FR3085524A1 - METHOD FOR LEARNING A PERSON'S LIFESTYLE AND DETECTING ABNORMALITIES - Google Patents

Abstract

The invention relates to a method for quantifying an activity of a person operating in a defined environment, in which is arranged at least one sensor connected to a central unit of a monitoring system, the method being executed by the unit. central, and comprising: a step of obtaining a generic sequence of results capable of being detected by said at least one sensor, said generic sequence being defined by at least one parameter; a step of determining a personalized value, said at least one parameter defining said generic sequence for the person operating in the defined environment, said at least one parameter being designated as personalized parameter when it takes the personalized value, and a personalized sequence, derived from said generic sequence, when it is defined by said at least one personalized parameter, the determining step being executed as a function of first detection results received from said at least one sensor; a monitoring step comprising a sub-step of acquiring second detection results received from said at least one sensor, a sub-step of searching for said personalized sequence among said second detection results; and, if said personalized sequence is not found during a predetermined time interval, a step of notifying an anomaly to a user.

Description

Method for learning a person's lifestyle and detecting anomalies

The present invention relates to the field of monitoring the activity at residence of a person with a physical and / or psychological state of weakness.

Nowadays, the increase in life expectancy, as well as the evolution of outpatient-oriented medicine lead to an increase in the number of people whose follow-up is provided at residence, as long as it is possible.

In order to provide this monitoring automatically, there are a large number of different devices. For example, there are portable connected objects, equipped with an alert button allowing the wearer, in case of difficulty, to notify the emergency services for intervention. Some of these devices include an accelerometer to detect a fall and send a signal to alert the emergency services.

A drawback of this type of device lies in the interaction between the person being monitored and the device they are wearing. In the event of loss of consciousness, in particular without falling, most of the known devices do not make it possible to issue an alert. In addition, this solution assumes that the person is able to remember to systematically wear the device, which can be a difficulty, given the profile of the people monitored.

Another known solution consists in installing an audio-video surveillance system. This solution is not well accepted by those under surveillance, because it is perceived as too intrusive.

Other solutions have been developed, by equipping with sensors a demarcated environment in which the person monitored evolves, for example their accommodation. The sensors used in this case are chosen to be perceived as less intrusive, and therefore better accepted. Such sensors can be, for example motion and activity detectors, as well as door opening detectors, sensors for detecting the use of household electrical appliances and more generally any sensor allowing directly or indirectly to retrieve a measurement linked to the environment or to the person, for example temperature, light, humidity sensors, etc.

Such a solution indeed allows non-intrusive surveillance, therefore better accepted, and more reliable, because the operation of the sensors does not require that they be worn by the person being monitored. Since this solution does not use audio / video sensors, the raw data from the sensors must be interpreted to analyze the activity of the person being monitored, and possibly detect an anomaly. Thus, by relying on the raw data, these monitoring systems operate first in "learning" mode of the person's lifestyle, and secondly, in "monitoring" mode.

Conventional systems implementing this solution have several drawbacks: in "learning" mode, the person's lifestyle is either deduced from the averages calculated from raw data from the sensors, or defined by preprogrammed scenarios, including the personalized parameters, adapted to the lifestyle of the person being monitored, must be entered manually when the system is put into service.

In the first case, the learning mode does not allow to benefit from the behavior models studied before; in the second case, manual entry is necessary, which increases the complexity, installation time and cost of the solution.

These conventional systems also have a drawback during their operation in “surveillance” mode, due to their purely reactive nature with respect to the events detected. Indeed, these conventional systems generate a posteriori alarms, following any anomalies, and do not allow slow degradation to be detected, for example in the event of depression or the onset of neurodegenerative disease.

The known solutions do not make it possible to quantify the activity of the person being monitored, and to detect fragility if, for example, the overall activity of the person is reduced, if a certain type of activity representative of the social bond is decreasing, or if slow changes in behavior occur, for example in the frequency and times of meals, or in the duration and regularity of sleep.

Thus, the solution proposed by the invention aims to provide a complete monitoring solution, capable of generating alerts as a preventive measure, in addition to the alerts generated in reaction to events requiring a specific intervention.

To solve one or more of the drawbacks mentioned above, the invention proposes a method for quantifying an activity of a person operating in a defined environment, in which is arranged at least one sensor connected to a central unit of a system of monitoring, the method being executed by the central unit, and comprising: a step of obtaining a generic sequence of results capable of being detected by said at least one sensor, said generic sequence being defined by at least one parameter; a step of determining a personalized value, of said at least one parameter defining said generic sequence for the person operating in the defined environment, said at least one parameter being designated as personalized parameter when it takes the personalized value, and a personalized sequence, derived from said generic sequence, when it is defined by said at least one personalized parameter, the determination step being executed as a function of first detection results received from said at least one sensor; a monitoring step comprising a sub-step of acquiring second detection results received from said at least one sensor, a sub-step of searching for said personalized sequence among said second detection results; and, if said personalized sequence is not found during a predetermined time interval, a step of notifying an anomaly to a user.

Characteristics or particular embodiments, usable alone or in combination, are:

• An indicator is associated with said personalized sequence, said indicator being able to take a first predetermined value if said personalized sequence is recognized in the second detection results received from said at least one sensor; the monitoring step further comprises a sub-step for assigning to said indicator an effective value, as a function of the result of the search sub-step; and the quantification method further comprises a step of transmitting to an user the effective value of said indicator.

The determination step further comprises a sub-step for identifying a time period during which the personalized sequence is found among the first detection results, with the number of occurrences greater than a first predefined threshold, known as the time range. custom, and the predetermined time interval is equal to the custom time range.

• The method further includes a step of evaluating the sensors.

• The method further comprises a first step of verifying the personalized parameters and, in the event of variation greater than a second predefined threshold, a first step of updating the personalized parameters, and of the personalized sequence.

• The method further comprises a second step of verifying the personalized time slot and, in the event of variation greater than a third predefined threshold, a second step of updating the personalized time slot.

• The first and second update steps are performed based on a predefined number of detection results from the second most recent detection results.

• An exclusion list is associated with each sensor, composed of identifiers of sensors arranged in said delimited environment so that simultaneous or quasi-simultaneous detection results from a sensor and from a sensor belonging to its list of associated exclusion indicates the presence of a second person.

• The plurality of sensors includes: presence sensors; motion sensors; accelerometers arranged on the doors; electrical consumption measurement devices arranged on electrical outlets supplying a first group of predetermined household appliances; accelerometers located on the doors of a second group of predetermined household appliances.

• The anomaly notification step includes: sending an alert message if at least one sequence of detection results received from said sensors indicates a variation of the personalized sequence of a first predefined type; the sending of a prevention message if a plurality of sequences of detection results received from said sensors indicate a variation of the personalized sequence of a second predefined type.

Another aspect of the invention relates to a computer program, comprising instructions for implementing the steps of a method as defined above during execution of the program by a computer.

The present invention also relates to a monitoring system configured to implement the quantification method as defined above.

The invention will be better understood on reading the description which follows, given solely by way of example and with reference to the accompanying drawings.

Brief description of the figures

FIG. 1 schematically represents an example of a monitoring system, according to an embodiment of the invention.

FIG. 2 schematically represents a central unit of the surveillance system, according to an embodiment of the invention.

Figures 3-5 and 7-8 schematically represent steps of a method for quantifying an activity of a person, according to different embodiments of the invention.

Figures 6 and 9 show examples of application of the quantification method according to an embodiment of the invention.

In FIG. 1 is illustrated an environment delimited ENV, in which a person to be monitored evolves. In the environment delimited ENV are arranged several sensors C1, C2, ... Cj, ... Cm, Cn, connected to a central unit CPU of a SYS surveillance system, via a home automation box B located in said environment delimited ENV. The sensors C1, ... Cn, are, as in conventional solutions, motion and activity detectors, as well as door opening detectors, sensors for detecting the use of household electrical equipment. and more generally any environmental or worn sensor, allowing direct or indirect feedback of a measurement linked to the environment ENV or to the person P, for example sensors of temperature, brightness, humidity, accelerometers placed on the doors or windows, or worn by the person, etc.

The home automation box B communicates with the central processing unit UC via an extended telecommunications network, for example the Internet. The central processing unit UC is connected to a SERV server hosting one or more databases, via COM communications means. This arrangement allows on the one hand to share the resources of the central processing unit and the databases between several monitored environments, and on the other hand to enrich the databases with raw data coming from several environments. These data can then be analyzed to evolve the behavior models used by the surveillance system.

Alternatively, in a configuration not shown, the home automation box B and the central processing unit UC are located within the delimited environment ENV. The two entities can be implemented on the same device, or on two separate devices, communicating with each other by a local network. Only anomalies notifications will be sent via the extended telecommunications network to supervisors and / or assistance and rescue organizations. In this way, additional protection of private data is ensured.

As an option, the home automation box B and / or the central processing unit UC can receive additional information from the extended telecommunications network, for example meteorological data, the calendar of holidays and public holidays, information concerning a programmed absence from the monitored person, entered by the supervisor or by another user, geolocation information of the monitored person in the event of an exit, etc.

During the installation of the monitoring system, operational data concerning the delimited environment ENV are transmitted to the central unit UC, for example the type of room where each sensor C1, C2 .. Cj .. is installed: kitchen, bedroom, living room, hallway, bathroom, garden, terrace, etc. One or more rooms can be designated as main living room (s). It is in this or these room (s) that the person spends the most time for their activities, and the sensors can be installed as a priority. Additional sensors of the same type or of a different type may also be installed. Thus, the system can analyze more closely the activity of the person being monitored, determining for example if they are watching television or if they are active, for example by cleaning. This differentiation allows a more precise quantification of the activity of the person, and in the event of a notable decrease in this activity, a slow deterioration in the state of the person could be detected. It should be noted that a purely reactive system would not trigger an alert in this case, since there are no urgent events.

The type of the various sensors is also informed, in addition to their distribution in the environment delimited ENV, thus giving a partial representation of the place of life of the person P.

The surveillance system can also use geolocation data of the monitored person, if he is equipped with a beacon, and / or of the delimited environment ENV, which makes it possible to correlate for example meteorological data specific to the location of the person being monitored and the actions taken by that person.

In FIG. 2 is schematically represented the central unit UC of the monitoring system SYS, according to an embodiment of the invention. The central unit includes a CPU, MEM memories to store data and computer programs, a Tempo time measurement unit, as well as RX communication interfaces to receive the detection results sent by the sensors Cj, and TX to notify possible anomalies to a user of the system, for example a relative of the monitored person or an assistance and rescue organization, as well as a communication interface with the home automation box B (not shown) and a COM communication interface with the external SERV server.

The monitoring system according to the invention operates initially in “learning” mode of the person's lifestyle, to determine a personalized sequence SEQPj of detection results capable of being received from said sensors Cj, said personalized sequence SEQ Pj characterizing the activity of the person P evolving in the environment delimited ENV. In a second step, the system operates in "monitoring" mode, which will be detailed later. In the event of an anomaly, a notification will be sent to a user of the system.

The quantification method according to the invention makes it possible to benefit from the behavior models studied beforehand. Such behavior patterns are defined for example for eating a meal, getting up, using the bathroom, cleaning, etc.

Each of the behaviors includes a series of movements of the person within the housing, and / or a series of actions exerted on the household appliances, and / or door openings / closings, which are detectable by the sensors Cj. Detection results, the order of detections, and time information for certain time-stamped results are analyzed to identify behavior.

A behavior model, capable of being identified and quantified by the monitoring system, is modeled by a sequence of results, in a certain order, the results being sent by certain sensors, for a certain duration.

A behavior model represents a general case, and a sequence SEQj of results capable of being detected by one or more sensors Cj, associated with this general behavior model, is a sequence SEQj called generic.

The definition of the generic sequences is carried out prior to the implementation of the invention. This definition can be, for example, the result of behavioral studies carried out by universities or specialized laboratories. An advantage of the solution proposed by the present invention is precisely to make it possible to benefit from these studies with substantial human contribution, in an automated monitoring system.

Each generic sequence SEQj is defined as a function of one or more parameters mj, nj. Special values, vmjp, vnj_p assigned to the parameters mj, nj, make it possible to obtain a specific instantiation SEQPj of the generic sequence SEQj. This particular instantiation SEQ Pj is thus representative of a particular instantiation of the behavior model, for a given monitored person P. In the remainder of this document, such a sequence is designated as a “personalized sequence” SEQ Pj. The particular values vmj_p, vnj_p allow, when they are assigned to the parameters mj, nj, to obtain the personalized sequence SEQ Pj, from the generic sequence SEQj. These parameters are designated in the remainder of this document as "personalized parameters" mj_p, nj_p.

For example, a person's sunrise can be modeled by a series of presence detections in one or more rooms in the ENV environment, for a certain period of time.

A meal is modeled by a presence detection in the kitchen, followed by a certain number of detections of the use of household appliances, signaled by devices for measuring electrical consumption arranged on electrical outlets supplying a first group. predetermined household appliances, and / or accelerometers arranged on the doors of a second group of predetermined household appliances, for example a refrigerator, a microwave oven, etc.

If we consider the survey modeled by a series of presence detections in one or more rooms of the ENV environment, for a certain duration, the generic sequence SEQj can be characterized by two parameters: a first parameter mj, linked to the number of detections presence, and a second parameter nj, linked to the duration during which the presence detection results are considered significant.

In one embodiment of the invention, the duration can be measured by counting a number of time intervals of predefined duration, for example ten minutes. In this example, the first parameter mj is the minimum number of presence detections during each time interval, and the second parameter nj is the number of successive intervals of predefined duration during which these detection results are expected.

For a first person P1, rising is manifested by a minimum of two presence detections per ten-minute interval, for at least one interval; for a second person P2, by a minimum of three presence detections per ten-minute interval, for at least two successive intervals.

Thus, for the first person P1, the generic sequence SEQj can be instantiated so as to obtain a personalized sequence SEQP1 representing the behavior of this person P1 if the first parameter mj takes the value vmj_1 equal to 2, and the second parameter nj takes the value vnj_1 equal to 1; for the second person P2, the generic sequence SEQj can be instantiated so as to obtain a personalized sequence SEQP2 representing the behavior of this person P2 if the first parameter mj takes the value vmj_2 equal to 3, and the second parameter nj takes the value vnj_2 equal to 2. The personalized parameters mj_p, nj_p respectively take the values (2, 1) for the first person P1 and (3, 2) for the second person P2.

The order of the parameters is important, so the values of the custom parameters mj_p, nj_p constitute 2-dimensional vectors in this example. A third dimension can be added, for example by introducing a third parameter representative of the time of day when the expected sequence is actually detected, or the seasonality of this sequence. More generally, N parameters can be grouped to constitute vectors of N-dimensional parameters.

The effective values of the personalized parameters mj_p, nj_p make it possible to adapt the generic sequence SEQj to each monitored person P, evolving in the environment ENV, in other words to customize the generic sequence SEQj to the person P and to the environment ENV.

Embodiments of the quantification method will now be described with reference to FIGS. 3, 4, and 5. The examples described illustrate variants of learning and monitoring a single behavioral model of the monitored person, but of course , a different instance of the method according to the invention can be executed for each behavior mode and each associated generic sequence SEQi of results capable of being detected by the sensor or sensors installed in the environment delimited ENV.

The SYS monitoring system operates, after installation, in learning mode. The central processing unit UC first obtains, in an obtaining step E100, the generic sequence SEQj representative of the behavior mode considered. This sequence SEQj is defined beforehand, and stored beforehand either by the same central processing unit UC, or by a separate entity, for example a server dedicated to the SERV application. Depending on the case, the obtaining step E100 can take different forms, for example a step of reading from memory of the central processing unit UC, a step of receiving via the communication means COM from one or more messages from the SERV server via a telecommunications network, an operator input step, etc.

Whatever the form of this step of obtaining E100, following its execution by the central unit UC, the latter is in possession of the generic sequence SEQj, of the parameters mj, nj defining it, including their type, as well as the sensor (s) Cj capable of reporting detection results relevant to the mode of behavior considered, and therefore for the associated generic sequence SEQj.

Then, the central unit executes a step E200 of determining one or more personalized value (s) vmj_p, vnjp, of the parameter (s) mj, nj defining the generic sequence SEQj for the person P evolving in the environment delimited ENV. This (s) parameter (s) are designated as custom parameter (s) mj_p, nj_p when it (s) takes / take the value (s) custom (s) vmjp, vnj p.

The central processing unit UC receives, as and when the determination step E200 of the quantification process is executed, first detection results r1, r2, .. rk, coming from the sensors Cj installed in the environment ENV .

These first results r1, r2, ..rk, are used by the monitoring system operating in “learning” mode. In this embodiment of the invention, the exploitation of the first results r1, r2, ..rk, is done according to statistical methods.

The central processing unit UC first accumulates a certain number of first results r1, r2, ..rk, during a predefined period or until a sufficient number of results is obtained in order to be able to apply statistical methods, and then performs successive tests, assigning to the parameters mj, nj, several sets of particular values, (vmj_1, vnj_1), (vmj_2, vnj_2), ... (vmj_k, vnj k), ... (vmj_p, vnj p),, .. (vmj_q, vnj q), etc.

Each set of particular values, (vmj k, vnj k) is applied according to the definition of the generic sequence SEQ j, and a maximum likelihood is sought among the representations obtained.

FIGS. 6a to 6d illustrate representations obtained for the example of modeling the rising of the two people P1, P2, in two respective environments ENV1, ENV2, the first parameter mj1, mj2, being the minimum number of presence detections during each interval time, and the second parameter nj1, nj2, being the number of successive intervals of predefined duration during which these detection results are expected. The curves of FIGS. 6a and 6c represent the interpretation of the first results r1, r2, ... rk, on the assumption that the parameters (mj1, nj1), respectively (mj2, nj2) take the personalized values from 1 to 6 for the parameter mj1, respectively mj2 and from 1 to 4 for the parameter nj 1, respectively nj2, for a given learning period.

The representations satisfying the criterion of maximum likelihood are those corresponding to the values (2,1) for figure 6a, respectively to the values (3,2) for figure 6c. Thus, these values will be assigned to the parameters (mj1, nj1), respectively (mj2, nj2), to obtain the personalized parameters mj_p1, nj_p1 and mj_p2, nj_p2, characterizing the personalized sequences associated with the lifting of people P1, P2, evolving in the two environments considered ENV1, ENV2.

Thanks to the personalized parameters (mj_p1, nj_p1), (mj_p2, nj_p2) thus determined, are then defined, from the generic sequence SEQj, the personalized sequences SEQPjl, SEQ_Pj2 modeling the rising of the first person P1, respectively of the second person P2.

In this example, the sequence SEQ Pjl is thus defined as a minimum of two presence detections per ten-minute interval, for at least one interval; and the sequence SEQ_Pj2, as a minimum of three presence detections per ten-minute interval, for at least two successive intervals.

The personalized sequences SEQ Pjl, SEQ_Pj2 are sought among the first results r1, r2, ... rk. A sequence SEQ Pjl, SEQ_Pj2 found, means that the person monitored P1, P2 has actually stood up. The surveys of people P1, P2 can be time-stamped. Figures 6b and 6d represent a determination of the hours of lifting people P1, P2, evolving in environments ENV1, ENV2, over 31 days, thanks to the use of personalized parameters (mj_p1, nj_p1), (mj_p2, nj_p2) and custom sequences SEQ Pjl, SEQ_Pj2.

Similarly, FIGS. 6e and 6f illustrate representations obtained for the example of modeling the meal of two people P1, P2, in two respective environments ENV1, ENV2.

Other techniques which analyze data sets and give predictions of statistical distribution can be used, for example machine learning algorithms, as described for example in:

- Ryszard S. Michalski: The Vision and Evolution of Machine Learning;

- Machine Learning - An Artificial Intelligence Approach - Editors: Michalski, R.S., Carbonell, J.G., Mitchell, T.M. (Eds.);

- Trevor Hastie, Robert Tibshirani, Jerome Friedman, The Elements of statistical learning, Springer 2009; and:

- Larry, Wasserman, All of Statistics. A Concise Course in Statistical Inference, Springer 2004.

Once the sequence SEQPj having the maximum likelihood has been found, a search is carried out among the sets of particular values, (vmj_1, vnj_1), (vmj_2, vnj_2), ... (vmj k, vnjk), ... (vmjp, vnj_p), ... (vmjq, vnjq), to identify the set applied to obtain this sequence SEQ Pj. The values (vmj_p, vnj p) constituting this set are designated as "custom values". The parameters mj, nj, taking respectively the personalized values vmj_p, vnj p, are designated as “personalized parameters” mj_p, nj_p, and the sequence SEQ Pj obtained according to the definition of the generic sequence SEQj, when it is defined by the parameters custom mj_p, nj_p, is referred to as "custom sequence".

In this embodiment, the personalized parameters mj_p, nj_p, and the sequence SEQ Pj are determined according to the so-called maximum likelihood method, a statistical method known in the state of the art.

Other statistical methods can also be applied, for example the so-called regression method. It is a statistical analysis which describes the variations of an endogenous variable associated with the variations of several exogenous variables. For example, a regression analysis can reveal a positive relationship between certain days of the week and the variables corresponding to the activity such as cleaning or social contact (visit of a home help). Thus, the generic sequence SEQj can find several instantiations SEP Pjl, SEQ_Pj2, ... SEQ_Pj7 different, depending on the days of the week. The learning mode of the monitoring system ensures the determination of all these instantiations, as well as the additional parameter which contributes to their differentiation.

For certain behavior models, the quantification method can also comprise a substep for identifying a time slot during which the personalized sequence SEQPj is found among the first detection results r1, r2, rk, with a number d 'occurrences higher than a first predefined threshold, called personalized time slot T_Pj. This time range can be determined according to known statistical methods, for example by the so-called “quantile” method; in this case, the threshold is defined according to the chosen quantile values.

The identification of the personalized time slot is particularly useful for the evaluation of slow degradations, during the operation of the system in "monitoring" mode, as will be explained below.

The determination step E200 of the personalized sequence SEQ Pj is executed by iteratively applying a statistical method, according to the first detection results r1, r2, ..rk, received by the central unit UC from the sensor or sensors Cj. Depending on the embodiments, the duration of this step can have a predefined maximum duration, for example three months.

The determination step can, in another embodiment, be executed as long as a predetermined condition relating to the personalized sequence (SEQ Pj) is not fulfilled. The latter case has the advantage of optimizing the learning time. The central processing unit UC evaluates the convergence of the results, the length of the learning period can vary greatly from one person to another or from one behavioral model to another.

Once the learning has been completed, the monitoring system switches to the "monitoring" operating mode proper. To do this, the central processing unit UC of the monitoring system SYS performs a monitoring step E300.

For each behavior model “j” of the monitored person P, the central unit UC has determined during step E200 the personalized sequence SEQ Pj associated with the behavior model of this person P. If the sequence SEQ Pj is obtained according to statistical methods, a statistical distribution and a variance of values are also determined to analyze the behavior of the person being monitored P.

The monitoring step E300 firstly comprises a sub-step of acquiring E310 of second detection results R1, R2, Rk received from said sensors Cj, and then a sub-step of searching E320 for said personalized sequence SEQPj among the second detection results R1, R2, Rk.

The central processing unit UC then seeks to assess whether the expected personalized sequence SEQ Pj has actually been received from the sensors Cj, taking account, where appropriate, of the abovementioned statistical data.

In the particular case where a personalized time slot T_Pj has been determined during the identification step E250, the predetermined time interval T_p, while the personalized sequence SEQ Pj is sought, is equal to this personalized time slot T_Pj.

If the personalized sequence SEQ Pj is not found during the personalized time slot T_Pj, which means for example that the person has not risen according to his habits, the central unit UC generates an alarm message to notify the 'anomaly detected. The monitoring system operates in reactive mode, in this case.

A behavior will be considered normal if the personalized sequence SEQ Pj is actually found among the second detection resultsRI, R2, .. Rk, received during the personalized time slot T_Pj. There will therefore be no anomaly notification to be sent as a reaction to an event.

The precise moment of detection of the personalized sequence SEQ Pj in this personalized time slot T_Pj can be evaluated over several days, weeks or months. If a drift of this moment of detection within the personalized time slot T_Pj is observed, for example a rising which takes place later and later, the central unit UC can deduce a slow deterioration of the general state of the person being monitored, and notify a supervisor of person P as a precaution. In these cases, the monitoring system operates in predictive mode, and sends a prevention message MP.

More generally, if several sequences of detection results received from said sensors (Cj) indicate a slow variation of the personalized sequence (SEQ Pj) called variation of second type, the monitoring system operates in predictive mode, and sends a prevention message MP. Such slow variations can be deduced, for example, if the duration and / or the frequency of outings of person P decreases from one week to another, or from one month to another. Another slow deterioration can be deduced from the decrease or increase in the number of meals, the segmentation of night rest, a change in the distribution of activity in the accommodation, etc.

The method according to the invention thus makes it possible to detect two types of anomalies, a first type of event character requiring immediate intervention, and a second type of anomaly, indicating a slow deterioration in the behavior of the person. The quantification method according to the invention thus provides a complete monitoring solution, unlike conventional methods.

In the examples described above, the monitoring step E300 is executed by comparing the second results actually received R1, R2, ... Rk, during the acquisition sub-step E310, with the personalized sequence of results SEQ Pj. The personalized parameters mj_p, nj_p defining this personalized sequence were determined during the determination step E200, on the basis of the first detection results r1, r2, ..rk. If there is a change in the behavior of the person being monitored P, these parameters may become obsolete, and therefore the monitoring system may lose precision and efficiency.

Custom sequences can also become obsolete, for example if the number, type or sensitivity of the sensors are changed, that is, if the monitored environment is changed. Such situations can occur if sensors are added or moved, if sensors are defective, and stop sending detection results, or send random results, or if the embedded software of the sensors is updated, which can change the detection thresholds.

To detect such situations, the central unit implements, in a preferred embodiment, a step of evaluating the E650 sensors. This step can include one or more of the following substeps:

• verification of the maintenance frames received from the sensors, if the sensors are configured to send periodically, for example every hour, such frames;

• validation of the values actually received, compared to the expected data ranges; thus for event type sensors, which send detection results "ON" or "OFF", the number of detections "ON" must be substantially equal to that of detections "OFF"; for temperature sensors, the detected temperature values must be in ranges corresponding to the environment (indoor or outdoor), for example between 10 ° C and 40 ° C for indoor use, and can possibly be correlated with meteorological data for outdoors;

• analysis of data traffic on each portion of the transmission chain, for example by researching the sequencing of exchanges if frames are missing, or duplicated; evaluation of the radio power "RSSI" on the local network between the sensors and box B;

• analysis of the level of electrical supply of the various elements;

• verification of the number of sensors and / or identifiers of active sensors;

• periodic movement along an evaluation path in the ENV environment, and analysis of the results of the detections sent by the sensors; this assessment path can be defined during the installation of the system, and then periodically carried out by a supervisor or a relative of the person being monitored. The detection results actually received are compared with the reference results obtained when defining the route.

• verification of statistical data concerning the sensors.

The particular example of a defective sensor detected using statistical methods is illustrated in Figure 9. In the left part of this figure are shown the results of detection during normal operation of a sensor. These results are used during the determination step E200, with the detection results of the other sensors as previously explained. Statistical data concerning this sensor can also be deduced, for example the average number of detections per day, the distribution of detections during a day. The central part of FIG. 9 indicates an absence of detections, which may correspond to an absence of the person or to a defective sensor. The right side indicates a higher number of detections, as well as a different distribution during the day; therefore the statistical data concerning this sensor will be modified. If the difference between the statistical data concerning this sensor obtained during the determination step E200 and the statistical data obtained during the verification step of the sensors E650 exceeds a predefined threshold, the sensor will be considered as defective, and its results detections will not be taken into account for the further execution of the quantification process. This is the case for the sensor that sent the results shown on the right side of Figure 9.

A faulty sensor identified induces a modification of the equipped environment, and therefore an update of the personalized parameters and of the SEQ Pj personalized sequence is timely.

After performing the sensor evaluation step, the central processing unit UC can deduce which sensors are considered to be reliable; the detection results sent by these sensors will be used to "relearn" the personalized parameters and the personalized sequence SEQPj, similar to the determination step E200. Advantageously, the sensor evaluation step allows the monitoring system to automatically detect a change in the ENV environment.

The SYS monitoring system can also detect a change in the behavior of the person being monitored. Indeed, if there is no defective sensor, or if one or more defective sensors were used marginally (or not used) in the behavior model corresponding to the sequence SEQ Pj, the parameters and the personalized sequence SEQ Pj previously determined always correspond to the expected sequence.

To determine whether an evolution occurs in the behavior of the monitored person P, the quantification method comprises, in particular embodiments, a first step of verification E600 of the personalized parameters mj_p, nj_p.

In the event of a variation of the parameters, greater than a second predefined threshold, the quantification method further comprises a first step of updating E700 of the personalized parameters mj_p, nj_p and of the personalized sequence SEQ Pj.

If the quantification method uses a personalized time slot T_Pj, as illustrated in FIG. 8, the latter is also verified, during a second verification step E610 and, in the event of variation of the personalized time slot T_Pj greater than a third threshold predefined, the method further comprises a second step of updating E710 of the personalized time slot T_Pj.

The joint use of an evaluation of the sensors and verification of the personalized parameters and of the SEQ Pj personalized sequence allows the system to react automatically both to a change in the equipped environment and in the behavior of the person being monitored.

It should be noted that the order of execution of the steps for verifying the parameters E600 and for evaluating the sensors E650 can change from one embodiment to another.

In a preferred embodiment, the first and / or the second updating step 700, 710 are executed as a function of a predefined number of detection results from the most recent second detection results R1, R2, Rk. In other words, the method again applies the statistical methods, similar to the determination step E200, to "relearn" the lifestyle of the person in the ENV environment according to the latest detection results. This method ensures an almost continuous updating of the parameters, based on a sliding learning "window".

In a preferred embodiment, in which the central processing unit UC has means for receiving information relating to a hospitalization of the person, the quantification method can detect a significant variation in the lifestyle of this person on his return. In this case, the anomaly notification will not be sent, and the personalized parameters will not be updated following the detection of significant variation. On the other hand, the system will monitor the difference in behavior detected with the old behavioral mode, and only if the difference observed remains present for more than a predefined duration, for example a convalescence duration estimated beforehand, the unit central unit UC will trigger the execution of update steps E700, E710. In this way, the execution of these updating steps is avoided if the evolution of the detected behavior indicates a return to the behavior mode initially modeled by the personalized sequence SEQ Pj. The operation of the system is thus optimized.

As already mentioned, the quantification method performs a step E400 of notification of anomaly to a user, if the expected personalized sequence SEQ Pj is not actually received during the monitoring step E300.

Result sequences can be complex, depending on the behavior modeled, and passing them as such to a user, who is not a specialist in the field of home monitoring, can cause problems of understanding. To avoid this complexity, the quantification method can use an indicator IN Dj associated with the personalized sequence SEQ Pj, and consequently with the associated behavior model. The indicator INDj can take a first predetermined value V1 if the personalized sequence SEQ Pj is recognized in the second detection results R1, R2, Rk received from the sensors Cj.

In these embodiments, the monitoring step E300 further comprises:

- a sub-step of attributing E330 to said indicator INDj with an effective value Vej, according to the result of the sub-step of research E320; and

a step of transmission E500 to a user of the effective value Vej of said indicator.

For example, in the case of monitoring the person's sunrise, a first indicator "LEVER" can be set up, and the value V1, for example "TRUE" logic, will be assigned to the indicator "LEVER" if the The corresponding SEQJ custom sequence was detected during the scheduled time slot. In this way, the user interface is simplified, ensuring a better user experience for the surveillance system.

With reference to FIGS. 6b and 6d, described above, when the personalized sequence SEQPjl, SEQ_Pj2 representative of the lifting of the person has been identified among the detection results as explained previously, the central processing unit UC can perform a time-stamped detection of the personalized sequence SEQPj, and a timestamped assignment of the value "TRUE" to the indicator "LEVER".

According to the modeled behavior modes, effective value Vej can take the value V1, thus indicating that the behavior of the monitored person does not present any anomaly, or any other value in the event of an anomaly, depending on the identified anomaly. This embodiment also has the advantage of minimizing message transmissions via telecommunications networks.

The quantification process is particularly suitable for monitoring a person living alone in the ENV environment. During social or medical visits, movement or presence sensors will send detection results regardless of the person who produces them, the person monitored and the visitor (s). This can happen both in learning mode and in detection mode. Thus, all the data received from the sensors is not relevant for monitoring the person P operating alone in the ENV environment.

Information on the presence of a second person in the circumscribed environment ENV, or the planned absence of the monitored person, for example for a vacation or for a hospitalization, can influence the relevance of the behavior model, as will be explained by the following.

In a particular embodiment of the invention, the quantification method makes it possible to select the relevant data from all the detection results received. To do this, the SYS monitoring system has the means to detect visits, and possibly means to receive information concerning the absences of the person to be monitored, which allows the central unit to first carry out an assessment. the relevance of the data sent by the sensors.

If information is obtained on a period of presence of a second person in the ENV environment, or on a period of programmed absence of the person being monitored, for example for holidays or for hospitalization, the data from sensors Cj are considered to be irrelevant during these periods and the determination step E200 is suspended for the duration of this presence. The methods of obtaining this information may vary depending on the learning and monitoring techniques used.

Thus, according to a first embodiment, each sensor Cj placed in the environment ENV is associated with an exclusion list LEj, composed of identifiers of sensors Ci, Ck, arranged in said environment delimited ENV, so that results simultaneous or quasi-simultaneous detection from a sensor Cj and a sensor Ci, Ck belonging to its associated exclusion list LEj indicate the presence of the second person P2.

According to variant embodiments, the exclusion lists LEj associated respectively with the sensors Cj can be predefined, for example during the installation of the sensors with respect to the rooms where the sensors are installed, or with respect to particular locations in the same room. In this variant, the sensors installed in other rooms than a given sensor Cj, will be part of the exclusion list associated with this sensor Cj. Similarly, sensors Ci, Ck, installed in the same room as the sensor Cj, but at a distance large enough that a single person cannot simultaneously or almost simultaneously generate presence detection results sent by the sensor Cj and by the sensors Ci, Ck. Of course, the distance depends on the types of sensors and the respective detection radii, specifications which are known during the installation of the system.

In another variant, the list is instantiated dynamically, for example by studying a simultaneity matrix, and on the assumption that the person is mainly alone in the ENV environment and knowing that simultaneous detections can be sent by the sensors for a single person in a dwelling.

In another embodiment, using preprogrammed scenarios, the periods of visits of the home helpers, as well as the outputs of the person being monitored, can be entered manually during the commissioning of the system, or during the updating operations. system day. The detection results sent by the sensors Cj during the periods thus filled in will be excluded from the detection results taken into account for learning the lifestyle of the person P, respectively for monitoring this person. These indications of presence calendar can also make it possible to refine the profile of simultaneity or quasi-simultaneity of the detection results for the constitution of the exclusion lists LEj, for example in a form of supervised learning of visits.

In a preferred embodiment of the invention, the inputs and outputs of people in the environment respectively can be identified by specific sequences of the detection results sent by sensors specially placed for this purpose, for example motion sensors near the entrance doors to the ENV environment, or accelerometers placed on these doors. Other detection means can be used, for example identifiers linked to visitor telecommunications equipment, such as smartphones, access badges or buttons to be activated during entry / exit.

Whatever the mode of detecting the presence of a second person in the environment ENV, during the execution of the determination step E200, in particular the acquisition of the detection results r1, r2, ..rk, from sensors, the execution in progress is suspended for the duration of this presence.

The results r1, r2, ..rk are therefore obtained by a selection of relevant results from all the results received, thanks to the ability of the central processing unit UC to detect the presence of a second person within the housing, and possibly to receive information on the absence of person P from the ENV environment. Within the meaning of the invention, a relevant result is that which represents the behavior of a person operating alone in the environment delimited ENV, excluding visits.

The implementation of the quantification process is particularly advantageous if the personalized values of the parameters and the personalized sequence SEQ Pj are determined by statistical or "machine learning" methods. Indeed, the detection results being selected according to their relevance, thanks to the detection of visits, the modeling of the behavior of the monitored person is more precise.

Similar to operating in “learning” mode, in this monitoring system having means for detecting visits, or means for receiving information concerning the absences of the person to be monitored, the central unit UC first performs a evaluation of the relevance of the detection results R1, R2, Rk.

If these results are deemed irrelevant, since they are not representative of the person's behavior during visits, or in the event of a planned absence, the monitoring step E300, in particular the acquisition sub-step E310, is interrupted.

If the method implements the first and / or the second updating step 700, 710 mentioned above, these steps are interrupted during the visits, in a similar manner to the suspension of the determination step E200, and of the sub-step E310, as shown in Figure 7.

If a visit is detected, the detection results will therefore not be taken into account for the learning of behavior patterns of the single person, nor for the monitoring corresponding to these behavior patterns. On the other hand, a detection of a visit received is a precious indicator of the activity of the person being monitored, which will be quantified according to another criterion, that of social ties.

To assess the social bond, the central processing unit UC can also perform a quantification of the person's exits from their accommodation. An output can be modeled, for example, by a sequence of door opening / closing detections, correlated with an absence of presence detection in the environment delimited ENV. An exit detected one day, which does not recur, will not be considered as an anomaly. On the other hand, if the person goes out less and less, this can show a drop in social ties.

Additional data can be taken into account for the evaluation of variations in the frequency of visits and / or outings, for example the day of the week, the calendar of public holidays, meteorology (rain, heat wave), etc. This data can be received by the central processing unit UC from the extended telecommunications network, according to known methods.

Variations in the frequency of visits and / or exits can thus be monitored by the central processing unit UC, and if a downward trend is identified, a preventive anomaly can be notified to a supervisor of the person P.

It should be noted that in this operation in predictive mode, the personalized sequences corresponding to the monitored behavior are found, so there is no urgency to react to a detected event. We consider a first person, who receives two visits per week on a regular basis, and a second person, who receives during a first period daily visits, and then a decreasing number of visits. The activity quantification method can send a prevention message MP, if the number of visits received by the second person reaches a certain threshold, predefined, for example three visits per week, or determined by learning the regularity of the visit. of the person being monitored. Thus, for the first person considered, a PM prevention message is not sent, because the number of visits received by this first person, although lower than that of the second person considered, is almost constant. On the other hand, a prevention message MP will be sent to a supervisor of the second person, to signify a slow deterioration in the behavior of this person.

More generally, the method according to the invention makes it possible to detect not only events, for example by comparing expected values with measured values, but also trends in the evolution of the measured values, thus ensuring a predictive character for the system of surveillance.

Claims (12)

1. Method for quantifying an activity of a person (P) evolving in a delimited environment (ENV), in which is arranged at least one sensor (Cj) connected to a central unit (UC) of a monitoring system (SYS), the method being executed by the central unit (UC), and comprising: A step for obtaining (E100) a generic sequence (SEQj) of results capable of being detected by said at least one sensor (Cj), said generic sequence (SEQj) being defined by at least one parameter (mj, nj); • a determination step (E200): o a personalized value (vmjp, vnj p), of said at least one parameter (mj, nj) defining said generic sequence (SEQj) for the person (P) operating in the delimited environment (ENV), said at least one parameter (mj, nj) being designated as personalized parameter (mj_p, nj_p) when it takes the personalized value (vmj p, vnj p), and o of a personalized sequence (SEQ Pj), resulting from said generic sequence (SEQj ), when it is defined by said at least one personalized parameter (mj_p, njjo), the determination step being executed as a function of first detection results (r1, r2, ..., rk) received from said at least a sensor (Cj) • a monitoring step (E300) comprising: o an acquisition sub-step (E310) of second detection results (R1, R2, ... Rk) received from said at least one sensor (Cj), o a search sub-step (E320) of said sequence personalized (SEQ Pj) among said second detection results (R1, R2, ... Rk); and, if said personalized sequence (SEQ Pj) is not found during a predetermined time interval (T_p), • a step of notification (E400) of anomaly to a user. 2. Quantification method according to the preceding claim, characterized in that: An indicator (INDj) is associated with said personalized sequence (SEQ Pj), said indicator (INDj) being able to take a first predetermined value (V1) if said personalized sequence (SEQ Pj) is recognized in the second detection results (R1, R2, Rk) received from said at least one sensor (Cj); • the monitoring step (E300) further comprises a sub-step for assigning (E330) to said indicator (INDj) an effective value (Vej), according to the result of the search sub-step (E320); and in that : • the quantification method further comprises a step of transmitting (E500) to a user the effective value (Vej) of said indicator. 3. Quantification method according to one of the preceding claims, characterized in that: • the determination step (E200) further comprises a sub-step of identification (E250) of a time period during which the personalized sequence (SEQ Pj) is found among the first detection results (r1, r2, rk ), with the number of occurrences greater than a first predefined threshold (P1), known as the personalized time slot (T_Pj), and in that: • the predetermined time interval (T_p) is equal to the personalized time slot (T_Pj). 4. Quantification method according to one of the preceding claims, further comprising: • a first verification step (E600) of the personalized parameters (mj_p, nj_p) and, in the event of variation greater than a second predefined threshold, • a first updating step (E700) o of the personalized parameters (mj_p, nj_p) , o of the personalized sequence (SEQ Pj). 5. Quantification method according to one of the preceding claims, further comprising a sensor evaluation step (E650). 6. Method of quantifying one of claims 3 or 4 when it depends on claim 3, further comprising: • a second verification step (E610) of the personalized time slot (T_Pj) and, in the event of a variation greater than a third predefined threshold, • a second updating step (E710) of the personalized time slot (T_Pj). 7. Quantification method according to one of claims 4, 5 or 6, characterized in that the first and the second updating step (700, 710) are executed according to a predefined number of detection results among the second results most recent detection (R1, R2, Rk). 8. Quantification method according to one of the preceding claims, characterized in that each sensor (Cj) is associated with an exclusion list (LEj), composed of sensor identifiers (Ci, Ck) arranged in said delimited environment (ENV ) so that simultaneous or quasi-simultaneous detection results from a sensor (Cj) and a sensor belonging to its associated exclusion list (LEj) indicate the presence of a second person (P2). 9. Method according to one of the preceding claims, in which the plurality of sensors (Cj) comprises: • presence sensors; • motion sensors; • accelerometers placed on the doors. • devices for measuring electrical consumption arranged on electrical outlets supplying a first group of predetermined household appliances; • accelerometers placed on the doors of a second group of predetermined household appliances. 10. Method according to one of the preceding claims, in which the anomaly notification step (E400) comprises: • sending an alert message (MA) if at least one sequence of detection results (SEQA) received from said sensors (Cj) indicates a variation of the personalized sequence (SEQ Pj) of a first predefined type ; • sending a prevention message (MP) if a plurality of sequences of detection results (SEQ D1, SEQ D2, ... SEQ D m) received from said sensors (Cj) indicate a variation of the personalized sequence (SEQ Pj) of a second predefined type. 11. A computer program comprising instructions for the execution of the steps of the quantification method according to any one of claims 1 to 10 when said program is executed by a computer. 12. Central unit (UC) for a system for monitoring (SYS) the activity of a person (P) operating in a defined environment (ENV), in which is arranged at least one sensor (Cj) connected to said central unit ( UC), including: Means for obtaining (COM, CPU, MEM) a generic sequence (SEQj) of results capable of being detected by said at least one sensor (Cj), said generic sequence (SEQj) being defined by at least one parameter (mj, nj); • means of acquisition (RX) of first detection results (r1, r2, ..rk) received from said sensors (Cj), • means of determination (CPU, MEM, TEMPO), according to first results of detection (r1, r2, ..., rk), o of a personalized value (vmj p, vnj p), of said at least one parameter (mj, nj) defining said generic sequence (SEQj) for the person (P) operating in the delimited environment (ENV), said at least one parameter (mj, nj) being designated as personalized parameter (mj_p, nj_p) when it takes the personalized value (vmj p, vnj p), and o of a personalized sequence (SEQ Pj), resulting from said generic sequence (SEQj), when it is defined by said at least one personalized parameter (mj_p, njjo), • acquisition means (RX) of second detection results (R1 , R2, ..Rk) received from said sensors (Cj), • search means (CPU, MEM, TEMPO) of said personalized sequence (SE Q Pj) among said second detection results (R1, R2, Rk); • anomaly notification means (TX) to a user, the central unit being configured to notify a user if said personalized sequence (SEQ Pj) is not found during a predetermined time interval (T_p).