FR3109081A1 - POSTURE EVALUATION METHOD, ASSOCIATED EVALUATION DEVICE - Google Patents

Abstract

The invention relates to a method of evaluating the posture of a user by a device comprising an accelerometer, the method comprising: A step of establishing scores of inclination and bending carried out at a current instant then at an immediately successive instant from the amplitudes of inclination and bending resulting from an acceleration measured by the accelerometer, A step of calculating the difference between the scores at l 'instant immediately successive at the current instant and at the current instant, If at least one of the differences between the scores is greater than a threshold, a step for establishing a score from the differences between the scores A step d assessment of the user's posture by classification of the established score. Figure to be published with the abstract: Fig. 5a

Description

POSTURE EVALUATION METHOD, ASSOCIATED EVALUATION SYSTEM

TECHNICAL FIELD OF THE INVENTION

The invention relates to the field of detecting and monitoring the posture of a user.

The invention relates more particularly to a method for evaluating the posture of a user over a time interval.

TECHNOLOGICAL BACKGROUND OF THE INVENTION

Today, many people suffer from back pain, in particular due to poor posture adopted in both static and dynamic positions. Poor posture is defined in relation to the spine, also called the spine, seen in profile which has the shape of an elongated "S".

illustrates the so-called physiological curvatures that condition the balance of the spine.

With reference to the , the CV vertebral column comprises three curvatures: a first in the cervical section CB1 called cervical lordosis, a second in the dorsal section CB2 called dorsal kyphosis and a third in the lumbar section CB3 called lumbar lordosis. The alternation of curvatures is manifested at the level of the various hinges of the CV spine: the cervico-dorsal hinge CCD, ie C7-D1, the dorso-lumbar hinge CDL, ie D12-L1, and the lumbosacral hinge CLS, ie L5-S1.

In poor posture, the CV spine is deformed so that it has curvatures different from the CB1, CB2, and CB3 curvatures shown in Fig. . For example, hyperkyphosis is an accentuation of the CB1 curvature backwards. Hyperlordosis is a forward accentuation of the CB2 curvature.

There are devices that can detect different positions and/or movements of an individual and emit a signal when the latter adopts a bad posture.

For example, patent application US 5,749,838 describes a posture detection device which can be attached to clothing or even to a belt of the user. The detection device comprises three lobes intended to be in contact with three determined areas of the user's back. Pressure is exerted on one of the three lobes as the curvature of the user's spine increases. The device emits a vibration when the pressure exerted on one of the lobes reaches a threshold value so as to inform the user that he has adopted a bad posture.

Furthermore, patent application US 2013207889 also describes a device for detecting the posture of a user. In particular, the detection device comprises an accelerometer suitable for determining the level of inclination of the user's pelvis as well as a microprocessor which analyzes the data from the accelerometer in real time so as to evaluate the posture of the user . When the posture of the user is considered to be bad, a signal, in the form of a vibration or a sound, is emitted by the device or by external means able to communicate with the detection device.

However, such detection devices do not make it possible to obtain results that are sufficiently reliable and sufficiently representative of the reality of the user's posture because they emit a signal as soon as the user's spine becomes deformed. Thus, as soon as the user changes posture by deforming his spine, for example by performing an action such as picking up an object, the detection devices of the prior art trigger an alarm. These nuisance alarms are not necessarily representative of the real risk incurred by the user.

There is therefore a need for a more reliable system for evaluating a user's posture.

In this context, the invention aims to solve the aforementioned problems by proposing a solution making it possible to obtain results that are more representative of the reality of a user's posture.

• Etablissement d’un premier score d’inclinaison et d’un premier score de flexion à un instant courant tc en fonction respectivement d’une amplitude d’inclinaison courante et d’une amplitude de flexion courante, l’amplitude d’inclinaison courante et l’amplitude de flexion courante étant calculées à partir d’une accélération courante mesurée par l’accéléromètre du dispositif d’évaluation à l’instant tc,
• Etablissement d’un deuxième score d’inclinaison et d’un deuxième score de flexion à un instant tc+1 immédiatement successif à l’instant courant tc en fonction respectivement d’une deuxième amplitude d’inclinaison et d’une deuxième amplitude de flexion, la deuxième amplitude d’inclinaison et la deuxième amplitude de flexion étant calculées à partir d’une deuxième accélération mesurée par l’accéléromètre du dispositif d’évaluation à l’instant tc+1 immédiatement successif à l’instant courant tc,
• Calcul de la différence entre le deuxième score d’inclinaison et le premier score d’inclinaison courante,
• Calcul de la différence entre le deuxième score de flexion et le premier score de flexion courante,
• Si la différence entre le premier score d’inclinaison courante et le deuxième score d’inclinaison est inférieure à un seuil d’inclinaison et si la différence entre le premier score de flexion courante et le deuxième score de flexion est inférieure à un seuil de flexion, l’étape d’établissement des scores d’inclinaison et de flexion à un instant tc+2 immédiatement suivant l’instant tc+1 est réitérée,
• Si la différence entre le premier score d’inclinaison courante et le deuxième score d’inclinaison est supérieure au seuil d’inclinaison et/ou si la différence entre le premier score de flexion courante et le deuxième score de flexion est supérieure au seuil de flexion, alors une étape d’évaluation de la posture de l’utilisateur à l’instant tc+1 immédiatement successif à l’instant courant tc est réalisée et comprend :
  • Une sous-étape d’établissement d’une note à partir des scores d’inclinaison et de flexion à l’instant tc+1 immédiatement successif à l’instant courant tc,
  • Une sous-étape de classification de la note établie dans une classe d’évaluation de la posture de l’utilisateur, la classe étant choisie en fonction de la valeur de la note.

The invention according to a first aspect relates to a method for evaluating the posture of a user by an evaluation device, the evaluation device comprising at least one accelerometer, the evaluation method being characterized in that it includes at least the steps of:

• Establishment of a first inclination score and a first flexion score at a current instant tc as a function respectively of a current amplitude of inclination and of a current amplitude of flexion, the current amplitude of inclination and the current flexion amplitude being calculated from a current acceleration measured by the accelerometer of the evaluation device at time tc,
• Establishment of a second inclination score and a second flexion score at an instant tc+1 immediately following the current instant tc as a function respectively of a second amplitude of inclination and of a second amplitude of flexion , the second amplitude of inclination and the second amplitude of flexion being calculated from a second acceleration measured by the accelerometer of the evaluation device at instant tc+1 immediately following current instant tc,
• Calculation of the difference between the second inclination score and the first current inclination score,
• Calculation of the difference between the second bending score and the first current bending score,
• If the difference between the first running lean score and the second lean score is less than a lean threshold and if the difference between the first running lean score and the second lean score is less than a lean threshold , the step of establishing the inclination and flexion scores at a time tc+2 immediately following the time tc+1 is repeated,
• If the difference between the first current lean score and the second lean score is greater than the lean threshold and/or if the difference between the first current bend score and the second bend score is greater than the bend threshold , then a step of evaluating the posture of the user at the instant tc+1 immediately following the current instant tc is carried out and comprises:
  • A sub-step of establishing a score from the inclination and bending scores at the instant tc+1 immediately following the current instant tc,
  • A classification sub-step of the rating established in a class of evaluation of the posture of the user, the class being chosen according to the value of the rating.

Thanks to the invention, the evaluation of the posture of a user at a given moment depends on the posture of the user at the previous moment. Thus, a posture is evaluated as being "bad" if it lasts over time and a deformity of short duration is not systematically considered as having caused a bad posture.

• le score de flexion est établi en utilisant un modèle d’apprentissage automatique d’établissement de score de flexion mettant en relation l’amplitude d’inclinaison courante et l’amplitude de flexion courante avec le score de flexion et en ce que le score d’inclinaison est établi en utilisant un modèle d’apprentissage automatique d’établissement de score d’inclinaison mettant en relation l’amplitude d’inclinaison courante et l’amplitude de flexion courante avec le score d’inclinaison.
• la note est établie en utilisant un modèle d’apprentissage automatique d’établissement de note mettant en relation les scores de flexion et d’inclinaison avec la note.
• la note est classifiée en mettant en relation la note avec une échelle de note ayant été utilisée pour l’apprentissage du modèle d’apprentissage automatique d’établissement de note.
• la méthode d’évaluation comprend une étape de détection d’un changement durable de posture, un changement durable de posture étant détecté lorsque :
  • la différence entre le deuxième score d’inclinaison à l’instant tc+1 et le premier score d’inclinaison courant à l’instant courant tc calculée à la sous-étape de calcul est supérieure au seuil d’inclinaison et/ou la différence entre le deuxième score de flexion à l’instant tc+1 et le premier score de flexion courant à l’instant courant tc calculée à l’étape de calcul est supérieure au seuil de flexion et que,
  • pour un instant tc+2 immédiatement successif à l’instant tc+1, la différence entre la différence entre un troisième score d’inclinaison établi à l’instant tc+2 et le deuxième score d’inclinaison établi à l’instant tc+1 est inférieure à un seuil d’inclinaison de maintien de posture ou la différence entre un troisième score de flexion établi à l’instant tc+2 et le deuxième score de flexion établi à l’instant tc+1 est inférieure à un seuil de flexion de maintien de posture.
• l’étape de détection d’un changement durable de posture est réalisée avant l’étape d’évaluation de la posture de l’utilisateur et en ce que l’étape d’évaluation de la posture de l’utilisateur n’est réalisée que lorsqu’un changement durable de posture est détecté à l’étape de détection d’un changement durable de posture.
• la méthode d’évaluation comporte une étape d’émission d’un signal d’inactivité si la différence entre le premier score d’inclinaison courante et le deuxième score d’inclinaison est inférieure à un seuil d’inactivité d’inclinaison et si la différence entre le premier score de flexion courante et le deuxième score de flexion est inférieure à un seuil d’inactivité de flexion, pendant un nombre d’instants successifs à l’instant courant tc, le nombre d’instants successifs étant supérieur à un seuil de durée d’inactivité.
• la méthode d’évaluation comporte une étape d’émission d’un signal de risque si la valeur absolue de la dérivée temporelle de l’amplitude de flexion courante est supérieure à un seuil de valeur pendant un temps supérieur à un seuil de durée.

In addition to the characteristics which have just been mentioned in the previous paragraph, the method according to a first aspect of the invention may have one or more additional characteristics among the following, considered individually or according to all technically possible combinations:

• the bend score is established using a bend scoring machine learning model relating the current tilt amount and the current bend amount to the bend score and that the score d The tilt is established using a machine learning tilt scoring model relating the current tilt amount and the current flexion amount to the tilt score.
• the score is established using a machine learning scoring model relating the bend and tilt scores to the score.
• the rating is classified by relating the rating to a rating scale that was used to train the rating machine learning model.
• the evaluation method comprises a step of detecting a lasting change in posture, a lasting change in posture being detected when:
  • the difference between the second inclination score at time tc+1 and the first current inclination score at current time tc calculated at the calculation sub-step is greater than the inclination threshold and/or the difference between the second bending score at time tc+1 and the first current bending score at current time tc calculated at the calculation step is greater than the bending threshold and that,
  • for an instant tc+2 immediately following instant tc+1, the difference between the difference between a third inclination score established at instant tc+2 and the second inclination score established at instant tc+ 1 is less than a posture maintenance inclination threshold or the difference between a third flexion score established at time tc+2 and the second flexion score established at time tc+1 is less than a threshold of posture holding flexion.
• the step of detecting a lasting change in posture is carried out before the step of evaluating the posture of the user and in that the step of evaluating the posture of the user is only carried out when a lasting change of posture is detected in the step of detecting a lasting change of posture.
• the evaluation method comprises a step of emitting an inactivity signal if the difference between the first current inclination score and the second inclination score is less than an inclination inactivity threshold and if the difference between the first current flexion score and the second flexion score is less than a flexion inactivity threshold, for a number of successive instants at the current instant tc, the number of successive instants being greater than a threshold duration of inactivity.
• the evaluation method comprises a step of emitting a risk signal if the absolute value of the time derivative of the current bending amplitude is greater than a value threshold for a time greater than a duration threshold.

Another aspect of the invention relates to a device for evaluating the posture of a user, comprising means configured to implement the evaluation method according to the invention.

Another aspect of the invention relates to a computer program product comprising instructions which, when the program is executed by a computer, lead the latter to implement the steps of the evaluation method according to the invention.

Another aspect of the invention relates to a computer-readable recording medium comprising instructions which, when executed by a computer, lead the latter to implement the steps of the evaluation method according to the invention. .

The invention and its various applications will be better understood on reading the following description and examining the accompanying figures.

BRIEF DESCRIPTION OF FIGURES

• à la figure 1, un schéma illustrant une vue latérale d’une colonne vertébrale présentant des courbures physiologiques ;
• à la figure 2, un schéma illustrant les plans de référence dans un corps humain ;
• à la figure 3, un schéma illustrant un dispositif d’évaluation de la posture d’un utilisateur selon un mode de réalisation de l’invention ;
• à la figure 4, un schéma illustrant le dispositif d’évaluation de la figure 3 lorsqu’il est fixé au niveau d’une zone déterminée de l’utilisateur ;
• aux figures 5a à 5e, des schémas en blocs illustrant les étapes d’une méthode d’évaluation de la posture d’un utilisateur selon un mode de réalisation de l’invention ;
• à la figure 6, un schéma en blocs illustrant la détection d’un changement de posture durable selon un mode de réalisation de l’invention ;
• aux figures 7a et 7b, des schémas illustrant la détection d’un changement de posture durable selon un mode de réalisation de l’invention,
• à la figure 8, un graphique représentant la détection d’une inactivité selon un mode de réalisation de l’invention ;
• à la figure 9, un organigramme illustrant la détection d’un geste à risque selon un mode de réalisation de l’invention.

Other characteristics and advantages of the invention will become apparent on reading the following description, with reference to the appended figures, which illustrate:

• in FIG. 1, a diagram illustrating a side view of a vertebral column exhibiting physiological curvatures;
• in FIG. 2, a diagram illustrating the reference planes in a human body;
• in FIG. 3, a diagram illustrating a device for evaluating the posture of a user according to one embodiment of the invention;
• in FIG. 4, a diagram illustrating the evaluation device of FIG. 3 when it is fixed at the level of a determined zone of the user;
• in FIGS. 5a to 5e, block diagrams illustrating the steps of a method for evaluating the posture of a user according to one embodiment of the invention;
• in FIG. 6, a block diagram illustrating the detection of a lasting change in posture according to an embodiment of the invention;
• in FIGS. 7a and 7b, diagrams illustrating the detection of a lasting change in posture according to one embodiment of the invention,
• in FIG. 8, a graph representing the detection of inactivity according to one embodiment of the invention;
• in FIG. 9, a flowchart illustrating the detection of a risky gesture according to one embodiment of the invention.

DETAILED DESCRIPTION

The present invention relates to a method and a device for evaluating the posture of a user.

In the rest of the description, it is recalled that when an action is assigned to a device, it is in fact performed by the microprocessor of the device controlled by instruction codes recorded in the memory of the device.

illustrates the reference planes in a human body.

With reference to the , the human body comprises three planes which pass through the center of gravity of the human body and which are perpendicular to each other.

The sagittal plane PS is a vertical plane that passes through the midline of the body and divides it into two symmetrical parts, right and left. The frontal plane PF is a vertical plane perpendicular to the sagittal plane PS which divides the body into two symmetrical parts, anterior (ventral) and posterior (dorsal). The transverse plane PT is a horizontal plane, parallel to the ground, which divides the body into two symmetrical parts, upper (on the side of the head) and lower (on the side of the feet).

In the invention, the posture of the user is evaluated on the basis of his inclination in the frontal plane PF and his flexion/extension in the sagittal plane PS.

In the remainder of the description, the term “flexion” will be used to designate both flexion and extension.

illustrates the evaluation device according to one embodiment of the invention.

As can be seen on the , the evaluation device 10 comprises various components grouped together in a housing 11 intended to be fixed on the back of the UTL user. Advantageously, the evaluation device 10 is removably attached to the back of the UTL user by bonding means 12, for example an adhesive layer adapted to stick to the skin of the UTL user. In a variant embodiment not shown, the evaluation device 10 is attached to clothing, a belt or even a bag.

illustrates the positioning of the evaluation device on a user's back.

The evaluation device 10 is fixed at the level of a determined zone Z of the back of the user UTL. Advantageously, the determined zone Z is at the level of the upper part of the back of the user UTL, between his spine and one of his shoulders, as illustrated in the . Such an area is easily accessible by the UTL user and allows accurate results to be obtained.

• un accéléromètre 12,
• un microprocesseur 14,
• des moyens de communication 15 sans fil,
• des moyens d’alimentation 16,
• un compteur de temps 17,
• une mémoire 18.

Furthermore, inside the box 11, the evaluation device 10 comprises:

• an accelerometer 12,
• a microprocessor 14,
• wireless communication means 15,
• supply means 16,
• a time counter 17,
• a memory 18.

Furthermore, the evaluation device 10 can include a gyrometer (not shown).

The accelerometer 12 is configured to measure the acceleration of the user UTL along three axes x, y, z, for example at a frequency of 26 Hz, then to transmit the measurements to the microprocessor 14. In an alternative embodiment not shown, the evaluation device 10 comprises at least two accelerometers 12.

The microprocessor 14 is configured to process the data from the accelerometer 12, for example at a frequency of 26Hz, in order to evaluate the posture of the user UTL.

The wireless communication means 15 ensure the transmission of the data processed by the microprocessor 14 to a remote terminal 19, for example a smartphone, a computer or even a television set. Advantageously, the means of communication 15 are formed by a Bluetooth interface, a Bluetooth Low Energy interface, a Wifi interface or another wireless network interface known to those skilled in the art.

The supply means 16, for example a battery, supply energy to the evaluation device 10.

The memory 18 is suitable for storing the data from the time counter 17, the data from the accelerometer 12, one or more physiological characteristics of the user UTL as well as a library of data relating to different postures.

is a block diagram illustrating the steps of the evaluation method 100 according to one embodiment of the invention.

In a calibration step 101, transformation matrices are calculated in order to replace the future signals measured by the evaluation device 10 in a fixed frame of reference. This fixed frame of reference, for example centered around the center of gravity of the user UTL, is the one that will be used subsequently in the evaluation method 100 for the acceleration measurements along the x, y, and z axes. An example of a fixed frame of reference is shown in And .

In an initialization step 102 carried out at an initial instant and represented at , initial tilt and flexion scores are established. Advantageously, the initialization step 102 comprises sub-steps 1020 to 1022.

In a sub-step 1020, an initial acceleration is measured by the accelerometer 12. Advantageously, the initial acceleration ALi is measured along the x, y and z axes shown in . Thus, ALxi corresponds to the linear acceleration along the x axis, ALyi corresponds to the linear acceleration along the y axis and ALzi corresponds to the linear acceleration along the z axis. The initial acceleration ALi = (ALxi, ALyi, ALzi) is then stored in the memory 18.

In a sub-step 1021, an initial inclination amplitude AIi and an initial flexion amplitude AFi are determined by the microprocessor 14. In particular, the initial inclination amplitude AIi and the initial flexion amplitude AFi are determined at from the initial acceleration measured during sub-step 1020. For example, the initial tilt amplitude and the initial flexion amplitude are determined using the Tilt/Twist method described in the article by Neil R. Crawford et al, “A new technique for determining 3-D joint angles: the tilt/twist method” (1999). Thus, the initial inclination amplitude AIi, is calculated using the following mathematical formula:

The initial flexion amplitude AFi is calculated using the following mathematical formula: With And

The values of initial inclination amplitude AIi and initial flexion amplitude AFi thus calculated are angular values corresponding to the angular deviation, respectively, in the frontal plane PF and in the sagittal plane PS expressed in radians with respect to the initial posture of the UTL user.

In a sub-step 1022, an initial inclination score SIi and an initial flexion score SFi are calculated, by the microprocessor 14, from, respectively, the initial inclination amplitude AIi and the flexion amplitude initial AFi measured.

The scores are preferably calculated using a polynomial formula resulting from the training of a machine learning model. This machine learning model, for example a Lasso regression model, may have been trained using supervised or unsupervised learning. For example, in the case of supervised learning, the model was trained with either tilt data (for initial tilt score SIi) or flex data (for initial flex score SFi) annotated. That is to say that the data included a label (also called “label”) comprising a value or an indication of the value that the score corresponding to a given amplitude must take. For example, it may be desired that the score have a value of 10 when the initial inclination amplitude along the x axis AIxi is 0.1 radians and 0 radians along the other axes. These values are indicative and at the choice of the operator training the model. The model thus trained with several correspondences of inclination amplitudes AI and inclination scores SI makes it possible to obtain a polynomial formula for calculating the inclination score SI as a function of the inclination amplitudes AI along each axis. So, for example, the initial inclination score SIi is calculated using the following polynomial formula:

Similarly, the initial bending score SFi, can be calculated using the following polynomial formula:

As indicated previously, the parametric values, i.e. the polynomial coefficients, used in the polynomial formulas above come from the training of a Lasso regression model.

In a step 103, the posture of the UTL user is evaluated. This step 103 comprises sub-steps 1030 to 1032 and is shown in . In sub-step 1030, a score N is established from the initial inclination SIi and flexion SFi scores. For example, the grade is established using the following polynomial formula:

As stated earlier, the parametric values, i.e. the polynomial coefficients, used in the polynomial formula above are derived from training a machine learning model for determining an N score, based on the Bending scores SFi and Tilt SIi. Thus, the model may have been trained so that the N score takes on values from 0 to 100 on a score scale, and for too much inclination to significantly increase the N score. Similarly, the model may have been trained to that an excessive SFi initial flexion score significantly increases the N score. Such learning makes it possible to link the SFi flexion and SIi inclination scores into a single N score, subsequently making it possible to indicate good or bad posture of the UTL user. The model may also have been trained so that the score N takes on values of with respect to some reference value.

Note N is then stored in memory 18.

In sub-step 1031, the posture of the user UTL at the initial instant ti is evaluated. The UTL user posture assessment includes relating the N rating to the rating scale that was used to train the rating learning model. Thus, the sub-step 1031 for evaluating the posture of the UTL user comprises comparing the score N with one or more thresholds included in the score scale and/or reference values, in such a way that the score N can be classified and thus determine a "quality" of the posture of the user UTL. For example, the N rating can be low on the rating scale, for example close to 0, for example between 0 and 30, thus indicating a “good posture” of the UTL user. The N rating can be high on the rating scale, for example close to 100, for example between 70 and 100, thus indicating “poor posture” of the UTL user. The N rating can also have an average value on the rating scale, for example between 30 and 70, indicating an "average posture", for example a posture not to be maintained for too long. It is clear to those skilled in the art that any other value scale and/or value thresholds can be used to qualify the score N and the posture of the UTL user.

The score N can also depend on one or more physiological characteristics of the user UTL, for example by weighting the score N by a coefficient according to the age, sex, or any other physiological characteristic of the user UTL.

In the sub-step 1032, the evaluation of the posture of the user UTL obtained during the sub-step 1031 is transmitted to the remote terminal 19 by the means of communication 15 wirelessly. This wireless transmission can be carried out using any known wireless protocol, such as one of the Bluetooth®, Wi-Fi®, LoRa®, SigFox® etc. protocols. Advantageously, the posture of the user UTL appears, on the display means of the remote terminal 19, as "good", "bad" or by any other indication of the evaluation of the posture by the user UTL according to the evaluation carried out in the previous sub-step 1031 .

In a step 104 shown in , the scores are recalculated at a time tc after the initial time ti, hereinafter called the current time tc, and compared with the initial scores. Advantageously, the step of calculating the current scores and the difference with the previous scores 104 comprises the following sub-steps 1040 to 1043.

In a sub-step 1040, a current acceleration is measured by the accelerometer 12. Advantageously, the current acceleration ALc is measured along the x, y and z axes. Thus, ALxc corresponds to the linear acceleration along the x axis, ALyc corresponds to the linear acceleration along the y axis and ALzc corresponds to the linear acceleration along the z axis. The current acceleration ALc = (ALxc, ALyc, ALzc) is then stored in memory 18.

In a sub-step 1041, a current inclination amplitude AIc and a current flexion amplitude AFc are determined by the microprocessor 14. As for the initial inclination and flexion amplitudes AIi and AFi in step 1021, the current tilt amplitude AIc and current flexion amplitude AFc are determined using the Tilt/Twist method.

Thus, the current inclination amplitude AIc, is calculated using the following mathematical formula:

The current flexion amplitude AFc is calculated using the following mathematical formula: With And

The current inclination amplitude values AIc and current flexion amplitude AFc thus calculated are angular values corresponding to the angular deviation, respectively, in the frontal plane PF and in the sagittal plane PS expressed in radians with respect to the initial posture of the UTL user.

In a sub-step 1042, and in the same way as in step 1022, a current inclination score SIc and a current flexion score SFc are calculated, by the microprocessor 14, from, respectively, the measured current inclination amplitude AIc and current flexion amplitude AFc.

The current inclination score SIc is calculated using a polynomial formula resulting from the training of a machine learning model as described in sub-step 1022. The current inclination score SIc can thus be calculated using the following polynomial formula:

Similarly, the current flexion score SFc, can be calculated using the following polynomial formula:

As indicated previously, the parametric values, i.e. the polynomial coefficients, used in the polynomial formulas above come from the training of a machine learning model, for example a Lasso regression model.

In a sub-step 1043, the difference between the current inclination Sic and flexion SFc scores calculated at the instant tc and the inclination and flexion scores calculated at the directly preceding instant is calculated by the microprocessor 14.

When the current time is a time directly following the initial time ti, i.e. when no bending and inclination score other than the inclination SIi and bending SFi scores have been calculated, the difference between the current inclination Sic and flexion SFc scores and the initial inclination SIi and initial flexion SFi scores is calculated by the microprocessor 14. By “difference” is meant the following calculation:

∆SI = SIc – SIi

∆SF = SFc – SFi

When the current instant is an instant not directly following the initial instant ti, that is to say when at least one other bending score and another inclination score other than the initial inclination scores SIi and initial flexion SFi were calculated, the difference between the current inclination SIc and flexion SFc scores and the current inclination scores at the previous time tc-1 SIc-1 and current flexion at the previous time tc-1 SFc-1 is calculated by the microprocessor 14. By “difference” is meant the following calculation:

∆SI = SIc – SIc-1

∆SF = SFc – SFc-1

Thus, if the current instant tc is the instant immediately following the initial instant, that is to say the instant t2, the sub-step 1043 consists in calculating the difference between the current scores SIc2 and SFc2 and the initial SIc1 and SFC1 scores.

If the current instant tc is instant t5, sub-step 1043 consists in calculating the difference between the current scores SIC5 and SFC5 and the previous current scores SIC4 and SFC4.

If |∆SI| is less than an inclination threshold and |∆SF| is less than a bending threshold, step 104 is repeated. The new scores SIc+1 and SFc+1 calculated at time tc+1 are then compared to the scores SIc and SFc calculated at time tc.

If, on the other hand, |∆SI| is greater than the inclination threshold and/or |∆SF| is greater than a bending threshold, then the method 100 continues at step 105 of recalculating the note.

In a step 105 shown in , the score N is recalculated according to the difference between the scores calculated in sub-step 1043. The recalculation of the score N is performed by the processor 14 in a sub-step 1050 according to the method described in step 103. Thus, the new score N is calculated from the current inclination SIc and current bending SFc scores. For example, the score is established using the following polynomial formula, as described in step 103:

As described in step 103, a polynomial formula for calculating the score N is derived from training a machine learning model.

The note N is then stored in the memory 18. It is possible to store all the calculated notes N together rather than overwriting the note N with each new calculation. This makes it possible to have access to the evolution of the N scores and therefore of the postures of the UTL user.

In sub-step 1051, the posture of the user UTL at the current instant tc is evaluated. The UTL user posture assessment includes relating the N rating to the rating scale that was used to train the rating learning model. Thus, the sub-step 1051 for evaluating the posture of the user UTL comprises the classification of the score N in a class representing the “quality” of the posture of the user UTL. This classification can be carried out for example by comparing the score N with one or more thresholds included in the score scale, so that the score N can be classified and thus determine a "quality" of the posture of the user UTL , using the same scale as in step 103 for evaluating the initial posture. In addition, the classification can be carried out using a classifier such as a trained automatic learning model, for example of the SVM (Support Vector Machine) type.

The score N can also depend on one or more physiological characteristics of the user UTL, for example by weighting the score N by a coefficient according to the age, sex, or any other physiological characteristic of the user UTL.

At sub-step 1052, the evaluation of the posture of the user UTL obtained during sub-step 1051 is transmitted to the remote terminal 19 by the means of communication 15 wirelessly. This wireless transmission can be carried out using any known wireless protocol, such as one of the Bluetooth®, Wi-Fi®, LoRa®, SigFox® etc. protocols. Advantageously, the posture of the user UTL appears, on display means of the remote terminal 19, as "good", "bad" or by any other indication of the evaluation of the posture by the user UTL.

In an optional step 106, at least the last score N calculated and/or the last evaluation of the posture of the user UTL can be transmitted to the remote terminal 19 by the wireless communication means 15 in order to display, by the terminal remote 19, a graphical representation of the score N and/or the last evaluation of the posture of the user UTL. In an alternative embodiment, all of the scores N calculated over a given period and/or all of the evaluations of the posture of the user UTL over a given period can be transmitted to the remote terminal 19 by the means of communication 15 wirelessly in order to display, via the remote terminal 19, a graphical representation of the scores N and/or the evaluations of the posture of the user UTL over the given period.

Change of posture sustainable

In one embodiment, the method 100 includes a step 107 of detecting a lasting change in posture, as shown in .

A lasting change in posture is defined as a significant change in posture and then maintaining the new posture for a fixed period of time.

First, a posture hold duration counter to determine if a new posture has been held long enough to consider that there is a lasting posture change is initialized. For example, the counter can be initialized to 0.

THE And illustrate score calculations and counter update, respectively. The counter, in the is first initialized to 0.

Secondly, it is checked, at a given current instant tc, that the difference |∆SI| calculated at sub-step 1043 is greater than an inclination threshold SEI1 and/or the difference |∆SF| calculated in step 104 is greater than a bending threshold SEF1. Thus, it is detected that the user UTL has made a significant change in posture, because the value of at least one of the current scores SIc or SFc is far from the value of the corresponding score SIc-1 or SFc-1 at the instant previous tc-1. For example at the a change in posture ΔS is represented between the score Sc-1 and the score Sc. It will be considered hereafter that this change ΔS is greater than the threshold of change in posture SE1.

If the difference |∆SI| between the current score SIc and the score SIc-1 calculated at the instant tc-1 preceding the current instant tc is not greater than the posture change threshold SEI1, and that the difference |∆SF| between the current score SFc and the score SFc-1 calculated at the instant tc-1 preceding the current instant tc is not greater than the posture change threshold SEI1, the score calculation step is performed again and difference with the previous scores 104, because this indicates that there is no significant difference between the previous and current scores and therefore that a change in posture has not taken place.

Then, when at least one of the differences |∆SI| and |∆SF| is greater than its respective threshold SEI1 or SEF1, i.e. when a change in posture is detected, it is checked that the posture is maintained.

An action is performed at an instant "immediately successive" to a previous instant, when the action was performed at the previous instant and no same action has taken place since the previous instant.

For this, it is checked, for the current time tc+1 immediately following the given time tc, that the difference between the current score Sc+1 and the score Sc is less than a posture maintenance threshold SE2, the difference having been calculated during a second score calculation step 104 following the first score calculation step 104. A preferred embodiment uses two different posture maintenance thresholds SE2 for each of the inclination SIc and flexion SFc scores . For example, if the difference |∆SI| between the inclination scores was greater than the SEI1 threshold over the interval [tc; tc-1], we check that the difference |∆SI| between the inclination scores is less than an SEI2 inclination bound over the interval [tc+1; ct]. In another example, if the difference |∆SF| between the flexion scores was greater than the SEF1 threshold over the interval [tc; tc-1], we check that the difference |∆SF| between the flexion scores is less than a flexion limit SEF2 over the interval [tc+1; ct]. Finally, in yet another example, if the difference |∆SI| between the inclination scores was greater than the SEI1 threshold over the interval [tc; tc-1] and the difference |∆SF| between the flexion scores was greater than the SEF1 threshold over the interval [tc; tc-1], we check that the difference |∆SI| between the inclination scores is less than an SEI2 inclination bound over the interval [tc+1; tc] and that the difference |∆SF| between the flexion scores is less than a flexion limit SEF2 over the interval [tc+1; tc] Thus, it is detected that the posture has been maintained, that is to say that the current score Sc+1 is not too far from the previous score Sc, during the time interval [tc+1; ct].

If the two aforementioned conditions for comparing the differences between the scores with respectively the threshold SE1 and the limit SE2 are verified over the time interval [tc+1; tc], the time counter C is incremented, for example by 1 if the steps 104 for calculating the scores are carried out every second. For example at the , the threshold SE2 is represented by the two lines having the most dotted lines, and at the instant tc+1 the score Sc+1 has not left the bounds formed by SE2. Thus, it is considered that the posture is maintained for a unit of time and the counter represented in is incremented by 1 at tc+1.

If the difference between the current score Sc+1 and the score Sc is not less than the terminal SE2, the time counter is reset, for example to 0 seconds, and a score calculation step 104 is performed again. Indeed, if the posture was not maintained, the posture hold duration counter must be reset. This is represented at the at time tc+3, the difference between the score Sc+3 and the score Sc+2 being greater than the threshold SE2 at the , the counter is reset to 0 at the .

If the value of the time counter C is less than a posture maintenance duration threshold SD, for example set at 10 seconds, the second step of calculating the scores 104 is carried out for the instant tc+2 immediately following the instant tc+1 immediately following the given current instant tc.

If the value of the time counter C is greater than the posture maintenance duration threshold SD, we conclude that there is a lasting change in posture of the user UTL. Indeed, this then indicates that the posture has been maintained for at least a determined time. When a change in posture is detected, the initialization step 102 is performed to determine a new initial score.

This embodiment makes it possible to save calculations of score N, the calculation of score N being carried out only when the posture is maintained for more than a certain time.

Inactivity

is a flowchart illustrating the determination of UTL user inactivity. Inactivity is defined as the absence of a change in posture for a determined time.

In order to detect inactivity of the user UTL, initially, an inactivity duration counter D is initialized, in order to determine if the immobility has been long enough to be considered as inactivity. For example, it can be initialized to 0 seconds.

For each of the Flexion SFc and inclination SIc scores, it is then checked that the difference between the current score Sc and the score Sc-1 calculated at the instant tc-1 preceding the current instant tc is less than a threshold d inactivity SE4 (respectively SEi4 for the inclination and SEF4 for the flexion), the difference having been calculated during a step 104 for calculating the scores and the difference with the preceding scores. If this condition is not verified, the time counter D is reset, for example to a value of 0 seconds, then the step of calculating the scores and the difference with the previous scores is performed again. Indeed, a large difference between Sc and Sc-1 indicates movement of the UTL user, which causes the idle time to reset.

If the aforementioned condition is verified, the time counter D is incremented by 1.

If the value of the time counter D is less than an inactivity duration threshold SE5, the step of calculating the scores and the difference with the previous scores is performed again for the instant tc+1 immediately following the current instant tc. Thus, the time during which the UTL user has remained inactive is counted.

If the value of the time counter D is greater than the inactivity duration threshold SE5, a state of inactivity of the user UTL is concluded. Advantageously, the microprocessor 14 transmits this information to the remote terminal 19 by the means of communication 15 wirelessly. An inactivity signal is then sent at a step 106 by the remote terminal 19 in order to inform the user 17 that he has been inactive for a long time.

In a variant, the device 10 uses the accelerometer 12 and the microprocessor 14 to count the steps of the user UTL and detect that the user UTL is inactive.

In another embodiment, the assessment device 10 includes the detection of inactivity associated with poor posture. The detection of inactivity is then carried out in the same way as described above, or in the variant described above. The N score is also calculated several times over the period of inactivity. When inactivity is detected for a time greater than the inactivity duration threshold SE5, and the calculated score is greater than a bad posture threshold SE6 during this time, then it is considered that a bad posture is maintained for a too long time. This inactivity duration threshold SE5 can for example be set at 45 minutes and the score threshold SE6 can for example be set at 70 when the score scale goes from 0 to 100. Advantageously, the microprocessor 14 transmits this information to the remote terminal 19 by means of communication 15 wirelessly. An inactivity signal is then transmitted at a step 106 by the remote terminal 19 in order to inform the user 17 that he has been inactive for a long time in a bad posture.

Risky gesture

is a graph representing the detection of a risky gesture for the UTL user. A risky gesture is defined as a major movement over a short period of time.

To determine a risky gesture of the user UTL, the current drift of the current flexion amplitude AFc is calculated as follows:

With dT the time elapsed between the current instant and the instant preceding the current instant (i.e. 1/26Hz=38 ms when the measurement frequency of the accelerometer 12 is 26 Hz), AFc the current flexion amplitude at the current instant and AFc-1 the amplitude of bending at the instant preceding the current instant.

If the absolute value of the current derivative |dAFc| of the current flexion amplitude AFc is greater than a derivative threshold SE7 for a duration greater than a time threshold SE8. The derivative threshold SE7 can for example be fixed at a value of 110 and the time threshold SE8 can for example be fixed at 270 ms. When the absolute value of the current derivative |dAFc| is greater than the value threshold SE7 for a duration greater than the time threshold SE8, it is determined that a risky gesture has been performed.

In order to avoid the detection of several risky gestures during the same acquisition, it is verified that between two detections of risky gestures a time greater than a detection duration threshold SE9 has elapsed. For example, the detection duration threshold SE9 can be set at 10 seconds, which means that 10 seconds must elapse between the detection of two risky gestures.

For example, at the a graph of the flexion amplitude and its derivative represents the detection of a risky gesture by the user UTL.

The graph represents the flexion amplitude AF as a function of time as well as its derivative dAF as a function of time. The value threshold SE7 is also represented. When the absolute value of the derivative dAF exceeds the threshold SE7, at the time mark 100, the time t during which the threshold of value SE7 is counted down. In the case shown in , the value threshold SE7 is exceeded for a time t equivalent to 11 time markers. In the case represented, the time threshold SE8 is for example 270 ms and the measurement frequency of the accelerometer is for example 26 Hz. Thus, the value threshold SE7 is exceeded during . This duration t is greater than the time threshold SE8 (423ms>270ms). Thus, a risky gesture is detected. At time marker 112, another risky gesture is performed. However, this risky gesture is performed in a time following the first detection that is less than the detection duration threshold SE9 (which is for example 10 seconds). Thus, this second risky gesture is not detected. This makes it possible to avoid detecting several risky gestures during the same acquisition.

When a significant difference has been detected between the scores at one instant and at the previous instant and this difference is small at the following instants, it is concluded that the UTL user has performed a risky gesture. Advantageously, the microprocessor 14 then transmits this information to the remote terminal 19 by the means of communication 15 wirelessly. A risk signal is then emitted 106 by the remote terminal 19 in order to inform the user 17 that he has performed a risky gesture.

Recommendations of exercises on remote terminal

In an advantageous embodiment, the remote terminal 19 recommends to the UTL user via the display on the display means of the remote terminal 19 the exercises to be carried out according to his current posture and/or his past postures. Preferably, the recommendations are based on the average values of the inclination and/or flexion scores and the average values of scores N passed over a previous time interval. For example, the exercise recommendations can be based on the inclination SIc and flexion SFc scores and/or on the N scores allowing the evaluation of the posture of the user. An exercise can thus consist in carrying out a physical action when the average of the inclination scores SI indicates for example a straight inclination and the average of the flexion scores is of degree 2.

Claims (11)

Method for evaluating (100) the posture of a user (UTL) by an evaluation device 10, the evaluation device 10 comprising at least one accelerometer (12), the evaluation method (100) being characterized in that it includes at least the steps of:

• Establishment (1042) of a first inclination score (SIc) and a first flexion score (SFc) at a current time tc as a function respectively of a current inclination amplitude (AIc) and of an amplitude of current flexion (AFc), the current amplitude of inclination (AIc) and the current amplitude of flexion (AFc) being calculated from a current acceleration (ALc) measured by the accelerometer (12) of the device evaluation (10) at time tc,
• Establishment (1042) of a second inclination score (SIc+1) and of a second flexion score (SFc+1) at a time tc+1 immediately following the current time tc according respectively to a second amplitude of inclination (AIc+1) and a second amplitude of flexion (AFc+1), the second amplitude of inclination (AIc+1) and the second amplitude of flexion (AFc+1) being calculated from a second acceleration (ALc+1) measured by the accelerometer (12) of the evaluation device (10) at instant tc+1 immediately following current instant tc,
• Calculation (1043) of the difference (|∆SI|) between the second inclination score (SIc+1) and the first current inclination score (SIc),
• Calculation (1043) of the difference (|∆SF|) between the second flexion score (SFc+1) and the first current flexion score (SFc),
• If the difference (|∆SI|) between the first current inclination score (SIc) and the second inclination score (SIc+1) is less than an inclination threshold (SEI1) and if the difference (|∆ SF|) between the first current flexion score (SFc) and the second flexion score (SFc+1) is less than a flexion threshold (SEF1), the step of establishing (1042) the inclination scores and of bending (Slc+2, SFc+2) at a time tc+2 immediately following time tc+1 is reiterated,

• If the difference (|∆SI|) between the first current inclination score (SIc) and the second inclination score (SIc+1) is greater than the inclination threshold (SEI1) and/or if the difference (| ∆SF|) between the first current flexion score (SFc) and the second flexion score (SFc+1) is greater than the flexion threshold (SEF1), then a posture evaluation step (105) of the user (UTL) at time tc+1 immediately following the current time tc is performed and includes:
  • A sub-step of establishing (1050) a score (N) from the inclination and flexion scores (Slc+1, SFc+1) at the instant tc+1 immediately following the current instant ct,
  • A sub-step of classifying (1051) the score (N) established in a class of evaluation of the posture of the user (UTL), the class being chosen according to the value of the score (N).

Method for evaluating (100) the posture according to the preceding claim, characterized in that the flexion score (SFc) is established by using an automatic learning model for establishing the flexion score relating the amplitude of current inclination (AIc) and the current flexion amplitude (AFc) with the flexion score (SFc) and in that the inclination score (SIc) is established using a machine learning model of establishment of inclination score relating the current inclination amplitude (AIc) and the current flexion amplitude (AFc) with the inclination score (SIc). Posture assessment method (100) according to any one of the preceding claims, characterized in that the score (N) is established by using a machine learning model for establishing a score relating the scores of flexion (SFc) and inclination (SIc) with the note (N). Method for evaluating (100) the posture according to the preceding claim, characterized in that the score (N) is classified by relating the score (N) to a score scale having been used for learning the model of machine learning of note setting. Evaluation method (100) according to any one of the preceding claims, characterized in that it comprises a step of detecting a lasting change in posture (107), a lasting change in posture being detected when:

• the difference (|∆SI) between the second inclination score (SIc+1) at time tc+1 and the first current inclination score (SIc) at current time tc calculated at the sub-step of calculation (1043) is greater than the inclination threshold (SEI1) and/or the difference (|∆SF|) between the second flexion score (SFc+1) at time tc+1 and the current first flexion score (SFc) at the current instant tc calculated in the calculation step (1043) is greater than the bending threshold (SEF1) and that,
• for an instant tc+2 immediately following instant tc+1, the difference (|∆SI) between a third inclination score (SIc+2) established at instant tc+2 and the second inclination score (SIc+1) established at time tc+1 is less than a posture maintenance inclination threshold (SEI2) or the difference (|∆SF|) between a third flexion score (SFc+2) established at time tc+2 and the second flexion score (SFc+1) established at time tc+1 is less than a posture maintenance flexion threshold (SEF2).

Evaluation method (100) according to the preceding claim, characterized in that the step of detecting a lasting change in posture (107) is carried out before the step of evaluating the posture (105) of the user (UTL) and in that the step of evaluating the posture (105) of the user (UTL) is only carried out when a lasting change in posture is detected at the step of detecting a change posture (107). Evaluation method (100) according to any one of the preceding claims, characterized in that it comprises a step of emitting an inactivity signal (106) if the difference (|∆SI|) between the first current tilt score (SIc) and the second tilt score (SIc+1) is below a tilt inactivity threshold (SEI4) and if the difference (|∆SF|) between the first flexion score (SFc) and the second flexion score (SFc+1) is less than a flexion inactivity threshold (SEF4), for a number of successive instants at the current instant tc, the number of successive instants being greater than an inactivity duration threshold (SE5). Evaluation method (100) according to one of the preceding claims, characterized in that it comprises a step of emitting a risk signal (106) if the absolute value of the time derivative (|dAFc|) of the current flexion amplitude (AFc) is greater than a value threshold (SE7) for a time greater than a duration threshold (SE8). Device for evaluating (10) the posture of a user (UTL), characterized in that it comprises means (12, 13, 14) configured to implement the evaluation method (100) according to any of the preceding claims. Computer program product comprising instructions which, when the program is executed by a computer, lead the latter to implement the steps of the evaluation method (100) according to any one of Claims 1 to 8. A computer-readable recording medium comprising instructions which, when executed by a computer, cause it to perform the steps of the evaluation method (100) according to any one of claims 1 to 8 .