FR3135945A1 - Method for detecting a type of road traveled by an equipped vehicle - Google Patents

Abstract

Title: Method for detecting a type of road traveled by an equipped vehicle Method for detecting detection of a type of road traveled by a vehicle coupling a phase of determining a type of road and a phase of learning the identification of a road type class integrating an evaluation of the road comfort felt by a user for a vehicle of given model and equipment. No figure for the abstract.

Description

Method for detecting a type of road traveled by an equipped vehicle Technical field of the invention

The invention relates to a method of detection, in particular intelligent detection, of a type of road traveled intended for a vehicle, in particular for a vehicle in a driving situation. The invention also relates to a system for detecting a route traveled. The invention also relates to a vehicle equipped with such a system and to a computer program product for implementing the steps of the detection method.

State of the prior art

In the automotive world, it is known to equip vehicles with controlled suspension systems. Classically, such systems require knowing the state and/or characteristics of a road traveled by an equipped vehicle in order to best control the system according to the characteristics specific to the road traveled.

To this end, it is possible to perform methods for detecting bad roads and/or obstacles. Such methods depend on detection systems carrying out explicit calculations of variables and intermediate characteristics, or “features” in English. It is therefore necessary to carry out fine-tuning and fine-tuning of the detection system in order to obtain an acceptable level of precision. Another disadvantage of such detection systems lies in the fact that they are not adapted to take into consideration, for a type of bad road or obstacle detected, the variability of the feeling or comfort of the driver of a type, or model. , from vehicle to another, in particular due to their different equipment. The result is that a road detected as being bad for a certain vehicle will also be considered bad, wrongly, for a separate vehicle more suited to traveling on such a type of road, for example due to particular equipment such as more suitable suspension systems. Conversely, a road identified as good, or passable, could be experienced as bad and uncomfortable depending on the type of vehicle used. Such methods thus have a lack of portability making them unsuitable for application to a set of different vehicles.

The present invention fits into this context and aims to propose a detection method suitable for a wide range of vehicles having distinct characteristics, for example private vehicles, utility vehicles or others, and at lower cost.

The present invention relates to a method for detecting a type of road traveled by a vehicle, in particular a vehicle equipped with a variable suspension system, the method being characterized in that it comprises a phase of determining a type of route traveled including:
- at least one step of acquiring, in real time or for a predetermined duration, data relating to the vehicle and/or a route traveled by the vehicle implemented via a detection module comprising at least one sensor and/or at least one detection means, such as a camera;
- a step of estimating a probability rate of belonging of the road traveled to at least one road type class, in particular bad roads, predefined according to said acquired data;
- a step of classifying the road traveled in one of the road type classes according to the estimated membership probability rates and extracting at least one level of comfort felt by a user previously associated with each class of road type as a function of at least one piece of equipment specific to the vehicle (10) capable of influencing the comfort felt by a user during the driving phase.

In particular, the method is an intelligent detection method.

In particular, the method according to the invention may comprise, prior to the execution of the determination phase, at least one supervised learning phase, specific to the equipped vehicle, of the identification of at least one road type class predefined and the association of at least one level of comfort felt by a user with each class of road type as a function of at least one piece of equipment specific to the vehicle capable of influencing the comfort felt by a user during the driving phase, the learning phase being implemented via a computer, a human-machine interface and/or a memory element.

In addition, the method may include a step of adjusting at least one ancillary system fitted to the vehicle, in particular a variable suspension system, so as to correct a level of comfort felt by the user, such a step being implemented. implemented when the comfort level associated with a road type class determined during the determination phase is lower than a predefined adapted comfort level threshold and/or when the determined road type class is included in a subset predefined class identified as requiring the execution of an adjustment step.

The method can, in addition, include a step of updating a map fitted to the vehicle so as to provide information on a specific road type class.

According to an example of execution of the method according to the invention, the classification step can be implemented only when the estimated probability rate is greater than a predetermined threshold of 60%.

According to an example of execution of the method, the estimation step can be executed by a network capable of processing time series of data and of being trained during a learning phase, in particular by a convolutional neural network.

The data acquisition step can be implemented by:
- detection and/or estimation of at least one characteristic relating to a portion of the route traveled located upstream of the vehicle, in particular a height of a road profile; and or
- measurement by at least one sensor of at least one characteristic relating to the vehicle in a driving situation on said road traveled, in particular suspension travel, wheel speed, longitudinal speed of the vehicle and/or acceleration of the vehicle.

Also, the determination phase may comprise, subsequent to the estimation step, a step of validating the estimated probability rate, the validation step being carried out on the basis of measurements, by at least one sensor, of at least one sensor. least one characteristic relating to the vehicle in a driving situation on said road traveled, in particular suspension travel, wheel speed, longitudinal speed of the vehicle and/or acceleration of the vehicle.

The present invention also relates to a system for detecting a type of road traveled for a vehicle in a driving situation. The system comprises hardware and/or software elements implementing the method described above, the hardware elements comprising at least one data processing unit, a module for detecting at least one characteristic relating to a route traveled and a module for control of at least one additional system fitted to the vehicle, in particular a variable suspension system. In particular, the detection system may be an intelligent detection system.

The invention further relates to a motor vehicle comprising at least one ancillary system capable of affecting a level of comfort felt by a user, such as a variable suspension system, and a detection system according to the invention , capable of controlling said ancillary system.

The invention can further extend to a learning method for a detection method according to the invention or for a detection system according to the invention, comprising at least one supervised learning phase, specific to the vehicle equipped, with the identification of at least one predefined road type class and the association of at least one level of comfort felt by a user with each road type class based on at least one piece of equipment specific to the vehicle capable of influence the comfort felt by a user during the driving phase, the learning phase being implemented via a computer, a human-machine interface and/or a memory element.

The invention further extends to a computer program product comprising program code instructions recorded on a computer-readable medium for implementing the steps of the method according to the invention when said program operates on a computer. Alternatively, the invention can extend to a computer program product downloadable from a communications network and/or recorded on a data medium readable by a computer.

Also, the invention relates to a data recording medium, readable by a computer, on which is recorded a computer program comprising program code instructions for implementing the detection method according to the invention or a medium computer-readable recording comprising instructions which, when executed by a computer, cause it to implement said detection method.

The invention finally relates to a signal from a data carrier, carrying the computer program product according to the invention.

Other details, characteristics and advantages will emerge more clearly on reading the detailed description given below, for information only and not limitation, in relation to the different examples of embodiment illustrated in the following figures:

There is a schematic representation of one embodiment of a detection system according to the invention.

There is a general flowchart of an example of execution of a method for detecting a type of road.

There is a schematic representation of an example of execution of a data acquisition step.

There is a flowchart detailing a second example of execution of the method according to the invention.

detailed description

There schematically illustrates an exemplary embodiment of a detection system 1, in particular intelligent detection, of a type of road traveled by a vehicle 10. Such a system is intended for a vehicle, in particular a motor vehicle. The vehicle 10 can be, for example, a private vehicle, a utility vehicle, a truck or a bus. In particular, the vehicle considered may be a connected and/or autonomous vehicle. Figures 2 to 4 illustrate examples and variants of execution of a detection method, in particular intelligent detection, of a type of road traveled by such a vehicle 10. The detection method can also be considered as being a method of operation of the detection system 1 according to the invention or as a method of operating the automobile vehicle 10 equipped with such a detection system 1.

Here we mean “road traveled” a road or a section of a road, a lane or a roadway at which the vehicle 10 is traveling in real time.

The vehicle 10 comprises the detection system 1, which comprises hardware and/or software elements capable of implementing, or designed to implement, the detection method according to the invention, as further explained below.

Said hardware elements comprise at least one data processing unit 2, a detection module 3 of at least one characteristic relating to a route traveled and a control module 4 of at least one annex system 5 equipping the vehicle 10, such as such as a variable suspension system 51, a driving assistance system 52 or a longitudinal speed management system 53.

The processing unit 2 comprises at least one computer 21 comprising hardware and software resources, more precisely at least one processor, or microprocessor, cooperating with memory elements 6 of the detection system 1 and/or of the vehicle 10. This computer 21 is capable of executing instructions for implementing a computer program. In particular, the processing unit 2 can provide at least one other function within the vehicle 10. For example, the computer 21 can be the on-board computer of the vehicle 10, or, in general, any computer 21 embarked in the vehicle 10 permanently or temporarily.

The detection module 3 can in particular be configured in order to detect and/or estimate at least one characteristic relating to a portion of the route traveled located upstream of the vehicle 10, such as a height of a road profile. Here we mean “road profile height” a height of a point on the road relative to the vehicle. In particular, the at least one characteristic considered can be detected and/or estimated on the basis of images of said portion of road.

For example, the detection module 3 may comprise at least one optical detection means 31, such as a camera, a scanning laser or a Lidar, capable of providing signals representative of an image of the route traveled which can then be transmitted to the processing unit 2 in order to process and/or archive them. Alternatively or in combination, the detection means 31 can be phonic, for example by ultrasound. Such detection means 31 can, preferably, be equipped at the front of the vehicle 10 or on a roof of the vehicle 10 in order to acquire data relating to a portion of the road placed upstream of the vehicle 10.

Alternatively or additionally, the detection module 3 may comprise at least one sensor 32 equipped on the vehicle 10 and capable of measuring at least one characteristic specific to the running of the vehicle 10, for example relating to the vehicle or to an assembly formed by the vehicle and the route traveled. Such a characteristic may, in a non-limiting manner, relate to suspension travel, wheel rotation speed and/or longitudinal speed of the vehicle.

The control module 4 is configured to control at least one annex system 5 fitted to the vehicle 10. Such an annex system 5 can be a variable suspension system 51, a driving assistance system 52, a speed management system longitudinal and/or a mapping system, such as a locating means 7 fitted to the vehicle 10. The control module 4 is in particular capable of carrying out the switching on, switching off, modification and/or adjustment of all or part of the parameters defined by the at least one annex 5 system considered. Advantageously, the control module 4 can be configured so as to control a plurality of ancillary systems fitted to the vehicle 10.

Optionally, the detection system according to the invention may also include all or part of the following options:
- a location means 7;
- a communication module 8.

The location means 7 can in particular allow the location of the vehicle 10. It integrates, for example, a system for approximate location of the vehicle 10 and/or high definition mapping of the road infrastructure. In particular, the approximate location of the vehicle 10 can be provided by a GPS type system, from the English acronym “Global Positioning system” allowing the extraction, from a mapping database, of information concerning the road infrastructure around the approximate position of the vehicle 10. Alternatively or in addition, the location means can be a location system on board the vehicle 10.

The communication module 8 notably allows the vehicle 10 to transmit and/or receive data from a database, a remote server, one or more connected devices such as a mobile telephone, or a vehicle itself equipped with a communication module 8 and a system according to the invention. Such communication is established via a low frequency or high frequency wireless link. It may, for example, be a wireless connection based on “cellular” or “Wifi” or “Bluetooth” technologies. As further explained below, the communication module 8 can be implemented in order to extract data relating to the external environment, in particular weather or road traffic conditions, or in order to transmit information relating to the road traveled in order to update mapping data.

The detection system 1 may, in addition, comprise a memory element 6 or a storage space constituting a recording medium readable by a computer or by the calculator 21 of the processing unit 2 comprising instructions which, when they are executed by the computer or the calculator 21, leading it to implement the method as described below.

One mode of execution of the detection method 100 of a type of road traveled by a vehicle 10 in a driving situation is described below with reference to Figures 2 to 4. Generally, the detection method 100 according to invention comprises a determination phase P1 of a type of road traveled by the vehicle and, prior to the execution of the determination phase P1, at least one supervised learning phase P2, specific to the vehicle 10 equipped, associating in particular a level of comfort felt by a user at different road type classes, especially bad roads. Such a learning phase P2 can also be defined as relating to a training method for a detection method and/or a detection system according to the invention.

Throughout the description below, a type of road, or a class of road type, may, in a non-limiting manner, relate to:
- one or more types of bad roads likely to cause pumping movements of the body when the vehicle 10 moves on a road whose spectral amplitude density of vertical stresses is maximum at low frequencies;
- one or more types of bad roads likely to generate wheel jolts or bounces when the vehicle 10 moves on a road whose spectral amplitude density of vertical stresses is maximum at high frequencies;
- paved roads or mountain paths;
- roads whose surface is of poor quality, for example due to bad weather, such as precipitation, frost, or due to wear;
- roads including obstacles, such as speed bumps, level crossings, potholes or bars, depending on their impact on comfort, for example depending on their dimensions, likely to cause vibrations or shocks in the vehicle 10, in particular presenting a high frequency of said obstacles, that is to say a degraded road with a short distance between obstacles.

The determination phase P1 comprises at least one acquisition step E1, in real time or for a predetermined duration, of data relating to a route traveled by the vehicle 10. According to an exemplary embodiment not detailed in section , the determination phase P1 may include a plurality of acquisition steps, relating to similar or different types of data. For example, said data may relate to one or more characteristics relating to the route traveled and/or the vehicle. Alternatively, the acquisition step E1 may comprise the acquisition of different distinct types of data. The acquisition step E1 is implemented via the detection module 3 as explained previously.

In order to classify the road traveled in at least one predetermined road type class, such as a bad road type, the acquisition step E1 can be carried out so as to include a detection sub-step and /or estimation E11 of at least one characteristic relating to a portion of the route traveled located upstream of the vehicle 10 and/or a measurement sub-step E12 of at least one characteristic relating to the vehicle 10 in a driving situation on the said route traveled. In other words, the acquisition step E1 can implement the acquisition of data so as to anticipate characteristics relating to a future portion of road included in the route traveled and/or so as to react to a portion of road on which the vehicle 10 moves, that is to say in contact with which is at least one of the wheels of the vehicle 10.

The detected and/or estimated data E11 on the basis of the portion of the route traveled located upstream of the vehicle 10 can, for example, relate to characteristics such as a height of a road profile relative to the vehicle, in particular to a wheel of the vehicle, or any other information from images of the road. For example, such data can be obtained via optical detection means, such as a camera.

Optionally but preferentially, the detected and/or estimated data E11 on the basis of the portion of road located upstream of the vehicle 10 can be associated with a confidence index α representative of a degree of reliability of the detection and/or of the estimate made. In particular, such a confidence index α can be defined on the basis of weather or road traffic conditions detected and/or depending on the type of road surface, likely to affect the detection and/or the estimation. For example, the confidence index α is defined depending on whether the road is paved, covered with snow, frost, leaves or sand. Advantageously, the use of a confidence index makes it possible to discard data defined as unreliable, likely to bias the classification of the road type, for example by discarding data presenting a confidence index lower than a threshold α predefined _min . For example, the confidence index can be defined by the processing unit 2 and/or by the detection module 3.

The measured data E12 can, in a non-limiting manner, be obtained by at least one sensor 32 and relate to characteristics such as the travel of the suspensions or of a wheel, for example of the center of a wheel, a wheel speed , a longitudinal speed of the vehicle 10 and/or an acceleration of the vehicle 10. The at least one sensor 32 can be included in the detection module 3. Alternatively, the sensor 32 can be an on-board sensor 32 specific to one of the additional systems 5 equipping the vehicle 10 and configured to transmit data to said detection module 3. Such sensors 32 may include an accelerometer, a deflection sensor, a deformation sensor associated with one of the wheels and/or a deflection sensor. pressure.

Preferably, the method according to the invention is at least carried out in anticipation, by estimation and/or detection E11 of data relating to the portion of road located upstream of the vehicle 10. As further explained below, with reference to there , the implementation of a measurement sub-step, in addition to a detection and/or estimation sub-step E11, can make it possible to improve the detection and classification of the route traveled by means of a step of confirmation.

Independently of the type of acquisition carried out, that is to say in anticipation or in reaction, the detection and/or estimation E11 of data on the one hand and the measurement of data E12 on the other hand are each carried out in at least one point on the route. Preferably, the data acquisition is carried out in at least two points, for example a point arranged on two lines representative of a virtual trajectory of the front wheels of the vehicle.

There illustrates a particular example of implementation of the acquisition step E1 in which the acquisition of data is carried out in anticipation via an optical detection means so as to allow an estimation of the heights of the profile of the portion of road located upstream of vehicle 10 by image processing. In the example illustrated, in a non-limiting manner, the data is collected at a plurality of points arranged in front of each of the front wheels of the vehicle 10. In this case, the data is collected, for each front wheel, at the level ten points. Preferably, for each front wheel, said points have a spacing at a fixed frequency so that data acquisition is not negatively affected by the longitudinal speed of the vehicle 10. The data relating to each of the twenty points of measurement, ten for each front wheel, are transmitted to the processing unit 2.

For each of the aforementioned points, the estimated value of the height of the profile of the road portion is associated with a confidence index α _x of said estimate. As explained above, such an index can be defined by the detection module 3 and/or the processing unit 2. This results in a set of data comprising forty incoming signals which represent the heights and the confidence indices α _x of the different points of the portion of the road located upstream of the two front wheels of vehicle 10.

Such a principle applies mutatis mutandis to the "reaction" operation explained above, according to which the data are directly obtained by measurements E12 representative of a set formed by the road traveled and the vehicle 10, with the difference that no index of trust is then included.

The processing unit 2 then executes an estimation step E2 of at least one probability rate of belonging of the road traveled to at least one predefined road type class, in particular bad road, based on the acquired data . As explained previously, when a confidence index is associated with the acquired data, such a probability rate is also defined on their basis.

The estimation of membership rates can be carried out by any type of network capable of carrying out a classification task from time series of data previously including a supervised P2 learning phase. The network considered receives as input the data acquired during the acquisition phase, and provides as output estimates of the rate of membership of the route traveled to the different classes of predefined road types corresponding to a probability rate that the route traveled is similar, or of the same type, as a road type class considered. In other words, such a type of network is previously trained during a supervised P2 learning phase, as further detailed below, in order to perform the identification of different classes of road types on the basis of different types of data and/or data ranges, then can subsequently be executed by the processing unit 2 during the implementation of the method in order to classify the road traveled in at least one of the classes identified during the learning phase.

According to an exemplary embodiment, the probability rates of belonging to different classes of road types can be calculated by a neural network, in particular by a convolutional neural network, or CNN. This type of models can be used to perform supervised classification of time series data. Such a principle is advantageously simple to implement and allows invariance of the processing by temporal translation. According to another non-limiting embodiment, the probability rates of belonging to different classes of road types can be calculated by a recurrent neural network, or RNN.

On the basis of the membership probability rates obtained, the processing unit 2 implements a step of classification E3 of the road traveled in a road type class and extraction of at least one level of felt comfort by a user previously associated, for example during at least one supervised learning phase P2 as explained below, with each class of road type depending on at least one piece of equipment specific to the vehicle 10 capable of influencing the comfort felt by a user while driving. We mean, according to non-limiting examples, “equipment” equipment likely to affect the level of comfort felt by a user such as a suspension system or tires.

For example, the route traveled is classified into a predefined road type class based on the highest estimated probability rate obtained. In particular, the classification step E3 can be implemented only when the estimated probability rate is greater than a predetermined threshold of 50%, or even 60% or even 75% in order to ensure the accuracy of the classification. . The processing unit 2 then extracts at least one level of comfort felt by a user associated with the road type class detected, that is to say previously defined, by calculation implemented by the processing unit 2 or by manual intervention of a user or technician, and recorded on the memory elements 6 of the detection system 1 and/or of the vehicle 10.

According to non-limiting example of execution, implemented for a vehicle 10 in a driving situation as illustrated in , at each calculation step the processing unit 2 forms, as input to the network, in particular the neural network, a sliding window of data. In this case, such a window includes the last forty values relating to the heights of road profiles and confidence indices. The neural network performs inference to obtain a probability distribution over different classes of roads, including bad roads, and then the type of road is determined by selecting the class with the highest probability rate.

The network used to extract features , in other words characteristics, relating to road type classes on the basis of the acquired raw data time series and define probability rates can be defined, in a non-limiting manner, by a basic block comprising :
- a 1D Convolution layer, or filter; Then
- a ReLU layer, for Rectified Linear Units, corresponding to a correction layer including the ReLU activation function; Then
- a MaxPooling layer, corresponding to a pooling layer; Then
- a batch normalization layer, or “Batch Normalization”.

Such a block can be repeated several times, for example twice, in order to obtain a succession of three blocks. According to a non-limiting exemplary embodiment, these three basic blocks can be assembled with different sizes of filters and used as input for a fully connected layer so as to give a distribution of probabilities of belonging to a road type class as output. For example, the sizes of filters used in the convolution layers can be defined as follows during the successive execution of the three basic blocks: 128, 256, 128. Similarly, the sizes of the kernels, or “ kernel ” in English, convolutions can be the following: 5, 3, 3. For the execution of the convolutions of each of the blocks, the padding applied to the input allows the output to keep the same temporal dimension and the step used is preferably 1 .

Following the implementation of the basic blocks, a softmax layer can be implemented. It provides a probability rate of membership of the route traveled for each of the possible classes, the sum of these probability rates amounting to 1. Conventionally, the softmax function takes as input a vector of dimension K: z= (z ₁ …z _K ) and is defined according to:

In the present exemplary embodiment, the network, in particular the convolutional neural network, receives as input a matrix of size 40x40 containing a time window containing 40 time steps of the 40 data signals described previously, relating to the profile height of the portion of road located upstream of vehicle 10 and with confidence indices. The network provides, as output, a distribution of the probability rates of belonging of the road to each of the classes of possible road types predefined during the learning phase P2. The classification step E3 as explained above is then executed, for example so that the final classification of the route traveled corresponds to the class for which the probability rate obtained at the output of the network is maximum.

The training of the network, for example of the convolutional neural network, is carried out, as explained previously, during at least one supervised learning phase P2, at least implemented prior to the determination phase P1 such that previously exposed. By “supervised” we mean that such a P2 learning phase requires, at least partially, human intervention.

The learning phase P2 aims on the one hand to train the network, in other words the model, so that it is capable of allowing the identification of a predefined road type class corresponding to a road traveled when the phase of determination P1 is executed, and, on the other hand, to train the model so that it is able to associate with the different classes of road types a level of comfort felt by the user.

The learning phase P2 can be implemented via a computer 21, a human-machine interface and/or a memory element. Once trained, the model can easily be implemented in a computer 21 of the vehicle 10, for example in the processing unit 2, during the implementation of the determination phase P1.

Training the network with a view to identifying a type of road includes a so-called “inference” phase, carried out according to a principle similar to what was explained with reference to the determination phase P1, from data for which the road type class is known beforehand, for example by manual, i.e. supervised, labeling of a learning database, so the preceding description applies mutatis mutandis . To constitute the learning base, data recordings relating to different types of roads are made, on the basis of one or more types of data relating to characteristics specific to the road traveled and/or to the vehicle, such as previously explained with reference to acquisition step E1.

These records, for which the desired result of the processing is not the model, are submitted to the model in order to train it to identify, that is to say to execute the estimation and analysis step. classification as previously explained, of a type of road on the basis of detected and/or estimated data, that is to say in anticipation, and/or on the basis of measured data, that is to say in reaction. The different classes of road types are thus defined as a function of at least one type of data, in particular as a function of ranges of values defined for at least one type of data considered, for example the height of the road profile. Then, depending on the class or the distribution of the probability rates of belonging of the road to each of the predefined classes of possible road types determined by the model, an estimate of the error made by the model is calculated.

In the case of networks, in particular convolutional neural networks, a loss or cost function, for example a cross-entropy function, can be used. Optionally but preferentially, a weighting can be applied to such a loss function in order to prevent overlearning, also called overfitting in English, or in order to allow artificially accelerated learning if the recordings of the different types of roads are not equally distributed in terms of number of examples per class.

Training the network also aims to identify and/or associate at least one estimated comfort level, representative of the user's comfort and the feeling of the road traveled, to the different classes of road types. Such training can be carried out on the basis of at least one criterion, which can be objective or subjective. Advantageously, training can be carried out on the basis of a plurality of objective and/or subjective criteria. The different classes of road types can be associated with at least one criterion, in particular with at least one range of values defined for the at least one criterion considered.

For example, the term “objective criterion” means criteria defined on the basis of values measured, estimated and/or calculated by detection means and/or sensors 32 on board the vehicle 10. For example, an objective criterion can be relative to transmissibility, that is to say the ratio between the displacement of the suspended mass in relation to the stress on the road which makes it possible to evaluate the vibration isolation characteristic of a given suspension. A criterion can also be defined based on the clearance of the wheels relative to the body when passing disturbances or obstacles, and in particular the reaching of the compression or relaxation stops which strongly penalize comfort, or even a criterion on the value effective, or RMS, of the vertical acceleration at the car's center of gravity.

When the training is based on objective criteria, it can advantageously be executed automatically and is defined specifically according to the equipment specific to the vehicle 10, or to a vehicle model 10, so that the evaluation of the associated felt comfort to the type of road identified is defined in an appropriate manner relative to the vehicle 10 equipped with the detection system and its equipment, such as its suspension system, and not in a generic manner.

Conversely, by “subjective criterion” we mean that learning is carried out in a supervised manner by a user or by an expert who can, for example, himself associate a predefined level of comfort with each of the different classes of types. of roads depending on the vehicle's equipment, such as the suspension system. The user or expert is then able to subjectively classify the type of road encountered based on their feelings about the response of the vehicle and the resulting comfort for passengers passing on this road.

Similar to what was explained above, during training relating to comfort, a loss function and, optionally, a weighting, can be implemented.

The learning phase P2 can thus advantageously be adapted according to the need and the different types of roads can thus be classified by the user or the expert according to what is desirable to detect, i.e. say according to the equipment of the vehicle 10 and/or according to the ancillary systems 5 capable of being controlled, in particular the suspension or shock absorber system, and according to a desired comfort level defined by the user or the expert.

During this P2 learning phase, the quality and consistency of training is essential to ensure precise and efficient identification of the different types. In this application, the method according to the invention advantageously makes it possible to obtain an overall precision of around 98% with a weighted F1 score of 97%.

In the case explained above, at least one learning phase P2 is implemented prior to the execution of the determination phase P1. A plurality of detection phases can be executed following such a learning phase P2, as illustrated by the dotted box P2’. For example, the P2 learning phase can be carried out before putting the vehicle into circulation or during maintenance or an overhaul of the vehicle.

Optionally, the method according to the invention may include an additional learning phase P2', implemented subsequently to the execution of the determination phase P1 of a road type. Such an additional learning phase P2 is executed similarly to what was explained previously and can be, for example, implemented by a user in order to refine or adapt the training of the identification and/or the level of comfort associated with different types of road to one's preferences or needs.

Thus, as illustrated in , in driving conditions, once the road type class specific to the road traveled has been identified, that is to say at the end of the classification step E3 of the determination phase P1, the unit of processing can extract, for example from a memory element, a level of comfort associated with the identified road type class which is then representative of a level of comfort felt, in real time, by the user while the vehicle evolves on the route traveled. The detection system 1 can then initiate, via the control module 4, an adjustment step E4 of at least one additional system 5 fitted to the vehicle 10 so as to correct the level of comfort felt by the user .

The adjustment step E4 can be implemented when the comfort level associated with the identified road type is lower than a predefined adapted comfort level threshold, for example when the determined road type is a bad road associated with a level of comfort deemed unsuitable during the learning phase. Such a threshold can be defined beforehand by an expert and/or adjusted by a user. Alternatively, the adjustment step can be implemented when the road type class identified during the determination phase P1 is included in a subset of classes previously identified, for example during the learning phase, as requiring a correction of felt comfort. For example, such a type of road may be a “bad road”.

By “correcting” we mean here that, when the estimated level of felt comfort is lower than the prerequisite threshold, it is necessary to adapt at least one annex 5 system in order to increase the comfort of a user. Such a principle can be achieved by adapting the suspension system so as to make them more flexible or rigid, by adapting the longitudinal speed of the vehicle, for example by reducing it.

For example, in the context of controlling a variable suspension or shock absorber system, the damping force can be modified, in particular increased, in various proportions when an obstacle, a speed bump or a bad road low frequency is detected, said proportions being adapted to the type of obstacle or road detected. Conversely, the damping force can be reduced when passing a high frequency bad road or a small percussion.

As explained previously, the annex system 5 in question can be a variable suspension system 51, a longitudinal speed control system and/or a driving assistance system 52. Note that the driving assistance functions driving may include collision warnings, driver information or alerts, emergency braking or even emergency steering. The correction carried out may then affect one or more of said systems.

As indicated above, advantageously, the classes of road types, in particular bad roads, can be adapted according to the desired use of the vehicle and/or the detection system, that is to say for example according to the annex system 5 referred to and/or according to an operating mode of the vehicle 10 among conventionally available modes such as a “sporty” or “comfort” driving mode for example. The method according to the invention thus allows, at the time of detection of the road type of the route traveled, the triggering of a control strategy of ancillary systems, such as the variable suspension system 51, specifically adapted to the type of road. determined during the determination phase P1 and the vehicle 10 equipped to ensure optimal comfort for a user.

Additionally or alternatively, once the road type class specific to the road traveled has been identified, the detection system 1 can initiate a step E5 of updating a map fitted to the vehicle 10 so as to provide information on a road type , in particular bad road, determined during the determination phase P1. In this sense, during the determination phase P1, for example during the acquisition step E1 and/or during the classification step E3, the acquired data can be associated with GPS data. Optionally, this data can also be associated with a timestamp, in particular in order to ensure suitable mapping when the determined road type has a temporary component, for example due to temporary obstacles or the presence of bad weather such as rain, snow or ice likely to affect the road class.

The road type class determined for the route traveled, as well as GPS data, and possibly timestamp data, can then be transmitted to other vehicles, to a database and/or to remote servers via of the communication module 8 in order to make this data accessible to other users, in particular other users using a similar or identical model of vehicle 10. It is, for example, then possible to associate in real time a GPS position and possibly a timestamp corresponding to an entry point and an exit point on the type of road detected. These data can, optionally, also be associated with a driving direction of the vehicle 10 on the route traveled and/or with the longitudinal speed of the vehicle 10.

Once this information has been transmitted to remote databases or servers, it may be subject to additional processing allowing in particular to filter, group and/or validate this information before making it available to other users.

There illustrates a second mode of execution of the method according to the invention, in which the determination phase P1 comprises, subsequent to the estimation step E2, a validation step E6 of the estimated probability rate carried out on the basis of measurements, by at least one sensor 32, of at least one characteristic relating to the vehicle 10 in a driving situation on the route traveled. In such an execution mode, the estimation step E2 as explained previously can be implemented in anticipation, on the basis of estimated and/or detected data relating to a portion of the route traveled located upstream of the vehicle 10, or in reaction, on the basis of measured data.

The validation step E6 can be executed simultaneously and/or subsequently with the estimation step E2. The validation step E6 is executed in a manner similar to what was explained above relating to the processing of data measured during the estimation step E2.

The classification step E3 of the road traveled in one of the predefined road type classes can then be executed conditionally only when the results of the estimation step E2 and the validation step E6 make it possible to identify an identical type of road. In other words, the classification step E3 is only implemented when the probability rates of belonging to a road type class obtained during the estimation step E2 and the validation step E6 are such that they make it possible to identify the same class of road type.

Alternatively, the validation step E6 can be executed in parallel with the estimation step E2 and classification E3, the validation step E6 reproducing said estimation steps E2 and classification E3 on the basis of data measured in view of validation. The method according to the invention then comprises two distinct classification stages, a primary one E3, as described with reference to the , and the other included in validation step E6. A classification of the route traveled is then recorded, for example on a memory element 6, and used for the adjustment E4 of an additional system only when the two classification steps result in the identification of the same class type of road. Such a principle advantageously makes it possible to further optimize the precision of the method according to the invention.

The present invention thus proposes a method for detecting a type of road traveled by a vehicle coupling a phase of determining a type of road and a phase of learning the identification of a class of road type integrating a evaluation of the road comfort felt by a user for a vehicle of given model and equipment. The method according to the invention can advantageously be transposed to any type of vehicle at lower cost, the execution of the learning phase making it possible to obtain a detection method adapted to the characteristics of the vehicle, in particular its model and/or its equipment. likely to impact driving comfort. The invention thus makes it possible to obtain a performance in detecting road types, in particular bad roads, refined and adapted to the equipped vehicle, while ensuring the transportability of the invention to a wide range of vehicles. The method according to the invention then allows more appropriate control of the ancillary systems linked to the comfort of the user equipping the vehicle.

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

Claims (11)

Method for detecting (100) a type of road traveled by a vehicle (10), in particular a vehicle (10) equipped with a variable suspension system (51), the method being characterized in that it comprises a phase for determining (P1) a type of route traveled comprising:
- at least one acquisition step (E1), in real time or for a predetermined duration, of data relating to the vehicle and/or a route traveled by the vehicle (10) implemented via a module detection (3) comprising at least one sensor (32) and/or at least one detection means (31), such as a camera;
- a step of estimating (E2) a probability rate of belonging of the road traveled to at least one road type class, in particular bad roads, predefined according to said acquired data;
- a step of classification (E3) of the road traveled in one of the road type classes according to the estimated membership probability rates and extraction of at least one level of comfort felt by a previously associated user to each class of road type according to at least one piece of equipment specific to the vehicle (10) capable of influencing the comfort felt by a user during the driving phase. Detection method (100) according to the preceding claim, further comprising, prior to the execution of the determination phase (P1), at least one supervised learning phase (P2), specific to the equipped vehicle (10), the identification of at least one predefined road type class and the association of at least one level of comfort felt by a user with each road type class as a function of at least one piece of equipment specific to the vehicle (10) capable of influencing the comfort felt by a user during the driving phase, the learning phase (P2) being implemented via a computer (21), a human-machine interface and/or or a memory item. Detection method (100) according to one of the preceding claims, further comprising a step of adjustment (E4) of at least one additional system (5) fitted to the vehicle (10), in particular a variable suspension system (51), so as to correct a level of comfort felt by the user, such a step being implemented when the level of comfort associated with a road type class determined during the determination phase (P1) is lower at a predefined adapted comfort level threshold and/or when the determined road type class is included in a subset of predefined classes identified as requiring the execution of an adjustment step (E4). Detection method (100) according to one of the preceding claims, further comprising a step of updating (E5) a map fitted to the vehicle (10) so as to provide information on a determined road type class. Detection method (100) according to one of the preceding claims, in which the classification step (E3) is implemented only when the estimated probability rate is greater than a predetermined threshold of 60%. Detection method (100) according to one of the preceding claims, in which the estimation step (E2) is executed by a network capable of processing time series of data and of being trained during a learning phase (P2), in particular by a convolutional neural network. Detection method (100) according to one of the preceding claims, in which the data acquisition step (E1) is implemented by:
- detection and/or estimation of at least one characteristic relating to a portion of the route traveled located upstream of the vehicle (10), in particular a height of a road profile; and or
- measurement by at least one sensor (32) of at least one characteristic relating to the vehicle (10) in a driving situation on said road traveled, in particular a travel of the suspensions, a wheel speed, a longitudinal speed of the vehicle (10) and/or acceleration of the vehicle (10). Detection method (100) according to one of the preceding claims, in which the determination phase (P1) comprises, subsequent to the estimation step (E2), a validation step (E6) of the estimated probability rate, the validation step (E6) being carried out on the basis of measurements, by at least one sensor (32), of at least one characteristic relating to the vehicle (10) in a driving situation on said route traveled, in particular a clearance of the suspensions, a wheel speed, a longitudinal speed of the vehicle (10) and/or an acceleration of the vehicle (10). System for detecting (1) a type of road traveled for a vehicle (10) in a driving situation, the system comprising hardware and/or software elements implementing the method according to one of claims 1 to 8, the hardware elements comprising at least one data processing unit (2), a detection module (3) of at least one characteristic relating to a route traveled and a control module (4) of at least one annex system (5 ) equipping the vehicle (10), in particular a variable suspension system (51). Motor vehicle (10) comprising at least one additional system (5) capable of affecting a level of comfort felt by a user, such as a variable suspension system (51), and a detection system (1) according to the claim 9, capable of controlling said annex system (5). A computer program product comprising program code instructions recorded on a computer readable medium for carrying out the steps of the method according to any of claims 1 to 8 when said program is running on a computer.