procedural automatism, can be more meaningful because it reveals the underlying intention of its author than the recurrent phenomenon studied in itself.Thus the typing of this type of memory allows the most accurate assessment of the event analyzed through different theories of cognitive sciences; - semantic memory (declarative and conscious), which is linked to the frontal and temporal lobes, aims to process the meaning of things and to store knowledge and vocabulary. It is strongly correlated with the monomodal behavioral analysis engines 6, in relation to the metaengine 10. Nevertheless, it is also present, in a coherent way, within the third module 12c. It is another bridge between the empirical collection of data, through the prism of the monomodal engines 6, and the typing of memory inherent in the cyber cognitive brain, as is also the case in the human brain. It is therefore possible to correlate, segmentation of individuals by segmentation of individuals, the way of structuring language based on the principle that a correctly carried out segmentation of individuals makes it possible to identify more relevant language types than academic language (e.g. generational language, by social groups, etc.).The more the meta-engine 10 learns, the more relevant it becomes and the more it is able to determine the emergence of future semantic concepts specific to a typical segmentation, by projecting the history of the statistical emergence of the development of the . language of a given segmented group (example: identify how a new concept develops, by which mechanisms it becomes predominant and predict its future predominance, according to the path of predominance taken statistically by concepts of the same nature analyzed in the past); - episodic memory (declarative and conscious) represents memories linked to the individual's history, including the endogenous and exogenous emotional context in which the events in question occurred. Physiologically, it is linked to the amygdala and the hippocampus. This memory therefore stores the multifactorial and temporal relationships between the independent variables, the dependent variables and the parasitic variables of a typical event. It is thus possible to call upon the history function of the research carried out, but also to determine a weight of this same history in the reaction, especially endogenous, of a given response to a studied event. Because the human being is made of his potentialities and his history; - implicit, unconscious memory, which influences the actions of individuals without their knowledge. It deals with complex automatisms, as opposed to procedural memory, which deals with ordinary automatisms. A stimulus to which a subject is, through their history, particularly sensitive generates an automatism that is complex to study. This memory represents the largest source of parasitic variables in the model and must be typed according to a probabilistic model. It is often the most meaningful and at the same time the most complex to determine by algorithm. Nevertheless, it should be noted that the human brain performs these automatic actions by drawing on its different memories. Statistical correlations are therefore a real lever for studying typical responses, again on a segmented population, because it is obviously not possible to study the complete histories of individuals one by one.

[0032] The fourth module 12d is a module encompassing all the motor actions voluntarily coordinated by the human brain (including in particular the cognitive analysis of gestures). This is the most complex module but is developed according to a "Minimum Viable Product" approach: - first of all universal gestures such as for example: shaking the head up and down to say yes, from right to left to say no (even if there are some cultural variants which are correlated to the cultural origin of the individual studied), raising the eyebrows and widening the eyes for astonishment, etc.; - then standardized gestures. Such as sign language or signs of distinction through cultural acquisition (falling into line in a class, sports chants, cultural ways of greeting each other, etc.); - gestures interpretable by behavioral theories (example: reflexive posture, defense, etc.); - all other gestures without interaction with the environment. They are correlated with historical data and other empirical data to help understand their meaning; - all other gestures interacting with the environment. They are correlated with historical data and other empirical data to help understand their meaning; - ultimately, it is possible to study and note, in particular for the same individual, an evolution (improvement or deterioration) of their movements, which can be correlated with an illness, a cure or a change in psychological state.

[0033] The fifth module 12e brings together the reception (hearing, decoding, understanding) and the transmission of information. This fifth module 12e is strongly correlated with the first module 12a of perception. The fifth module 12e is configured to analyze the language through the meta-engine 10 and according to an expert system which deals with the particularity of the different spoken languages. A voice-to-text conversion function makes it possible to convert the spoken language (which is captured) into written language, digitized and intelligible for the semantic analysis engines used in the system 2.

[0034] In the sixth module 12f, executive functions are like typical algorithms that an individual will operate with obvious intentionality. For example, looking to the right and left of the street before crossing, resisting the urge to fall asleep at the wheel, etc. These functions are coded according to the theories of cognitive science, and particularly behavioral science, in order to be able to be identified within the framework of the analyzed data. This assumes for several of them to have the analysis of the interaction between the individual studied and his environment. Among them, we can cite in particular those which deal with: - organization and planning (steps necessary to achieve an objective); - flexibility (adaptation to constraints, to the environment, etc.); - abstraction (connecting principles and knowledge); - judgment (evaluate according to a standard or morality); - self-control (control of one's actions and emotions in the face of the often exogenous constraints encountered); - creativity.

[0035] The meta-engine 10 comprises a self-learning statistical analysis module 14. The self-learning statistical analysis module 14 is configured to analyze and correlate the input data from the single-modal behavioral analysis engines 6, and to provide an analysis of human behavior from this analyzed and correlated input data and from data processed by the modules 12a-12f. Preferably, the self-learning statistical analysis module 14 is a multi-factorial and temporal statistical analysis module. More preferably, the self-learning statistical analysis module 14 is configured to statistically synthesize the understanding of the cyber-cognitive analysis platform 8 into a comprehensible and probable statistical model (along different axes and in near real time), to provide data explaining the analysis, and to aggregate historical data to evolve its statistical model.

[0036] Preferably, the meta-engine 10 is configured to aggregate the input data via a pivot model 16 for analyzing and interpreting human behavior data. The pivot model 16 takes into account, on the one hand, the latest work in cognitive science, and more specifically that concerning the analysis of human behavior, and, on the other hand, the technical limitations of artificial intelligence algorithms and cognitive computing. The pivot model 16 makes it possible to communicate to the meta-engine 10 the various input data collected by the single-modal behavioral analysis engines 6.

[0037] The pivot model 16 is provided with an “evaluation” indicator. Such an “evaluation” indicator makes it possible to indicate, when one of the monomodal behavioral analysis engines 6 knows how to recognize it, the phenomenon triggering the transmitted data (for example: the emotion of surprise generated by an exploding firecracker, when the phenomenon studied is an emotion). When the monomodal engine in question 6 does not know how to recognize the “evaluation” indicator, it is the metaengine 10 which, by correlating all the data of a single phenomenon at its disposal, indicates an “evaluation” indicator associated with a probability of certainty. Consequently, the higher the number of monomodal engines 6, the more the “evaluation” indicator of a given analyzed event is probably certain. The problem increases, however, in the case of endogenous stimuli.Indeed, identifying an exogenous stimulus on a video (for example) is relatively simple (for example: the case of the exploding firecracker), but identifying an endogenous stimulus requires having even a summary psychological interpretation of the subject (individual), who generated the phenomenon studied, caused by the endogenous stimulus in question. For example, an event with an unlikely "evaluation" indicator falls into the probabilistic model of implicit memory mentioned previously, in connection with the third module 12c.

[0038] The meta-engine 10 therefore takes into account all of these elements to conclude on a probability of certainty of the “evaluation” indicator of the event studied.

[0039] According to a preferred embodiment of the invention, the self-learning statistical analysis module 14 is configured to implement a hierarchical hidden Markov model (HHMM) structure. Such an HHMM structure makes it possible to correlate the input data from the single-modal behavioral analysis engines 6 with different states of the human individual studied (such as, for example, their emotions, their needs, their short-term and long-term motivation, etc.), and thus to confer explainability and multimodality to the underlying problem of multivariate analysis and integration of multiple sensors, all within a comprehensible structure. Some states are less likely to change than others (short-term motivation changes more frequently than long-term motivation, for example) and can induce mechanisms on the other states.

[0040] HHMM modeling is therefore perfectly suited to the aforementioned problem for several reasons: - it allows the categories of states to be hierarchized in the form of layers, from the least flexible (upper layers) to the most flexible (lower layers); - it allows, for each state, to construct an HMMM which has a specific structure relating to the state considered; - it allows each state of the same layer to be linked, opening the way to multimodality; - the structure of an HHMM can illustrate psychological, socio-logical or cognitive theories in its design, in order to statistically formalize a representation.

[0041] Labeling is also possible to label the different states of the hierarchical hidden Markov chain structure. Each of the labels conditions the value of the transition probabilities (depending on the subcategory of the population concerned), but the psychological profile in particular also conditions the order of the layers of each HHMM (a hypersensitive profile will let itself be governed by its emotions for example).

[0042] The method for analyzing human behavior, implemented by the analysis system 2 according to the invention, will now be described. The method comprises the following steps: - a step 20 of capturing input data relating to human behavior, by at least one of the data sensors 4; - a following step 22 of acquisition, by at least one of the monomodal behavioral analysis engines 6, of the input data relating to a com human behavior which have been captured by the data sensor(s) 4; a following step 24 of transmission to the cyber-cognitive analysis platform 8, by said at least one monomodal behavioral analysis engine 6, of the acquired input data; a following step 26 of aggregation, by the meta-engine 10, of the transmitted input data; a next step 28 of consolidation, by the meta-engine 10, of the aggregated input data; a next step 30 of transmission to the modules 12a-12f, by the metaengine 10, of the consolidated input data; a next step 32 of analysis and correlation, by the self-learning statistical analysis module 14, of the input data transmitted to the cyber-cognitive analysis platform 8; and a next step 34 of providing, by the self-learning statistical analysis module 14, an analysis of human behavior from the analyzed and correlated input data and data processed by the modules 12a-12f.

Claims

Claims

1. System (2) for analyzing human behavior, the system (2) comprising a plurality of sensors (4) of input data relating to human behavior, a plurality of single-modal behavioral analysis engines (6) connected to the data sensors (4), and a cyber-cognitive analysis platform (8) connected to the single-modal behavioral analysis engines (6), each single-modal behavioral analysis engine (6) being capable of acquiring the input data relating to human behavior;the cyber-cognitive analysis platform (8) comprising a meta-engine (10) and at least two modules (12a-12f) configured to reproduce cognitive functions of a human brain, the meta-engine (10) being connected to said at least two modules (12a-12f) and being configured to aggregate the input data from the plurality of single-modal behavioral analysis engines (6), to consolidate said input data, and to transmit said consolidated input data to said at least two modules (12a-12f); the meta-engine (10) being provided with a self-learning statistical analysis module (14), said self-learning statistical analysis module (14) being configured to analyze and correlate the input data from the plurality of single-modal behavioral analysis engines (6), and to provide an analysis of human behavior from said analyzed and correlated input data and data processed by said at least two modules (12a-12f).;

2. A human behavior analysis system (2) according to claim 1, wherein the self-learning statistical analysis module (14) is a multi-factorial and temporal statistical analysis module.

3. System (2) for analyzing human behavior according to claim 1 or 2, wherein the cyber-cognitive analysis platform (8) comprises six modules (12a-12f) configured to reproduce cognitive functions of a human brain; a first module (12a) reproducing a perception function; a second module (12b) reproducing an attention function; a third module (12c) reproducing a memory function; a fourth module (12d) reproducing a motor function; a fifth module (12e) reproducing a language function; a sixth module (12f) reproducing executive functions.

4. A system (2) for analyzing human behavior according to any preceding claim, wherein the meta-engine (10) is configured to aggregate the input data via a pivot model (16) for analyzing and interpreting human behavior data.

5. A human behavior analysis system (2) according to any preceding claim, wherein the self-learning statistical analysis module (14) is configured to statistically synthesize the understanding of the cyber-cognitive analysis platform (8) into a comprehensible and probable statistical model.

6. System (2) for analyzing human behavior according to any one of the preceding claims, wherein the self-learning statistical analysis module (14) is configured to provide data explaining the analysis.

7. A system (2) for analyzing human behavior according to any one of the preceding claims, wherein the self-learning statistical analysis module (14) is configured to aggregate historical data to evolve its statistical model.

8. A human behavior analysis system (2) according to any preceding claim, wherein the self-learning statistical analysis module (14) is configured to implement a hierarchical hidden Markov chain structure.

9. A system (2) for analyzing human behavior according to any one of the preceding claims, wherein the cyber-cognitive analysis platform (8) is an integrated platform.

10. A method for analyzing human behavior, implemented by a system (2) comprising a plurality of sensors (4) of input data relating to human behavior, a plurality of monomodal behavioral analysis engines (6) connected to the data sensors (4), and a cyber-cognitive analysis platform (8) connected to the monomodal behavioral analysis engines (6), each monomodal behavioral analysis engine (6) being capable of acquiring the input data relating to human behavior; the cyber-cognitive analysis platform (8) comprising a meta-engine (10) and at least two modules (12a-12f) configured to reproduce cognitive functions of a human brain, the meta-engine (10) being connected to said at least two modules (12a-12f) and being configured to aggregate the input data from the plurality of monomodal behavioral analysis engines (6), to consolidate said input data, and to transmit said consolidated input data to said at least two modules (12a-12f); the meta-engine (10) being provided with a self-learning statistical analysis module (14), said self-learning statistical analysis module (14) being configured to analyze and correlate the input data from the plurality of single-modal behavioral analysis engines (6), and to provide an analysis of human behavior from said analyzed and correlated input data and data processed by said at least two modules (12a-12f), the method comprising the following steps: - a step (20) of capturing input data relating to human behavior, by at least one of the data sensors (4); - a step (22) of acquisition, by at least one of the monomodal behavioral analysis engines (6), of input data relating to human behavior captured by said at least one data sensor (4); - a step (24) of transmission to the cyber-cognitive analysis platform (8), by said at least one monomodal behavioral analysis engine (6), of the acquired input data; - a step (26) of aggregation, by the meta-engine (10), of the transmitted input data; - a step (28) of consolidation, by the meta-engine (10), of the aggregated input data; - a step (30) of transmission to said at least two modules (12a-12f), by the meta-engine (10), of the consolidated input data; - a step (32) of analysis and correlation, by the self-learning statistical analysis module (14), of the input data transmitted to the cyber-cognitive analysis platform (8); and - a step (34) of providing, by the self-learning statistical analysis module (14), an analysis of human behavior from said analyzed and correlated input data and data processed by said at least two modules (12a-12f).

Images