IT201800021418A1 - METHOD OF CONTROL OF EQUIPMENT FOR GENERATION AND CONVEYANCE OF AN AIR FLOW ON THE BASIS OF IMAGE PROCESSING AND SYSTEM INCLUDING THE EQUIPMENT - Google Patents

Description

METHOD OF CONTROL OF EQUIPMENT FOR GENERATION AND CONVEYANCE OF AN AIR FLOW BASED ON IMAGE PROCESSING AND SYSTEM INCLUDING THE EQUIPMENT.

DESCRIPTION

TECHNOLOGICAL BACKGROUND OF THE INVENTION

Field of application

The present invention refers in general to the sector of equipment for generating and conveying air flows configured to improve the living comfort of an environment by one or more users. Such equipment includes, for example, fans, air conditioners, air purifiers, air vents, fan heaters, ionizers, humidifiers and the like.

In particular, the invention relates to a method of controlling an apparatus for generating and conveying an air flow on the basis of image processing through the use of convolutional neural networks and to a relative system which includes the the aforementioned equipment.

Known art

Fans configured to operate autonomously, i.e. without direct guidance or control by a user, are now commercially available for residential use.

Some known fans include a timer and are configured to independently activate or deactivate based on a time set by the user. Other fans include temperature sensors and can activate or deactivate based on the temperature detected in the environment in which they are located.

Also known are fans with an air purification function configured to independently adjust the intensity of the air flow based on the purity of the air revealed.

Some known fans are also provided with remote control devices, for example a remote control, which, in addition to allowing a user to remotely activate or deactivate an air flow, allow to regulate the conveyance of the air flow in a specific direction, the strength of the air flow and the respective temperature.

Furthermore, fans equipped with devices for detecting the presence of one or more users within an environment are known. These fans are equipped with appropriate presence detection sensors, for example infrared, and are configured to activate following the detection of a presence in the environment.

The fans of the known type mentioned above suffer from limitations and drawbacks.

In particular, the use of a remote control to regulate the conveyance of the air flow in a specific direction within an environment is inconvenient in many circumstances.

Furthermore, the known fans or air conditioners, while detecting the presence of one or more individuals within an environment, are unsuitable for distinguishing such individuals from one another and are unable to cope with the different needs of more living comfort. users present at the same time within the same environment.

SUMMARY OF THE INVENTION

The purpose of the present invention is to devise and make available a method for controlling an apparatus for generating and conveying an air flow inside an environment which allows to at least partially overcome the drawbacks mentioned above in relation to the solutions. of known fans.

This purpose is achieved by means of a control method of an equipment for generating and conveying an air flow within an environment in accordance with claim 1.

In particular, the invention refers to a method for controlling an equipment for generating and conveying an air flow based on image processing performed with a trained neural network.

Preferred embodiments of this control method are described in dependent claims 2-13.

The present invention also relates to a system which implements the above method in accordance with claim 14.

BRIEF DESCRIPTION OF THE DRAWINGS

Further characteristics and advantages of the control method of an apparatus for generating and conveying an air flow inside an environment and of the related system according to the invention will result from the following description of preferred embodiment examples, given to indicative and non-limiting title, with reference to the attached figures, in which:

- Figure 1 illustrates, with a flow chart, a method for controlling an equipment for generating and conveying an air flow inside an environment on the basis of image processing in accordance with the invention;

- Figure 2 illustrates a block diagram of an example of a system for generating and conveying an air flow within an environment that implements the method of Figure 1;

- Figure 3 illustrates, in a logic diagram, an example of implementation of a neural network, including convolutional levels, used in the method of the invention;

Figure 4 illustrates, with a flow diagram, a training method of the neural network of Figure 3;

Figure 5A illustrates, with a view from above, the system of Figure 2 in which the image acquisition means is integral with the means for conveying an air flow in an environment hosting a user, in which the user is on the left side of the field of view of the image acquisition medium and is not hit by the air flow;

Figure 5B illustrates, with a top view, the system of Figure 2 in which the image acquisition means is integral with the means for conveying an air flow in an environment hosting a user, in which the user is in the center of the field of view of the imaging medium and is hit by the air flow.

In the above figures, the same or similar elements will be indicated by the same numerical references.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

With reference to Figure 1, a control method of an equipment 10 for generating and conveying an air flow inside an environment in accordance with the present invention is indicated with the numerical reference 100.

In particular, the invention relates to a control method of an equipment 10 for generating and conveying an air flow based on image processing performed with a trained neural network.

In the following, this method 100 for controlling an equipment 10 for generating and conveying an air flow is also simply indicated as a method.

With reference to Figure 2, an electronic system for generating and conveying an air flow inside an environment on the basis of an image processing is indicated as a whole with the numerical reference 1000.

This 1000 system is chosen from the group consisting of: fans, air conditioners, air purifiers, air vents, fan heaters, ionizers, humidifiers and the like.

System 1000 includes the aforementioned equipment 10 for generating and conveying an air flow inside the environment. This equipment comprises an air flow generation member 11 and an air flow conveyor member 12 in at least one direction.

Furthermore, the system 1000 comprises digital image acquisition means 21 operatively associated with said apparatus 10 and configured to acquire, for example continuously or at predetermined time intervals, at least a digital image of a portion of the interior of a range of action of the generating and conveying apparatus 10. In the example of Figure 2, these means 21 for acquiring digital images are electrically connected to the equipment 10 for generating and conveying an air flow through a wired (wired) or wireless (wireless) connection.

Such image acquisition means take the form, for example, of one or more cameras 21. These cameras 21 are suitable for acquiring a more or less large portion of the environment, based on their number, their position and the geometry of the lens. which characterizes the field of view (Field of View or FOV).

The cameras 21 are configured to acquire images in grayscale, or preferably in the visible spectrum with color coding (for example, RGB). The cameras 21 can be chosen to operate in the visible or infrared spectrum, in the thermal radiation spectrum or in the ultraviolet spectrum, or they are configured to complete the optical information on the acquired image through the use of a dedicated channel. depth (for example, RGB-D).

Furthermore, each camera 21 is configured to output a matrix, which represents the image of the at least a portion of the environment, having a predetermined number of rows and columns, and a channel for each detected color frequency (e.g. RGB : 3 channels, RBG_thermal: 4 channels).

In addition, the system 1000 includes an electronic processing unit 22 connected to the digital image acquisition means 21 and to the said air flow generation member 11 and the air flow conveyor member 12 of the equipment 10.

In the example of figure 2, such an electronic processing unit 22 is included in the apparatus 10, but in other embodiments it can be external to such apparatus and electrically connected to the latter by means of a wired connection or wireless (wireless).

In greater detail, this electronic processing unit 22 comprises at least a processor 23 and a memory block 24 associated with the processor for storing instructions. In particular, this memory block 24 is connected to the processor 20 through a data communication line or bus 26 (for example PCI) and is constituted, for example, by a service memory, of the volatile type (e.g. of the SDRAM type). , and from a non-volatile system memory (eg SSD type).

Furthermore, the electronic processing unit 22 comprises a data communication interface 27, for example of the wireless type, configured to connect said processing unit 22 to a data communication network 28, for example the Internet network, to allow the processing unit to communicate with remote electronic devices, such as servers or portable devices (smartphones, tablets, laptops) associated with one or more users.

In addition, this electronic processing unit 22 comprises an input / output interface block 25 connected to the at least one processor 23 and to the memory block 24 through the communication bus 26. This interface block 25 is configured to connect the electronic unit 22 to the cameras 21 to acquire the images of the environment.

This electronic processing unit 22 is configured to process the digital image of at least a portion of the environment to obtain a descriptor for this portion of the environment.

Furthermore, advantageously, the electronic processing unit 22 is configured to generate at least one control signal S1 of the aforementioned air flow generating member 11 or air flow conveying member 12 on the basis of the descriptor of the portion d 'environment.

With reference to Figure 1, the operating steps of the control method of an apparatus 10 for generating and conveying an air flow inside an environment, on the basis of an image processing, are described in greater detail below. , implemented through the 1000 system.

In an example of embodiment, the electronic processing unit 22 of the system 1000 is designed to execute the codes of an application program that implements the method 100 of the invention.

In a particular embodiment, the processor 23 is configured to load, in the memory block 24, and execute the codes of the application program which implements the method 100 of the present invention.

Method 100 comprises a symbolic start phase STR and a symbolic end phase ED.

In the more general example of implementation, method 100 includes an acquisition phase 101 of at least one digital image of a portion of the environment within an action range of the generation and conveying equipment 10. This acquisition is carried out by means of the acquisition means 21 of digital images mentioned above operationally associated with the equipment.

Furthermore, the method comprises a processing step 102, by an electronic processing unit 22, of at least one acquired digital image.

Advantageously, this processing provides for the execution of at least one convolution operation on the digital image acquired through a trained convolutional neural network 300.

The use of neural networks in image processing has the advantage of greater robustness in relation to plays of light and color, partial vision of the users hosted in the environment, disturbances and errors and ensures a higher degree of generalization than analytical techniques of image processing. This allows the 1000 system implementing the method to function even in new or unexpected environments without requiring a specific calibration. This increases the ease of use for the user.

It should be noted that this neural network 300 used for the execution of the method includes at least one convolutional layer. Furthermore, such at least one convolution operation is performed by "forward" execution of the neural network.

An example of such a neural network 300 is shown in Figure 3 and will be described in more detail below.

Furthermore, the method 100 includes a step of obtaining 103, by the electronic processing unit 22, at least one descriptor of the portion of the environment on the basis of the aforementioned processing.

The method 100 comprises a generation step 104, by the electronic processing unit 22, of at least one control signal S1 of the air flow generating member 11 or of the air flow conveying member 12 included in the apparatus 10 on the basis of the descriptor of the room portion.

In a particular embodiment, the aforementioned generation step 104 comprises the steps of generating at least one control signal of the air flow generation member 11 and at least one control signal of the air flow conveying member 12. air.

In an example of embodiment, the aforementioned at least one control signal S1 is an on or off signal of the equipment 10.

In a further embodiment example, the at least one control signal S1 is a signal for changing the direction of conveying 12 of the air flow.

In an example of implementation, the descriptor of the portion of the environment is a unique descriptor chosen from the group consisting of:

- a user's position within the environment; - an aesthetic identifying feature of the user; - a gesture by the user representative of a climatic preference.

In a first example of implementation, in a first variant, the descriptor of the portion of the environment is a unique descriptor that specifies a position of a user within the environment. According to a particular embodiment of the system 1000 in which the apparatus 10 for generating and conveying the air flow is a fan, a television camera 21 is fixed integrally to this apparatus 10 and can be moved with it, for example with the main drive shaft of the visual field 54 oriented in the direction and in the direction of the air flow 53 indicated in figures 5A and 5B by means of arrows.

In particular, with reference to these figures 5A-5B, the user 59 can be in the left field of view of the camera (as in figure 5A) or centered (figure 5B). Since the FOV (field of view) 52 of the camera 21 captures a much larger portion of the environment than that crossed by the air flow 53, only when the user 59 is centered in the FOV is he hit by the air flow itself . In all other cases, the control method is configured to control the rotation of the air flow conveyor 12, and of the camera 21 integral with it, in the direction necessary to bring the user 59 to the center of the visual field.

Variations can be introduced without altering the sense of the present invention. For example, it is possible to provide a fixed camera (ie not integral with the air flow conveyor) and move the flow conveyor 12 according to known directions preset during installation.

In the absence of users inside the room, the descriptor of the room portion has a null value and consequently the system 1000 of the invention behaves in accordance with a suitable pre-set rule: for example, in the case of a fan 10, it is decided that the air flow generator 11 remains in a deactivated state.

Following the entry of a user into the field of view of the camera 21, the environment descriptor is configured to detect the presence of that user in relation to the field of vision itself. If the user is in a first (left) portion of the image, defined with respect to a median position of the acquired image, then the method provides for activating the air flow conveyor 12 and, consequently the camera 21 is moved (rotated) to the left until the user is perfectly centered with respect to the air flow and the field of view of the camera 21.

If the user is centered, the environment descriptor will report, again in the form of appropriate neuronal activations, the presence of the user in the center of the image to activate a control logic that stops the motion of the conveyor 12 and maintains the flux generation unit 11 is switched on.

Note that variants of the aforementioned control methodology may be introduced depending on the purpose assigned to the air flow in the specific case of use.

For example, if the equipment 10 is a fan, it is desirable for the user to be hit by the generated air flow. In the event that the equipment 10 to be controlled is an air conditioner, it may instead be desirable that the flow of air generated is never directly oriented in the direction of the user and an attempt is made to avoid it as much as possible.

Furthermore, the control logic based on the environment descriptor can be suitably accompanied by the use of traditional control logic, without changing the purposes of the present invention. In particular, the control logic that uses the environment descriptor can be accompanied by the use of timers or sensors that allow you to create more sophisticated control logics.

In a second variant of the aforementioned first example of implementation, the descriptor of the portion of the environment is a unique descriptor configured to specify the positions of one or more users within the environment. In this case, the environment descriptor is able to specify, again in the form of neuronal activations, the presence of no one, one or more users within the environment.

In the event that there are no users present within the environment or there is only one, the control logic substantially coincides with the one detailed above in relation to the first variant.

In the event that two users are present in the same environment at the same time, the control logic can for example distribute the air flow to the two users in alternating phases. In a first phase, during a first time interval t1, a first user is centered with respect to the field of view of the camera 21 and the air flow. After this first time interval t1, in a second phase, the air flow conveyor 12 rotates to center the second user and to maintain this position for a second time interval t2. Once this second time interval t2 has also elapsed, the control logic is configured to repeat the aforementioned phases cyclically. These first t1 and second t2 time intervals are programmable to have the same duration or duration different from each other.

It should be noted that the same considerations are applicable, with the appropriate adaptations, even to the case of three or more users present in the environment at the same time.

In a second example of realization, the descriptor of the portion of the environment expresses an aesthetic characteristic of the user. This characteristic is expressed in the form of neuronal activations of a plurality of output neurons ("embedding") of the network 300. This group of neurons constitutes a vector that in an N-dimensional space encodes a certain aesthetic aspect of the user. This vector can be compared with other vectors, for example by calculating a distance in a Euclidean space.

According to a first preferential variant, the apparatus 10 for generating and conveying the air flow, for example the aforementioned fan, is configured to acquire and store data relating to the physical appearance of a user in the memory block 24 of the system 1000 .

This physical aspect is stored in the form of a vector of neuronal activations of the last output layer of the network 300.

This storage can take place, for example, as soon as a new unknown user enters the field of view of camera 21. System 1000 embeds (coding) the user's physical appearance and compares it with those in memory. If the distance between the embedding of the new user and those stored is greater than a predetermined threshold, then the new user is a person not known to the 1000 system. In this case, the relevant embedding is stored in a new memory cell.

The physical appearance may be related to a characteristic that varies over time (for example, the user's clothing) or a characteristic that is stable over time (for example, the user's face) without altering the general meaning of the present invention. When the user is present within the field of view of camera 21, one or more preferential settings for that specific user can be implemented.

In the case of a new user never seen before, recognized as described above, the climatic preferences are initialized to the pre-set standard values and can be subsequently modified by the user himself through known methods (push-button panel, remote control, smartphone application) or, as described later, through gestures of control.

For example, if a user has a specific need for air free of pollen or allergens, then an air purifier 1000 which implements the method of the present invention is configured to recognize it among several users occupying the room to direct the flow of purified air only towards its direction.

In a second variant of this second example, the ventilation system 1000 is configured to memorize the aesthetic appearance of two or more users and, moreover, to memorize the climatic preferences of each user. In this way, the ventilation system can generate a micro-climate tailored to each occupant.

According to a third variant of the second embodiment example, the descriptor of the user's aesthetic characteristic is able to identify not a single individual, but classes of users, for example based on their clothing or their age.

This solution is particularly advantageous in the case of large distributed ventilation / conditioning systems where users are free to move around in large rooms. In this way it is possible to direct a colder or warmer air flow towards users based on the clothing worn by the users themselves.

According to a third example of implementation, the descriptor of the portion of the environment is a gesture of the user representative of a climatic preference of the user himself. For example, an upward thumb shown by the user when addressing the camera 21 of the fan 10 is adapted to express a desire to increase the air flow generated by the air flow generating member 11. Conversely, an inch facing downwards may represent the desire to reduce the generated air flow.

This gesture representative of a climatic preference is expressed in the form of the neuronal activations of an output neuron, representative of the specific gesture. The convolutional neural network can be designed to interpret sequences of frames, for example by using a final layer of the RNN or LSTMo GRU type known to an expert in the field, placed downstream of the convolutional sequence.

Alternatively, it is possible to use a “hierarchical neural attention encoder” to obtain a similar but computationally more efficient result. The gesture can be a more or less explicit gesture.

In a fourth embodiment example, the descriptor of the environment portion is generated on the basis of the combination of a first and a second descriptor chosen from the group consisting of:

- a user's position within the environment; - an aesthetic identifying feature of the user; - a gesture by the user representative of a climatic preference.

In a fifth example of implementation, the descriptor of the portion of the environment is generated on the basis of the combination of:

- a user's position within the environment; - an aesthetic identifying feature of the user; - a gesture by the user representative of a climatic preference.

In this fifth and more complete example it is possible to create complete and pleasant user experiences. For example, when a new user enters an environment, his appearance is memorized, based on the temperature and clothing, the system assumes a climatic preference. The user can correct it with a single gesture. The new preference expressed is then associated with its aesthetic appearance. The user moves around the environment and the 1000 system that implements the method recognizes and follows it, conveying it with a customized air flow.

With reference to Figure 3, an example of a convolutional neural network 300 which can be used in all the examples of embodiment of the method 100 of the present invention is described.

This neural network includes at least the following layers:

- an input layer 301 configured to receive the entire digital image or the sum of the digital images or at least a sub-portion of the digital image acquired with the cameras 21;

- at least one convolutional layer conv 1;

- at least one fully connected layer 303a;

- an output layer 304 with more neurons available which, depending on the environment descriptor, take on a different meaning.

In particular, the output layer 304 has at least one neuron whose activation or combination of activations is capable of describing at least one of the following environment descriptors:

- a user's position within the environment;

- an aesthetic identifying feature of the user;

- a gesture by the user representative of a climatic preference.

In greater detail, the network 300 comprises a convolution block 302 consisting, for example, of twenty-two convolutional layers conv 1, conv 2, conv 3, ..., conv 22 in cascade also of the DepthWise Convolution (DW) type, of type known to an expert in the field. The convolutional level input is connected to the output of a respective convolutional layer with linearity of the ReLU and BatchNorm type of the type known to the expert in the field.

As is known, in a convolutional layer of a neural network, each neuron is connected only to a few neighboring neurons in the previous layer. The same set of weights (and local connection layout) is used for each neuronal connection. Conversely, in a fully connected layer of the network, each neuron is connected to each neuron in the previous layer and each connection has its own weight.

In the example of Figure 3, the neural network 300 consists of two fully connected layers 303a and 303b. These two layers are similar to convolutions having a kernel that covers the entire input layer (input) of the neural network 300. They can therefore be interpreted as two additional convolutions configured to provide a global meaning to the input layer. In any case, specific embodiments of the step of the processing method 102 and control 103 of the method do not alter the generality of the present invention.

For the correct functioning of the method 100 it is necessary that the neural network 300 is a network trained to generate at least one, two or all three of the mentioned environment descriptors. A training procedure 400 of the network 300 is described with reference to Figure 4.

The training method 400 provides an initial step 401 for defining a position and orientation of the digital image acquisition means 21.

The method for obtaining the most general network possible involves the acquisition 402 at the input of a plurality of digital images by the camera 21 of the air flow conveying generation apparatus 10 configured to film different situations in which different users move in the environment, each with different physical characteristics and who sometimes make a gesture of control.

This plurality of images is collected in environments of different appearance and with different degrees of illumination. The occupants are also required to be dissimilar to each other, in different poses and attitudes and to have a good variance in appearance. This leads the 1000 system to generalize the desired recognition characteristics well.

The next phase involves the 403 annotation of the plurality of digital images acquired. This annotation is performed by associating an appropriate label or code to each digital image acquired.

In particular, to train the network to return an environment descriptor as in the first example of a neural network embodiment described above, a bounding box is drawn around the occupants in order to highlight their position in the camera's field of view. In addition, a unique identifier is associated with each individual user, and, in the event he is performing a control gesture, the type of gesture is noted.

At this point, the training method 400 provides for the 405 initialization of the neural network 300 by associating it with the weights of the neural connections in a random or predefined way.

In a preferred embodiment example of the training method 400, this training phase of the neural network 300 also comprises a step 404 for increasing the number of images that can be used for training by performing further processing operations on the original images acquired.

This is achieved, for example, by performing rotations of each image, selecting sub-portions of the images or correcting, for each image, at least one chromatic channel.

The advantage achieved by this phase of increase 404, is to make available to the neural network 300 a greater number of images to be used for training and, therefore, to improve learning by the network itself.

The subsequent training step 406 of the network 300 takes place by means of a back-propagation method of the type known to a person skilled in the art. In the case of the invention, the SGD (Stochastic Gradient Descent) method was used. In particular, one proceeds to train at least one level of the neural network 300 by modifying the weights associated with the network on the basis of the labels of the plurality of classified digital images. A loss is calculated at each backpropagation cycle, calculating the error between the classes predicted by the network in the training phase and the real ones. The loss for the recognition of people and gestures can be, for example, a Euclidean loss.

In particular, with reference to the second example of realization of a neural network described above, an analogous method is followed, with the difference that the system is forced to learn a vector of typical characteristics of the users. In this case, the training takes place on various types of people or different clothing so that the network learns to represent ("encoding") the different characteristics with a vector. To calculate the "loss" used to update the weights in the backpropagation phase, for example, the so-called "triplet loss" can be used, ie by presenting to the network two different images of the same occupant and a third image depicting a different occupant. At each backpropagation cycle the “loss” is calculated by computing the two vector distances in Euclidean space. In this way the network is induced to internally adjust weights so that the distance between the two images of the same occupant is ideally zero and the distance between the images of different users is maximized.

Note that, as often happens in “deep learning” applications, the vector encoding the characteristics is not necessarily interpretable by a human being.

As highlighted above, the method 100 of the present invention and the related electronic system 1000 have numerous advantages and achieve the intended purposes.

Since the modeling of the various scenarios is not explicitly coded by a human being but learned automatically by the machine, a first advantage is that the system has high cognitive abilities that allow it to autonomously undertake actions hitherto unthinkable.

For example, the 1000 system can recognize the individual users present in an environment and memorize the preferences of each of them.

For example, the air flow conveying generation equipment can recognize areas occupied by users from unoccupied areas and consequently distribute the air only towards users. Or an air conditioner that generates cold air can carefully avoid directing the air towards users. Or a fan can distinguish whether a certain user is near or far and then adjust the focus and intensity of the air jet accordingly.

For example, the fan can recognize one or more users in an environment and consequently create customized microclimates. In the case of only one user present in the room, the fan can follow the user himself during his movements within the room.

A second advantage is that the system of the invention can interpret signals suggested by users. For example, a certain gesture by a user can cause the ventilator to exclude it (or include it in ventilation). Or another gesture can adjust the intensity of the air flow.

The improvement of comfort, customized air conditioning and energy saving are the main benefits of the present invention.

To the embodiments of the method and system of the invention, a person skilled in the art, to meet contingent needs, may make changes, adaptations and replacements of elements with other functionally equivalent, without departing from the scope of the following claims. Each of the features described as belonging to a possible embodiment can be realized independently of the other described embodiments.

Claims (15)

CLAIMS 1. Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment on the basis of an image processing, said apparatus comprising a generating member (11) of the air flow and a member (12) for conveying the air flow in at least one direction, in which the method comprises the steps of: - acquire (101) at least a digital image of a portion of the environment within a range of action of the apparatus (10) for generating and conveying by means of acquisition means (21) of digital images operatively associated with said apparatus ; - process (102), by an electronic processing unit (22), the at least one digital image acquired through the execution of at least one convolution operation on the digital image using a trained neural network (300); - obtain (103), from the electronic processing unit (22), at least one descriptor of said portion of the environment on the basis of said processing; - generating (104), by the electronic processing unit (22), at least one control signal (S1) of said air flow generating member (11) or of said conveying member (12) of the air flow on the basis of said room portion descriptor. Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment according to claim 1, wherein said generating step (104) comprises the steps of generating at least one control signal of said air flow generation member (11) and at least one control signal or of said air flow conveying member (12). Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment according to claim 1, wherein said room portion descriptor is a univocal descriptor selected in the group consisting of: - a user's position within the environment; - an aesthetic identifying feature of the user; - a gesture by the user representative of a climatic preference. Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment according to claim 1, wherein said room portion descriptor is generated on the basis of the combination of a first and a second descriptor chosen from the group consisting of: - a position of a user within the environment; - an aesthetic identifying feature of the user; - a gesture by the user representative of a climatic preference. Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment according to claim 1, wherein said room portion descriptor is generated on the basis of the combination of: - a user's position within the environment; - an aesthetic identifying feature of the user; - a gesture by the user representative of a climatic preference. Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment according to claim 1 or 3, further comprising a step of training the neural network (300) to recognize a user's position within the environment. Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment according to claim 1 or 3, further comprising a step of training the neural network (300) to recognize an aesthetic identifying feature of the user. Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment according to claim 1 or 3, further comprising a step of training the neural network (300) to recognize a gesture of the user representative of a climatic preference. Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment according to claim 1 or 3, further comprising a step of training the neural network (300) to simultaneously recognize a position of a user within the environment, an aesthetic identifying characteristic of the user, a gesture of the user representative of a climatic preference by activating special output neurons dedicated to each of the three recognition classes . Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment according to claim 1, wherein said at least one control signal (S1) is a signal switching the appliance on or off. Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment according to claim 1, wherein said at least one control signal (S1) is a signal to change the direction of air flow (12). Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment according to claim 1, wherein said neural network (300) comprises at least one layer (layer ) convolutional and said at least one convolution operation is carried out by "forward" execution of the neural network. Method (100) for controlling an apparatus (10) for generating and conveying an air flow inside an environment according to claim 12, wherein said convolutional neural network (300) comprises at least the following layers : - an input layer (301) configured to receive the entire digital image or at least a sub-portion (down-sample) of said digital image; - at least one convolutional layer (conv 1); - at least one fully connected layer (303a); - an output layer (304) with at least one neuron whose activation or combination of activations describes at least one of the following environment descriptors: - a position of a user within the environment; - an aesthetic identifying feature of the user; - a gesture by the user representative of a climatic preference. 14. System (1000) for generating and conveying an air flow within an environment on the basis of image processing, including: - equipment (10) for generating and conveying an air flow inside the environment comprising an organ for generating the air flow (11) and a member (12) for conveying the air flow in at least one direction; - means (21) for acquiring digital images operationally associated with said equipment (10) to acquire at least one digital image of the environment; - an electronic processing unit (22) connected to said digital image acquisition means (21) and to said air flow generation unit (11) and air flow conveyor (12), said unit processing electronics comprising at least one processor (23) and a memory block (24) associated with the processor for storing instructions, said processor and said memory block being configured to carry out the steps of the method according to any one of claims 1- 13. System (1000) for generating and conveying an air flow inside an environment on the basis of an image processing according to claim 14, in which said system is selected from the group consisting of: fans, air conditioners, air purifiers, air vents, fan heaters, ionizers, humidifiers.