Method and system for determining the context in which a vehicle is traveling

Abstract

TITLE: Method and system for determining the context in which a vehicle operates. A method for determining the context in which an ego vehicle equipped with a driver assistance system operates, the method comprising at least one prior learning phase and a phase for determining the context in which the vehicle operates, including a step of acquiring data relating to at least one primary parameter relating to the vehicle's speed and one secondary parameter relating to the context in which the vehicle operates, and a step of determining the context in which the ego vehicle operates based on the acquired data. Abstract figure: Figure 2

Description

Title of the invention: Method and system for determining the context in which a vehicle is traveling

[0001] The invention relates to a method for determining the traffic context of a vehicle. The invention further relates to a system for determining the traffic context of a vehicle.

[0002] In the automotive industry, it is known to equip vehicles with driver assistance systems, or AD AS from the English "Advanced Driving Assist System", capable of assisting or managing vehicle driving functions such as vehicle positioning in road infrastructure, for example in the absence of markings, or vehicle speed management based on an estimation of the context of the environment in which the vehicle is traveling.

[0003] Determining the context, i.e. the environment, in which the vehicle is operating can provide essential information to driver assistance systems, particularly when fusing information from at least one sensor, corresponding to a combination operation of the data obtained by the vehicle necessary for the implementation of the vehicle's driver assistance functions.

[0004] For example, characteristics of traffic lanes, particularly their width, can vary depending on whether one is in a city or on a highway. Knowledge of the environment can therefore aid fusion. In particular, lane detection algorithms can virtually divide the main lane into multiple lanes, the number of lanes and the lane width varying according to the context in which the ego vehicle is located.

[0005] Similarly, the maximum authorized vehicle speed, or safety distances relative to other vehicles on the road, vary depending on the context in which the vehicle is operating and can be estimated by the driver assistance system based on an estimated context. Therefore, it is important to know the context in which the vehicle is operating.

[0006] Traditionally, determining the context in which a vehicle operates, for example, to determine whether the vehicle is operating in a city or on a highway, is based on analyzing data relating to the vehicle's speed. However, it has been found that the reliability of determining the context based on speed alone is insufficient. Alternatively, it is known to estimate such a context based on the vehicle's geolocation data and a map database.

[0007] The invention falls within this context and aims to provide a method for determining the context in which the vehicle is traveling, allowing for improved detection and determination of the context in which the vehicle is traveling at low cost, in particular by using equipment conventionally installed in the vehicle, and without resorting to a mapping database and geolocation.

[0008] The invention relates to a method for determining the context in which an ego vehicle equipped with a driver assistance system is operating, the method comprising a phase for determining the context in which the vehicle is operating, when it is in motion, comprising: - a data acquisition step comprising at least one primary parameter, relating to the speed of the ego vehicle determined via a speed sensor, and at least one secondary parameter relating to the context in which the ego vehicle is operating; and - a step of determining, by a processing unit, the context in which the ego vehicle circulates by discrimination between at least two classes of possible contexts by application of a statistical supervised learning algorithm trained to operate a distinction between at least two classes of possible contexts from the acquired data so as to determine a result corresponding to a class of context in which the ego vehicle circulates.

[0009] In particular, at least one secondary parameter relating to the environment in which the ego vehicle operates comprises at least one of the following: - a speed of at least one secondary vehicle circulating in the road infrastructure, determined by means of a detection device; - information relating to at least one secondary vehicle, such as a number of static vehicles, including parked vehicles, and / or a number of secondary vehicles of a specific type; - information relating to at least one element, fixed or mobile, of the road infrastructure, such as the length and / or spacing of lane marking lines, information provided by traffic signs or traffic lights and / or the presence or number of specific road infrastructure elements, such as traffic lights, speed bumps, roundabouts or other; - a number of pedestrians detected.

[0010] According to an example of execution, the supervised learning statistical algorithm is a linear discriminant analysis type algorithm, also called "Fisher Discriminant Analysis", trained to differentiate between two classes of contexts or to differentiate between a considered class of context and a set comprising classes of contexts distinct from the considered class.

[0011] For example, for a number of possible context classes strictly greater than two, the process includes a plurality of iterations of the step of determining the context in which the vehicle is traveling, each of said steps being configured to perform the discrimination between a considered context class and a set comprising context classes distinct from the considered class.

[0012] For example, the step of determining the context in which the vehicle is traveling by the linear discriminant analysis type algorithm includes the implementation of a scalar product between a vector comprising the acquired data and a predefined reference vector, and then the implementation of a substep of exploiting the result of such a scalar product.

[0013] In particular, the exploitation of the result of the scalar product includes: - comparing said result with at least one reference threshold; or - applying a statistical filter to said result, in particular a Bayesian filter or a Kalman filter, using at least one previous result of determining a context class in which the vehicle travels determined during a previous execution of the determination phase.

[0014] For example, the method according to the invention includes comparing said result with at least one reference threshold and includes, based on at least one previous result of determining a context class in which the vehicle is traveling determined during a previous execution of the determination phase: - adjusting at least one reference threshold based on at least one previous result of determining a context class; and / or - the recording or input of at least one piece of information relating to a driver profile, such as a preferred route or a probability of driving in a predetermined context class, and the adjustment of at least one reference threshold according to said profile.

[0015] In particular, the data acquisition and the step of determining the context in which the vehicle is traveling are carried out: - at a predefined time interval; or - at a predefined distance interval.

[0016] Optionally, the method includes, prior to the phase of determining a context in which the vehicle travels from among at least two possible context classes, a phase of training the supervised learning statistical algorithm, configured to determine a model, in particular at least one reference vector, allowing the best possible discrimination between at least two possible distinct context classes in which the vehicle is likely to travel or at least one considered context class and a set of the other context classes in which the vehicle is likely to travel, based on reference data. derived from a database and / or acquired beforehand, said reference data includes a primary parameter relating to a speed of the ego vehicle and at least one secondary parameter representative of the context in which the ego vehicle evolves.

[0017] Optionally, the method further includes a step of using at least one result of determining a class of context in which the vehicle is traveling by the driver assistance system including the modification of at least one setting of the driving of the ego vehicle, such as a display or projection of information, a vehicle speed, an acceleration and / or a position of the vehicle in the road infrastructure.

[0018] The invention also extends to a system for determining a context in which an ego vehicle is circulating, the system comprising hardware and / or software elements implementing the determination method according to the invention, the hardware elements comprising at least one data processing unit, at least one sensor capable of extracting data relating to the ego vehicle and a plurality of data capture means, capable of extracting data relating to the ego vehicle and / or the environment outside the ego vehicle.

[0019] The invention further extends to a motor vehicle comprising a determination system according to the invention.

[0020] The present invention may be further extended to a computer program product comprising program code instructions recorded on a computer-readable medium to implement the steps of the process as described above when said program is run on a computer. Alternatively, the invention may relate to a computer program product downloadable from a communication network and / or recorded on a computer-readable and / or computer-executable data medium, the program comprising instructions which, when the program is executed by the computer, cause the computer to implement the process according to the invention.

[0021] The invention also relates to a computer-readable data recording medium on which is recorded a computer program comprising program code instructions for implementing the process according to the invention, or to a computer-readable recording medium comprising instructions which, when executed by a computer, cause the computer to implement the process according to the invention.

[0022] Finally, the invention may relate to a signal from a data carrier carrying the computer program product as described above.

[0023] Other details, features and advantages will become clearer upon reading the detailed description given below, by way of example and not limitation, in relation to the various embodiments illustrated in the following figures:

[0024] Fig. 1 schematically represents an embodiment of a vehicle equipped with a system for determining the context in which the vehicle is traveling.

[0025] The [Fig.2] is a general flowchart for the execution of a method for determining a context in which the vehicle is moving.

[0026] Fig. 3 is a flowchart detailing alternatives to the steps of the determination process.

[0027] Fig. 4 is a graphical representation illustrating the output probability density functions of a supervised learning statistical algorithm for two classes of contexts.

[0028] Figure 1 schematically illustrates a motor vehicle 1 according to an embodiment of the invention. The vehicle 1 can be of any type. In particular, it can be a passenger car, a commercial vehicle, a truck, or a bus. The vehicle 1 can also be an autonomous vehicle.

[0029] Throughout the description below, the vehicle 1, comprising the means for implementing the invention, may also be referred to as the "ego" vehicle to distinguish it from other surrounding vehicles. Such a designation does not in itself impose any technical limitations on the motor vehicle. Vehicles circulating on the road infrastructure that are distinct from the ego vehicle 1 will also be referred to as secondary vehicles. Secondary vehicles may be of any type, including bicycles or motor vehicles, for example, motorcycles, cars, trucks, buses, electric bicycles, or other electric means of locomotion.

[0030] The vehicle 1 comprises hardware and / or software elements capable of implementing a method 100 for determining a driving context of the vehicle 1 ego when it is in motion, as illustrated in Figures 2 and 3 and detailed below. The determination of a "context" is understood to mean the determination of a specific environmental class from among a plurality of classes included in a predefined classification.

[0031] Throughout the description below, the context under consideration relates to a classification distinguishing different traffic zones for vehicle 1 ego in which traffic conditions and / or road infrastructure vary, for example, comprising a plurality of classes selected from a "city" or "city center" class, an "urban area" or "peri-urban" class, and a "highway" class. It is understood that such a principle can be extended to other classes, for example, to include "mountain" or other areas. According to alternative embodiments not detailed herein, the context may aim to distinguish classes relating to road conditions or other. It is understood that the description applies mutatis mutandis to such alternatives.

[0032] The vehicle 1 ego is equipped with at least one driving assistance system 2, or AD AS, from the English "Advanced Driving Assist System" capable, without limitation, of adapting the driving conditions of the vehicle 1 according to a given traffic context, for example capable of adapting a position of the vehicle 1 on the road, of estimating a position of marking lines delimiting lanes within the road traveled by the vehicle 1 ego and / or capable of adapting a speed, or an acceleration, of the vehicle 1 ego.

[0033] For example and without limitation, the driving assistance system 2 includes a Contextual Adaptive Cruise Control system, also known by the acronym Contextual ACC or Contextual ACC from the English "Contextual Adaptive Cruise Control", allowing the speed of vehicle 1 to be adapted according to the context in which vehicle 1 is traveling.

[0034] The vehicle comprises a context determination system 10 and / or the vehicle 1 comprising a plurality of data capture means 3 configured to extract data relating to the vehicle 1 and / or data relating to the external environment, for example, to secondary vehicles and / or to fixed or mobile elements present in the external environment. In particular, the data capture means 3, the context determination system 10, and / or the vehicle 1 comprise at least one speed sensor 31 capable of determining the speed of the vehicle 1 and advantageously its acceleration.

[0035] Also, the data capture means 3, the context determination system 10, and / or the vehicle 1 may include at least one detection means 32, for example, an image capture means, configured to extract data relating to the environment outside the vehicle 1 ego or to elements or vehicles surrounding the vehicle 1 ego. For example, the detection means 32 may be a radar and / or a lidar and / or a camera. Preferably, the vehicle 1 ego includes a plurality of detection means 32, for example, equipped at the front or on the sides of the vehicle 1 ego, in particular at the side doors or the rearview mirrors. As further detailed below with reference to the method 100 according to the invention, these are, without limitation, capable of performing at least one of the following: - detect the speed or acceleration of a secondary vehicle; - detect a number of secondary vehicles of a particular type, such as trucks or cyclists, or a number of pedestrians; - detect and identify road signs and / or speed limit signs; - detect topography and / or geometry of the road or road infrastructure surrounding the vehicle 1 ego.

[0036] The determination system 10 also includes a data processing unit 4. The processing unit 4 comprises at least one computing unit with hardware and software resources, more specifically at least one processor or microprocessor, cooperating with memory elements of the vehicle 1. Alternatively, the processing unit 4 comprises such memory elements. This computing unit is capable of executing instructions for the implementation of a computer program. The processing unit 4 is particularly capable of receiving data from the data capture means 3, in particular from at least one speed sensor 31 and at least one detection means 32, and from the driver assistance system 2.

[0037] An example of the execution of the method 100 for determining a traffic context of the vehicle 1 ego during the driving phase is described below with reference to Figures 2 and 3. The method 100 for determining a traffic context of the vehicle 1 ego can also be considered as a method 100 for operating the determination system 10 according to the invention or as a method 100 for operating a motor vehicle 1 equipped with such a system.

[0038] Generally, the method 100 for determining a driving context for vehicle 1 ego includes a prior learning phase H01 executed at least once, for example prior to putting vehicle 1 into circulation, and a determination phase H02 of a context in which vehicle 1 ego circulates, implemented during driving phases of vehicle 1 ego. Advantageously, the learning phase H01 can be common to a plurality of vehicles 1 ego, that is to say, it can be executed once for a series of vehicles, whether of the same model or not, and it is not necessary to execute it individually for each vehicle 1 ego.

[0039] The H01 learning phase enables the training of a supervised learning statistical algorithm to distinguish between a plurality of driving context classes from input data. The input data are, for example, extracted from a driving database and / or come from observations made from a vehicle.

[0040] In particular, the H01 learning phase includes the comparison between different parameters, taken from at least one driving database and / or from measurements taken on the vehicle, with situations representative of real driving conditions, or real terrain, for example obtained using a satellite positioning system, also known by the acronym GNSS for " Global Navigation Satellite System”, and a map database and / or identified by an operator.

[0041] It should be noted that, during the learning phase H01 and during the implementation of a subsequent phase of determination H02 of a context in which the vehicle 1 ego circulates, described below, the different parameters are organized in the same way.

[0042] The objective of the H01 learning phase is to train a supervised learning statistical algorithm to distinguish between at least two distinct classes of driving contexts, that is, between two distinct classes of driving contexts or between a given class and a set comprising a plurality of classes distinct from the given class. Unless otherwise stated, the following description refers to a distinction between two distinct classes of driving contexts; however, it is understood that the description extends mutatis mutandis to a distinction between a given class of context and a set comprising a plurality of classes of contexts distinct from the given class, as further detailed below.In the description below, a "pair" will be defined as a pair of distinct classes opposed in the supervised learning statistical algorithm so as to distinguish between said classes, or a pair comprising a considered class and the set formed by the "remaining" classes, distinct from the considered class.

[0043] For example, the supervised learning statistical algorithm makes it possible to distinguish the type of traffic zone of vehicle 1 from among several classes, the classes being, according to a non-limiting example of an embodiment, selected from a "city" or "city center" class, a "suburban" class, corresponding for example to traffic in the suburbs or on the outskirts of a city, and a "highway" class. It is understood that said classes can also extend to other traffic zones, such as "mountain" or other zones.

[0044] According to a preferred embodiment, the training phase H01 relates to a linear discriminant analysis algorithm, in particular an FDA algorithm, from the English "Fisher Discriminant Analysis". Alternatively, the statistical machine learning algorithm trained during the training phase H01 may be any other type of classification algorithm, for example: - an SVM algorithm, from the English "Support Vector Machine"; - a logistic regression; - a "k nearest neighbors" or KNN algorithm, from the English "K-Nearest neighbors".

[0045] Subsequently, the description is made in connection with the use of an "FDA" type algorithm. This algorithm has the advantage of being very simple and inexpensive to implement, since during the inference phase, it is reduced to a simple product scalar between a vector of observed data and a reference vector constructed during the training phase. Its drawback is that it does not allow the distinction of more than two classes of contexts.

[0046] Thus, when there are only two classes W1 and W2, the FDA algorithm is trained in such a way as to allow the distinction between the two classes, i.e., W1 versus W2. The H01 learning phase therefore allows the definition of a reference vector which, when associated with observational data, makes it possible to determine the most probable context among the contexts W1 and W2.

[0047] To operate the distinction between more than two classes of contexts using the FDA, it is necessary to train the algorithm several times in order to obtain a reference vector per pair of opposing classes, and by extension per pair comprising a considered class and a set comprising the remaining classes, corresponding to a vector of different parameters K of data having optimal values ​​in order to differentiate two distinct classes, when only two classes are opposed, or in order to differentiate a considered class and a set comprising the remaining classes.

[0048] For example, and without limitation, in the case of a distinction between three classes W1, W2 and W3, three training runs of the supervised learning statistical algorithm are necessary in order to establish three reference vectors respectively trained to allow, when associated with observational data, the determination of the most probable context between: - the class Wl and the set formed by the other classes, namely W2 and W3; - class W2 and the set formed by the other classes, namely W1 and W3; and - class W3 and the set formed by the other classes, namely W2 and Wl.

[0049] Note that in this case, Nl iterations would be sufficient, N here corresponding to number of classes, the context being of class W3 when the observation of classes W1 and W2 concludes each time that the context does not belong to these classes.

[0050] In the case of the FDA algorithm, the training performed during the learning phase makes it possible to obtain a reference vector for which the dot product with observational data minimizes the standard deviation and maximizes the difference between the mean between two classes or between a given class and the set including the remaining classes, i.e., for a given pair. This reference vector is implemented during a context determination H02 phase.

[0051] According to one execution mode, the training phase H01 includes defining at least one representation of a probability density function of the outputs of the trained statistical algorithm for each pair of context classes considered, or each pair consisting of the class considered and a set of other possible classes. This probability density can be represented, for example, as The form of a table or graph, as illustrated in [Fig. 4], allows for easy analysis of this data. This principle is notably implemented in the execution of a supervised learning statistical algorithm of the linear discriminant analysis type, specifically a FDA algorithm.

[0052] For example, after defining at least one reference vector, at least one rolling database and / or the measurements used for the training phase and / or other measurements specifically related to one of the classes or sets of classes to be discriminated against can be processed by the trained supervised learning statistical algorithm, so as to construct a histogram based on the algorithm's results for each class or set of classes considered, and to represent the probability density function of the algorithm's output for that class or set of classes. Such a representation then makes it possible to set a threshold for discriminating between the classes, or a pair comprising a considered class and the remaining associated classes, and to associate a level of reliability for the context determination performed during the H02 context determination phase.

[0053] Figure 4 represents the probability distribution of the outputs of the trained supervised learning statistical algorithm, shown here as dashed lines for a "city" driving context and as dotted lines for a "highway" driving context. It is clear that a threshold value close to 4 allows for a clear distinction between the two "city" and "highway" contexts.

[0054] The determination phase H02 of the process 100 is carried out after the vehicle 1 ego has been put into service, in particular during periods when the vehicle 1 ego is in operation, corresponding, for example, to periods during which an engine of the vehicle 1 ego is running. As detailed below, the determination phase H02 can be repeated at defined time or distance intervals.

[0055] The determination phase H02 of a traffic context of vehicle 1 ego comprises, initially, a data acquisition step E01. Generally, said data relates to a primary parameter Kxl relating to a longitudinal speed of vehicle ego and to at least one secondary parameter Kx2 relating to a parameter representative of the context in which the vehicle is moving.

[0056] In particular, at least one secondary parameter Kx2 is selected from an acceleration of vehicle 1 ego, a speed or acceleration of at least one secondary vehicle circulating in the road infrastructure, information relating to at least one secondary vehicle and / or information relating to at least one element, fixed or mobile, of the road infrastructure and / or any other parameter representative of the context in which the vehicle is circulating.

[0057] The primary parameter Kxl, relating to the longitudinal speed of the vehicle 1 ego is determined by means of at least one speed sensor 31.

[0058] The secondary parameter(s) Kx2 representative of the context in which the vehicle operates may include: - an acceleration of vehicle 1 ego, determined via the speed sensor 31 and / or the braking sensors; - information relating to one or more secondary vehicles, and / or one or more fixed or mobile elements of the road infrastructure, and / or any other parameter representative of the context in which the vehicle is traveling, acquired by at least one data capture means 3, in particular an image capture means of the image capture means and / or a data capture means 3, in particular of a type distinct from that used for measuring speed data.

[0059] For example, the at least one secondary parameter Kx2 may include, but not be limited to, one or more of the following parameters: - a speed or maximum speed of at least one secondary vehicle circulating in the road infrastructure, obtained by at least one detection means 32; - a number of pedestrians detected, notably detected via an image capture device; - a number of secondary vehicles of a specific type, for example heavy goods vehicles or bicycles; - a number of static secondary vehicles, including parked ones; - the length and / or spacing of marking lines for a traffic lane; - information provided by road signs or traffic lights; - the presence or number of specific road infrastructure elements, such as traffic lights, speed bumps, roundabouts or other.

[0060] Indeed, the speed of vehicle 1 ego or secondary vehicles typically varies depending on the context. For example, the speed of vehicle 1 ego, or secondary vehicles traveling on the road infrastructure, is typically higher on highways than in urban areas. Similarly, acceleration and braking tend to be more frequent in urban areas than on highways. Using secondary data relating to at least one parameter representative of the context in which the vehicle is traveling, in addition to the vehicle's speed, allows for a more informed decision regarding the context class in which vehicle 1 ego is likely to be traveling. Using separate data capture methods increases the reliability of the determined context class.The greater the amount of secondary data, the greater the reliability of the context determination method according to the invention, particularly when the data sources are diverse.

[0061] The presence or number of pedestrians, bicycles or heavy goods vehicles tends to vary depending on the context in which vehicle 1 is traveling. Similarly, a greater number of traffic lights, roundabouts or speed bumps is classically observed in town than on motorway.

[0062] According to one execution mode, the acquisition of data relating to the various primary and secondary parameter(s) is carried out at predefined time intervals or in real time.

[0063] According to an alternative execution, the acquisition of data relating to the various primary and secondary parameter(s) is carried out at predefined distance intervals.

[0064] The process 100 then includes a determination step E02, by a processing unit 4, of a context in which the vehicle 1 ego circulates on the basis of the acquired data.

[0065] The determination step E02 includes a substep of applying the trained supervised learning statistical algorithm E021 to the acquired data, also referred to as the inference substep. This step makes it possible, in particular, to discriminate between classes among a plurality of classes to be determined.

[0066] According to a particular and preferred example, the executed supervised learning statistical algorithm is a linear discriminant analysis algorithm, specifically a Fisher Discriminant Analysis (FDA) algorithm. During the execution of such an algorithm, the acquired data are organized into a vector whose structure is identical to that of the reference vector established during the training phase H01. The processing unit 4 is configured to perform the dot product between the reference vector specific to the pair under consideration—namely, the pair of classes considered or the pair formed by the class under consideration and the set comprising the remaining classes—determined during the training phase H01, and the vector of acquired data corresponding to the primary and secondary parameters.The scalar resulting from the operation is used in a substep E022 of the exploitation of the result of said supervised learning statistical algorithm to determine the context in which the vehicle evolves among the considered pair. When the number of classes considered is greater than two, the determination step must be implemented several times, each implementation of the step using a reference vector relative to a pair as defined above and allowing the probabilities of belonging of the context to the considered class or to the set formed by the other classes.

[0067] According to a particular execution mode, the exploitation E022 of the result of the scalar product includes the comparison of the resulting scalar with a predefined reference threshold Srx.

[0068] For example, in the previously mentioned case of the distinction between three classes W1, W2 and W3, the method according to the invention comprises three implementations of step E02 with: - the realization of a scalar product between the reference vector allowing the class Wl to be distinguished from the other classes; - the realization of a dot product between the reference vector allowing the class W2 to be distinguished from the other classes; - the realization of a scalar product between the reference vector allowing to distinguish the class W3 from the other classes.

[0069] For each pair considered, namely a pair of classes or a pair formed by a class considered and a set including the remaining classes, the result of the scalar product obtained can be compared with a reference threshold Srx specific to each pair or be treated by a Bayesian filtering, as described below, in order to determine the probabilities of belonging to the context in which the vehicle evolves for each pair considered.

[0070] Note that in practice, for a number N of classes, Nl implementations, or iterations, of such an exploitation step E22 is sufficient to determine the context in which the vehicle is operating, as explained above.

[0071] According to one embodiment, the exploitation E022 of the result of the dot product is carried out by comparing this result with a reference threshold Srx. The reference threshold Srx can be obtained from the table or graph representing the probability density functions of the outputs of the supervised learning statistical algorithm for each context class, as shown in [Fig.4].

[0072] The reference threshold Srx corresponds, for example, to an intersection value between two curves of distinct context classes, to a value between two curves without intersection, or to a value minimizing the risk of false detection. Such an example of execution then corresponds to an ML detection principle, or "ML detection" from the English "Maximum Likelihood Detection".

[0073] When several reference vectors are established during the learning phase in order to differentiate a plurality of classes for an FDA-type algorithm, a reference threshold Srx is associated with each of these reference vectors.

[0074] The context determination step E02 and / or the substeps of applying the supervised learning statistical algorithm E021 to the acquired data and exploiting the result E022 can be applied at regular time intervals.

[0075] Alternatively, they can be applied at regular distance intervals. Indeed, the context in which a vehicle operates is more likely to vary over a spatial horizon than over a temporal horizon. For example, it is rare It is likely that there will be a change of context between two iterations of a location-determination algorithm implemented at short time intervals for a vehicle traveling at low speed in the city, or for a vehicle traveling on a highway. Implementing the algorithm at regular distance intervals therefore makes more efficient use of computing resources than implementing it at regular time intervals.

[0076] According to another embodiment, the E022 results exploitation substep includes the application of a statistical filter using at least one prior determination of the context class in which the vehicle 1 travels, determined during a prior execution of the method 100 according to the invention.

[0077] Such statistical filtering advantageously allows for consideration of previous results in determining the traffic context, in other words, the previous decisions of the processing unit 4, in order to reduce potential oscillations between distinct context classes due to estimation errors, or between a given class and the remaining sets of classes, for example, when the result of the dot product obtained after the execution of the supervised learning statistical algorithm is close to or equal to at least one reference threshold Srx. The determination of the traffic context of vehicle 1 ego is thus made more robust by allowing the determination of membership in a detected context class through previously acquired data based on Bayesian probabilities at each decision, i.e., at each determination step E02.

[0078] Thus, in such an execution mode, the determination phase H02 of a vehicle 1 traffic context initially comprises the implementation E021 of a dot product between a vector containing the acquired data and a reference vector, as described above. This value can be assigned to the distribution function of the outputs of the trained supervised learning statistical algorithm, in order to determine a probability p(FDA = ^context) that the dot product has the value obtained for each traffic context class. Applying a Bayesian filter or a Kalman filter, taking into account the context probability obtained during a previous execution of the process, makes it possible to obtain the probability p(contextdFDA = k) of belonging to each context class as a function of the result of the dot product and the previous measurements.It is then possible to compare these probabilities to make a final decision, that is, to determine the final traffic context class of vehicle 1 ego, by selecting the most probable class. The processing unit 4 is thus capable of performing the fusion of a provisional decision to determine a context class at time t with one or more previous decisions to determine a class of . context executed at one or more instant(s) tx corresponding to at least one previous execution of the determination phase H02 and / or the determination process 100 according to the invention.

[0079] Such an execution method makes it possible to obtain a result using a MAP detector, from the English "Maximum a Posteriori Detector", rather than via a "Maximum Likelihood" detector. The MAP detector has the advantage of making the best use of the output probabilities of the dot product and the previous decisions.

[0080] Determining the context class based on the MAP detector therefore allows for an improvement in the reliability and robustness of the determination compared to simple comparison with at least one reference threshold Srx.

[0081] The method 100 for determining a context according to the invention optionally includes, in addition, a step E04 of using the determined context class in which the vehicle 1 is operating by the driver assistance system 2, for example, for the fusion of data from different sensors within the framework of driver assistance systems, such as lane modeling. In particular, during such a step, the driver assistance system 2 is capable of modifying at least one driving and / or vehicle setting. For example, such a driving setting may relate to at least one of the following: - a display, via a Human Machine interface, of information relating to the context or determined from the context, for example relating to traffic lanes; - a projection of context-related or context-determined information, for example within the vehicle or onto the roadway, by means of a projection device, for example the projection of traffic lane markings; - the modification of a vehicle 1 ego speed, acceleration and / or maximum driving speed of vehicle 1 ego; - the modification of a position of vehicle 1 ego in the road infrastructure.

[0082] Optionally, the determination method 100 includes adjusting the data acquisition interval E052, in other words their acquisition frequency, according to at least one previous result of determining a context class in which the vehicle 1 travels, determined during a previous execution carried out at time tx, of the method 100.

[0083] For example, the time or distance interval between successive data acquisitions can be adapted according to the specific context in which vehicle 1 is traveling. In this case, such an interval can be increased when it is determined, for example, at a time tx, that vehicle 1 is traveling on a highway. Alternatively or Additionally, such an adjustment is made when the same context class is determined after a plurality of successive executions of the determination phase H02, for example, when the number of successive repetitions Zres of obtaining the same context class result is greater than or equal to a predefined repetition threshold Zsmin. Such an adjustment is particularly advantageous in the case of data acquisition performed at defined intervals, since it reduces the load on the determination system, and therefore allows the process according to the invention to be executed at a lower frequency when the context does not require it.

[0084] Such an adjustment E052 is implemented by recording at least one result of determining a context class on at least one memory element after a previous execution of process 100 and then at least one substep of comparing the previous result with pre-recorded data, associating the different context classes with predefined distance and / or time intervals of data acquisition and / or comparing a number of repetitions of obtaining the same context class result with the predefined repetition threshold Zsmin.

[0085] Optionally, the method 100 according to the invention includes the adjustment E053 of at least one reference threshold Srx, discriminating between a pair comprising two context classes or a considered class and a set comprising the remaining classes, based on the prior determination of a context class. The preceding description applies mutatis mutandis. In particular, the reference threshold Srx can be adjusted based on a repetition frequency of a context class and / or based on a driver profile. For example, the reference threshold Srx can be adjusted when it is detected that the vehicle 1, for a given driver, is traveling in a defined context class at a frequency greater than or equal to a reference frequency Fkx. For example, said threshold is adjusted when it is detected that the defined driver is traveling more than 70%, or even more than 80%, of the time, or of the trips made, in one of the context classes.

[0086] In order to enable such an adjustment E053, the method 100 may include, beforehand, the recording or input of at least one piece of information relating to a driver profile, such as a preferred route, addresses and / or a probability of traveling in a calculated context class and then the adjustment E053 of at least one reference threshold Srx discriminating between at least two context classes according to said profile.

[0087] The present invention thus proposes a method and a system for determining a context class in which the ego vehicle circulates, advantageously optimizing the determination of the context class in which the vehicle for the purpose of using such a determination by a driver assistance system, and at a lower cost.

[0088] The present invention cannot, however, be limited to the means and configurations described and illustrated herein and it also extends to any equivalent means or configuration and to any technically operative combination of such means insofar as they ultimately fulfill the functionalities described and illustrated in this document.

Claims

Demands

1. Method (100) for determining a context in which a vehicle (1) ego equipped with a driving assistance system (2) is traveling, the method (100) comprising a phase of determining (H02) a context in which the vehicle (1) is traveling, when it is in a driving situation, comprising: - a step of acquiring data (E01) of which at least one primary parameter (Kxl), relating to a speed of the vehicle ego determined by means of a speed sensor (31), and at least one secondary parameter (Kx2) relating to the context in which the vehicle (1) ego is traveling;and - a determination step (E02), by a processing unit (4), of the context in which the vehicle (1) ego circulates by discrimination between at least two classes of possible contexts by application of a supervised learning statistical algorithm trained to operate a distinction between at least two classes of possible contexts from the acquired data so as to determine a result corresponding to a class of context in which the vehicle (1) ego circulates.;

2. A method (100) for determining a context according to the preceding claim, wherein the at least one secondary parameter (Kx2) relating to the environment in which the vehicle (1) ego is traveling comprises at least one of: - a speed of at least one secondary vehicle traveling in the road infrastructure, determined by means of a detection means (32); - information relating to at least one secondary vehicle, such as a number of static vehicles, in particular parked, and / or a number of secondary vehicles of a specified type;- information relating to at least one element, fixed or mobile, of the road infrastructure, such as the length and / or spacing of lane marking lines, information provided by traffic signs or traffic lights and / or the presence or number of specific road infrastructure elements, such as traffic lights, speed bumps, roundabouts or other - a number of pedestrians detected.;

3. A method (100) for determining a context according to any one of the preceding claims, wherein the supervised learning statistical algorithm is a linear discriminant analysis type algorithm, also referred to as "Fisher Discriminant Analysis", trained to differentiate between two classes of contexts or to differentiate between a considered class of context and a set comprising classes of contexts distinct from the considered class.

4. A method (100) for determining a context according to the preceding claim, wherein, for a number of possible context classes (N) strictly greater than two, the method (100) comprises a plurality of iterations of the step of determining the context in which the vehicle (1) is traveling, each of said steps being configured to perform the discrimination between a considered context class and a set comprising context classes distinct from the considered class.

5. Method (100) of determining a context according to claim 3 or 4, comprising, the step of determining (E02) the context in which the vehicle (1) is traveling by the algorithm of the linear discriminant analysis type includes the implementation of a dot product between a vector comprising the acquired data and a predefined reference vector, then the implementation of a substep of exploiting (E022) the result of such a dot product.

6. Method (100) of determining a context according to the preceding claim, wherein the exploitation (E022) of the result of the dot product comprises: - comparing said result with at least one reference threshold (Srx); or - applying a statistical filtering (E023) to said result, in particular a Bayesian filtering or a Kalman filtering, using at least one prior result of determining a context class in which the vehicle (1) travels determined during a prior execution of the determination phase (H02).

7. A method (100) for determining a context according to the preceding claim, comprising comparing said result with at least one reference threshold (Srx) and comprising, based on at least one prior result for determining a context class in which the vehicle (1) circulates determined during a previous execution of the determination phase (H02): - the adjustment (E053) of at least one reference threshold (Srx) according to at least one previous result of determining a context class; and / or - the recording or input of at least one piece of information relating to a driver profile, such as a preferred route or a probability of circulating in a predetermined context class, and the adjustment (E053) of at least one reference threshold (Srx) according to said profile.

8. A method (100) for determining a context according to any one of the preceding claims, wherein the data acquisition (E01) and the step (E02) of determining the context in which the vehicle is traveling are carried out: - at a predefined time interval; or - at a predefined distance interval.

9. A method (100) for determining a context according to any one of the preceding claims, comprising, prior to the determination phase (H02) of a context in which the vehicle (1) circulates among at least two possible context classes, a learning phase (H01) of the supervised learning statistical algorithm, configured to determine a model, in particular at least one reference vector, enabling the best possible discrimination between at least two possible distinct context classes in which the vehicle (1) is likely to circulate or at least one context class considered and a set consisting of the other context classes in which the vehicle (1) is likely to circulate, from reference data from a database and / or acquired beforehand,said reference data comprising a primary parameter (Kxl) relating to a speed of the ego vehicle and at least one secondary parameter (Kx2) representative of the context in which the ego vehicle evolves.

10. A method (100) for determining a context according to any one of the preceding claims, further comprising a step (E04) of using at least one result for determining a context class in which the vehicle (1) is operating by means of the driver assistance system (2) comprising modifying at least one driving setting of the vehicle (1), such as a display or a projection of information, a vehicle speed (1), an acceleration and / or a vehicle position (1) in the road infrastructure.

11. System for determining (10) a context in which a vehicle (1) ego is moving, the system comprising hardware and / or software elements implementing the method (100) for determining according to any one of the preceding claims, the hardware elements comprising at least one data processing unit (4), at least one sensor capable of extracting data relating to the vehicle (1) ego and a plurality of data capture means, capable of extracting data relating to the vehicle (1) ego and / or the environment outside the vehicle (1) ego.

12. Motor vehicle (1) comprising a determination system according to the preceding claim.

Images