Method for selecting sensors from a set of sensors

Abstract

The present invention relates to a method for selecting sensors from a set of sensors present in an environment, the selection method comprising, upon detecting a certain behaviour of a living being by means of sensors selected prior to behaviour detection: - modifying the selection of the selected sensors used by the behaviour detection, the selection of the sensors being modified according to a confidence score relating to the behaviour detection and the nature of the sensors.

Description

Method for selecting sensors from a set of sensors

[0001] The present invention relates to the field of sensor selection in an environment. The present invention applies in particular, but not limited to, the detection of the behavior of a living being in an environment, in order to be able to provide suitable services.

[0002] The invention is situated in the context of understanding human behavior, in particular the prediction and detection of activities and actions or the location and identification of an occupant of a house, a building, a factory, based on different types of sensors. There are currently methods for detecting human activities based on the combination of ambient sensors and behavioral understanding systems based on sound and visual modalities placed in the user's room, combining the sensors or not. There are also behavioral understanding systems based on modalities from equipment worn by the user such as smartwatches and augmented reality headsets. Generally, the data captured by these different modalities are complementary for understanding human behavior.There is therefore work that seeks to combine different types of modalities to improve the performance of human behavior recognition. For example, video, audio, and physiological data (worn sensors) are used to identify human emotions. However, the applicant is interested on the one hand in improving the protection of users with regard to their personal data and on the other hand in reducing the carbon footprint of its products and services and the present invention aims to propose a solution that can improve the behavior of a living being, while taking these two parameters into account.

[0003] The present invention relates, according to a first aspect, to a method for selecting sensors from a set of sensors present in an environment, the selection method comprising, following a detection of a behavior of a living being by means of sensors selected prior to a behavior detection: - a modification of selection of the selected sensors used by the behavior detection, the modification of selection of the sensors being a function of a confidence score relating to the behavior detection and of a character of the sensors.

[0004] According to a second aspect, the present invention relates to a method for recognizing the behavior of a living being in an environment comprising a plurality of sensors, the recognition method comprising:- a detection of a behavior of a living being based on data received from sensors selected prior to the detection;- a modification of the selection of the selected sensors used by the behavior detection, the modification of the selection of the sensors being a function of a confidence score relating to the behavior detection and of a character of the sensors.

[0005] In some embodiments, a modified subset of selected sensors provided by the selection modification includes the selected sensors used in behavior detection subsequent to the selection modification.

[0006] According to certain embodiments, said selection modification is iterated as long as the confidence score differs from a determined threshold. In particular, the selection modification is carried out as long as the confidence score is lower than a threshold to provide a subset of selected sensors making it possible to achieve a minimum confidence score for the detection of behavior of a living being. And possibly, in a complementary or alternative manner, the selection modification is carried out as long as the confidence score is higher than a threshold to reduce the disadvantages linked to a characteristic of the sensors, for example to reduce the intrusiveness of the sensors, when detecting the behavior of a living being while using a subset of selected sensors making it possible to achieve an optimal confidence score for the detection of behavior of a living being.

[0007] In some embodiments, the selection modification relating to a sensor is a function of a level of a character of the sensor.

[0008] According to certain embodiments, the modification of selection of the activated sensors comprises an operation relating to at least one sensor among the following operations: - an addition to the selected sensors of a sensor among the set of sensors present in the environment, - a removal of a sensor from the selected sensors, - a replacement of a sensor from the selected sensors by a sensor among the set of sensors present in the environment.

[0009] In some embodiments, when the selection change includes removing a sensor from the selected sensors, the selection change triggers a deactivation of the sensor.

[0010] According to some embodiments, the characteristic on which the sensor selection modification is a function is a sensor intrusiveness characteristic.

[0011] According to certain embodiments, the selection modification comprises an operation relating to a sensor depending on the level of the intrusiveness of the sensor and the nature of the operation among the following: - an addition of a low intrusiveness level sensor; - a removal of a high intrusiveness level sensor; - a replacement of a high intrusiveness level sensor by a low intrusiveness level sensor.

[0012] According to certain embodiments, said confidence score relating to the detection of said behavior is determined from one or more of: - a quality indicator of a behavior recognition model used during said behavior detection, - an evaluation of a history of data of detected behaviors - an evaluation of data relating to said environment, - a reception of at least one piece of information from a living being present in said environment.

[0013] According to certain embodiments, some of the set of said sensors are associated with an intrusive nature, said selection of activated sensors selecting in priority the least intrusive sensors.

[0014] According to certain embodiments, when said behavior is detected with a confidence score greater than a determined threshold, - a deactivation of at least one sensor among the activated sensors if - the function associated with said deactivated sensor is not used to detect said behavior or - the power consumption associated with the sensor is high, or - the level of intrusiveness of the sensor is high.

[0015] According to certain embodiments, the selection method comprises - obtaining, for at least some of said sensors, a level relating to its intrusive nature, said obtaining being received from a user or determined from the function associated with said sensor.

[0016] According to certain embodiments, the method for recognizing a behavior of a living being in an environment comprising a plurality of sensors associated with one or more functions, one or more sensors being activated, said recognition method comprising: a) - an execution of a model for detecting a behavior of said living being from data received from said activated sensors, b) - a modification of a selection of the activated sensors, according to a confidence score obtained following the execution of the model, the confidence score relating to a relevance of said detection, c) - an execution of said model for detecting a behavior of said living being from data from the sensors of said modified selection, steps b) and c) being iterated one or more times according to said confidence score.

[0017] The features presented in isolation in the present application in connection with certain embodiments of the method of the present application can be combined with each other according to other embodiments of the present method.

[0018] According to another aspect, the present invention also relates to a computer program comprising instructions for executing the steps of the selection method and / or the recognition method according to the invention, according to any one of the embodiments, when said program is executed by a computer.

[0019] According to another aspect, the present invention also relates to a computer-readable recording medium on which is recorded a computer program comprising instructions for carrying out the steps of the method according to the invention, according to any of its embodiments.

[0020] According to another aspect, the present invention also relates to a sensor selection device for executing the steps of the selection method according to the invention, according to any one of its embodiments. Thus, the present invention relates to a sensor selection device from a set of sensors present in an environment, the selection device comprising one or more processors configured to, following detection of a behavior of a living being by means of sensors selected prior to behavior detection: - modify the selection of the selected sensors used by the behavior detection, the modification of the selection of the sensors being a function of a confidence score relating to the behavior detection and a character of the sensors.

[0021] According to another aspect, the present invention relates to a device for recognizing the behavior of a living being in an environment comprising a plurality of sensors, the recognition device comprising:- a communication device configured to communicate with sensors present in an environment, and- one or more processors configured to:+ detect a behavior of a living being based on data received from sensors selected prior to the detection by means of the communication device;+ modify the selection of the selected sensors used by the behavior detection, the modification of the selection of the sensors being a function of a confidence score relating to the behavior detection and of a character of the sensors.

[0022] According to another aspect, the present invention relates to a system for detecting the behavior of a living being comprising:- one or more connected sensors comprising at least a subset of activated sensors,- a communication device configured to communicate with said sensors, and comprising one or more processors configured to- execute a model for detecting a behavior of said living being from data received from said activated sensors,- modify a selection of the activated sensors, according to a confidence score obtained following the execution of the model, the confidence score relating to a relevance of said detection,- execute said detection model from the data from the sensors of said modified selection, according to one or more iterations, the number of iterations being a function of said confidence score obtained following the iteration, the selection of the activated sensors being modified during each iteration.

[0023] Other characteristics and advantages of the present invention will emerge from the description given below, with reference to the accompanying drawings which illustrate an exemplary embodiment thereof without any limiting character.

[0024] Represents a system according to certain embodiments of the invention,

[0025] Represents a method according to certain embodiments of the invention,

[0026] The figure represents a device according to certain embodiments of the invention.

[0027] This disclosure may apply to the detection of behavior in an environment and in particular to human or animal behavior, more generally to the behavior of a living being.

[0028] By environment, we can mean a domestic environment, an industrial environment, a professional environment, an environment in a vehicle such as a car and more generally any type of environment where it may be relevant to detect human behavior.

[0029] Human behavior detection can be applied in many areas. These applications include providing behavior-based services, assisting technicians with maintenance operations by providing them with real-time displays of relevant information, reducing energy consumption, and managing home automation.

[0030] To do this, many devices such as sensors are now present in the environments mentioned above. Among these sensors, we can cite, but not limited to: - ambient sensors such as door and window opening devices, volumetric sensors, surveillance cameras, microphones - sensors worn by the user, such as smart watches, augmented reality headsets, medical monitoring devices such as alert systems (connected bracelets for example), motion or fall sensors.

[0031] The sensors considered in the present disclosure are connected sensors, that is to say, whose state, operation, start-up, standby, stop, can be controlled remotely by one or more other devices. Such control devices can, for example, be access gateways controlling a network to which these sensors are connected. The control of these connected sensors can also be carried out by software present at network operators, for example, or at service providers. In such a context, the gateway controlling the local network to which the sensors are connected can constitute an interface between the local network and an external network, of the WAN (for "wide area network") type, such as a cellular network.

[0032] The represents an example system implementing the present disclosure.

[0033] An access gateway P1 controls at least one local network R1. The local network R1 is an Ethernet-type network, for example, and may be a wired or wireless network. According to certain embodiments, certain sensors may be connected by a protocol other than the Ethernet protocol, in particular WIFI, and may, for example, be connected via Bluetooth, Zigbee, Z-Wave, LoRaWAN, NFC, Thread, Sigfox, or even cellular technology such as 5G, 4G, or 3G. The choice of network type may depend on the specific requirements of the application, such as range, power consumption, bandwidth, and the availability of existing communication infrastructures. Combinations of several of these technologies are often used to meet the complex needs of connected smart sensor networks such as IoT (Internet of Things) sensors.

[0034] Sensors C1, C2, C3 are connected to the gateway P1 via the R1 network, here a WIFI network. Sensors C4 and C5 are connected to the gateway P1 via a Bluetooth network. Sensors C6 and C7 are connected to the gateway P1 via a Zigbee network.

[0035] The P1 gateway can also be connected to an R2 network and therefore serve as an interface between the Ci sensors or other equipment connected to the P1 gateway (not shown, such as TVs, multimedia decoders, etc.), and an external network such as a cellular network. This is the case when the access gateway is an access gateway provided by a telecommunications operator, the gateway allowing the devices in the environment to be connected to the operator's IR network infrastructure and to access the internet network, for example. The R2 network can be a cellular network.

[0036] According to some embodiments, additional sensors, C i , not shown, are connected to the P1 gateway via the R2 network.

[0037] C sensors i can be characterized by different parameters. They are notably associated with a function. By function, we can understand, in an illustrative and non-exhaustive manner: - image capture (or video capture), - sound capture (or audio capture), - motion capture (presence detector for example), - opening capture (door or window opening), - fall capture, - activity capture, - position capture, - emotion capture.

[0038] Sensors can be sensors worn by a user, such as smartwatches or augmented reality headsets, ambient sensors, or from behavioral understanding systems based on sound and visual modalities placed in the room where the user is located.

[0039] One or more functions can be associated with a single sensor. For example, a smartwatch can be associated with position detection, activity detection, or emotion detection.

[0040] C sensors i can also be characterized by their intrusive nature. Thus, a sensor can be associated with a level of intrusiveness.

[0041] According to certain embodiments, this level is binary and can be equal to “1” if the sensor is considered to be intrusive (or if its function is intrusive) and equal to “0” if the sensor is considered to be non-intrusive.

[0042] In some embodiments, this level is not binary but may be defined on a scale, for example from 0 to 10, with "0" representing a non-intrusive level and "10" representing a very intrusive level.

[0043] For example, a camera may be considered to be highly intrusive and therefore either have a level equal to "1" in a binary system or equal to "10" in a level system comprising a scale from 0 to 10 as mentioned above.

[0044] According to this same example, a door opening detector can be considered as having a non-intrusive character and therefore either have a level equal to "0" in a binary system or equal to "0" in a level system comprising a scale from 0 to 10 as mentioned above.

[0045] The intrusiveness of each sensor is a property that can be defined a priori by the model, or that can be indicated by the user via a configuration means (for example a user interface on a network device or on the gateway), or that can be derived from a database, or that can be any combination of these different possibilities. In particular, it is conceivable that the user can indicate that a sensor is never used by the model (and therefore has infinite intrusiveness), or that a sensor is always used by the model (and therefore has zero intrusiveness). This indication by the user can be given prior to the implementation of the method, for example when connecting the sensors to the gateway or in real time during the initial or subsequent selection of the sensors, for example by a “dialogue” with the user on a mobile phone or in an augmented reality headset.

[0046] C sensors i can also be characterized by their energy consumption. Indeed, the multiplicity of devices such as C sensors i contributes to the increase in the carbon footprint of the environments in which they are used and it is therefore desirable to reduce their energy consumption, while benefiting from their function.

[0047] This disclosure helps to control the use of these sensors in a relevant manner, reducing their intrusiveness and energy consumption.

[0048] The present invention can thus be used to recognize the behavior of a living being, while using the available sensors in a relevant manner, that is to say while controlling the intrusive side provided by these sensors and the energy consumption of these sensors.

[0049] In the following description, we will take a human being as an example of a living being, but this application can also be applied to an animal.

[0050] The gateway P1 can thus implement a method as described later in the. The method can, in other embodiments, be implemented remotely, in the infrastructure of a network operator or at a service provider, these being connected to the gateway P1 via the network R2. The gateway P1 can then transmit the data from the sensors C i , in real time, to remote devices on the R2 network.

[0051] The system relates to a system for detecting the behavior of a living being comprising:- one or more connected sensors comprising at least a subset of activated sensors,- a communication device configured to communicate with said sensors, and comprising one or more processors configured to- execute a model for detecting a behavior of said living being from data received from said activated sensors,- modify a selection of the activated sensors, according to a confidence score obtained following the execution of the model, the confidence score relating to a relevance of said detection,- execute said detection model from the data from the sensors of said modified selection, according to one or more iterations, the number of iterations being a function of said confidence score obtained following the iteration, the selection of the activated sensors being modified during each iteration.

[0052] The depicts a method according to certain embodiments of the invention.

[0053] As mentioned above, the method as described below can be implemented by the access gateway P1 or in remote infrastructures or servers connected to the gateway P1 or in a combination of both.

[0054] The gateway P1 implements a method for selecting sensors from a set of sensors present in an environment and associated with one or more functions, one or more sensors being activated to enable detection of a behavior of a living being, the method comprising: - a modification of the selection of the activated sensors from the set of sensors as a function of a confidence score relating to said detection of said behavior, said modification being carried out one or more times as a function of the confidence score.

[0055] The gateway P1 can also implement a method for recognizing a behavior of a living being in an environment comprising a plurality of sensors associated with one or more functions, one or more sensors being activated, said method comprising: a) - an execution of a model for detecting a behavior of said living being from data received from said activated sensors, b) - a modification of a selection of the activated sensors, according to a confidence score obtained following the execution of the model, the confidence score relating to a relevance of said detection, c) - an execution of said model for detecting a behavior of said living being from data from the sensors of said modified selection, steps b) and c) being iterated one or more times according to said confidence score.

[0056] In the environment of the, a plurality of sensors C iare installed, or are in operational mode, in the P1 gateway environment, i.e. they can communicate as mentioned above with the P1 gateway according to different communication protocols and are identified by the P1 gateway.

[0057] According to certain embodiments, during a step E1, the method as described can be triggered permanently, that is to say it can always be active.

[0058] According to some embodiments, the method can be activated on demand, for example at the request of an energy supplier, or any service provider, or at the request of the user or owner of the access gateway P1.

[0059] One of the objectives or applications of implementing this method is to recognize the behavior of a user. Thus, according to certain embodiments, the present method can be triggered when a behavior recognition request is requested or activated. This request can be made by a service or energy provider or any other request.

[0060] A selection of activated sensors may have been previously carried out or may be carried out during a step E2. By selection of activated sensors, we mean more precisely an activation of a selection of sensors. By activation, we mean the switching on, the fact of making a sensor operational and / or in operation, so that the function associated with it is activated and can be detected by the access gateway P1.

[0061] This first selection of sensors, or initial selection of sensors, may include an SE subset j C sensors i . The SE subset j can understand all C sensors i but more particularly includes part of the C sensors i . This selection of sensors from the initial subset can be done in different ways:- by a selection by the user based on his preferences and the use he makes of the sensors in his environment. Indeed, the sensors C ican be used by the user of the access gateway for applications other than behavior recognition, and may therefore have been selected for these applications, - by an initial selection dedicated to behavior recognition and taking into account one or more parameters. Among these parameters, one can select a sensor per function (for example, a single camera, a single microphone, a single opening detector). One can also select a sensor according to the level of its intrusiveness, favoring sensors whose intrusiveness level is the lowest. For example, one can select sensors whose intrusiveness level is lower than a given level. One can also select sensors according to their energy consumption. For example, one can select sensors whose energy consumption is lower than a given level. Of course, one can combine these criteria.Sensors can also be selected based on their geographical location within the P1 access gateway environment. Indeed, the P1 gateway may be located in an environment with several rooms (this is the case in a domestic or industrial environment), and thus the sensors initially activated may be those of the room where a user is most frequently located.

[0062] A user interface, on the gateway P1, or on a control device such as a computer connected to the network R1, may for example allow the user to select the sensors to be activated, the gateway or the device then being able to activate the selected sensors. Alternatively, the user may also activate the sensors one by one, by actuating them manually.

[0063] According to some embodiments, a home automation system may also be present in the environment where the gateway P1 is located and the home automation system may have selected a set of sensors from among the sensors Ci.

[0064] According to the present disclosure, an activated sensor is a sensor that is either active or in standby mode, but if in standby mode, it is triggered when it detects an action. For example, surveillance cameras may be in standby mode but are automatically activated upon an intrusion. Thus, they are active within the meaning of the present disclosure.

[0065] A non-activated sensor is a sensor that does not transmit data, in other words, it is not configured to transmit data, for example, it is switched off (in the electrical sense) or it is in sleep mode but does not activate when data is detected, or when data changes, so to speak, it does not automatically switch from sleep mode to data capture mode.

[0066] We therefore place ourselves in the context where a request for behavior detection is received. This request for behavior detection can be the triggering of the process as mentioned previously or can be received at the end of step E2. Step E2 advantageously allows us to benefit from an initial selection of activated sensors.

[0067] In a step E3, a behavior detection of at least one user present in the environment of the gateway P1 is carried out from the data provided by the sensors of the initial selection of sensors. A human behavior detection model can allow the recognition of human behavior, from the data from the activated sensors and algorithms.

[0068] The data provided by sensors is heterogeneous and may depend on the function associated with the sensor. A door opening sensor can transmit binary information, indicating the status of the door, either open or closed. A sensor accelerometer worn by an individual transmits data relating to a three-dimensional acceleration vector "(x,y,z). A camera transmits an image stream with a format that depends on the camera, a microphone transmits an audio stream, etc.

[0069] Data can be pre-processed before being used by the behavior detection model. This includes normalization, time synchronization, and concatenated data. Pre-processing can also include detecting the segmentation of a camera image or extracting words from a sound stream.

[0070] The behavior detection model used is configured to process data from sensors of different nature or is configured to process multimodal situations. Machine learning, neural network or transformer based approaches can be used in such models. The paper entitled "Accommodating missing modalities in time-continuous multimodal emotion recognition" by Rodrigues, Lefebvre, Cumin and Crowley, published on November 16, 2023 (https: / arxiv.org / pdf / 2311.10119.pdf) describes a method for recognizing a person's emotions from video, audio and physiological data.

[0071] In step E4, the detected behavior is evaluated, or in other words, the quality of the detected behavior is evaluated. This evaluation can determine whether or not the detected behavior is close to the actual behavior of the user whose behavior was performed.

[0072] This assessment may, for example, result in the determination and assignment of a confidence score for the detected behavior.

[0073] When the sensors selected and used for detection do not provide reliable detection of behavior, then the confidence score associated with the detection is low. Conversely, when the sensors selected and used for detection provide reliable detection of behavior, then the confidence score associated with the detection is high.

[0074] The quality assessment, or the determination of a confidence score Sc, can be determined from one or more of: - a quality indicator of the behavior recognition model used during behavior detection, - an evaluation of a history of detected behavior data - an evaluation of data relating to the environment. This data can for example be data relating to the room (or the place in an industrial or other environment) where the user is located. If the behavior is compatible with the room in which the user is located, then the quality can be high. - a reception of at least one piece of information from a human being present in the environment. This information can for example confirm or deny that the detected behavior is indeed consistent with the actual behavior of the user.

[0075] At the end of step E4, we can therefore obtain information relating to the quality of the behavior detection.

[0076] For example, we can set a confidence score threshold S1: - if the confidence score is lower than S1, then the detected behavior is not of good quality, - if the confidence score is higher than or equal to the threshold S1, then the detected behavior is of good quality. For example, we can choose S1 equal to 5 on a confidence score scale of 1 to 10.

[0077] Depending on the quality obtained, the set of sensors used to determine the behavior can be modified.

[0078] Thus, the selection of activated sensors from the set of sensors is modified based on a confidence score relating to the detection of the behavior, the modification being carried out one or more times depending on the confidence score. These modification steps are carried out during steps E5 and E6 described below. These modifications may be an addition or a removal, or both, of activated sensors.

[0079] If, at the end of step E4, the behavior detection is evaluated as being of good quality, then we move on to step E5. During step E5, we modify the set of activated sensors so as to delete, from the set of sensors activated and used during the detection of the behavior by the model, at least one sensor according to one or the other or several of: - the function associated with the deleted sensor is not related to the detected behavior, - the power consumption associated with the sensor is high, - the level of intrusiveness of the sensor is high.

[0080] In other words, removing one or more sensors from the set of activated sensors consists of deactivating at least one sensor among the activated sensors.

[0081] Deactivation may include sending messages or commands to the removed sensors to deactivate them, such as commands to sleep, or commands to turn them off.

[0082] We can prioritize removing the most intrusive sensors, as long as the confidence score is higher than the S1 threshold. We can also prioritize removing the most energy-consuming sensors.

[0083] In step E5, one can also replace a sensor with another sensor rather than removing a sensor. Thus, one can keep the same number of sensors but replace a very intrusive sensor with a less intrusive sensor or replace a very energy-consuming sensor with a less energy-consuming sensor, while maintaining a confidence score higher than the threshold S1 and possibly lower than the score obtained with the initial set of sensors or with the set of sensors in a subsequent iteration.

[0084] More specifically, and for example, when the initial selection includes two cameras, located in two different rooms, for example the living room and the kitchen, it may be unnecessary to keep both cameras active if the detected user behavior is cooking. The camera located in the living room can be disabled.

[0085] As another example, when the initial selection includes a camera and a microphone, located in the same room, it may not be necessary to keep the microphone active if the detected user behavior is working on a computer. The microphone can then be disabled and thus removed from the list of active sensors.

[0086] As another example, when two devices can detect the same behavior, for example if there are two cameras, the more energy-consuming camera can be removed from the list of active sensors.

[0087] According to another example, if the detected behavior is that the user is having a telephone conversation, then if the initial list of sensors included at least one camera and one microphone, the camera can be deactivated and the microphone retained, taking into account the energy-consuming nature of the devices, a camera consuming more than a microphone. According to another embodiment, if the non-intrusive nature is taken as a priority, the microphone can be deactivated (to prevent the telephone conversation from being heard) and the camera can be kept active. It can be noted that the intrusive nature can be related to the detected function.

[0088] The following three criteria can be ranked in order of priority:- the function associated with the removed sensor is not related to the detected behavior,- the power consumption associated with the sensor is high,- the level of intrusiveness of the sensor is high.

[0089] This priority order can be used to select certain sensors over others.

[0090] Following the deletion, or replacement, of one or more sensors from the list of activated sensors, the behavior determination model may be executed again. One or more iterations of steps E3 to E5 may be necessary to refine the list of activated sensors giving a confidence score greater than or equal to the threshold S1. In particular, one or more iterations may be implemented to modify the selection of sensors, either to reduce the intrusiveness level of the selected sensors, or to reduce the power consumption of the selected sensors, or to reduce the number of sensors. The above order of priority may be used to determine which sensor to prioritize or which type of sensor to prioritize.

[0091] According to some embodiments, a level of intrusiveness is associated with at least some of the set of said sensors, said selection of activated sensors prioritizing the least intrusive sensors.

[0092] According to certain embodiments, an energy consumption level is associated with at least some of the set of said sensors, said selection of activated sensors selecting in priority the least energy-consuming sensors.

[0093] In some embodiments, said modification of sensor selection includes adding, removing, or replacing sensors based on detected behavior, and based on either a level associated with the intrusiveness of a sensor or a level of power consumption associated with a sensor.

[0094] As illustrated in, following step E5, a measurement test of an overall intrusive level NivI of all the selected sensors and / or a measurement test of an overall level NivE of electrical consumption of the selected sensors can be carried out. Depending on the result of these tests, it can be determined whether to repeat steps E3 to E5.

[0095] The overall power consumption level NivE or the overall intrusive level NivI can be calculated respectively by adding the power consumption levels of the selected sensors or by adding the intrusive levels of the selected sensors.

[0096] Alternatively, one may also wish to perform one or more iterations to reduce the number of sensors selected while maintaining an equivalent overall power consumption level or an equivalent overall intrusive level.

[0097] Changing the selection of activated sensors can be done as long as the confidence score is higher than the determined threshold S1. Thus, a lower confidence score, but higher than S1, can be obtained in subsequent iterations with a reduced number of sensors, or a lower number of intrusive sensors, or a higher number of less energy-consuming sensors.

[0098] Thus, we can return to step E3 using real-time data. In order to avoid replaying the model on the actual data observed during step E3, the data from the sensors and transmitted to (or retrieved by) the model during step E3 can be saved and reused during subsequent iterations of the model.

[0099] In other words, step E5 can be iterated with the updated set of sensors as long as the confidence score is greater than the threshold S1. When the confidence score reaches the threshold S1, according to certain embodiments, iteration can continue to be carried out: to reduce the intrusiveness of the selected sensors by replacing certain sensors with other less intrusive sensors or to reduce the energy-consuming nature of the selected sensors by replacing certain sensors with other less energy-consuming sensors.

[0100] Indeed, even if the selection of sensors, to remove them, during step E5, can be carried out from one iteration to another by first selecting the most intrusive sensors then the less and less intrusive sensors, additional iterations can make it possible to reduce the energy footprint by keeping sensors having the same level of intrusiveness but a lower energy footprint.

[0101] In the same way, if the selection of sensors, to remove some during step E5, can be carried out from one iteration to another by first selecting the most energy-consuming sensors then the less and less energy-consuming sensors, additional iterations can make it possible to reduce the intrusive side by keeping sensors having the same energy footprint but a lower level of intrusiveness.

[0102] Similarly as mentioned previously alternatively one may want to reduce the number of sensors involved.

[0103] At the end of step E5, when a score of either a correct overall intrusive level, or a correct overall energy consumption level, or a minimum number of sensors is reached, we move on to step E7 in which we can record the list of sensors associated with the detected behavior. This can advantageously allow the activation of sensors corresponding to the detection of a determined desired behavior. Thus, we obtain a connection or correspondence of a set of activated sensors with a behavior, usable for example by an application of a service provider, or of a telecommunications operator.

[0104] If, at the end of step E4, the behavior detection is evaluated as being of poor quality, then we move on to step E6. During step E6, we modify the set of activated sensors so as to add one or more sensors to increase the confidence score and reach a confidence score at least equal to the threshold S1. During step E6, we modify the set of activated sensors so as to add to the set of sensors activated and used during the detection of the behavior by the model, at least one sensor according to one or the other or several of: - the function associated with the sensor, - the power consumption associated with the sensor, - the level of intrusiveness of the sensor.

[0105] You can add one or more sensors simultaneously or one by one.

[0106] In step E6, in addition to adding sensors to improve the confidence score, one sensor can also be replaced by another sensor in order to improve the intrusiveness of all the sensors or in order to improve the energy performance of all the sensors. This replacement can be carried out following the addition of sensors or simultaneously. In fact, it is possible, initially, to select the sensors to obtain a confidence score greater than or equal to S1, for example by taking less energy-consuming or less intrusive sensors first and in a second step, for example during subsequent iterations, to replace some of the selected sensors with sensors, for example having the same function, but a lower intrusiveness or a lower energy-consuming character.

[0107] Specifically, and for example, if the model detects behavior consistent with a person being in the kitchen, for example, using volumetric sensors enabled, the model may indicate a confidence score of 4 for behavior consistent with a person eating. An additional sensor, such as a microphone, may then be enabled to determine whether the person is actually eating or cooking. A microphone may be selected over a camera if it is considered that less intrusive sensors are preferred and a microphone is classified as a less intrusive sensor than a camera.

[0108] According to other embodiments, one or more added sensors may be added based on their energy consumption or based on their function and based on the detected behavior with a confidence score lower than S1. One aim being to improve the confidence score of the detected behavior, the added sensor(s) are selected based on the assumed behavior (assumed because the confidence score associated with the detected behavior is low), to allow above all to increase this score and among the sensors possibly allowing to increase this score, a selection of the least intrusive or least energy-consuming sensors is carried out.

[0109] Following the addition, or replacement, of one or more sensors from the list of activated sensors, during step E6, the behavior determination model can be executed again. Thus, one can return to step E3 using real-time data. One or more iterations of steps E3 to E6 may be necessary to refine the list of activated sensors giving a confidence score greater than or equal to the threshold S1. The modification of the selection of activated sensors can be carried out as long as the confidence score is greater than the determined threshold S1.

[0110] In order to avoid replaying the model on the real data observed during step E3, the data from the sensors and transmitted to (or recovered by) the model during step E3 can be saved and reused during subsequent iterations of the model.

[0111] Step E6 is iterated with the updated set of sensors as long as the confidence score is lower than the threshold S1. When the confidence score reaches the threshold S1, according to certain embodiments, iteration can continue to be carried out: to reduce the intrusiveness of the selected sensors by replacing certain sensors with other less intrusive sensors or to reduce the energy-consuming nature of the selected sensors by replacing certain sensors with other less energy-consuming sensors.

[0112] Indeed, even if the selection of sensors, during step E6, can be carried out from one iteration to another by first selecting the least intrusive sensors then the increasingly intrusive sensors, additional iterations can make it possible to reduce the energy footprint by keeping sensors having the same level of intrusiveness but a lower energy footprint.

[0113] In the same way, if the selection of sensors, during step E6, can be carried out from one iteration to another by first selecting the least energy-consuming sensors then the increasingly energy-consuming sensors, additional iterations can make it possible to reduce the intrusive side by keeping sensors having the same energy footprint but a lower intrusive level.

[0114] At the end of a plurality of iterations of the loop E3-E6, the confidence level reaches a confidence level greater than or equal to S1. We then move on to step E5 described previously.

[0115] At the end of the method, all the sensors making it possible to obtain a detection of a behavior can be recorded, step E7. Thus, it is possible, for example, to associate, with a behavior to be detected, a set of sensors, in the environment considered, here in the environment of the gateway P1.

[0116] This information linking the selection of sensors allowing behavior detection and minimizing intrusiveness or energy consumption can be applied in the field of smart homes, understanding occupant behavior and can provide adapted services in real time, such as, for example, redirecting a call to messaging, optimizing the energy consumption of home appliances, dynamic management of lighting and shutters, displaying information on a screen, etc.

[0117] In the field of augmented technicians, understanding behavior can make it possible to display the information they need in real time in augmented reality: the user guide for a tool, an assembly procedure, a route to a particular area of ​​a factory.

[0118] In these different areas, the confidentiality criterion is important and the adaptive aspect of this disclosure can make it possible to comply with users' expectations by giving them control and thus giving them increased confidence in the system.

[0119] La represents an example of at least one part of hardware architecture 30 of the gateway P1, allowing the implementation of a method according to the present invention and as represented for example in. This hardware architecture is that of a computer. Other hardware architecture elements are present in the gateway P1 and not represented here.

[0120] The hardware architecture 30 comprises one or more processors 31 (only one is shown in the) implementing a method according to the present disclosure, a read-only memory 32 (of the “ROM” type), a rewritable non-volatile memory 33 (of the “EEPROM” or “NAND Flash” type for example), a rewritable volatile memory 34 (of the “RAM” type) a communication interface 35 with the gateway 2. The read-only memory 32 constitutes a recording medium in accordance with an exemplary embodiment of the invention, readable by the processor or processors 31 and on which is recorded a computer program Prog in accordance with an exemplary embodiment of the invention comprising instructions for executing steps of the sensor selection method according to the invention. Alternatively, the computer program Prog is stored in the rewritable non-volatile memory 33.

[0121] The computer program Prog may enable the gateway P1 to implement at least part of the method in accordance with the present disclosure and as illustrated for example in.

[0122] This computer program Prog can thus define functional and software modules, configured to implement the steps of a selection method in accordance with an exemplary embodiment of the invention, or at least part of these steps. These functional modules rely on or control the hardware elements 31, 32, 33, 34, 35 of the access gateway P1 cited above.

Claims

Method for selecting sensors from a set of sensors present in an environment, the selection method comprising, following detection of a behavior of a living being by means of sensors selected prior to behavior detection: - a modification of selection of the selected sensors used by the behavior detection, the modification of selection of the sensors being a function of a confidence score relating to the behavior detection and of a character of the sensors. The selection method of claim 1 wherein a modified subset of selected sensors provided by the selection modification comprises the selected sensors used in behavior detection subsequent to the selection modification. Selection method according to one of the preceding claims in which the selection modification is iterated as long as the confidence score differs from a determined threshold. Selection method according to one of the preceding claims in which the selection modification relating to a sensor is a function of a level of a character of the sensor. Selection method according to one of the preceding claims in which the modification of selection of the sensors comprises an operation relating to at least one sensor among the following operations: - an addition to the selected sensors of a sensor from the set of sensors present in the environment, - a removal of a sensor from the selected sensors, - a replacement of a sensor from the selected sensors by a sensor from the set of sensors present in the environment. Selection method according to one of the preceding claims wherein, when the selection modification comprises a removal of a sensor from the selected sensors, the selection modification triggers a deactivation of the sensor. Selection method according to one of the preceding claims in which the character on which the modification of selection of the sensors is a function is a character of intrusiveness of the sensors. Selection method according to the preceding claim, in which the selection modification comprises an operation relating to a sensor depending on the level of intrusiveness of the sensor and the nature of the operation among the following: - an addition of a low intrusiveness level sensor; - a removal of a high intrusiveness level sensor; - a replacement of a high intrusiveness level sensor by a low intrusiveness level sensor. Selection method according to one of the preceding claims in which said confidence score relating to the detection of the behavior is determined from one or the other or more of:- a quality indicator of a behavior recognition model used during the detection of behavior,- an evaluation of a history of data of detected behaviors- an evaluation of data relating to the environment,- a reception of at least one piece of information from a living being present in said environment. Method for recognizing the behavior of a living being in an environment comprising a plurality of sensors, the recognition method comprising:- a detection of a behavior of a living being based on data received from sensors selected prior to the detection;- a modification of the selection of the selected sensors used by the behavior detection, the modification of the selection of the sensors being a function of a confidence score relating to the behavior detection and of a character of the sensors. Computer program comprising instructions for executing the steps of the selection method according to one of claims 1 to 9 and / or of the recognition method according to claim 10 when said program is executed by a computer. Device for selecting sensors from a set of sensors present in an environment, the selection device comprising one or more processors configured to, following detection of behavior of a living being by means of sensors selected prior to behavior detection: - modify the selection of the selected sensors used by the behavior detection, the modification of the selection of the sensors being a function of a confidence score relating to the behavior detection and of a character of the sensors. Selection device according to the preceding claim in which the processor is configured so that modifying the selection of a sensor modifies the activation state of the sensor. Device for recognizing the behavior of a living being in an environment comprising a plurality of sensors, the recognition device comprising:- a communication device configured to communicate with sensors present in an environment, and- one or more processors configured to:+ detect a behavior of a living being based on data received from sensors selected prior to detection by means of the communication device;+ modify the selection of the selected sensors used by the behavior detection, the modification of the selection of the sensors being a function of a confidence score relating to the behavior detection and of a character of the sensors.

Images