PROCEDURE FOR DETERMINING INTENTIONS OF A DRIVER IN A VEHICLE AND CORRESPONDING SYSTEM

Description

DESCRIPTION of the industrial invention entitled: “Procedure for determining the intentions of a driver in a vehicle and corresponding system” by: Stellantis Europe SpA, of Italian nationality, Corso Giovanni Agnelli 200, 10135 Turin Designated Inventors: Silvano MARENCO, Paola DALMASSO, Stefano MANGOSE, Filed on: **** DESCRIPTION TEXT Field of invention The present invention relates to systems and methods for controlling self-driving vehicles. Known technique In the field of self-driving road vehicles, for some years now there have been significant developments are underway. A common criterion for describing the level of The range of a road vehicle is based on an SAE classification (“Society of Automotive Engineering”) from 0 to 5. The most modern cars equipped with devices such as adaptive cruise control, and / or the system automatic lane centering systems are classified as level 1. Level 2 autonomous vehicles can contain, for example, more than a dozen sensor devices. In general, vehicles of levels 0 to 3 have increasing comfort and safety features, but still require that the The driver must be alert and ready to intervene if necessary. A big leap. occurs at level 4, with vehicles allowed to be used in restricted areas, without any intervention by a driver. At level 5, a car can be equipped with more than 30 additional sensors in order to obtain complete control over the driving, essentially without human intervention. Level 5 is a full level autonomy, in which the vehicle is able to operate without any intervention from the driver in any environment (urban, suburban, rural, highway and even off-road) at any reasonable speed and under any conditions environmental. Various technologies are used to detect the surrounding environment. A The first type of sensor is the so-called “LIDAR”, which uses a Laser technology for measuring the distance to an object. LIDAR sensors They perceive the environment surrounding the vehicle in three dimensions. They are safe in obstacle detection and allow the calculation of the vehicle's position thanks to 3D mapping. The use of video cameras to analyze is also widespread the environment surrounding the vehicle, in particular to detect road signs and traffic lights. Video cameras are sometimes used in conjunction with controllers electronics programmed with algorithms capable of categorizing obstacles. In various types of autonomous vehicles, already proposed in the past, they are also widely used satellite navigation (GNSS) antennas that allow you to detect the position of the vehicle with an accuracy close to the centimeter. Radar devices are also used to determine the position and the speed of surrounding objects and for long-distance vision. Autonomous vehicles They are often also equipped with odometers to estimate and confirm the position of the vehicle and its speed, as well as inertial measurement devices, to detect the accelerations and rotations of the vehicle, in order to confirm the information on the vehicle position and to improve its accuracy. In support of the network of Sensors are also commonly used in various types of communication systems. The present invention relates to self-driving vehicles, and has in in particular the control of self-driving vehicles which include the use of one or more sensors of the driver's vital parameters and any additional occupants of each vehicle, and in particular the use of sensors for EEG (encephalogram) configured to monitor brain activity driver and occupants of the vehicle. For example, in known solutions of this type, the output signals from the sensors EEGs are used to generate a risk assessment profile in view of to prevent the autonomous vehicle from switching from a self-driving mode to a manual driving mode in a condition where the risk profile it seems excessive. The above solution is susceptible to further improvements, both from point of view of the way in which EEG sensors are used, both from the point of view of the actions that can be activated on the basis of the findings made by these sensors, in order to improve the level of safety of the driver and the occupants of the reference vehicle and of vehicles in its vicinity vicinity. Purpose and summary One purpose of one or more embodiments is to help provide a solution that takes into account the above-mentioned problems. According to one or more forms of implementation, this aim can be achieved by means of a process having the characteristics set out in the claims that follow. Claims are an integral part of the technical teaching here provided with reference to the forms of implementation. In particular, forms of implementation of this solution concern a process for determining the intentions of a driver in a vehicle including the stages of associating first synaptic weight values ​​to respective combinations of detectable characteristics and intentions; associate second values ​​of synaptic weight to respective combinations of characteristics and driving contexts detectable; acquire an EEG of the driver; extract a plurality of features from the EEG; estimate a composite score for each intention detectable on the basis of the characteristics, of the first values of synaptic weight, and of the second values ​​of synaptic weight; identify the score major composite, together with its respective intention. In various forms of implementation, the procedure includes checking whether the maximum composite score value is greater than an associated threshold value to a respective intention; to store the intention of the message in a message driver detected. In various forms of implementation, the procedure includes a phase of acquire a reference EEG of the driver, in particular obtained during a settings initialization phase customization. In various embodiments, the process comprises a filtering step the encephalogram acquired on the basis of the reference encephalogram. In various forms of implementation, the procedure includes the phases of determine a driving context, in particular on the basis of data acquired from V2X devices and ADAS systems; in response to not determining any context of guide, initialize the synaptic weight values, in particular to a value equal to 1; in response to successfully determining a driving context, update the values of synaptic weight based on the detected driving context. In various forms of implementation, the procedure includes the phases of determine whether the driver is distracted; in response to determining that the driver is distracted, memorize a state and resume execution of the process from the stage of acquiring an encephalogram. In various embodiments, the synaptic weight values ​​have respective values initially determined by means of a training procedure. In various embodiments, the stage of determining whether the driver is distracted is carried out on the basis of data obtained from video cameras processed by artificial vision procedures. Furthermore, various forms of implementation of the solution described here concern a control system for determining the intentions of a driver in a vehicle comprising a control unit configured to implement the process described above; one or more encephalogram sensors; a transceiver; a or more cameras; and one or more sensors. Furthermore, various forms of implementation of the solution described here concern a vehicle incorporating the control system described above. Brief description of the figures The invention will now be described with reference to the attached figures, provisions by way of example only and not limited to, where: - Figure 1a shows a block diagram of a vehicle comprising a control system implemented in accordance with forms of implementation of the solution described here; - Figure 1b shows a device comprising a plurality of sensors for encephalography usable in embodiments of the solution here described; - Figure 2 shows a flowchart of a process for detect the intentions of a driver while driving a vehicle; - Figure 3 shows a flowchart of a process for train an artificial neural network. Detailed description In the following description, one or more specific details are illustrated, aim to provide an in-depth understanding of examples of embodiments of this description. Embodiments may be obtained without a or more specific details or with other processes, components, materials, etc... In other cases, known operations, materials or structures are not illustrated or described. in detail so that certain aspects of the forms of implementation are not made unclear. A reference to "an implementation form" under this description is meant to indicate that a particular configuration, structure, or feature described with reference to the embodiment is included in at least one form of implementation. Therefore, phrases like "in one form of implementation" which may be present in one or more points of this description not necessarily refer to the same form of implementation. Furthermore, particular conformations, structures or characteristics can be combined in any suitable way in one or more embodiments. The references used here are provided simply for convenience and therefore they do not define the scope of protection or the scope of the forms of implementation. In the following, terms such as “first” and “second” are used for to distinguish one element from another and not to indicate a sequential order, unless otherwise indicated. Also, relative position terms such as "vertical" and "horizontal", or "front" and "back", when used, are intended as relative to each other and do not have to be absolute, and refer only to one possible device location associated with such terms depending on of the device's orientation. For example, when two elements are indicated as generally "vertical" and "horizontal" are not necessarily so oriented to the ground unless specifically indicated, but instead are oriented orthogonally to each other. As anticipated, this description concerns a procedure for determine the intentions of a driver in a vehicle. In this regard, Figure 1a shows a block diagram depicting a vehicle 10 including a control system 100 that implements a level 1 or higher autonomous driving system, capable of controlling, in a manner if known, operating components of the vehicle (engine, transmission, steering, accelerator, brakes, etc.). The control system 100 comprises a control unit control 110, one or more encephalogram sensors 120, a transceiver 130 capable of communicating with similar vehicle communication modules that find nearby, one or more video cameras 140, and one or more sensors 150. As illustrated, the control unit 110 is coupled to one or more sensors for encephalogram 120, to the transceiver 130, to one or more video cameras 140, and to one or more sensors 150 via a system bus 105 such as, for example example, a CAN, CAN-FD, LIN, or Automotive Ethernet bus, so as to transmit and receive data with each device connected to the system bus 105. The one or more 120 EEG sensors allow you to perform on the driver of the vehicle 10 an electroencephalography (EEG), that is a non-invasive technique invasive technique used to record the electrical activity of the brain. For this purpose, the one or more EEG sensors 120 may comprise a plurality of electrodes that can be placed on the scalp of the driver or occupants of the vehicle vehicle and configured to detect voltage fluctuations generated by activity neuronal. The plurality of electrodes included in the one or more sensors for 120 EEG may include disposable electrodes, or electrodes reusable. The electrodes can be supplied in a band or cap in in order to facilitate use by the driver, or in other types of devices wearables such as earphones. The one or more sensors for 120 EEG can detect different types of signals, including alpha waves, beta waves, gamma waves, delta waves, and theta waves. A possible example of a device including sensors for EEG 120 is shown in Figure 1b, where the plurality of EEG sensors 120 are carried by a temporal band 121 and by two parietal bands 122 which form part of a cap-like structure, covered by a headdress 123 (illustrated with line (dotted line). The sensors 120 are connected to a wire harness 124 which is received inside the seat structure 125 (where a retractor 126 is provided) elastic recall) and which is connected to the control unit 110. In the example specifically illustrated here, on one side of the seat 125 there is a hook 127 for receive the device when it is not in use. This solution is only an example. Any alternative solution capable of combining EEG sensors can be used by the driver and occupants of the vehicle. For example, ear sensors that can be placed in the ears have already been proposed users' ears and capable of perceiving users' brain activity, which can connect with the control unit 110 via a wireless connection wires. Any other known solution can be used for this purpose. It is also necessary consider that, although the preferred solution is the one that provides a By monitoring the driver's brain activity, it is also possible to predict that such monitoring is also carried out for some or all of the occupants people present in the vehicle. The transceiver 130 is configured to exchange data with devices or systems external to the vehicle 10 in which the control system 100 is provided. In In particular, the transceiver 130 is coupled to the system bus 105 for exchange data with peripherals connected to the control system 100. In various embodiments, the transceiver 130 supports various V2X standards for data exchange with other vehicles and / or with infrastructures such as, for example, 3GPP Cellular V2X or IEEE 802.11p. Additionally, the transceiver 130 is configured to exchange data, preferably via the Internet, with one or more processing systems remote. Remote computing systems may include database servers, file servers, or other systems. The one or more video cameras 140 are arranged inside the passenger compartment of the vehicle 10 and are configured to acquire one or more video streams, sending the data acquired to the control unit 110. In particular, the one or more video cameras 140 are arranged in the passenger compartment in such a way as to allow monitoring, by means of computer vision techniques known in themselves, of parameters physiological of the driver. The one or more sensors 150 are arranged in the vehicle 10, for example inside the passenger compartment and include, for example, heart rate sensors, and blood glucose sensors. In various embodiments, the one or more sensors 150 send data collected from respective measurements to the control unit 110 via the system bus 105. The sensors 150 may also include other sensors arranged externally of the passenger compartment such as radar, lidar and video cameras to create assistance systems advanced driving, Advanced driver-assistance systems (ADAS), such as example attention monitoring systems, Driver Attention Alert, and / or systems lane keeping systems, Lane Keeping System, and / or Cruise Assist systems. Each sensor 150 can exchange data via the system bus 105 with control unit 110. In various embodiments, the control unit 110 comprises a microcontroller 111, a RAM memory 112, and a non-volatile memory 113. The control unit 110 is configured to run one or more resident programs in non-volatile memory 113 by one or more microcontroller processors 111. For example, one or more programs can be loaded and / or executed by means of an operating system resident in non-volatile memory 113. In various embodiments, the control unit 110 performs procedures aimed at to the detection of physiological parameters of the driver of the vehicle 10, in particular by analyzing data acquired by one or more sensors for 120-megapixel EEG, analysis of video streams acquired from one or more video cameras 140 using artificial vision procedures, and through analysis of acquired data from one or more sensors 150. Specifically, said extraction procedures of physiological parameters include procedures for detecting the degree of attention, distraction, and driver fatigue. In various forms of implementation, the control system 100, in particular via a specific software resident in the memory of the control unit 110, It is configured to detect a driving context according to procedures known in itself. In particular, the driving context is detected on the basis of data acquired from V2X devices such as traffic lights, and sensors 150, in particular from ADAS systems. Therefore, as will be better detailed in the following description, the control system 100 is configured to process signals in output from the EEG sensors 120 in order to determine the cognitive state of the driver, with reference to the degree of attention / distraction, sense of fatigue and the degree of drowsiness, and the driver's intentions. For this purpose, the transceiver 130 is configured to connect and exchange data with devices equipped with V2X (Vehicle-to-Everything) connectivity such as, for example, connected traffic lights, forwarding the information obtained, to example regarding a traffic light state or an intersection topology, contained in received SPAT messages and / or MAP messages. The data obtained, for example example regarding the presence of a crossroads, a pedestrian, a cyclist, of heavy traffic, or a traffic light, are sent via the system bus 105 to control unit 110. Sensors 150, in particular ADAS systems, can send data via the system bus 105 to the control unit 110. For example, the data sent contain information regarding the activation of one or more ADAS systems. The control unit 110 is further configured to run a program to determine driver intentions, resident in non-volatile memory 113, and which implements a procedure 200 for determining the intentions of a driver in a vehicle. In this regard, Figure 2 shows a flowchart of a procedure 200 for determining the intentions of a driver while driving a vehicle. In a first step 201 the control system 100 performs a procedure of identification of a reference EEG through techniques per se note. In particular, in phase 201 the control system 100 acquires by means of the sensors for 120 encephalogram an EG driver encephalogram, and in The following calculates an encephalogram based on the acquired EG reference, or baseline EB. Subsequently, in a step 202 the control system 100 associates a predetermined weight WI for each combination of characteristics F and i,ji detectable intentions I , the index i being between 1 and a number of j characteristics N, and the index j being between 1 and a number of intentions The detectable M. In the example considered, the intentions I correspond to: jj - lane change I ; - overtaking I ; - turn right or left I ; - departure I ; - acceleration I ; - deceleration I ; and - arrest I . In various embodiments, the value of each WI weight is associated with a i,j respective combination of characteristic Fi and intention Ij is determined following of a neural network training procedure. For example, the WI weights are i,j obtained through training methods such as, for example, the retro-method error propagation coupled to stochastic gradient descent. In In particular, the neural network training procedure will be better detailed in the following of this description. Subsequently, in a step 203 the control system 100 associates a weight predetermined WC to each combination of features F and contexts of i,ki detectable guide C , the index i being between 1 and the number of features k N, and the index k being between 1 and a number of driving contexts C k detectable T. In the example considered, the detectable driving contexts C are: k - intersection C; - presence of pedestrians C; - presence of cyclists C; - heavy traffic C; - speed control C ; and - regular driving C . In various embodiments, the value of each WC weight associated with i,k a respective combination of feature F and context C is determined to ik following a training procedure of a neural network. Even in this case, WC weights are obtained through training methods such as, for example i,k for example, the error back-propagation method coupled with descent stochastic gradient. In particular, the network training procedure neural will be further detailed later in this description. Subsequently, in a step 204 the control system 100 acquires from the respective plurality of sensors for encephalogram 120 data regarding the driver's brain activity, obtaining an EG encephalogram. Consequently, in a step 205 the control system 100 performs a filtering of data regarding the EG driver's brain activity acquired in the previous stage 204. In particular, the filtering of the acquired EG encephalogram in phase 204 it allows to remove noise and artifacts from the acquired signals. For example, noise can be introduced into an EEG due to the eye movement, muscle tension, or electromagnetic interference. in this regard, the EG filtering uses the baseline, or baseline EEG, EB obtained in stage 201. Subsequently, in a step 206 the control system 100 extracts features F highlights from the driver's brain activity data, normalizing the the results obtained. For example, in various forms of implementation the extraction of characteristics is carried out by means of well-known procedures such as, for example, Forward Selection, Backward Elimination, Recursive Feature Elimination (RFE), etc. In particular, the normalization of data acquired by one or more sensors for 120 EEG it is advantageous as different sensors could produce output values ​​with different scales and / or units of measurement. Subsequently, in a step 207 the control system 100 checks whether the driver of vehicle 10 is distracted. In particular, the control system 100 is configured to check if the driver is distracted by techniques per se notes using data collected by 120 EEG sensors, the data provided from video cameras 140, and further data provided by sensors 150 located in the passenger compartment of the vehicle 10. Based on the collected data, the 100 control system assigns a value to a weight W between 0 and 1. In particular, if the control system 100 D detects that the driver is distracted, the value associated with the weight W is equal to 0. D In response to detecting that the driver of vehicle 10 is distracted, the system control 100 performs a step 208 in which a message is output Q having a value of determinate intention equal to “no intention”, in particular due to the fact that the driver is in a state of distraction and therefore the control system 100 is not able to determine with sufficient accuracy accuracy the driver's intention. Conversely, if in phase 207 the control system 100 verifies that the driver is not distracted the execution of procedure 200 continues in a stage 209. Specifically, in step 209 the control system 100 detects a context of driving C and consequently checks whether a driving context C has been kk successfully detected. If not, the control system 100 executes a step 210 in which the WC synaptic weights associated with each combination of features F and i,ki detectable driving contexts C are re-initialized, in particular to a value equal to k to 1. Conversely, if a driving context C is detected in phase 209, k control system 100 performs a step 211 in which the associated WC synaptic weights i,k for each combination of features F and detectable driving contexts C are ik updated, and at the same time the information, in particular the synaptic weights, obtained regarding driving context C and driver intentions I are kj combined, in particular by performing a product 𝑊 𝐼 ⋅ 𝑊 𝐶 . 𝑖 , 𝑗 𝑖 , 𝑘 Subsequently, in a phase 212 the control system 100 makes an estimate of a composite score S for each detectable intention I, where the index j jj indicates the driver's intention considered and is between 1 and the number of detectable intentions Ij M. In various implementations, the composite score Sj is obtained by means of the following expression: O ∑ 𝑊 ⋅ 𝑊 𝐼 ⋅ 𝑊 𝐶 ⋅ 𝐹 𝐷 𝑖 = 1 𝑖 , 𝑗 𝑖 , 𝑘 𝑖 𝑆 = O O ∑ 𝑊 𝐼 ⋅ 𝑊 𝐶 𝑖 , 𝑗 𝑖 , 𝑘 𝑖 = 1 Subsequently, in a step 213 the control system 100 identifies and select the intention I having the highest composite score S, jj previously calculated in step 212. Subsequently, in a step 214 the control system 100 checks whether the composite score of the selected value is greater than a threshold value TS. j In particular, in various implementation forms a threshold value TS is associated with j each detectable intention I , therefore the comparison between composite score S jj and the threshold value TS is performed by considering the j indices having the same value. j If so, the control system 100 executes a step 215 in which a Q message containing the driver's intention is produced at the output detected. On the contrary, if not, the control system 100 performs phase 208 again, thus indicating in the Q message that it is not possible detect a driver's intention, for example by including in the message the value “no intention”. As anticipated, the value of the weights of the intentions WIi,je of the contexts of WC guide is initially acquired, for example in stages 202 and 203 of the i,k method 200 described above, from the non-volatile memory 113 of the system control 100. The values ​​of the weights of the WI intentions and the WC driving contexts i,ji,k stored in the memory 113 of the control system 100 are initially determined following a training procedure of a neural network. In this regard, a training procedure is shown in Figure 3 300 usable to obtain starting values ​​of the WI intention weights and i,j WC driving contexts, usable in the method 200 to determine i,k intentions of a driver driving a vehicle described above. It is noted that the training procedure 300 described here can be performed in a system of any processing, and does not necessarily have to be performed by the system of control 100. In fact, the purpose of the training procedure 300 is to provide a set of synaptic weights related to WI intentions and WC driving contexts having i,ji,k values ​​that allow the 200 process to function optimally. In a first phase 301, they are provided with training data D from use in the training process. The training data D is obtained from one or more 120 EEG sensors, and contain information regarding brain waves such as delta waves, theta waves, alpha waves, beta waves, and / or gamma waves, and in general data containing one or more EG EEGs and / or one or more reference ER EEGs. Next, in a step 302 the training data D is filtered. In particular, the training data D is processed in such a way as to suppress, at least partially, disturbances in one or more EG EEGs. For this purpose, training data D may contain one or more EEGs of ER reference, on the basis of which the training data D is processed. Subsequently, in a step 303 the training data D is again processed to extract a plurality of features, such as, for example, power theta wave power, beta wave power, alpha wave power, ratio between alpha waves and beta waves, theta-gamma coupling, ERD (Event-Related Desynchronization), P300 amplitude, and SMR (SensoriMotor Rythm). Next, in a step 304 the training data D is split in a plurality of data sets, including training data, validation, and test data. In particular, during the execution of the validation process training 300 it is possible to use subsets comprising a portion of the training data D, or exchange the training data and the training data validation, so as to prevent over-calibration of synaptic weights stringent, commonly called “overfitting,” which in an artificial neural network causes poor performance with data not included in the training data D. In a step 305 the training data D is scaled to improve the performance of the neural network you intend to train. For example, the data from training D can be scaled by normalization, or by standardization (z-score). Subsequently, in a stage 306 the model is trained based on the training data D. In particular, the model, including the set of weights synaptic input related to WI intentions and WC driving contexts, is trained using i,ji,k methods such as, for example, the error back-propagation method (backpropagation) coupled with stochastic gradient descent (stochastic gradient descent), for a predetermined duration of time. Subsequently, in a stage 307 the trained model is applied to a set of test data, and the results obtained are stored. Subsequently, in a 308 step the results obtained from the data set of tests are evaluated, and in particular a value assumed by a function is evaluated of indicative cost of the relevance of the results obtained with respect to the expected results. Based on the values ​​obtained from the cost function in step 308, in a stage 309 the model is calibrated to reduce the value obtained from the function of cost, until it is minimized. In particular, the calibration of the model can understand how to vary, therefore increase or decrease, the value of each weight synaptic included in the model. For this purpose, in a step 310 the performance of the model is evaluated, for example example by checking whether the cost function evaluated in the previous step 308 returns a value close to a respective minimum point. If not, in a step 311 the training data D is filtered and processed again, for example by selecting some features salient through Principal Component Analysis. Consequently, the training procedure 300 is repeated starting from step 303 above described, for example until an optimal, indicative value is obtained of the fact that the desired level of performance is achieved, in the phases of evaluation of the cost function 308 and 310. Alternatively, the training procedure 300 may be repeated from phase 306 by changing one or more parameters related to the training model used. If so, procedure 300 continues to a phase 312 in which the set of synaptic weights related to WI intentions and WC driving contexts is i,ji,k stored, for example to be used in a run of the 200 method for detecting intentions of a driver in a vehicle above described. In light of the above, the characteristics of this solution are clear, as are its advantages. In summary, the solution described here includes a process 200 capable of predicting a driver's intentions a vehicle based on neural activity recorded through sensors for encephalogram 120. In particular, the identification of a driving context and / or of the driver's state of distraction contribute to increasing robustness, reliability, and adaptability in real-world use scenarios. For this purpose, the control system 100 described above comprises means of V2X communication such as, for example, the respective transceiver 130, and 150 sensors for identifying a driving context. Furthermore, the system control 100 includes EEG sensors 120 that are easily wearable by the driver and allow the detection of brain activity. The 200 procedure for detecting the intentions of a driver above described is able to achieve this goal on the basis of data regarding the driver's brain activity, data regarding the driving context, and data regarding the driver's state of distraction, extracting features from the acquired EG encephalograms, calibrating synaptic weights based on the acquired data regarding, driving context and driver's state of distraction, calculating finally a composite score based on synaptic weights and brain activity registered. Therefore, the solution described here advantageously allows a self-driving vehicle to prevent accidents thanks to increased knowledge of the driver's intentions, and at the same time to increase the robustness and reliability of forecasts thanks to the combination of data from brain activity, driving context data, and distraction state data. Naturally, without prejudice to the principle of the invention, the details of construction and implementation forms of the lighting system may be widely varied from what is described and illustrated, purely for information purposes example, without thereby departing from the scope of this solution. The scope of protection is defined by the attached claims.

Claims

1. Procedure (200) for determining a driver's intentions in a vehicle (10) comprising carrying out the phases of: - associate (202) first values ​​of synaptic weight (WI ) to respective i,j combinations of characteristics (F ) and detectable intentions (I); ij - associate (203) second synaptic weight (WC) values ​​to respective i,k combinations of features (F ) and detectable driving contexts (C ); ik - acquire (204) an encephalogram (EG) of the driver; - extract (206) a plurality of features (F ) from said the encephalogram (EG); - determine (207) whether the driver is distracted; - estimate (212) a composite score (S ) for each intention j detectable (I) on the basis of said characteristics (F ), of said first weight values ji synaptic (WI ), of said second synaptic weight values ​​(WC ), and of said state of i,ji,k driver distraction (W ); D - identify (213) the highest composite score (S ), together with the j respective intention (I). j 2. Process (200) according to claim 1, comprising the steps Of: - check (214) whether the maximum composite score value (S ) is j greater than a threshold value (TS ) associated with a respective intention (I ); jj - store (215) in a message (Q) the driver's intention (I) j detected. 3 . Procedure (200) according to any of the preceding claims, comprising a step of acquiring (201) an encephalogram of reference (ER) regarding the driver.

4. Method (200) according to claim 3, comprising a filtering phase (205) said acquired encephalogram (EG) (204) on the basis of called reference encephalogram (RE).

5. Process (200) according to claim 4, comprising the steps Of: - determine (209) a driving context (C ), in particular on the basis of k data acquired by V2X devices and ADAS systems and / or sensors (150); - in response to not determining (209) any driving context (C), k initialize (210) said synaptic weight values ​​(WC ), in particular to a value i,k equal to 1; - in response to successfully determining (209) a driving context (C ), k update (211) said synaptic weight (WC) values ​​based on the context of i,k guide (Ck) detected. 6 . Procedure (200) according to any of the preceding claims, comprising a step of storing (208) a state and resume the execution of the procedure (200) from said acquisition phase (204) an encephalogram (EG) in confirmation to determine (207) that the driver is distracted.

7. Procedure (200) according to any of the above claims, wherein said synaptic weight values ​​(WI, WC) have respective i,ji,k values ​​initially determined by means of a procedure (300) of training.

8. Procedure (200) according to any of the above claims, wherein said step of determining (207) whether the driver is distracted is carried out on the basis of data obtained from video cameras (140) processed by artificial vision procedures and data obtained from ADAS systems and sensors (150). 9 . Control system (100) for determining a driver's intentions in a vehicle (10) comprising: - a control unit (110) configured to implement the process (200) according to any of claims 1 to 8; - one or more encephalogram sensors (120); - a transceiver (130); - one or more video cameras (140); and - one or more sensors (150). 10 . Vehicle (10) comprising a control system (100) according to the claim 9.