Device and method for monitoring the posture of a subject

Abstract

The invention relates to a method and a processing device (T1) for monitoring the posture of a subject (UR1). The method comprises: obtaining sensor data (DT1) representative of the subject (UR1); generating, from the sensor data, a model (ML1) defining anatomical points (PT1) on the subject over time; determining, from the model (ML1), kinematics variables (V1) representative of the change in the posture of the subject; determining, from the kinematics variables (V1), at least one non-linear variability indicator (V2) representative of a temporal structure of the kinematics variables (V1); analysing (S10) the at least one non-linear variability indicator (V2) by comparing it with a respective reference value (REF2); and generating a posture setpoint (CMD1) on the basis of a result of the analysis.

Description

[0001] Description

[0002] Title of the invention: Device and method for tracking the posture of a subject

[0003] Domain

[0004] The invention relates to the field of analysis and monitoring of a subject's posture and relates more particularly to a processing device and method for monitoring a subject's posture, in particular with a view to evaluating and / or correcting the subject's posture.

[0005] Prior art

[0006] It is now generally accepted that moving is widely beneficial for people, whether physically, psychologically, cognitively, immune-related, etc., and at any age. Physical activity, and more generally movement, therefore constitute major public health issues and now occupy a central place in supporting individuals, particularly for athletes and people with specific needs, or for the management of injuries, pathologies, post-operative follow-ups, etc.

[0007] In both professional and amateur sports, injury is a major issue for athletes and practitioners. There is currently a significant need for athlete care to prevent injuries (preparation, re-athleticization, etc.), assess their physical and psychological condition, and optimize their performance. Injury prevention and management are key issues in the monitoring and care of athletes at all levels.

[0008] In addition to supporting athletes, movement is a key element in supporting individuals, particularly in a preventative or curative context (after injury, for example). Healthcare offerings are therefore increasingly moving toward an active approach to health. Many public programs thus aim to promote physical activity in all its forms (sport, play, work, etc.).

[0009] Monitoring a person's posture, i.e. the way they stand or move in an environment, can provide a great deal of information about the person's health or condition (e.g. in the case of injury, pathology, etc.) or about their functioning (e.g. their performance, habits, etc.). There is therefore a growing need today, particularly among professionals but also among the general public, to assess the quality of people's movement, in various contexts, including for the purposes of assisting with physical activity, gaming, performance monitoring, rehabilitation or even re-athleticization. In this context, movement analysis techniques have grown significantly in recent years, but current solutions currently present limitations and constraints that remain a hindrance to the development of this sector.

[0010] Thus, current posture (or movement) analysis techniques generally do not offer satisfactory results, particularly in terms of reliability and precision. For example, applications currently available on smartphones for posture tracking do not allow for a detailed analysis of a posture, whether static or dynamic. These solutions often induce positioning errors in the subject's anatomical parts and are generally limited to comparing the captured movements to reference models without being able to assess the quality of the movement in all its complexity.

[0011] In addition to limitations in terms of accuracy and reliability, some current solutions require complex, expensive and / or time-consuming equipment to implement. Some current techniques require, for example, the deployment of a significant number of sensors, or the installation of reflective targets or markers that are positioned on the target subject itself. The complexity of installing, adjusting and using the necessary equipment can then prove problematic, or even prohibitive, in practice. For example, placing markers on the subject can be tedious and presents the risk of leading to measurement errors in the event of unexpected movements of the markers. The economic, human and / or time cost of implementing these techniques as well as the training required for users limit the deployment of these solutions.

[0012] Statement of the invention

[0013] One of the objects of the present invention is to solve at least one of the problems or deficiencies of the technological background described above.

[0014] Another object of the present invention is to enable efficient monitoring of the posture of a subject in terms of precision and reliability, in particular with a view to optimizing the health and / or performance of individuals, for example for the purposes of assisting with physical activity, physical preparation, play, performance monitoring, rehabilitation and / or re-athleticization.

[0015] Another object of the present invention is to enable accurate and reliable analysis of the posture, static or dynamic, of a subject with limited cost and complexity of implementation. Another object of the present invention is to effectively assist in correcting the posture of a subject.

[0016] To this end, a first aspect of the invention relates to a processing method (or tracking method), implemented by a processing device, for tracking the posture of a subject, said method comprising: a) obtaining sensor data representative of the subject; b) generating, from the sensor data, a model defining anatomical points of the subject over time; c) determining, from the model, kinematic variables representative of the evolution of the posture of the subject over time; d) determining, from the kinematic variables, at least one non-linear variability indicator representative of a temporal structure of the kinematic variables; e) analyzing said at least one non-linear variability indicator by comparison with a respective reference value; and f) generating a posture setpoint as a function of a result of the analysis.

[0017] The method according to the invention may include other characteristics which may be taken separately or in combination, in particular among the following embodiments which are presented for illustration purposes only and may be combined or associated unless otherwise stipulated.

[0018] In a particular example, the sensor data obtained in a) comprises image data captured by a plurality of image capture devices, the image data from each image capture device defining a respective 2D view of the subject over time; generating b) comprising:

[0019] - estimation of a position of the anatomical points of the subject over time in the 2D view from each image acquisition device; and

[0020] - generation of the 3D model by triangularization of the estimated positions of the anatomical points in each 2D view.

[0021] In a particular example, generation b) includes:

[0022] - assigning, to each estimated position of the anatomical points of the subject in the 2D views, a respective confidence score representative of a level of confidence in said estimated position; the model being generated from the estimated positions of the anatomical points from which are excluded, by a first filtering, each estimated position of an anatomical point whose confidence score does not reach at least a first threshold value. According to a particular example, the generation b) of the model comprises:

[0023] - comparison of first and second confidence scores associated respectively with first and second estimated positions for the same anatomical point in 2D views from the image acquisition devices; the model being generated from the estimated positions of the anatomical points to which a second filtering is applied according to which, if the difference between said first and second confidence scores reaches at least a second threshold value, the one among the first and second estimated positions having the lowest confidence score is excluded.

[0024] According to a particular example, said at least one non-linear indicator determined in d) comprises a first sequence of values ​​over time; the analysis e) comprising:

[0025] - comparison of the first sequence of values ​​of said at least one non-linear indicator over time with a second sequence of reference values ​​respectively; the result of analysis e) being a function of a result of said comparison of the first and second sequences.

[0026] According to a particular example, generation f) comprises at least one of:

[0027] - restitution, as a first posture instruction, by means of a user interface, of posture information specifying how to correct the subject's posture; and

[0028] - sending, as a second posture instruction, a command causing a movement assistance device to be configured to correct the subject's posture.

[0029] In a particular example, analysis (e) includes:

[0030] - checking whether said at least one non-linear variability indicator meets a conformity criterion with respect to the respective reference value; wherein, if the conformity criterion is not met, the posture instruction generated in f) is configured to correct the posture of the subject.

[0031] According to a particular example, if a non-linear indicator of variability - called a non-compliant indicator - does not meet the compliance criterion, analysis e) includes:

[0032] - determination of at least one kinematic variable, called target kinematic variable, associated with the non-compliant indicator and of a correction to be applied to said at least one target kinematic variable to improve the non-compliant indicator compared to the reference value according to the compliance criterion; said posture instruction being generated in f) as a function of the correction to be applied to said at least one target kinematic variable to correct the posture of the subject.

[0033] According to a particular example, the determination of said at least one target kinematic variable and the correction to be applied is carried out by an optimization method taking as input: - a trend objective of the non-compliant indicator;

[0034] - a function defining a relationship of the non-compliant indicator with a plurality of associated kinematic variables; and

[0035] - constraint data defining movement constraints that the model must respect; the optimization method delivering as output setpoint data defining said at least one target kinematic variable and the correction to be applied to modify the non-compliant indicator according to the trend objective; the posture setpoint being generated in f) from the setpoint data.

[0036] According to a particular example, the method comprises, prior to calculation c):

[0037] - recognition, from the model, of a movement made by the subject; and

[0038] - selection, depending on the recognized movement, of the kinematic variables to be calculated in c).

[0039] According to a particular example, said at least one nonlinear indicator of variability comprises at least any one of: an entropy value or an entropic indicator; a Lyapunov exponent; a result of a fractal analysis; a superrogation indicator; a geometric approach indicator; and an indicator of recurring correlations.

[0040] In a particular embodiment, the different steps of the method according to the first aspect of the invention are determined by computer program instructions. Consequently, a second aspect of the invention relates to a computer program on an information medium (or recording medium), this program being capable of being implemented in a processing device, or more generally in a computer, this program comprising instructions adapted to the implementation of the steps of the processing method according to the first aspect of the invention. In particular, the second aspect of the invention relates to a computer program comprising instructions for the execution of the steps of the method according to the first aspect of the invention when said program is executed by a computer.

[0041] Thus, the method of the invention can be implemented by means of a non-volatile memory storing computer program instructions and by means of a processor executing these instructions.

[0042] This program may use any programming language, and may be in the form of source code, object code, or code intermediate between source code and object code, such as in a partially compiled form, or in any other desirable form.

[0043] A third aspect of the invention relates to an information medium (or recording medium) readable by a computer, and comprising instructions of the computer program according to the second aspect of the invention.

[0044] The information carrier may be any entity or device capable of storing the program. For example, the carrier may include a storage medium, such as a rewritable non-volatile memory or ROM, for example a CD-ROM or a microelectronic circuit ROM, or a magnetic recording medium, for example a hard disk or a memory in the form of a USB key or the like.

[0045] On the other hand, the information carrier may be a transmissible carrier such as an electrical or optical signal, which may be conveyed via an electrical or optical cable, by radio or by other means. The program according to the invention may in particular be downloaded from a network such as the Internet.

[0046] Alternatively, the information carrier may be an integrated circuit in which the program is incorporated, the circuit being adapted to perform or to be used in the performance of the method in question.

[0047] A fourth aspect of the invention relates to a processing device (or tracking device) configured to implement the method according to the first aspect of the invention. The fourth aspect of the invention relates in particular to a processing device configured to carry out posture tracking of a subject, said device comprising a memory associated with a processor configured to implement the steps of the processing method according to the first aspect of the present invention.

[0048] According to one example, the invention according to the fourth aspect provides a treatment device for assisting a subject in performing a posture, said device comprising:

[0049] - obtaining module configured to obtain sensor data representative of the subject;

[0050] - first generation module configured to generate, from the sensor data, a model defining anatomical points of the subject over time;

[0051] - first determination module configured to calculate, from the model, kinematic variables representative of the evolution of the subject's posture over time;

[0052] - second determination module configured to determine, from the kinematic variables, at least one non-linear indicator of variability representative of a temporal structure of the kinematic variables;

[0053] - analysis module configured to analyze said at least one non-linear indicator of variability by comparison with a respective reference value; and

[0054] - second generation module configured to generate a posture instruction based on a result of the analysis. Note that the various embodiments mentioned above (as well as those described below) in relation to the processing method of the invention as well as the associated advantages apply in a similar manner to the processing device of the invention.

[0055] In particular, for each step of the processing method, the processing device of the invention may comprise a corresponding module configured to carry out said step. According to one embodiment, the invention is implemented by means of software and / or hardware components. In this regard, the term “module” may correspond in this document to a software component, a hardware component or a set of hardware and software components.

[0056] A software component corresponds to one or more computer programs, one or more sub-programs of a program, or more generally to any element of a program or software capable of implementing a function or a set of functions, as described below for the module concerned. Such a software component can be executed by a data processor of a physical entity (terminal, server, gateway, router, etc.) and is likely to access the hardware resources of this physical entity (memories, recording media, communication buses, electronic input / output cards, user interfaces, etc.).

[0057] Similarly, a hardware component is any element of a hardware assembly capable of implementing a function or set of functions, as described below for the module concerned. It may be a programmable hardware component or one with an integrated processor for running software, for example an integrated circuit, an electronic card, etc.

[0058] The present invention as defined above and described below in particular examples advantageously makes it possible to accurately and reliably monitor the posture of a subject. In particular, it is possible to evaluate the quality of a posture, static or dynamic, of a subject over time and to guide or improve this posture by generating a posture instruction adapted to the behavior of the subject. Such monitoring makes it possible in particular to evaluate and improve the health of the subject and / or his performance, and this for various purposes depending on the case, such as for example for assistance with physical activity, physical preparation, play, physical rehabilitation and / or re-athleticization. The invention further allows accurate and reliable monitoring of the posture, static or dynamic, of a subject and this with limited cost and complexity of implementation. The present invention can also advantageously offer diagnostic assistance.

[0059] In particular, the analysis of at least one non-linear indicator of variability determined from kinematic variables advantageously makes it possible to finely evaluate the quality of a subject's movement in order to take into account complex aspects that are not perceptible on the sole basis of kinematic variables. Depending on the result of this analysis, a posture instruction can be advantageously generated, for example to adapt, improve and / or correct the subject's posture and thus help to improve the subject's health and / or performance.

[0060] The invention therefore offers a powerful tool, both to movement professionals such as health and / or sports professionals (doctors, therapists, physiotherapists, physical trainers, sports coaches / trainers, etc.) and to the general public, to analyze and improve the posture of a subject.

[0061] Brief description of the drawings

[0062] Other characteristics and advantages of the present invention will emerge from the description of the particular and non-limiting exemplary embodiments of the present invention below, with reference to the appended figures 1 to 18, in which:

[0063] [Fig. 1] Figure 1 schematically represents a processing device cooperating with at least one associated device, according to at least one embodiment of the invention;

[0064] [Fig. 2] Figure 2 represents an example of implementation of the processing device of Figure 1 according to at least one embodiment of the invention;

[0065] [Fig. 3] Figure 3 schematically represents the processing device (or monitoring device) of Figure 1, according to at least one embodiment of the invention;

[0066] [Fig. 4] Figure 4 represents, in the form of a diagram, the steps of a processing method (or monitoring method), according to at least one embodiment of the invention;

[0067] [Fig. 5] Figure 5 schematically represents the obtaining of sensor data (image data) during the processing method of Figure 4, according to at least one embodiment of the invention;

[0068] [Fig. 6] Figure 6 schematically represents the obtaining of sensor data (inertial data) during the processing method of Figure 4, according to at least one embodiment of the invention;

[0069] [Fig. 7] Figure 7 schematically represents the obtaining of sensor data (EMG data) during the processing method of Figure 4, according to at least one embodiment of the invention;

[0070] [Fig. 8] Figure 8 schematically represents the generation of a model defining anatomical points during the processing method of Figure 4, according to at least one embodiment of the invention;

[0071] [Fig. 9] Figure 9 schematically represents the generation of a model defining anatomical points during the processing method of Figure 4, according to at least one embodiment of the invention; [Fig. 10] Figure 10 schematically represents the calculation of kinematic variables during the processing method of Figure 4, according to at least one embodiment of the invention;

[0072] [Fig. 11] Figure 11 schematically represents the calculation of kinematic variables during the processing method of Figure 4, according to at least one embodiment of the invention;

[0073] [Fig. 12] Figure 12 schematically illustrates different degrees of posture complexity that can be represented by a non-linear indicator of variability determined during the processing method of Figure 4, according to at least one embodiment of the invention;

[0074] [Fig. 13] Figure 13 schematically illustrates different degrees of posture complexity that can be represented by a non-linear indicator of variability determined during the processing method of Figure 4, according to at least one embodiment of the invention;

[0075] [Fig. 14] Figure 14 schematically illustrates different degrees of posture complexity that can be represented by a non-linear indicator of variability determined during the processing method of Figure 4, according to at least one embodiment of the invention;

[0076] [Fig. 15] Figure 15 schematically represents the determination of at least one non-linear indicator of variability (approximation entropic indicator) during the processing method of Figure 4, according to at least one embodiment of the invention;

[0077] [Fig. 16] Figure 16 schematically represents the determination of at least one non-linear variability indicator (GEM type indicator for “Goal Equivalent Manifold” in English) during the processing method of figure 4, according to at least one embodiment of the invention;

[0078] [Fig. 17] Figure 17 schematically represents the generation of a posture instruction during the processing method of Figure 4, according to at least one embodiment of the invention; and

[0079] [Fig. 18] Figure 18 schematically represents the generation of a posture instruction during the processing method of Figure 4, according to at least one embodiment of the invention.

[0080] Description of embodiments

[0081] Examples of implementations of the invention will now be described in the following with reference to Figures 1-18. Unless otherwise indicated, elements common or similar to several figures bear the same reference signs and have identical or similar characteristics, so that these common elements are generally not described again for the sake of simplicity.

[0082] The terms "first(s)", "second(s)", etc.) are used in this document by arbitrary convention to identify and distinguish different elements (such as operations, threshold values, etc.) implemented in the embodiments described below.

[0083] The invention proposes to track and adapt the posture of a subject. In this context, a posture designates in the present disclosure a positioning or a movement of the subject considered, or more precisely of all or part of the subject's body. In other words, the posture being tracked can be a static posture (position of the body at a given instant) or a dynamic posture (movement of the body). Generally speaking, depending on the precision of the motion sensor(s) used, a posture can therefore in certain cases appear to be static or dynamic. A posture can provide information on the position of one or a plurality of anatomical points of a subject at a given instant or over time.To the extent that a given individual is normally subject to variations, even tiny ones, in the positioning of the different parts of his body due to at least natural or involuntary movements (muscle tension, breathing, tremors, joint repositioning, etc.), a posture can generally be considered dynamic over time provided that the sensor data used are sufficiently precise to represent the movements in question.

[0084] The quality of a subject's posture can be influenced by various factors, including physical, psychological and environmental. It has been found that a good understanding and therefore good monitoring of posture is desirable to effectively guide a subject in performing a posture, movement or physical activity. In the present disclosure, the subject subject to posture monitoring can be any person who may have various physical conditions or states (child, adult, elderly person, injured person, person with a disability, etc.), or more generally a living being, including an animal, such as a dog or a horse for example. The invention relates in particular to a treatment method (or monitoring method), and a corresponding treatment device (or monitoring device), for monitoring the posture of a subject.The method is based in particular on the determination of kinematic variables from a model defining anatomical points of the subject, and on the analysis of at least one non-linear indicator of variability determined from the kinematic variables. A posture instruction can thus be generated based on a result of this analysis, in order to adapt, correct and / or improve the posture of the subject.

[0085] The invention, according to its various embodiments, thus implements a processing method, implemented by a processing device, for monitoring the posture of a subject, said method comprising: a) obtaining sensor data representative of the subject; b) generating, from the sensor data, a model defining anatomical points of the subject over time; c) determining, from the model, kinematic variables representative of the evolution of the posture of the subject over time; d) determining, from the kinematic variables, at least one non-linear variability indicator representative of a temporal structure of the kinematic variables; e) analyzing said at least one non-linear variability indicator by comparison with a respective reference value; and f) generating a posture setpoint as a function of a result of the analysis.

[0086] The invention, according to its various embodiments, also relates to a computer program, and an information medium containing such a program, for executing the steps of the treatment method according to the invention.

[0087] Other aspects and advantages of the present invention will emerge from the exemplary embodiments described below with reference to the drawings mentioned above.

[0088] Figure 1 schematically represents a processing device T1 (also called tracking device or device) capable of cooperating with at least one associated device 20, also called control device, to carry out posture tracking of a subject denoted UR1 of the invention according to particular embodiments. The processing device T1 and said at least one control device 20 together form a processing system (or tracking system) denoted SY1. It is assumed by way of example that the associated device(s) 20 are distinct from the processing device T1 although variants are possible in which the associated device(s) are part of the processing device T1.

[0089] As an example, it is subsequently assumed that the processing device T1, also called device, is configured to monitor the posture of the user UR1 in a context which may vary depending on the case. The subject UR1 may be, for example, an athlete, an injured or convalescing person, or more generally any person whose posture (or movement) we wish to evaluate and improve, for the purposes, for example, of helping with physical maintenance, improving health, preventing injuries, playing and / or performance.

[0090] As illustrated in Figure 1, the processing device T1 may use at least one posture sensor (or motion sensor) 12 to obtain sensor data DT1 representative of the subject UR1, or more precisely, of the posture PS1 of all or part of the subject UR1. The sensor(s) 12 may be distinct from the device T1 or be part of the latter depending on the case. The nature of the sensor data DT1 recovered by the processing device T1 may vary depending on the case, depending in particular on the type of sensor 12 used. Examples of such sensors, denoted 12a, 12b and 12c, are described below.

[0091] According to one example, the processing device T 1 is configured to receive, from at least one image acquisition device 12a, image data DT1a representative of the posture PS1 of the subject UR1.

[0092] According to one example, the processing device T 1 is configured to receive, from at least one inertial sensor 12b positioned on the subject UR1, inertial data DT1b representative of the posture PS1 of the subject UR1.

[0093] According to one example, the processing device T 1 is configured to receive, from at least one EMG (for “electromyogram”) type sensor 12c positioned on the subject UR1, EMG data DT1c representative of the posture PS1 of the subject UR1.

[0094] The processing device T 1 can thus be configured to receive, as sensor data DT1, at least one of the image data DT1a, the inertial data DT1b and the EMG data DT1c as previously described, or any combination of at least two of these types of data. The configuration and use of these sensors 12a-12c, as well as the associated sensor data, are described in more detail in the following description of the tracking method according to various embodiments. The processing device T 1 can be configured to receive and process in real time all or part of the sensor data DT1 received from the sensor(s) 12.

[0095] According to one example, the processing device T 1 is configured to obtain all or part of the sensor data DT 1 by accessing a memory (for example, the memory 6) in which the data is stored. For example, it is possible to record sensor data DT 1 and then configure the processing device T 1 to retrieve and process this data later.

[0096] In the following, it is assumed that the sensor data DT 1 defines a posture PS1 of the subject UR1 over time, which in particular makes it possible to analyze a movement or an evolution of the posture PS1 over time, although variants are possible where the sensor data DT1 only defines the posture PS1 of the user UR1 at a given instant, for example to track the spatial positioning (or disposition) of the subject at a given instant.

[0097] As illustrated, the processing device T1 comprises in this example at least one processor 4 and a memory 6. This memory 6 may comprise various types of memory, in particular a volatile memory (of the RAM type) and a non-volatile memory. The non-volatile memory may comprise a read-only memory (of the ROM type) and / or a rewritable non-volatile memory. This memory 6 may comprise in particular an operating system 10 executable by the processor 4 to operate the processing device T1. The memory 6 constitutes a recording medium (or information medium) according to particular embodiments, readable by the processing device T1, and on which is recorded a computer program PG1 conforming to various particular embodiments. This computer program PG1 comprises instructions for executing the steps of the tracking method of the invention according to particular embodiments.The steps of this method are represented, in a particular embodiment, in figure 4 described later.

[0098] Thus, the processor 4 is configured to execute the instructions of the computer program PG1 in order to carry out steps of the tracking method of the invention according to particular exemplary embodiments. For this purpose, the processor 4 may include integrated memory, an input / output interface, and various circuits known to those skilled in the art. In particular, the processor 4 may use a volatile memory internal to the processing device T1 (this memory being able to be part of the memory 6 or be a separate memory) to carry out the various operations and functions necessary for the operation of the processing device T1, including to execute the computer program PG1 during the implementation of the processing method of the invention.

[0099] As shown in Figure 1, the memory 6 is capable of storing various data that may be used during the execution of the tracking method. Thus, the memory 6 can store a model ML1 defining anatomical points PT1 of the subject UR1. Positions PN1 can be predicted (or estimated) and stored for each anatomical point PT1 of the model ML1, possibly in association with respective confidence scores SC1. The memory 6 can also be used to store kinematic variables V1, one or more non-linear indicators of variability V2, and a reference value REF2. According to a particular example, the memory 6 can also be used to store a linear indicator of variability V3 and a reference value REF3. The nature and use of the aforementioned data are described in more detail later in particular examples.

[0100] The processing device T 1 may have various forms depending on the case, and may in particular be a computer, a server, a smartphone, a tablet, or more generally a processing device comprising the means configured to carry out the steps of the monitoring method of the invention.

[0101] As shown in Figure 1, the processing device T1 is further configured to cooperate with, or control, at least one control device 20. In particular, the processing device T1 is able to send a posture instruction CMD1 to one or more control devices 20. The number and type of control devices 20 used may vary depending on the case. A posture instruction CMD1 is a command (or instruction) intended to control a control device 20 in order to guide or assist the subject UR1 in the execution of a posture, a movement or a physical activity. By way of example, this posture instruction CMD1 makes it possible to adapt, correct and / or improve the posture PS1 of the subject UR1 through the control device(s) 20.

[0102] The type of each posture instruction CMD1 depends in particular on the type of the control device 20 concerned as well as the objective sought. Examples of such control devices, noted 20a, 20b, 20c and 20d (figure 1), are described below.

[0103] According to one example, the processing device T 1 is configured to send a first posture instruction CMD1 to a user interface 20a-20c to cause the restitution of posture information IF1 specifying how to adapt, correct and / or improve the posture PS1 of the subject UR1. As illustrated, this user interface 20 can take various forms, the manner in which the information IF1 is rendered depending on the type of user interface used.

[0104] According to one example, the processing device T 1 is configured to send at least one first posture instruction, more precisely denoted CMD1a, to a display device 20a to cause the display of posture information IF1 (in visual form). According to one example, the processing device T 1 can send at least one first posture instruction CMD1a to a sound device 20b to cause the emission of posture information IF1 in sound form. According to one example, the processing device T 1 can send at least one first posture instruction CMD1a to a light device 20c to cause the emission of posture information IF1 in light form (for example by activating a light indicator or by adapting the color of an emitted light). Several user interfaces 20 among those described can also be used in combination.

[0105] By way of example, a control device 20 cooperating with the processing device T 1 may be a smartphone, tablet, computer or equivalent, comprising a display device 20a as indicated above.

[0106] According to one example, the processing device T 1 is configured to send at least one second posture instruction, more precisely denoted CMD1b, to a control device 20 taking the form of a movement assistance device 20d (also called a physical conditioning device). This second posture instruction CMD1b then causes a setting (or a configuration) of the movement assistance device 20d to adapt, correct and / or improve the posture PS1 of the subject UR1.

[0107] The assistance device 20d may be any device configured, under the control of the processing device T 1 , to facilitate, guide and / or intensify the performance of postures, movements and / or physical activities by a subject or individual. For example, the assistance device 20d may be or include a weight training machine, a treadmill, an elliptical trainer, an exercise bike, a rowing machine, a multi-gym station, etc. This type of device is intended, for example, for improving cardiovascular fitness, muscle strengthening, weight loss, developing proprioception, or more generally for physical preparation or training or re-athleticization.

[0108] The assistance device 20d may be more or less complex depending on the case. In particular, it may be configurable according to at least one operating parameter (for example, a running speed of the belt for a treadmill, a resistance force for an exercise bike or an elliptical bike, a duration of a training program, etc.). The processing device T1 can thus control the configuration of such an assistance device 20d by sending a second posture instruction CMD1b.

[0109] According to one example, the posture sensor(s) 12 used to obtain the DT1 sensor data are part of one or more control devices 20.

[0110] According to one example, the processing device T1 cooperates with at least one sensor 12 and at least one control device 20, these elements T1, 12 and 20 all being part of (or forming) the same device or system. By way of example, a terminal such as a smartphone or other may comprise the processing device T1, at least one sensor 12 and a display device 20a as previously described.

[0111] Figure 2 shows an exemplary embodiment of the system SY1 as described with reference to Figure 1. In the example of Figure 2, a plurality of cameras 12a are arranged around the subject UR1 to capture the posture PS1 of the latter over time. In this case, the processing device T1 obtains or retrieves image data DT1a comprising images representing the subject UR1 over time.

[0112] The processing device T 1 can thus be configured to send at least one of a first posture instruction CMD1a and a second posture instruction CMD1b, or any combination of the different examples of posture instructions envisaged.

[0113] According to one example, the movement assistance device 20d may comprise at least one of the user interfaces 20a-20c. In this case, a first posture instruction CMD1a and a second posture instruction CMD1b may be sent by the processing device T1 to cause both a restitution of information IF1 by the assistance device 20d and the configuration of said assistance device 20d.

[0114] The processing device T 1 can be configured to generate and send one or more posture instructions CMD1 in real time to control the control device(s) 20. Alternatively, the posture instruction(s) CMD1 can be generated and sent during a phase subsequent to that of capturing the posture of the subject UR1.

[0115] It should be noted that the processing device T1 shown in Figure 1 constitutes only an exemplary embodiment, other implementations being possible within the scope of the invention. Those skilled in the art will further understand that certain elements of the processing device T1 are described here only to facilitate understanding of the invention, implementations of the invention being possible without these elements.

[0116] As shown in figure 3 according to a particular embodiment, the processor 4 controlled by the computer program PG1 (figure 1) implements a certain number of modules, namely: an MD2 obtaining module, a first MD4 generation module, a first MD6 determination module, a second MD8 determination module, an MD10 analysis module and a second MD12 generation module.

[0117] Specifically, the MD2 acquisition module can be configured to obtain representative DT1 sensor data from the UR1 subject.

[0118] The first MD4 generation module can be configured to generate, from the DT1 sensor data, an ML1 model defining PT1 anatomical points of the UR1 subject over time.

[0119] The first determination module MD6 can be configured to calculate, from the ML1 model, kinematic variables V1 representative of the evolution of the posture PS1 of the subject UR1 over time.

[0120] The second determination module MD8 can be configured to determine, from the kinematic variables V1, at least one non-linear indicator of variability V2 representative of a temporal structure of the kinematic variables V1.

[0121] The MD10 analysis module can be configured to analyze said at least one non-linear indicator of variability V2 by comparison with a respective reference value REF2.

[0122] The second generation module MD12 can be configured to generate a posture instruction CMD1 based on a result of the analysis carried out by the analysis module MD10. The configuration and operation of the modules MD2-MD12 of the device T1 will appear more precisely in the exemplary embodiments described below.

[0123] The tracking method implemented by the processing device T1 as previously described with reference to Figures 1-2 is now described according to particular embodiment examples in conjunction with Figures 3-18. To do this, the device T1 performs steps S2-S12 (Figure 4) by executing the computer program PG1.

[0124] For example, it is assumed that the processing device T 1 is used to monitor and improve the posture PS1 of the subject UR1, the latter being able to stand on any surface, or possibly, he can use a movement assistance device 20 as appropriate.

[0125] During an obtaining step S2, the device T 1 obtains sensor data DT 1 representative of the subject UR1. This sensor data DT1 may be generated by various types of sensors 12 capable of capturing the posture PS1 of the subject UR1 over time. The sensor data DT 1 may be received from one or more posture sensors 12 or retrieved from a memory, for example the memory 6 (FIG. 1). According to one example, the sensor data DT 1 received by the processing device T 1 comprises image data (or video data) DT1a representative of the subject UR1. As an example, FIG. 5 illustrates the obtaining S2 of image data DT1a representing a subject UR1 performing a squat movement, other postures however being possible. The image data DT1a may comprise images IG1 representing all or part of the subject UR1 over time.These image data DT1a can be generated by at least one image sensor 12a, i.e. at least one image acquisition device such as a camera or equivalent. The processing device T1 can obtain (receive, determine, etc.) these image data DT1a in various ways depending on the case. Obtaining image data DT1a during step S2 is advantageous in that it makes it possible to deduce therefrom an accurate or faithful representation of the posture PS1 of all or part of the body of the subject UR1. From image data DT1a, a model of what is visible in the image can be generated, which provides a more complete representation than that provided by other types of sensor, such as those described below.

[0126] The use of a plurality of cameras 12a (as for example in FIG. 2) advantageously makes it possible to capture the posture PS1 of all or part of the subject UR1 in space over time. It is thus possible to obtain a detailed visual representation and minimize the risks of occulting an area of ​​interest of the body of the subject UR1. By using multiple cameras 12a positioned appropriately in space, it is possible to acquire images (or a video) capturing the part or parts of the subject UR1 that it is desired to follow and analyze, such as for example a joint (knee, ankle, etc.) or a limb (lower limb or upper limb). In particular, it is possible to use a plurality of cameras 12a to make a 3-dimensional (3D) acquisition of the subject UR1.

[0127] According to an example illustrated in Figure 6, the sensor data DT 1 received by the processing device T1 comprises inertial data DT1b representative of the subject UR1 over time. To do this, one or more inertial sensors 12b (Figure 1) can be positioned on the subject UR1 (on his body or his clothing) in order to acquire this inertial data, which is then received and processed by the processing device T 1. This inertial data DT1b can comprise for example acceleration data 30 acquired by at least one accelerometer 12b and / or gyroscopic data 32 acquired by at least one gyroscope 12b. In other words, the inertial data DT1b are for example acquired by means of at least one of an accelerometer-type sensor 12b and a gyroscope-type sensor 12b, said at least one sensor being positioned on the subject UR1.As illustrated in FIG. 6, at least one accelerometer 12b may be used to generate acceleration data 30, for example linear acceleration data representative of an acceleration along one axis or along several axes (for example along two axes X and Y, or along three axes X, Y and Z) as a function of time t. It is possible to use three accelerometers 12b to measure the acceleration in the space of anatomical points PT1 of the subject UR1 over time. Similarly, at least one gyroscope 12b may be used to generate gyroscopic data 32, for example angular velocity data representative of an angular velocity (for example in degrees per second) as a function of time t.

[0128] According to one example, the DT1 sensor data received by the processing device T1 comprises DT1c EMG data (electromyogram data) representative of the subject UR1 over time. More specifically, this DT1c data defines an EMG activity of the subject UR1 over time. To do this, one or more EMG sensors 12c may be positioned on the subject UR1 to acquire this DT1c EMG data, which is then received and processed by the processing device T1.

[0129] Electromyography (EMG) is used to measure the electrical activity of the UR1 subject's muscles. This electrical activity can be captured using electrodes (20c EMG sensors) placed in contact with the subject, for example, surface electrodes placed on the skin or needle electrodes inserted into the subject's muscle tissue. The DT1c EMG data thus collected can be used, in particular, to analyze the electrical response of muscles to nerve stimulation, providing relevant information on the PS1 posture or the state of the UR1 subject, for the monitoring and management of various physical or medical conditions.

[0130] According to one example, the sensor data DT1 obtained (S2, FIG. 4) by the processing device T 1 comprises at least one of the sensor data DT1 a, DT1 b and DT1 c described above, or any combination of at least two of these types of sensor data.

[0131] In one example, DT1b inertial data and / or DT1c EMG data are obtained at S2 (Figure 4) in addition to the DT1a image data to further improve the capture of the PS1 posture of the subject UR1. The DT1b-DT1c sensor data produced by the inertial and EMG sensors 12b-12c may be useful for posture analysis but, unlike the DT1a image data, provide a fragmentary representation of the subject's state in certain localized areas. It may be necessary to position multiple sensors 12b and / or 12c on the subject to obtain a sufficiently complete representation of the area of ​​the body that is desired to be analyzed.

[0132] It is subsequently considered by way of example that image data DT1a are obtained during the obtaining step S2 by means of a plurality of image capturing devices 12a (for example cameras), these data DT1a being subsequently used during the tracking method as described below. During a generation step S4 (FIG. 4), the processing device T1 generates, from the sensor data DT1 obtained in S2, a model ML1 defining anatomical points (or parts) PT1 of the subject UR1 over time. FIGS. 8 and 9 represent examples of models ML1 generated during the generation step S4. According to an example, the image data DT1a produced by each camera 12a comprise images IG1 defining respective 2D views (or representations) of the subject UR1 over time.Thus, during generation S4 (figure 4), the processing device T1 estimates (or predicts) the position PN1 of anatomical points PT1 of the subject UR1 over time in the 2D views (image IG1) from each camera 12a. Each position prediction constitutes a hypothesis of the position at which a respective anatomical point PT1 is located. A 2D model (taking for example the form of a 2D skeleton as illustrated in figures 8-9) can thus be generated from each image IG1 produced by a camera 12a. The device T1 can then generate a 3D model ML1 by triangularizing the estimated positions PN1 of the anatomical points PT1 in each 2D view (each image IG1). In other words, a 3D model is constructed from the 2D captures of each camera 12a. This 3D model allows us to precisely represent the PS1 posture of subject UR1 in space.

[0133] Alternatively, it is possible to generate a two-dimensional or even one-dimensional ML1 model to represent the UR1 subject. A 1D or 2D model allows the UR1 subject to be modeled in only one or two dimensions, which may be sufficient and advantageous in some cases, for example due to the more limited resource requirements than for generating a 3D model.

[0134] The PT1 anatomical points used in S4 to construct the ML1 model can be adapted as appropriate depending on the posture to be analyzed and the desired objective. These PT1 anatomical points can, for example, define joints (elbow, knee, ankle, etc.) and / or characteristic points of the subject's body (for example, elements of the face or different parts of a limb).

[0135] Prediction of PN1 positions of PT1 anatomical points can be performed by running a prediction algorithm from the IG1 images defined in the DT1a image data. This algorithm can use an artificial intelligence-based prediction model to accurately estimate the PN1 position of each PT1 anatomical point of interest.

[0136] The prediction of the PN1 positions can also take into account the relative position of the cameras 12a with respect to each other. A calibration step may be necessary prior to the steps S2-S4 of obtaining to guarantee satisfactory accuracy of the prediction of the PN1 positions, and therefore the generation of a quality ML1 model. It is thus possible to construct in S4 (figure 4) a 3D ML1 model presented for example in the form of a 3D skeleton representative of the PS1 posture of the subject as represented in figures 8-9. This ML1 model can make it possible to model the evolution of the respective PN1 position of the anatomical points PT1 over time.

[0137] According to an example, the device T1 assigns, to each estimated position PN1 of the anatomical points PT1 of the subject UR1 in the 2D views, a respective confidence score SC1 which is representative of a level of confidence in said estimated position PN1. In other words, a respective confidence score SC1 is associated with each position estimate PN1. The model ML1 is then generated (S4, figure 4) taking into account the confidence scores SC1. In particular, the model ML1 can be generated (S4) from the estimated positions PN1 of the anatomical points PT1 from which are excluded, by a first filtering, each estimated position PN1 of an anatomical point PT1 whose confidence score SC1 does not reach at least a first threshold value SL1. In other words, this filtering based on SC1 confidence scores makes it possible to generate a good quality ML1 model using only the estimated PN1 positions whose confidence score reaches at least the first SL1 threshold value.This SL1 value can be adapted on a case-by-case basis. By avoiding taking into account the estimated PN1 positions whose reliability is considered insufficient, it is thus possible to advantageously generate a precise and reliable ML1 model.

[0138] It may be better to reach a compromise by limiting the number of anatomical points in the ML1 model to keep only the most reliable points to ensure good model quality. If a PT1 anatomical point has an insufficient confidence score in all 2D views, for example because it is not visible to any camera, this point is then absent (excluded) from the final 3D model (partial model).

[0139] It may indeed be difficult for a 12a camera to capture all parts of the subject corresponding to the PT1 anatomical points of interest. In some cases, problems with annoying reflection or even limited image quality may make it difficult to detect a PT1 anatomical point from an IG1 image. The greater the number of 12a cameras used, the lower the risk of occultation but the higher the cost in resources and implementation complexity. The prediction of PN1 positions advantageously makes it possible to build a quality ML1 model regardless of the posture of the UR1 subject, including in the case where PT1 anatomical parts of the UR1 subject are occulted or difficult to identify in certain IG1 images. However, the quality of the position predictions varies depending on different factors.Depending on the number and position of the cameras 12a, some PT1 anatomical parts may not be visible in the 2D views produced by the cameras 12a, which may result in more or less reliable estimates of certain anatomical points. Taking into account the SC1 confidence score as previously described makes it possible to generate a complete and good quality ML1 model.

[0140] According to one example, to generate the model ML1 (S4, figure 4), the device T 1 may retain, for a given anatomical point PT1 at a given instant, only a single position estimate PN1 from the 2D view of a respective camera 12a, that is to say the best estimate among those available, in order to further improve the quality of the model ML1. For example, if a first camera 12a visualizes an anatomical point PT1 better than a second camera 12a at a given instant (in the event of occultation for example), it may be preferable to retain only the estimated position PN1 obtained from the image IG1 of the first camera 12a.

[0141] According to an example, during the generation S4 (figure 4), the processing device T1 compares first and second confidence scores SC1 associated respectively with first and second estimated positions PN1 for the same anatomical point PT1 in 2D views from the cameras 12a. The model ML1 is then generated from the estimated positions PN1 of the anatomical points PT1 to which a second filtering is applied according to which, if the difference between said first and second confidence scores SC1 reaches at least a second threshold value SL2, the one among the first and second estimated positions PN1 having the lowest confidence score is excluded (by taking it into account to generate the model ML1). In other words, only the estimated position PN1 with the highest confidence score SC1 is used to generate the model ML1. This second filtering can be repeated over time to improve the quality of the model ML1.

[0142] It is thus possible to compare the respective SC1 confidence scores obtained for PN1 position estimates of the same anatomical point PT1 (representing for example a knee or other) in several 2D views, or in several 2D skeletons from several cameras 12a, and to eliminate the estimated PN1 position with the lowest confidence score, but only in the case of a strong disparity between the respective SC1 confidence scores, which may indicate a significant risk of error. In this way, it is possible to maximize the number of PN1 position estimates taken into account during the S4 generation and thus further improve the quality of the ML1 model. The second threshold value SL1 can for example be defined in the model of the prediction algorithm used to generate the ML1 model.

[0143] During a determination or calculation step S6 (figure 4), the processing device T1 determines (or calculates), from the model ML1 generated in S4, kinematic variables (or data) V1 representative of the evolution of the posture PS1 of the subject UR1 over time. The number and type of the kinematic variables V1 determined in S6 can be adapted as appropriate. Generally, the kinematic variables V1 characterize the dynamics of the posture PS1 of the subject UR1 over time. According to one example, the kinematic variables V1 determined in S6 comprise at least one of speed data and acceleration data calculated from the estimated positions PN1 of the anatomical points PT1 of the model ML1 over time. In other words, the kinematic variables V1 can characterize speeds and / or accelerations of points of the model ML1, for example anatomical points PT1.These can be angular and / or translational speeds and accelerations depending on the case.

[0144] A kinematic variable V1 can thus define, for example, a speed component or an acceleration component in a given direction, for example along an X, Y or Z axis.

[0145] According to one example, the kinematic variables V1 determined in S6 comprise at least any one of the following variables, or a combination of at least two of these variables: a position, an angle, a speed and an acceleration. In other words, the kinematic variables V1 may characterize positions, angles, speeds and / or accelerations of points of the model ML1, for example anatomical points PT1. These may be, for example, angular or translational positions as the case may be, or angular and / or translational speeds and accelerations as the case may be.

[0146] According to one example, at least a portion of the kinematic variables V1 determined in S6 is sensor data DT1 obtained in S2. In other words, it is possible to determine a kinematic variable V1 from a sensor data DT1 without any intermediate processing, using the model ML1 generated in S4. Alternatively, a kinematic variable V1 is determined in S6 by processing from the sensor data DT1 defined in the model ML1.

[0147] As an illustration, Figure 10 schematically represents V1 kinematic variables calculated in S6 over time, these variables representing a displacement (in meters) of a center of mass of the ML1 model in space. In this example, the three curves represent translational movement components of the center of mass in 3 dimensions, namely along X, Y and Z axes respectively (lateral, vertical and anteroposterior plane). In the example considered, the center of mass corresponds to the pelvis of subject UR1 while the latter performs a squat movement. These V1a kinematic variables show in this case that:

[0148] - the movement of the center of mass along Z is negligible over time t which means that the subject UR1 is centered laterally;

[0149] - the center of mass of subject UR1 moves along Y over time t between a reference position “0” and a low position corresponding to - 0.7 meters; and

[0150] - during the squat movement, the center of mass moves back slightly (by 0.1 meters) along X due to the backward displacement of the buttocks of subject UR1 (slight backward movement) then returns to its initial position along X. For illustration purposes, Figure 11 schematically represents a kinematic variable V1 determined in S6 over time, this variable representing an angular displacement (in degrees) along a given direction of a joint, namely a knee of subject UR1. In this example, the curve therefore represents the evolution of an angular movement component as a function of time (according to a percentage of a walking cycle) while subject UR1 performs a walking movement. As this figure shows, the evolution of the knee flexion angle has two peaks during a complete walking cycle. Indeed, at the beginning of the cycle, the heel touches the ground, then the knee passes under the subject's body (1 erangular peak). Following this first flexion, the knee extends into a posterior position (almost zero flexion) before reaching a second, larger angular peak during which the subject's toe leaves the ground and the knee is brought forward until the knee returns to the initial position.

[0151] During a determination step S8 (figure 4), the processing device T1 determines, from the kinematic variables V1 obtained in S6, at least one non-linear variability indicator V2 representative of a temporal structure of the kinematic variables V1 (or of at least one variable V1 among those previously calculated in S6). One or a plurality of these indicators V2 can be determined as the case may be. A non-linear variability indicator V2 is a statistical or mathematical indicator characterizing in a complex manner the manner in which one or more kinematic indicators V1 vary over time. Unlike a linear indicator which would assume proportional and constant relationships, a non-linear variability indicator captures more complex dynamics such as chaos, non-proportional interdependence, or emergent behaviors.These nonlinear indicators may include, for example, sampling entropy, the Lyapunov exponent, fractal dimension, and other methods from chaos theory and dynamical systems. These indicators have been found to be particularly effective for analyzing time series or data where traditional linear models fail to capture the full underlying dynamics, thus providing a deeper understanding of the systems under study.

[0152] In particular, a nonlinear indicator of variability is a mathematical tool that allows revealing complex structures, dynamics, and behaviors within the variability of complex dynamic systems, beyond what linear indicators can capture. It has been observed that this type of indicator is particularly effective for monitoring the behavior of a human body. Indeed, these nonlinear indicators allow analyzing non-proportional dynamic relationships and emergent behaviors, such as the interactions of kinematic variables in complex dynamic systems, such as a moving human body for example. From such indicators, one can advantageously detect patterns and temporal structures in data where linear models are insufficient.

[0153] According to one example, each kinematic variable V1 defines a component of movement (for example a position, speed or acceleration, angular or in translation) in one dimension over time. The device T1 thus determines in S8 (figure 4) at least one non-linear indicator of variability V2 by projection of the kinematic variables V1 into a space of dimensions N, where N >1. From this or these indicators V2, it is advantageous to reconstruct kinematic variables in a space of dimension N (N>1).

[0154] For illustrative purposes, Figure 12 includes different representations C1-C6 of the evolution of signals over time as well as the respective value of a Lyapunov exponent representative of the variability of the signal considered. This example shows several signals with different signal temporal structures (periodic, non-periodic or random). Signals C1, C3 and C5 on the one hand, and signals C2, C4 and C6 on the other hand, vary in the same range of values ​​(30 and 80, respectively) and have the same average (equal to 0) for both groups. However, their temporal structures may or may not differ within the same group, which is reflected in the values ​​of the Lyapunov exponents LyE indicated by the reference sign 34 (varying between -0.001 and 0.564 in this example).For example, although varying in different ranges, signals C1 and C2 have the same Lyapunov exponent value because they are characterized by the same temporal structure (purely periodic signals, LyE = 0). Signals C3 and C4 (chaotic) each have a certain degree of repetition (repetition of a variation pattern), which is reflected by an increase in their Lyapunov exponent LyE values ​​(close to 0.1). Signals C5 and C6 are completely random so that each point in time is uncorrelated with other points, resulting in even higher Lyapunov exponent LyE values ​​(close to 0.5).

[0155] For illustration purposes, Figure 13 shows other examples of signals (A), (B) and (C) (left) representative of kinematic variables V1 in one dimension and respectively periodic, chaotic and random, as well as their corresponding spatial representations in a 3-dimensional space.

[0156] Figure 14 reproduces the right column of Figure 13 in 2 dimensions to place the elements on a plane representing the predictability (x axis) and the complexity (y axis) of the signal. It appears that random signals (C) and periodic signals (A) are respectively little and very predictable although weakly complex, unlike chaotic signals (B) which are complex with variable predictability lying between the first two.

[0157] According to one example, the non-linear variability indicator(s) V2 determined in S8 (figure 4) comprise at least one of: - an entropy value or an entropic indicator;

[0158] - an exponent of Lyapunov;

[0159] - a result of a fractal analysis (for example a fractal analysis value or statistical self-affinity value or a DFA rectified fluctuation analysis in English);

[0160] - a surogate indicator;

[0161] - a geometric approach indicator, for example UCM for “Uncontrolled Manifold”, TNC for “Tolerance-Noise-Covariation”, MIP for “Minimum Intervention Principle” or GEM for “Goal Equivalent Manifold”; and

[0162] - an indicator of recurrent correlations (RQA type for “Recurrence Quantification Analysis”).

[0163] In one example, a plurality of nonlinear indicators among those mentioned in the present disclosure may be combined for use in accordance with the processing method of the invention.

[0164] Each type of nonlinear variability indicator provides a type of information about how V1 kinematic data vary over time. In particular, they can provide information about the regularity or predictability of a movement.

[0165] Note that the UCM (“Uncontrolled Manifold”) indicator is considered a non-linear indicator insofar as it analyzes the variability of the coordination of elements in a multidimensional space, geometrically represented by a manifold. This variability defines a subspace within which variations in the system elements do not significantly affect the overall performance of the task. The UCM approach thus distinguishes two types of variability: (i) variability along tolerance directions, where fluctuations have no significant impact on task performance, reflecting the inherent flexibility of a redundant system, and (ii) variability along error directions, where variations directly influence performance and require corrections to achieve the objective.By integrating the complex and dynamic interactions between the different dimensions of the system, this method advantageously overcomes the limitations of traditional linear approaches. It thus makes it possible to identify non-proportional and emergent relationships that characterize redundant and adaptive motor systems.

[0166] Entropy is a mathematical tool used to quantify the regularity and unpredictability of fluctuations in time series data. In practice, the presence of repetitive patterns of fluctuation in a time series makes it more predictable than a time series in which such patterns are absent. Entropy thus reflects the probability that similar observations will not be followed by other similar observations. A time series containing many repetitive patterns has a relatively low entropy; on the contrary, a less predictable process has a higher entropy. Using an entropy value advantageously allows us to characterize a level of disorganization, or unpredictability, in the variations of the V1 kinematic variables, which advantageously allows us to identify relevant postural evolutions and thus improve the subject's posture analysis.

[0167] For illustrative purposes, Figure 15 represents an example in which an approximate entropy indicator V2-1 (also called ApEn or "approximate entropy" in English) is determined in S8 (Figure 4) as a non-linear indicator of variability V2 from a time series of angular knee flexion V1. As indicated in (A), the time series is first divided into short vectors of similar length m. One of these vectors is represented in Figure 15(A) by the arrow between the points u(44) and u(45). Then, for each vector thus determined, the processing device T 1 determines the number of other vectors which are similar to said vector considered. The vectors are considered to be similar to the original vector when their tails and their tips are contained in a band of width r above and below those of the original vector, here u(44) ± r and u(45) ± r.The vectors that were found to be similar to the original vector are represented by the other arrows. As illustrated in Figure 15(B), this operation can then be repeated for vectors that are one unit longer than the shorter ones (e.g., for a vector spanning between u44 and u46 as illustrated in 15(B)), i.e., of length m+1. The ApEn is then obtained by calculating the natural logarithm of the relative prevalence of repeating motifs of length m compared to those of length m+1.

[0168] The largest Liapunov exponent, also known simply as the "Liapunov exponent," is a measure of the sensitivity to initial conditions in a dynamical system. It can be used to quantify the chaotic nature of a nonlinear dynamical system. It measures how the trajectories of a dynamical system diverge from slightly different initial conditions. In a chaotic system, small initial variations can lead to very different trajectories, which means that the system is sensitive to initial conditions. Using a Liapunov exponent (or function) advantageously allows us to estimate the stability of an equilibrium point, i.e., the stability of the kinematic variables V1.

[0169] To calculate the Liapunov exponent, two initial trajectories that are very close in state space (usually, very close points in the time series) can be selected. These trajectories can then be tracked over time. The Liapunov exponent is then calculated as the limit (e.g., by taking the average over all possible trajectories) of the exponential growth of the deviations between these trajectories. As a result, a positive Liapunov exponent indicates sensitivity to initial conditions and suggests that the system is chaotic. In this case, small initial perturbations can lead to exponential divergences of the trajectories, making long-term prediction difficult. A zero exponent indicates that the trajectories remain close over time, meaning that the system is stable and predictable.Conversely, a negative exponent indicates that the trajectories converge over time, meaning that the system is stable and tends to return to an equilibrium state. The result RS1 determined in S10 (figure 4) by the processing device T 1 can therefore include an indication of the sign of the obtained Liapunov exponent, or an indication of whether the obtained Liapunov exponent is positive (> 0), negative (> 0) or zero (= 0).

[0170] DFA (Detrended Fluctuation Analysis) is a fractal analysis technique used to quantify fluctuations in a time series. This technique involves decomposing a signal into small time scales and then calculating the relationship between fluctuations and the size of the time scales. The DFA alpha coefficient can then be used to assess the presence of fractality in a time series. An alpha coefficient value between 0.5 and 1 indicates positive autocorrelation (or persistence) in the time series, suggesting a fractal structure. The closer the alpha value is to 1, the more the series is considered fractal and therefore exhibits similarities or repeating patterns.In practice, an autocorrelated time series is characterized by a positive correlation between successive values, i.e., high values ​​are likely to be followed by high values ​​and low values ​​by low values. Conversely, an alpha coefficient value between 0 and 0.5 indicates anti-correlation (or anti-persistence) in the time series. This is characterized by an inverse relationship between successive values, i.e., when the preceding value is high, the following value is likely to be low and vice versa. GEM is a method, or indicator, for representing movement in a multidimensional space.This indicator allows, after defining a task objective, to see how one or more kinematic variables are associated with the success of this objective by looking at the "good" variability (beneficial to the success of the task objective) and the "bad" variability (leading to a deterioration in performance).

[0171] For illustrative purposes, Figure 16 shows an example in which a GEM indicator V2-2 is determined in S8 (Figure 4) as a nonlinear indicator of variability V2. Part (A) of Figure 16 shows the GEM indicator for maintaining a constant walking speed (v). The center point represents the preferred average operating point. Each triangle represents a combination of stride time and stride length. Points that lie exactly on the diagonal line of the GEM indicator have reached the same speed and satisfy the objective, namely in this example maintaining a constant walking speed v. Part (B) of Figure 16 shows the time series of the 5T (deviation tangential to the GEM) and 5P (deviation perpendicular to the GEM) deviations for the data set shown in (A).

[0172] During an analysis step S10 (figure 4), the processing device T1 analyzes said at least one non-linear variability indicator V2 (determined in S8) by comparing it with a respective reference value REF2. A result RS1 is produced from this analysis S10.

[0173] According to an example, the device T1 determines a single non-linear indicator of variability V2 in S8 (figure 4) and compares this to a reference value REF2 to determine the result RS1.

[0174] Alternatively, the device T1 determines a plurality of non-linear indicators of variability V2 in S8 (figure 4) and compares each of them to a respective reference value REF2 to obtain the result RS1. The use of a combination of non-linear indicators of variability V2 of different types can for example advantageously make it possible to better characterize the variability of the kinematic values ​​V1 in their complexity and thus improve the result RS1 of the analysis S10. Each non-linear indicator of variability V2 can give a type of information so as to obtain a more complete analysis result covering various aspects of the variability, such as the regularity and predictability of the posture, in view for example of a task objective.

[0175] According to one example, the device T1 verifies (S10, figure 4) whether said at least one non-linear variability indicator V2 determined in S8 meets one (or at least one) conformity criterion CR1 with respect to its respective reference value REF2. For example, it is checked whether a non-linear variability indicator V2 is greater than or less than its respective value REF2 or whether the difference between these two values ​​is greater than or less than a threshold value. The result RS1 determined in S10 is then a function of whether the conformity criterion is met or not.

[0176] According to an example, the verification of the conformity of a non-linear indicator of variability V2 to a plurality of conformity criteria CR1 can be carried out in order to obtain the result RS1. For example, it can be verified whether a Lyapunov exponent representative of the variability respects a plurality of conformity criteria CR1.

[0177] The result RS1 may be a function of whether a plurality of non-linear indicators of variability V2 comply with a respective conformity criterion CR1 with respect to a respective reference value REF2. In other words, the device T1 may carry out successive comparisons over time to verify that the evolution of a non-linear indicator of variability V2 complies with criteria CR1 over time.

[0178] As an example, the device T 1 can check whether a Lyapunov exponent LyE determined in S8 figure 4) as a nonlinear indicator of variability V2 respects the following two conformity criteria CR1: (1) LyE > 0 (which indicates that the movement of the subject UR1 is not purely periodic or stereotyped) and (2) LyE < 0.5 (which indicates that the movement of the subject UR1 is not too random). It has indeed been observed that a too low LyE value suggests that the subject UR1 is not in good condition (injury, pathology, etc.) which induces a stereotyped behavior. Conversely, a too high LyE value is not desirable since the random nature of a movement suggests certain disorders or disorders in the subject UR1.A high degree of variability is necessary for an athlete to adapt to any situation in a sport, for example rugby, but excessive variability can occur in the case of uncontrollable or disorganized movements, which may result from a disorder. By checking the conformity of this Lyapunov exponent LyE to the two criteria above (in relation to the reference values ​​REF2 equal to 0 and 0.5), we can thus advantageously ensure that the subject movement UR1 presents a satisfactory variability.

[0179] According to an example, the processing device T 1 determines in S8 (figure 4) a plurality of non-linear indicators of variability V2, each of them being compared in S10 with a respective reference value REF2 to obtain the result RS1 of said analysis S10. It has notably been observed that the combination of an entropic indicator V2 in combination with a Liapunov exponent LyE makes it possible to effectively analyze the variability due to their complementarity. Indeed, these two indicators have different time series characteristics. For example, a signal may have a high regularity (low entropy) but may have diverse values ​​in terms of sensitivity to the initial conditions (Liapunov exponent).These two indicators can therefore provide useful information on different and particularly complementary aspects (regularity and sensitivity to initial conditions) of the subject's posture, and thus further improve posture analysis.

[0180] According to one example, the processing device T1 determines the GEM indicator (step S8, Figure 4) which gives us access to the time series of deviations tangential and perpendicular to the task objective (as illustrated in Figure 16(B)). It is then possible to apply one or more nonlinear analysis techniques, such as DFA, on these deviation time series. The GEM indicator provides information on the variability that is beneficial or detrimental to the success of the task objective, while the other nonlinear analysis techniques provide relevant information on the temporal structure of the variability. This can further improve the relevance of the posture analysis.

[0181] According to one example, a non-linear indicator V2 determined in step S8 (FIG. 4) comprises a first sequence of values ​​over time. Thus, the device T1 can determine a first sequence of values ​​defining the evolution of a non-linear variability indicator V2 over time. In this case, the device T1 compares, for example, this first sequence of values ​​of said at least one non-linear indicator V2 over time with a second sequence of reference values ​​REF2. Thus, the values ​​of this non-linear indicator V2 over time can be compared to respective reference values ​​REF2. The result RS1 of the analysis S10 can then be determined based on a result of the comparison of the first and second sequences.In other words, by comparing the values ​​of a non-linear variability indicator V2 with respective reference values ​​REF2 over time, it is advantageous to finely analyze the temporal structure of the kinematic data V1, and therefore the way in which they vary over time. During these comparisons, the device T 1 can for example verify the conformity of each value of the first sequence with a respective conformity criterion CR1 with respect to the corresponding reference value REF2.

[0182] For example, the device T 1 can check whether a Lyapunov exponent LyE determined in S8 figure 4) as a non-linear indicator of variability V2 increases over time, or even increases according to a minimal growth rate. It has been observed that an injured person typically exhibits stereotyped behavior with a low level of variability. Thus, one can check for example whether a convalescent patient increases his LyE over time which indicates an increase in variability, synonymous with recovery.

[0183] According to one example, during the tracking method, the processing device T 1 also determines at least one linear indicator V3 of variability from the kinematic variables V1 and compares this linear indicator of variability with a second reference value REF3. The result RS1 of analysis S10 is then also a function of this comparison. Taking into account at least one non-linear indicator V2 combined with at least one linear indicator V3 makes it possible to further improve the quality of the analysis S10. It is thus possible to use linear indicators in addition to non-linear indicators to further improve the posture analysis.

[0184] According to a particular example, at least one non-linear indicator V2 of variability is determined in S8 (figure 4) and a result RS1 is determined in S10 from this or these indicators V2. The processing device T1 can then apply one or more linear analyses to the result RS1 (and vice versa). This result RS1 can then be used during step S12 (figure 4) described below.

[0185] Still with reference to Figure 4, during a generation step S12, the processing device T1 generates a posture instruction CMD1 based on the result RS1 of the analysis S10. This posture instruction CMD1 can then be sent to a control device 20 (Figure 1) to guide the subject UR1 in a posture, a movement or a physical activity. Sending such an instruction UR1 can for example advantageously make it possible to adapt, improve and / or correct the posture UR1 of the subject UR1. The nature and the number of posture instructions CMD1 can vary depending on the case.

[0186] The posture instruction CMD1 is for example intended for the subject UR1 and / or any other user in order to assist the subject in his postures or movements. This instruction can for example be transmitted to the subject UR1 himself or to a professional to help him in his care of the subject UR1 (for example for the development of a training program). The posture instruction CMD1 can for example give an instruction for the subject UR1 to increase or decrease the intensity of a physical exercise, to lengthen or shorten a physical exercise, or to adapt a posture or a movement.

[0187] According to an example, during the generation step S12, the device T1 restores (or renders), as a first posture instruction CMD1a, by means of a user interface, posture information IF1 specifying how to correct or improve the posture of the subject UR1. In other words, the device T1 can send a first posture instruction CMD1a to a user interface to trigger the restitution, in an appropriate form, of posture information IF1 specifying an adaptation, correction or improvement of the posture of the subject UR1. According to a variant, the posture information IF1 specifies a posture, other than the posture PS1, to be performed by the subject UR1, for example to improve or maintain his physical condition.

[0188] As already indicated, the IF1 posture information can be rendered, under the control of the T1 device, in various forms, the manner in which the IF1 information is rendered being dependent on the type of user interface used.

[0189] According to one example, the processing device T 1 sends at least a first posture instruction, more precisely denoted CMD1a (figure 1), to a display device 20a to cause the display of posture information IF1 (in visual form). According to one example, the processing device T 1 can send a first posture instruction CMD1a to a sound device 20b to cause the emission of posture information IF1 in sound form (for example by beeps or via a synthesized voice). According to one example, the processing device T 1 can send a first posture instruction CMD1a to a light device 20c to cause the emission of posture information IF1 in light form (for example by activating a light indicator or by adapting the color of an emitted light). Several user interfaces 20 among those described can also be used in combination.

[0190] For example, the processing device T 1 can send a first instruction command CMD1a to cause the display of information IF1 comprising an indication to correct or improve the posture PS1 of the subject UR1. The information IF1 can for example indicate to the subject UR1 and / or to a third party (a health professional or other) that a part of the body of the subject UR1 must be moved or repositioned in a given direction for correction purposes in order to accomplish a given task (for example a jump, a run, a squat movement, etc.). The information IF1 can for example be displayed on a screen or in holographic form.

[0191] Figure 17 represents a particular example of embodiment in which the device T1 sends (S12, figure 4) at least one posture instruction CMD1a to cause the display of a first posture information IF1 on a screen (or display device) 20a and / or to cause the restitution of a second posture instruction IF1 in the form of light indications by means of light sources 20c arranged around the subject UR1. The screen 20a and / or the light sources 20c can thus transmit instructions to the user UR1 so that he adapts, improves and / or corrects his posture PS1. The reference 24 in figure 18 indicates for example the position of the feet of the subject UR1 in an initial default position.

[0192] Figure 18 shows an exemplary embodiment in which a screen 20a is used to display (S12, Figure 4) posture information IF1 and a treadmill system is configured (S12) according to a configuration CF1 in response to a command CMD1 sent as a posture instruction.

[0193] According to an example, during the generation step S12 (figures 1 and 4), the device T1 sends, as a second posture instruction CMD1b, a command causing a setting (or a configuration) of a movement assistance device 20d (also called a physical conditioning device) to adapt, correct or improve the posture PS1 of the subject. This command can trigger a setting to allow the realization of a posture other than the posture PS1.

[0194] As already indicated, the assistance device 20d may be any device configured, under the control of the processing device T 1 , to facilitate, guide and / or intensify the performance of postures, movements and / or physical activities by a subject or individual (weight training machine, treadmill, elliptical bike, exercise bike, rowing machine, multi-gym station, etc.). The assistance device 20d may be more or less complex depending on the case. In particular, it may be configurable according to at least one operating parameter (for example a running speed of the belt for a treadmill, a resistance force for an exercise bike or an elliptical bike, a duration of a training program, etc.). The processing device T 1 may thus control the configuration of at least one parameter of the assistance device 20d by sending a second posture instruction CMD1b.

[0195] As an example, the device T1 sends in S12 (figures 1 and 4) a posture instruction CMD1b to a device 20d of the treadmill type to adapt an operating parameter such as the speed of movement of the treadmill or an angle of inclination of the treadmill. As already indicated, the device T1 can check in S10 (figure 4) whether a nonlinear indicator of variability V2 determined in S8 complies with a conformity criterion CR1 with respect to its respective reference value REF2. According to an example, if the conformity criterion is not complied with, the posture instruction generated in S12 (figure 4) is configured to correct the posture PS1 of the subject. In other words, the posture instruction PS1 specifies a posture correction to be carried out in the form of posture information IF1 and / or a configuration command CMD1b.

[0196] The manner in which the posture PS1 must be corrected may be determined in various ways by the device T1 depending on the case. According to one example, during the analysis S10 (figure 4), the processing device T1 determines whether a non-linear variability indicator V2, called non-compliant indicator V2a, does not meet a compliance criterion CR1. If so, the device T1 determines at least one kinematic variable V1, called target kinematic variable V1a, associated with the non-compliant indicator V2a and determines a correction to be applied to said at least one target kinematic variable V1a to improve the non-compliant indicator V2a with respect to the compliance criterion CR1 (i.e. with respect to the reference value REF2).In other words, the device T 1 identifies at least one target kinematic variable V1a on which to act to correct the non-compliant indicator V2a as well as a correction to be applied to this target kinematic variable V1a in order to improve the non-compliant indicator V2a according to the compliance criterion CR1 considered. The posture instruction CMD1 generated in S12 (figure 4) can then be a function of the correction to be applied to said at least one target kinematic variable V1a to correct the posture of the subject UR1.

[0197] The non-compliant indicator V2a can in fact be linked to a plurality of kinematic variables V1. It is then necessary to identify the kinematic variable(s) V1 on which it is necessary to act to correct the non-compliant indicator V2a according to the conformity criterion CR1 and thus improve the posture PS1 of the subject UR1.

[0198] According to one example, during the analysis step S10 (figure 4), the processing device T1 determines said at least one target kinematic variable V1a as well as the correction to be applied to it by an optimization method (for example of the “optimal control” type) taking as input:

[0199] - a non-compliant indicator trend objective V2a (by an increase or decrease objective, or an objective of reaching a target value or a range of values);

[0200] - a function defining a relationship of the non-compliant indicator V2a with a plurality of associated kinematic variables V2; and

[0201] - constraint data defining movement constraints that the ML1 model must respect. According to an example, the processing device T 1 determines a GEM indicator as a non-linear indicator V2 of variability as previously described. The trend objective can therefore aim to minimize the “bad” variability (variability deleterious to the success of the task objective) because it is considered too high (non-compliant indicator). The definition of the GEM provides information on the correlation (relation function) of this indicator with several kinematic variables V1, such as the speed and position of anatomical parts of the subject. The constraint data to be respected can include, for example, biomechanical rules, such as angular amplitudes or movement speeds that the human body cannot (or must not) exceed.By combining these three elements in an optimization tool, it is advantageously possible to obtain optimal posture options respecting these criteria and thus to generate a relevant posture setpoint. The optimization method thus used in S10 (figure 4) can deliver as output setpoint data defining said at least one target kinematic variable V1a and the correction to be applied to it to modify the non-compliant indicator V2a according to the trend objective to reach or tend towards the trend objective. The posture setpoint CMD1 can then be generated in S12 (figure 4) from the setpoint data thus determined.

[0202] According to one example, the trend objective provides for example the maximization, or minimization, of the non-compliant indicator V2a or the modification of this indicator so that it is included in a range of values. The function defining a relationship of the non-compliant indicator V2a with a plurality of associated kinematic variables V2 can result from or be deduced from the very definition of the non-compliant indicator V2a. Finally, the constraint data taken into account during the optimization method can vary according to the case: they can for example define movements that certain parts of the body of the subject UR1 or certain of its anatomical points PT1 cannot carry out (for example an angular displacement beyond which a knee cannot move, etc.). It is thus possible to effectively correct a non-compliant indicator V2a and thus advantageously improve a posture or a movement.By correcting a posture to move towards a goal, we can beneficially help the UR1 user in their posture or movement.

[0203] According to an example, prior to the calculation step S6 (figure 4), the processing device T1 performs a recognition, from the ML1 model generated in S4, of a movement performed by the subject UR1. This recognition of the movement can be performed for example by means of a prediction model based on artificial intelligence. The device T1 then selects, depending on the recognized movement, kinematic variables V1 to be calculated during the calculation step S6. Indeed, it may be useful not to calculate the same kinematic variables V1 according to the posture or the movement performed by the user UR1. In other words, the relevant kinematic variables may differ depending on the posture considered. It is thus possible to adapt (or select) the kinematic variables V1 that are calculated in S6 according to the situation in which the subject UR1 is placed.By calculating only the relevant kinematic variables, it is possible to advantageously avoid unnecessary processing, obtain resource savings (time, processing) and improve the performance of the tracking process.

[0204] According to one example, the device T 1 further determines a task objective to be performed and selects the kinematic variables V1 to be calculated based on both the recognized movement and the task objective, in order to further improve the performance of the method.

[0205] In general, the present invention advantageously makes it possible to precisely and reliably monitor and adapt the posture of a subject. In particular, it is possible to evaluate the quality of a posture, static or dynamic, of a subject over time and to guide or improve this posture by generating a posture instruction adapted to the behavior of the subject. Such monitoring makes it possible in particular to evaluate and improve the health of the subject and / or his performance, and this for various purposes depending on the case, such as for example for assistance with physical activity, physical preparation, play, physical rehabilitation and / or re-athleticization. The invention also allows precise and reliable monitoring of the posture, static or dynamic, of a subject and this with limited cost and complexity of implementation. The present invention can also offer diagnostic assistance.

[0206] It has been observed that variability in biological systems, far from being mere noise or error, actually constitutes an important functional property. A certain degree of variability, called "optimal variability," reflects a dynamic balance between stability and adaptability, ensuring flexible and efficient responses to disturbances. This optimal state is characterized by a deterministic chaotic structure, indicating a high capacity for adaptation. Conversely, deviations in this variability—whether marked by excessive rigidity or overly random behavior—can be indicators of pathological dysfunctions, such as motor or cognitive disorders. It has indeed been observed that if a person is injured, blocked, or presents certain disorders or disorders, they may exhibit stereotyped behavior that results in a low level of variability in their kinematic variables.Conversely, excessive variability in kinematic variables suggests other disorders or disturbances that result in disorganized or uncontrollable movements. In theory, a relatively high variability in kinematic variables is desirable to allow the subject to adapt to all situations during physical exercise, but within certain limits in order to maintain good control of movements. However, it has been observed that linear indicators, such as standard deviation for example, have limitations in their ability to analyze and monitor complex dynamic systems, particularly because these indicators are blind to temporal structures. They are inherently incapable of capturing the richness of dynamic behaviors present in such systems. These linear measures are based on the assumption that variations are proportional and additive, which corresponds to a linear model of reality.However, complex dynamic systems, especially those with a deterministic chaotic structure, are characterized by nonlinear relationships between variables and by emergent behaviors that cannot be explained by the simple sum of the parts. For example, the standard deviation, which measures the dispersion of data around a mean, is a linear indicator that ignores temporal correlations or interdependencies between variables. However, it has been observed that these interdependencies are crucial for understanding the dynamics of complex systems, especially those with nonlinear or chaotic behaviors. Moreover, nonlinear systems are characterized by emergent behaviors, i.e., properties that cannot be explained by the simple sum of the parts. These behaviors result from complex and nonlinear interactions between variables.

[0207] The conventional approach is based on a static perspective where each measurement is considered independent. It therefore plans to use exclusively linear indicators to study and monitor such systems, which can lead to an incomplete, or even potentially erroneous or misleading, understanding of the dynamics at play. Consider, for example, an athlete balancing on a soccer ball: his center of pressure exhibits high variability, yet the system is stable. This apparent paradox is explained by the fact that the standard deviation, a linear indicator, does not capture the complexity of the interactions at play. The present invention takes into account these limitations and deficiencies inherent in the conventional approach.

[0208] Unlike the standard linear approach, the present invention is based on a different perspective according to which the measurements are considered as interdependent. By analyzing at least one non-linear indicator of variability based on kinematic variables, it is advantageous to evaluate in detail and reliably the quality of a subject's movement in order to take into account complex aspects not perceptible on the sole basis of kinematic variables. Depending on the result of this analysis, a posture instruction can be advantageously generated, for example to adapt, improve or correct the subject's posture and thus help to improve the subject's health and / or performance.

[0209] The invention therefore offers a powerful tool, both to movement professionals such as health and / or sports professionals (doctors, therapists, physiotherapists, physical trainers, sports coaches / trainers, etc.) and to the general public, to analyze and improve the posture of a subject.

[0210] As understood by those skilled in the art, all the embodiments and variants described above, some of which have been deliberately simplified to facilitate explanations, constitute only non-limiting examples of implementation of the present disclosure. In particular, those skilled in the art may envisage any adaptation or combination of the embodiments and variants described above, in order to meet a particular need.

[0211] The present invention is therefore not limited to the exemplary embodiments described above but extends in particular to a monitoring method which would include secondary steps without thereby departing from the scope of the present invention. The same would apply to a processing device, or more generally to a processing system, for implementing such a method.

Claims

CLAIMS 1. Processing method, implemented by a processing device (T1), for tracking the posture of a subject (UR1), said method comprising: a) obtaining (S2) sensor data (DT1) representative of the subject; b) generating (S4), from the sensor data, a model (ML1) defining anatomical points (PT1) of the subject over time; c) determining (S6), from the model, kinematic variables (V1) representative of the evolution of the posture of the subject over time; d) determining (S8), from the kinematic variables, at least one non-linear variability indicator (V2) representative of a temporal structure of the kinematic variables (V1); e) analyzing (S10) said at least one non-linear variability indicator by comparison with a respective reference value (REF2); and f) generating (S12) a posture setpoint (CMD1) as a function of a result of the analysis.

2. The method of claim 1, wherein the sensor data obtained in a) comprises image data (DT1a) captured by a plurality of image capture devices (12a), the image data from each image capture device defining a respective 2D view of the subject over time; generating b) comprising: - estimation of a position (PN1) of the anatomical points (PT1) of the subject over time in the 2D view from each image acquisition device; and - generation of the 3D model by triangularization of the estimated positions of the anatomical points in each 2D view.

3. Method according to claim 2, in which generation b) comprises: - attribution, to each estimated position (PN1) of the anatomical points of the subject in the 2D views, of a respective confidence score (SC1) representative of a level of confidence in said estimated position; the model (ML1) being generated from the estimated positions of the anatomical points among which are excluded, by a first filtering, each estimated position of an anatomical point whose confidence score does not reach at least a first threshold value (SL1).

4. Method according to claim 3, in which the generation b) of the model comprises: - comparison of first and second confidence scores associated respectively with first and second estimated positions for the same anatomical point in 2D views from the image acquisition devices; the model being generated from the estimated positions (PN1) of the anatomical points (PT1) to which a second filtering is applied according to which, if the difference between said first and second confidence scores reaches at least a second threshold value, the one among the first and second estimated positions having the lowest confidence score is excluded.

5. Method according to any one of the preceding claims, wherein said at least one non-linear indicator (V2) determined in d) comprises a first sequence of values ​​over time; the analysis e) comprising: - comparison of the first sequence of values ​​of said at least one non-linear indicator over time with respectively a second sequence of reference values; the result of the analysis e) being a function of a result of said comparison of the first and second sequences.

6. Method according to any one of the preceding claims, in which generation f) comprises at least one of: - restitution, as a first posture instruction (CMD1a), by means of a user interface (20a-20c), of posture information (IF1) specifying how to correct the subject's posture; and - sending, as a second posture instruction (CMD1b), a command causing a setting of a movement assistance device (20d) to correct the subject's posture.

7. Method according to any one of the preceding claims, in which the analysis e) comprises: - checking whether said at least one non-linear variability indicator (V2) meets a conformity criterion (CR1) with respect to the respective reference value; wherein, if the conformity criterion is not met, the posture instruction generated in f) is configured to correct the posture of the subject.

8. Method according to claim 7 in which, if a non-linear indicator of variability (V2a) - called non-compliant indicator - does not respect the conformity criterion (CR1), the analysis e) comprises: - determination of at least one kinematic variable (V1a), called target kinematic variable, associated with the non-compliant indicator (V2a) and of a correction to be applied to said at least one target kinematic variable to improve the non-compliant indicator compared to the reference value according to the conformity criterion; said posture instruction being generated in f) as a function of the correction to be applied to said at least one target kinematic variable to correct the posture of the subject.

9. Method according to claim 8, in which the determination of said at least one target kinematic variable (V1a) and of the correction to be applied is carried out by an optimization method taking as input: - a non-compliant indicator trend objective; - a function defining a relationship of the non-compliant indicator with a plurality of associated kinematic variables; and - constraint data defining movement constraints that the model must respect; the optimization method delivering as output setpoint data defining said at least one target kinematic variable and the correction to be applied to modify the non-compliant indicator according to the trend objective; the posture setpoint being generated in f) from the setpoint data.

10. Method according to any one of the preceding claims, in which the method comprises, prior to calculation c): - recognition, from the model, of a movement made by the subject; and - selection, depending on the recognized movement, of the kinematic variables to be calculated in c).

11. Method according to any one of the preceding claims, wherein said at least one non-linear indicator of variability (V2) comprises at least any one of: an entropy value or an entropic indicator; a Liapunov exponent; a result of a fractal analysis; a superrogation indicator; a geometric approach indicator; and a recurring correlation indicator.

12. Computer program (PG1) comprising instructions for executing the steps of a processing method according to any one of the preceding claims when said program is executed by a computer.

13. Treatment device (T1) for assisting a subject (UR1) in performing a posture (PS1), said device comprising: - obtaining module (MD2) configured to obtain sensor data (DT1) representative of the subject; - first generation module (MD4) configured to generate, from the sensor data, a model (ML1) defining anatomical points (PT1) of the subject over time; - first determination module (MD6) configured to calculate, from the model, kinematic variables (V1) representative of the evolution of the subject's posture over time; - second determination module (MD8) configured to determine, from the kinematic variables, at least one non-linear variability indicator (V2) representative of a temporal structure of the kinematic variables; - analysis module (MD10) configured to analyze said at least one non-linear indicator of variability by comparison with a respective reference value (REF2); and - second generation module (MD12) configured to generate a posture instruction (CMD1) based on a result (RS1) of the analysis.

Images