Movement mode recognition via motion sensor

Abstract

The invention relates to a method for recognizing an individual's mode of movement. The method comprises obtaining (P6) a first structured set of correlations between reference data extracted from a signal of historical interest from sensors worn by the individual, the reference data being pre-classified by possible mode of movement of the individual. The method further comprises recognizing (P7) a common mode of movement of the individual.Recognition (P7) relies on: - structuring a second set of correlations between current data extracted from a current signal of interest, originating from the sensors worn by the individual, the current data being indicative of the individual's current movement, - coupling the first set with the second set, and - inferring an indicative value of the individual's current movement mode during the current time interval based on the second set coupled with the first set. Abstract figure: Figure 1.

Description

Title of the invention: Movement mode recognition by motion sensor technical field

[0001] This disclosure falls within the field of data science. More specifically, this disclosure relates to a method for recognizing an individual's mode of movement and to a corresponding computer program and recording medium. Previous technique

[0002] In the context of digital health services for vulnerable people, for example the elderly, people with disabilities, or isolated people, a permanent challenge is to assess, automatically if possible, their well-being and health.

[0003] In this context, it appears necessary to have a good understanding of the modes of locomotion and transport that these individuals use on a daily basis, in order to provide a robust measure of autonomy. This measure is a key indicator for assessing the physical, social, and mental well-being of individuals by the medical community, family, and the individual themselves during remote monitoring, remote assistance, or remote rescue services.

[0004] Indeed, the use and frequency of use of modes of locomotion and transport can reveal variations and describe a person's habits and degree of autonomy. It is obvious that the absence of use of means of transport or, conversely, regular use of transport are indicators of a person's well-being, and these characteristics can influence their health.

[0005] Thus, to analyze modes of transport, an analysis system can focus on the movements made by the person themselves, but also on the vibrations or accelerations that the person undergoes in a transport situation, for example under the effect of a propulsion or traction mode of a vehicle.

[0006] A method for analyzing modes of locomotion or transport is known and described in M. Leodolter, P. Widhalm, C. Plant and N. Brandie, "Semi-supervised segmentation of accelerometer time series for transport mode classification", 2017 5th IEEE International Conference on Models and Technologies for Intelligent Transportation Systems (MT-ITS), Naples, 2017, pp. 663-668, doi: 10.1109 / MTITS.2017.8005596. In this document, the authors propose a method for recognizing modes of transport (walking, car, bus, train) based on accelerometer data. Their approach proposes automatic semi-supervised segmentation. Signals are grouped, or "clustered," using an algorithm called DTW, for "Dynamic Time Warping." However, such an algorithm requires comparing all signals in the knowledge base with each signal to be classified, which is very costly in terms of time and computing power; therefore, this algorithm is not suitable for a rapid analysis of a movement situation.

[0007] A method for analyzing modes of locomotion is also known and described in Taimoor Afzal, Gannon White, Andrew B. Wright and Kamran Iqbal - Locomotion Mode Identification for Lower Limbs using Neuromuscular and Joint Kinematic Signs, 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2014, doi:10.1109 / EMBC.2014.6944518. In this document, the authors propose a comparative study to identify different modes of locomotion (standing position, walking, ascending stairs, descending stairs, ascending ramp, descending ramp) with electromyographic signals (i.e. obtained by analysis of the muscles and the nerve cells that control them), accelerometric signals (i.e. representative of body movements) and goniometric signals (i.e. obtained by angular measurements at the level of the hip).These three sources of information are coupled and classified using a classification method similar to linear discriminant analysis. However, such an approach requires the use of multimodal information, which necessitates processing data from different capture devices. Furthermore, linear discriminant analysis relies on numerous characteristic variables, which implies the implementation of extensive signal processing.

[0008] Furthermore, a method for analyzing modes of locomotion and transportation is known and described in A. Jahangiri and HA Rakha, "Applying Machine Learning Techniques to Transportation Mode Recognition Using Mobile Phone Sensor Data," in IEEE Transactions on Intelligent Transportation Systems, vol. 16, no. 5, pp. 2406-2417, Oct. 2015, doi: 10.1109 / TITS.2015.2405759. The authors use machine learning techniques to recognize modes of transportation (bicycle, car, walking, running, bus) from an accelerometer, a gyroscope, and a rotation sensor. This is a traditional signal processing approach for describing singularities, selecting them, and classifying them. 165 features are calculated from the captured signals, including the mean, energy, and spectral entropy. Summary

[0009] This disclosure improves the situation.

[0010] A method is proposed, implemented by a data processing circuit, for recognizing the movement pattern of an individual, the method comprising: - obtaining a first structured set of temporal and multimodal correlations between reference digital data extracted from a temporal signal multidimensional data of historical interest, derived from sensors worn by the individual, with the reference digital data pre-classified by possible mode of individual movement, and - recognition of an individual's mode of movement over a current time interval, the recognition being based on: — a structuring of a second set of temporal and multimodal correlations between current numerical data extracted from a multidimensional temporal signal of current interest, originating from said sensors worn by the individual, the current numerical data being indicative of a movement of the individual during the current time interval, — a coupling of the first set of correlations with the second set of correlations, and — an inference of an indicative value of the individual's mode of movement during the current time interval based on the second set of correlations coupled with the first set of correlations.

[0011] By "mode of movement of the individual", we mean a mode of locomotion or a mode of transport.

[0012] Locomotion refers to the movement of an individual primarily by their own means, that is to say, by their muscular energy. Examples include walking, running, cycling, or skiing.

[0013] Transport refers to the movement of an individual primarily using a form of energy other than the individual's own muscular energy. Examples include road, rail, air, and sea transport.

[0014] It is considered that certain modes of locomotion or transport are previously defined and identified as "possible modes of movement of the individual".

[0015] The proposed method makes it possible to robustly recognize an individual's mode of locomotion or transport at a given moment, even if the number of sensors that can be used for this purpose is limited. For example, the inventors obtained good results using only four wearable sensors: an accelerometer, a gyroscope, a magnetometer, and a barometer. The robustness of the recognition is inherent in the coupling between two sets of correlations, one relating to reference data and the other to current data.

[0016] The proposed method also has the advantages of being inexpensive in terms of resources and computation time, and is capable of evolving to take into account new modes of transport or locomotion without requiring complete relearning. This results from the fact that the recognition of a mode of locomotion or transport is done by reference to situations already produced and not by exhaustive and massive learning of all existing modes of locomotion and transport.

[0017] The proposed method has the additional advantage of being both locally and customized in its implementation. Indeed, both the reference digital data and the current digital data relate to the movements of a given individual, originate from sensors worn by that individual, and can also be processed exclusively locally for the purpose of recognizing the current movement pattern of that individual.

[0018] According to another aspect, a computer program is proposed comprising instructions for implementing all or part of a process as defined herein when this program is executed by a processor. According to another aspect, a non-transient, computer-readable recording medium is proposed on which such a program is recorded.

[0019] The features described in the following paragraphs may optionally be implemented independently of each other or in combination with each other:

[0020] In some examples, the method further includes a determination, on the basis of the second set of correlations coupled with the first set of correlations, of a degree of similarity between the current numerical data and the reference numerical data, and the inference of the indicative value of the mode of movement of the individual during the current time interval is based on said degree of similarity.

[0021] Thus, it is possible, for example, to compare current digital data with different reference digital data, each indicative of a given mode of transportation, such as walking, scootering, driving, etc. The individual's current mode of transportation can thus be identified, or recognized, as being analogous to the mode of transportation corresponding to the most similar reference digital data. It is therefore possible to identify any known current mode of transportation, such as walking in this example. It is also possible to identify previously unknown current modes of transportation—that is, modes of transportation to which no reference digital data is associated—by measuring their similarity to already known modes of transportation.Thus, for example, current numerical data associated with an individual's bus trip may be identified as having more similarities with reference numerical data associated with an individual's car trip than with other available reference numerical data. Conversely, due to this coupling, the dynamic processing of reference numerical data associated with a car trip can evolve to better recognize future bus trips by the individual.

[0022] In some examples, the method further includes an association of a measure confidence in recognizing the mode of movement, a comparison of the confidence measure with a first threshold and a second threshold, and if the result of the comparison indicates that the confidence measure is between the first threshold and the second threshold, an alert to the individual in order to obtain, by interaction with a human-machine interface, an identification of the individual's mode of movement and to update the first set of correlations according to the identified mode of movement.

[0023] The confidence measure can, for example, be correlated with the degree of similarity, and / or be based on a degree of reliability of the reference numerical data or current numerical data.

[0024] Such a double thresholding makes it possible to obtain confirmation of the individual only in case of reasonable doubt about the recognized mode of movement, in order to definitively identify the current mode of movement and to improve in the future the robustness of the recognition process while requiring the least possible effort from the individual.

[0025] In some examples, the process further includes, if the result of the comparison indicates that the confidence measure is both above the first threshold and the second threshold, a presentation to the individual of a service recommendation adapted to the recognized mode of travel.

[0026] In some examples, the method further includes, if the result of the comparison indicates that the confidence measure is between the first threshold and the second threshold and after obtaining the identification of the individual's mode of travel by interaction with the human-machine interface, a presentation to the individual of a service recommendation adapted to the identified mode of travel.

[0027] Thus, after recognizing or obtaining confirmation of an individual's usual mode of transportation, it is possible to recommend one or more services adapted to that mode. For example, certain modes of locomotion or transport require the individual's constant attention. To this end, upon recognition, for example, of car travel as a usual mode of transportation, it is possible to suggest restricting the functions of a terminal, such as switching to hands-free mode.

[0028] In some examples, the service recommendation is based on a model of individual preferences.

[0029] For example, an individual may wish to keep a history of their walking activities, possibly including the number of steps or distance covered, as well as a separate history of their running or cycling activities. Thus, upon detection of one of the corresponding modes of locomotion, it is possible to signal the start and type of activity in progress and to offer the individual the option of recording the data relating to that ongoing activity.

[0030] In some examples, the process further includes an evaluation of the individual's response to the presented service recommendation and an update of the individual's preference model based on the evaluation. This allows for the personalization of service recommendations.

[0031] In some examples, the method further includes implementing the recommended service upon receipt of an acceptance signal resulting from an interaction between the individual and a human-machine interface. Alternatively, the recommended service can, for example, be automatically implemented after the expiration of a predefined delay.

[0032] In some examples, the method further includes a pre-classification of historical signals of interest by possible mode of movement of the individual, the historical signals of interest being derived from a history of digital information from sensors of equipment worn by the individual, the digital information describing movements caused or undergone by said sensors. Brief description of the drawings

[0033] Other features, details and advantages will become apparent from reading the detailed description below and from analyzing the accompanying drawings, in which: Fig. 1

[0034] [fig.l] shows an example of a general algorithm of a computer program for the implementation of the proposed process, according to one embodiment. Fig. 2

[0035] [fig-2] shows the structure of a model based on two recurrent neural networks Siamese doors, according to a method of implementation. Fig. 3

[0036] [fig.3] shows a processing circuit for implementing the proposed process, according to a particular embodiment. Description of the implementation methods

[0037] The present invention aims to address the aforementioned drawbacks and to provide a robust prediction of modes of locomotion (e.g., walking, running, roller skating, snowshoeing) and transport (e.g., by scooter, bicycle, skateboard, vehicle, train, sea, river, air, space).

[0038] To this end, a predictive model is established by learning, structuring temporal and multimodal correlations between signals of interest from sensors.

[0039] The predictive model, which may be based, for example, on recurrent neural networks, consists of two coupled dynamic systems. Such a model makes it possible, from raw signals of interest from said sensors at a current instant, to recognize an associated mode of movement of the individual. The recognition of a mode The prediction of locomotion or transportation is based on similarity with previously learned situations. Such a predictive model represents an optimization in terms of cost and computation time compared to known methods relying on exhaustive and massive training of all known modes of locomotion and transportation. Furthermore, if an individual uses a given mode of transportation for the first time, for example, a Segway, the predictive model is still able to identify this new mode of transportation by analogy with a similar situation already learned, such as another mode of transportation, for example, a scooter.

[0040] Thus, it is possible to provide a cohort of individuals with a default version of the predictive model, initially capable of recognizing a small number of generic modes of locomotion and transport. Then, subject to specific authorization granted by an individual, the model can evolve locally, based on data from the sensors of one or more connected objects worn by that individual, so as to progressively recognize additional modes of locomotion and transport commonly used by that individual.

[0041] Learning is optionally adaptive through interaction with the individual. In general, such personalization through supervision of learning has the advantage of improving the robustness of subsequent recognition of an individual's mode of locomotion or transport.

[0042] The proposed predictive model offers robust prediction even if the signals of interest come from a limited list of sensors, for example, a list including an accelerometer, a gyroscope, a magnetometer, and a barometer. In one implementation example, such sensors are integrated into a single connected object worn by an individual, such as a smartwatch.

[0043] The recognized mode of locomotion or transport can then be used as a situational reference to propose or recommend a service, for example a so-called "connected health" service, to the individual. The design of the end-to-end model, that is, from raw data to the promotion of a suitable service, constitutes a further point of innovation.

[0044] This invention makes it possible to power many applications and services that can be linked to connected health for monitoring vulnerable people or monitoring residents in their use of the smart home or smart city.

[0045] An example of an application is the measurement of autonomy for vulnerable individuals in connected health services. This measurement makes it possible to inform the family and medical professionals about changes in the physical, mental, and social health of the person being monitored, particularly through remote monitoring services.

[0046] Another example of a possible application is a future service for the industry of the future, also known as "Industry 4.0," which would make favorable use of the analysis of modes of transport in indoor or outdoor environments.

[0047] Similarly, services adapted to the situation of a lone worker can be provided based on their modes of locomotion and transport as perceived from a distance, near a danger zone. For example, an alert can be issued upon the arrival of a worker in a risk zone if the mode of locomotion used and identified is potentially too fast.

[0048] Reference is now made to [fig. 1], which illustrates an example of an embodiment of the process which is the subject of the invention.

[0049] In IP, a data collector continuously records digital information from up to four information sources, collected by a single terminal. In one example implementation, the terminal is a phone or a smartwatch. The four information sources are four different sensors: an accelerometer, a gyroscope, a magnetometer, and a barometer, all integrated into a single wearable device. For example, the wearable device and the terminal collecting the digital information could refer to the same entity. Alternatively, they could be two separate entities configured to communicate with each other. The collected digital information describes the movements caused or experienced by the sensor from which it originates.

[0050] In P2, in parallel with data collection, a process allows obtaining and, optionally, selecting the signals of interest to be processed for each information source. For example, it is possible, through such filtering, to process only the numerical information from one, two, or three of the information sources.

[0051] In P3, the raw signals of interest are, if necessary, subjected to temporal segmentation. In an example implementation, after segmentation, a signal of interest from a given information source can be represented as a time series consisting of consecutive measurement points. Each measurement point corresponds to an average of the numerical data collected over an analysis time window. For example, the analysis windows can be one-minute segments. For example, two consecutive analysis windows, corresponding to two consecutive measurement points, can partially overlap. For example, the analysis windows can be 20-second segments with a 10-second overlap between two consecutive analysis windows.In the remainder of this document, all signals of interest are considered to be subject to temporal segmentation into one-minute segments, each associated with a corresponding instant.

[0052] In P4, digital processing can be performed on the signals of interest collected in P2 and, optionally, segmented in P3. The processing includes, but is not limited to, low-pass filtering to denoise the information, and normalization. To standardize the data, data resampling to synchronize sources, amplitude calculations, gravity, and orientation are performed. Reorientation of signals of interest from inertial sensors (accelerometer and gyroscope) is also possible, since connected objects such as a phone or a watch are not oriented in the same way depending on the mode of locomotion or transport.

[0053] In P5, the signals from P4 are pre-classified by mode of locomotion and transport. In an example implementation, several modes of locomotion and several modes of transport can be predefined. For example, four possible modes of locomotion can be considered: walking, running, roller skating, and snowshoeing, associated with a number from 1 to 4 in a first information domain. For example, nine possible modes of transport can be considered: cycling, skateboarding, scootering, train travel, a rolling vehicle, maritime transport, river transport, air transport, and space transport, associated with a number from 1 to 9 in a second information domain.

[0054] In an example implementation, sensor databases, denoted SEN_DB, can be constituted to store respectively histories of pre-classified signals of interest for the two informational domains describing, respectively, modes of locomotion and modes of transport.

[0055] Furthermore, inertial and magnetometric data can be collected to describe locomotion situations, such as walking or running, in a database relating to modes of locomotion, denoted LOC_DB, and to describe transport situations, such as maritime or air, in a database relating to modes of transport, denoted TRAN_DB.

[0056] In P6, a MOD model is loaded which structures temporal and multimodal correlations between the different sources of information in order to infer in P7, on the basis of the signals of interest from P4 up to a current time, the most probable mode of locomotion and / or the most probable mode of transport of an individual at said current time.

[0057] In an example implementation, the MOD model is previously learned from the SEN_DB, LOC_DB and TRAN_DB databases.

[0058] Reference is now made to [fig.2], which illustrates the structure of such a MOD model in an example implementation.

[0059] In this example, the MOD model uses two Siamese gated recurrent neural networks. Gated recurrent neural networks are also known to those skilled in the art as "Gated Recurrent Units" or "GRUs" and are described, for example, in Cho, Kyunghyun; van Merrienboer, Bart; Gulcehre, Caglar; Bahdanau, Dzmitry; Bougares, Fethi; Schwenk, Holger; Bengio, Joshua (2014). "Leaming Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation". arXiv:1406.1078.

[0060] At a given time, a first network (1) has, as inputs, a first set of input vectors called reference vectors (x't-n, ..., x't-1, x't), respectively associated with a corresponding time tn, ..., t-1, t. This first set of input vectors is previously associated, in P5, with a particular mode of transport or locomotion of the individual. As outputs, the first network generates a first reference output vector h't characteristic of this particular mode of transport or locomotion of the individual. It is thus possible, for example, to define a plurality of first networks, designed to generate respectively a first reference output vector h't characteristic of a respective particular mode of transport or locomotion of the individual.

[0061] In parallel, still at the current time, a second network (2) has a set of input vectors (xt-n, ..., xt-1, xt) indicating, respectively, a time segment of the signals of interest originating from P4. Time t is the current time corresponding to the most recently obtained input vector. At the output, the second network, whose operating principle is otherwise identical to that of the first network, generates a second output vector ht characteristic of a mode of transport or locomotion of the individual at the current time.

[0062] Thus, for two multidimensional time-domain signals x't and xt respectively supplied as input, the model generates at least one similarity measure between a first output vector h't generated by a first network (1) and the second output vector ht generated by the second network (2). The similarity measure is, for example, normalized between 0 and 1. Thus, two similar signals, corresponding to the same mode of locomotion or transport, have maximum similarity.

[0063] For example, when the individual is walking, the signals from the sensors show similarities with reference signals previously from the sensors and previously associated with the individual's walking activity. The similarity measure between a first output vector h't characteristic of walking and the second output vector ht takes a value close to 0.

[0064] According to another example, one can imagine that the individual is currently moving using a Segway, but that no reference signal is associated with the Segway as a means of locomotion. In such a situation, it is possible that the signals from the sensors will exhibit similarities with reference signals previously from the sensors and previously associated with a mode of locomotion similar to the Segway (for example, a scooter). Thus, the similarity measure between the first output vector h't characteristic of a displacement of the individual on a scooter, on the one hand, and the second output vector ht characteristic of the individual's movement at the current moment, on the other hand, takes in this example a low value, close to 0.

[0065] Conversely, two distant signals, corresponding to radically different modes of locomotion or transport (for example running and driving) have minimal resemblance and the similarity measure takes a value close to 1.

[0066] According to yet another example, we can imagine that the individual is walking, at the present moment, during a train journey. In such a situation, it is possible that the second output vector ht has a high level of similarity both with a first output vector h't characteristic of the individual's walking activity and with another first output vector h't characteristic of the individual's train journey.

[0067] In a known manner, the operation of recurrent neural networks with gates is based on internal vectors called gates, and more particularly on an update gate and a reset gate, as well as on an activation function, respectively denoted z't, r't and ÿt for a first neural network and zt, rt and ^t for the second neural network.

[0068] In simple terms, within a neural network, these internal vectors and this activation function control the data that are used to determine an output vector. That is to say, the data included in the set of input vectors, i.e., (xt-n, ... xt-1, xt) taking the example of the second neural network, are filtered so that only the data deemed relevant for determining the mode of locomotion or transport at the current time are taken into account.

[0069] The general equations defining the update and reset gates and the activation function are shown below. These equations include not only the input vector xt associated with the current time, but also the output vector ht-1 associated with the time immediately preceding the current time. It is in this sense that the determination of the output vector ht relies, in particular, on a time recurrence. Furthermore, the equation of the activation function involves the reset gate.

[0070] In these equations and the following ones, the values ​​of the quantities W and b (for example Wir and bir) result from a machine learning process and are therefore evolving as new input vectors are obtained and new output vectors are generated. (¾¾ + èty + Mr^-l + ^r) - ton / i(^xt 4- 4- + ^))

[0071] If we rely solely on the above equations, the two neural networks operate in a totally decoupled manner.

[0072] A particular feature of the proposed MOD model is that it includes a coupling between: - on the one hand, temporal and multimodal correlations between signals of interest from P4, and - on the other hand, the temporal and multimodal correlations between pre-classified reference interest signals by mode of transport or locomotion from P5.

[0073] Taking up the implementation example of [fig.2], to implement this coupling, we define a coupling vector, denoted and, as a function of the output vector ht-1, associated with the time preceding the current time, generated at the output of the second neural network, and of the corresponding output vector h't-1 generated at the output of the first neural network.

[0074] In other words, the coupling vector is determined exclusively based on data obtained in the past and allows the dynamics specific to each mode of transport to be integrated into the similarity measure.

[0075] An intermediate gate vector is also defined, denoted t for the first neural network and ^t for the second neural network. Each intermediate gate vector is a function of the coupling vector and the activation functions respectively defined within the two neural networks.

[0076] The output vector ht is then determined on the basis of the coupling vector, on the basis of the update gate defined within the second neural network, as well as on the basis of the activation functions respectively defined within the two neural networks.

[0077] The general equations defining the coupling vector, the intermediate gate vector and the output vector are shown below. (1 ~~ + - (1 — 4-

[0078] Thus, in general, the MOD model structures and establishes a coupling between - on the one hand, a first set of temporal and multimodal correlations between digital reference data extracted from pre-classified reference signals of interest by mode of transport or locomotion, from P5, and - on the other hand, a second set of temporal and multimodal correlations between numerical data extracted from the signals of interest from P4.

[0079] In P7, the MOD model recognizes a current mode of movement of the individual, more precisely a mode of locomotion in P7A and / or a mode of transport in P7B, by inference, on the basis of a similarity between the output vectors ht and h't, that is to say for example a distance between these vectors in an n-dimensional space, n denoting the number of coordinates of each of these vectors in this space.

[0080] Generally, in P7, the recognition of the mode of locomotion, in the first informational domain and / or of the mode of transport, in the second informational domain, at the current time is carried out according to a history of pre-classified signals of interest from P5.

[0081] In P8, a confidence measure Z is associated by the MOD model with the recognition of the situation. For example, one can calculate a maximum likelihood value whether the recognized mode of locomotion or transport corresponds to the individual's mode of locomotion or transport at the current time.

[0082] The higher this value is, the more confident the MOD model is about the robustness of the recognition in P7 at the current time.

[0083] The confidence measure is determined based on the history of pre-classified signals of interest from P5. For example, if a particular mode of locomotion is recognized (e.g., running), the confidence measure can be higher the greater the number of pre-classified signals of interest associated with that particular mode of locomotion. Indeed, an individual is more likely to be using, at the current moment, a mode of locomotion that they are accustomed to using.

[0084] Furthermore, taking up the example of [fig.2], the confidence measure can be determined on the basis of a comparison, or a ratio, between the distance between the output vectors ht and h't on the one hand and a threshold value on the other hand.

[0085] In P9, if the confidence measure Z calculated in P8 is greater than a threshold value, a service adapted to the recognized mode of locomotion or transport can be determined.

[0086] In an example implementation, on a scale of 0 to 1, the threshold value can be set at 0.8. It can also be planned to solicit interaction with the individual when the measurement Z is between a first threshold value A and a second threshold value A'. Conversely, if the measurement Z is less than the second threshold value A', no service is offered and the individual is not contacted. A new iteration of the process is then initiated from PI.

[0087] This interaction based on a double thresholding can make it possible to resolve a doubt The current situation is then categorized among the modes of locomotion and transport defined in the MOD model, which is updated accordingly. A service adapted to the categorized situation can then be determined.

[0088] In P10, a recommendation for the service determined in P9 can be presented to the individual. In one implementation example, the service recommendation is presented via the best distribution channel for the individual, given their recognized or labeled mode of locomotion or transport, for example, on the screen of a smartphone or a smartwatch. For example, if the predicted mode of locomotion is the use of a scooter, the services associated with this mode of locomotion in a PREF preference model could be, in order of preference, activating airplane mode on a terminal or deactivating a gesture interaction with the terminal. Other service examples could be described for each situation according to the individual's preferences and habits.

[0089] These preferences can be defined by the individual himself or by a trusted third party (e.g. family, close friend, or medical staff) or automatically by learning from services used in the past in correlation with recognized modes of locomotion or transport.

[0090] For example, if the individual is a semi-dependent person, an additional feature may stipulate that if the recognized mode of transport corresponds to a predefined list of modes of transport (for example, train, or car if the individual always drives), then this recognition automatically triggers geographical tracking of the individual, for example, by geolocation using a GPS receiver. This makes it possible to detect potential risky situations while respecting the individual's privacy because their location is not continuously recorded.

[0091] In PI 1, the individual's reaction or response to the service recommendation can be evaluated. The proposed service may or may not be activated by the individual. If activated, the recommended service can then be executed in P12.

[0092] In P13, the individual's PREF preference model can be updated according to the recognized situation and, where appropriate, according to the activation or non-activation of a recommended service following this recognition.

[0093] Reference is now made to [Fig. 3], which represents an example of a processing circuit for implementing the above-mentioned process. Such a processing circuit CT (100) comprises at least one CPU (101) connected to a non-transient MEM (102) on which is recorded a program containing instructions for implementing the above-mentioned process.

[0094] The processing circuit may further include a COM communication interface (103) with at least one third-party device. The COM communication interface (103) can for example be controlled to receive the current multidimensional time signal of interest or the current digital data extracted from it, or to transmit the inferred value indicative of the individual's mode of movement at the current time.

Claims

Demands

1. A method, implemented by a data processing circuit, for recognizing the movement mode of an individual, the method comprising: - obtaining (P6) a first structured set of temporal and multimodal correlations between reference digital data extracted from a multidimensional temporal signal of historical interest, originating from sensors worn by the individual, the reference digital data being pre-classified by possible movement mode of the individual, and - recognizing (P7) a movement mode of the individual during a current time interval, recognition (P7) relying on: — structuring a second set of temporal and multimodal correlations between current digital data extracted from a multidimensional temporal signal of current interest, originating from said sensors worn by the individual,The current numerical data being indicative of an individual's movement during the current time interval, — a coupling of the first set of correlations with the second set of correlations, and — an inference of an indicative value of the individual's mode of movement during the current time interval based on the second set of correlations coupled with the first set of correlations.

2. A method according to claim 1, further comprising: - a determination, on the basis of the second set of correlations coupled to the first set of correlations, of a degree of similarity between the current numerical data and the reference numerical data, and - wherein the inference of the indicative value of the mode of movement of the individual during the current time interval is based on said degree of similarity.

3. A method according to claim 1 or 2, further comprising: - an association (P8) of a confidence measure with the recognition of the mode of movement, - a comparison (P9) of the confidence measure with a first threshold and a second threshold, and if the result of the comparison indicates that the confidence measure is between the first threshold and the second threshold, an alert to the individual in order to obtain, through interaction with a human-machine interface, an identification of the individual's mode of movement, and to update the first set of correlations according to the identified mode of movement.

4. Method according to claim 3, wherein if the result of the comparison indicates that the confidence measure is both above the first threshold and the second threshold, a presentation (P10) to the individual of a service recommendation adapted to the recognized mode of travel.

5. A method according to claim 3 or 4, further comprising, if the result of the comparison indicates that the confidence measure is between the first threshold and the second threshold and after obtaining the identification of the individual's mode of movement by interaction with the human-machine interface, a presentation (P10) to the individual of a service recommendation adapted to the identified mode of movement.

6. A method according to claim 4 or 5, wherein the service recommendation is based on a model of individual preferences.

7. A method according to claim 6, further comprising an evaluation (PI 1) of an individual's response to the presented service recommendation and an update (P 13) of the individual's preference model based on the evaluation.

8. A method according to any one of claims 4 to 7, comprising an implementation (P 12) of the recommended service upon receipt of an acceptance signal from an interaction of the individual with a human-machine interface.

9. A computer program comprising instructions for carrying out the method according to any one of claims 1 to 8 when this program is executed by a processor.

10. A non-transient, computer-readable recording medium on which a program is recorded for the implementation of the method according to any one of claims 1 to 8 when this program is executed by a processor.

Images