Method and device for determining a comfortable speed for a vehicle traveling on a roadway using deep neural networks

Abstract

The present invention relates to a method and device for determining the comfort speed of a vehicle traveling on a roadway, the method being implemented by at least one processor and comprising the following steps: - obtaining (41) a first data point representing traffic conditions on the roadway from an image of the roadway; - obtaining (42) a second data point representing a driver's pose from an image of the driver; - obtaining (43) a third data point representing a current operating state of the vehicle from at least one on-board sensor of the vehicle; - obtaining (44) the vehicle's comfort speed as output from a trained model for determining vehicle comfort speed when the first, second, and third data points are presented as input to the trained model for determining vehicle comfort speed. Figure 4 (for the abstract)

Description

Title of the invention: Method and device for determining a comfortable speed of a vehicle traveling on a roadway using deep neural networks Technical field

[0001] The present invention relates to determining the speed of a vehicle. In particular, the present invention relates to a method and a device for determining a comfort speed of a vehicle traveling on a roadway, in particular an autonomous vehicle. Technological background

[0002] Contemporary vehicles incorporate an increasing number of on-board systems that contribute to vehicle control by assisting the driver and / or ensuring the comfort of the driver and passengers of the vehicle. Some of these systems are known as AD AS systems (from the English “Advanced Driver-Assistance System” or in French “Advanced Driving Assistance System”).

[0003] Among the most well-known AD AS systems, some influence the speed of the vehicle traveling on a roadway. This is the case, for example, we can cite the ISA (Intelligent Speed ​​Adaptation) systems, ACC (Adaptive Cruise Control), PCC (Predictive Cruise Control), ISA (Intelligent Speed ​​Adaptation).

[0004] Some of these AD AS systems need a prediction of a vehicle speed or a set speed to influence the actual speed of the vehicle. These AD AS systems therefore need to know a comfort speed of the vehicle which can be determined so that the driver and passengers are in optimal comfort.

[0005] These AD AS systems which determine a vehicle speed can help the driver to make a driving decision.

[0006] Some of these AD AS systems determine a vehicle speed based on data relating to the operation of the vehicle. Others determine a vehicle speed based on the terrain on which the vehicle is traveling. Others determine a vehicle speed based on data collected on the driver of the vehicle and his reactions to events while driving the vehicle.

[0007] Existing systems therefore determine a vehicle speed based on either data relating to the roadway on which the vehicle is traveling, either data relating to the vehicle or data relating to the driver's behavior.

[0008] The performance of these existing systems is acceptable but needs to be improved to optimize the comfort of the driver and passengers of the vehicle.

[0009] Summary of the present invention

[0010] An object of the present invention is to solve at least one of the problems of the technological background described above.

[0011] Another object of the present invention is to determine a comfort speed of a vehicle which ensures comfort for the driver and passengers of the vehicle.

[0012] Another object of the present invention is to improve AD AS systems that use vehicle speed predictions or set speeds.

[0013] Another object of the present invention is to improve road safety.

[0014] According to a first aspect, the present invention relates to a method for determining a comfort speed of a vehicle traveling on a roadway, the method being implemented by at least one processor and comprising the following steps: - obtaining initial data representative of traffic conditions on the roadway from an image of the roadway; - obtaining a second data representative of a pose of a driver of the vehicle from an image of the driver; - obtaining a third data item representative of a current operating state of the vehicle from at least one on-board sensor of the vehicle; - obtaining the vehicle comfort speed as output from a trained vehicle comfort speed determination model when the first, second and third data are presented as input to the trained vehicle comfort speed determination model.

[0015] The method determines a comfort speed of a vehicle based on a trained model for determining the comfort speed of the vehicle during a learning phase when data representative of traffic conditions on the roadway on which the vehicle is traveling, data representative of a pose of a driver of the vehicle and data representative of a current operating state of the vehicle are presented as input to said model.

[0016] Taking all of these data into account for determining a comfort speed of the vehicle ensures that the comfort speed is optimal whatever the traffic conditions on the roadway and in particular whatever the state of the roadway, whatever the position of the driver and whatever the operating state of the vehicle.

[0017] The vehicle comfort speed determination model provides a more robust driving experience because the method provides a lower comfort speed when the The road surface is only damaged when the road surface is smooth and in good condition.

[0018] The method makes it possible to personalize the comfort speed according to the driver's behavior (pose) which may have different poses depending on whether his driving is sporty or not.

[0019] The method is also advantageous because the determined comfort speed depends on the vehicle's capabilities. It therefore helps prevent undue vehicle breakdowns and accidents.

[0020] The method is advantageous because it is modular in the sense that the first, second or third data can be used by other on-board systems of the vehicle. For example, the second data can be used by a system for monitoring the attention of a vehicle driver, also known as a driver attention monitor or driver fatigue warning system, abbreviated to DMS (from the English "Driver Monitoring System"). The first data can also be used for a lane detection system or an on-board object detection system of the vehicle.

[0021] The method is flexible because obtaining the first, second and third data is not limited to particular methods. Thus, methods adapted to specific needs can be used to obtain the first, second and third data.

[0022] The method is advantageous because it allows the parallelization of the steps of obtaining the first, second and third data because these data are obtained from independent means. This makes it possible to limit the latency time to obtain a comfortable speed from these first, second and third data obtained.

[0023] According to a variant, the first data is obtained at the output of a trained model for predicting traffic conditions on the roadway when the image of the roadway is presented as the input of the trained model for predicting traffic conditions on the roadway.

[0024] According to a variant, the road traffic condition prediction model is implemented by a deep convolutional neural network.

[0025] According to a variant, the first data is representative of a segmentation mask of the image of the roadway, of a mapping of irregularities of the roadway, of a class identifier of type of the roadway and of a class identifier of meteorological conditions in an environment of the vehicle.

[0026] According to a variant, the second data is obtained at the output of a trained model for estimating the two-dimensional pose of a vehicle driver supplied as input by the image of the driver.

[0027] According to a variant, the two-dimensional driver pose estimation model vehicle is implemented by a neural network.

[0028] According to one variant, the vehicle comfort speed determination model is implemented by a deep convolutional neural network.

[0029] According to a second aspect, the present invention relates to a device for determining a comfort speed of a vehicle traveling on a roadway, the device comprising a memory associated with a processor configured for implementing the steps of the method according to the first aspect of the present invention.

[0030] According to a third aspect, the present invention relates to a vehicle, for example of the automobile type, comprising a device as described above according to the second aspect of the present invention.

[0031] According to a fourth aspect, the present invention relates to a computer program which comprises instructions adapted for executing the steps of the method according to the first aspect of the present invention, in particular when the computer program is executed by at least one processor.

[0032] Such a computer program may use any programming language, and be in the form of source code, object code, or intermediate code between source code and object code, such as in a partially compiled form, or in any other desirable form.

[0033] According to a fifth aspect, the present invention relates to a computer-readable recording medium on which is recorded a computer program comprising instructions for carrying out the steps of the method according to the first aspect of the present invention.

[0034] On the one hand, the recording medium may be any entity or device capable of storing the program. For example, the medium may comprise a storage means, such as a ROM memory, a CD-ROM or a microelectronic circuit type ROM memory, or a magnetic recording means or a hard disk.

[0035] Furthermore, this recording medium may also be a transmissible medium such as an electrical or optical signal, such a signal being able to be conveyed via an electrical or optical cable, by conventional or hertzian radio or by self-directed laser beam or by other means. The computer program according to the present invention may in particular be downloaded from an Internet-type network.

[0036] Alternatively, the recording medium may be an integrated circuit in which the computer program is incorporated, the integrated circuit being adapted to perform or to be used in performing the method in question. Brief description of the figures

[0037] Other features and advantages of the present invention will become apparent from the following: Description of the particular and non-limiting examples of embodiment of the present invention below, with reference to the appended figures 1 to 4, in which:

[0038] [Fig.l] schematically illustrates a vehicle 10, according to a particular and non-limiting exemplary embodiment of the present invention.

[0039] [Fig.2] schematically illustrates a process for determining the comfort speed of a vehicle traveling on a roadway according to a particular and non-limiting exemplary embodiment of the present invention;

[0040] [Fig.3] schematically illustrates a device configured to determine a comfort speed of a vehicle traveling on a roadway, according to a particular and non-limiting exemplary embodiment of the present invention;

[0041] [Fig.4] illustrates a flowchart of the different steps of a method for determining the comfort speed of a vehicle traveling on a roadway, according to a particular and non-limiting exemplary embodiment of the present invention.

[0042] Description of the examples of embodiment

[0043] A method and a device for determining the comfort speed of a vehicle will now be described in the following with joint reference to FIGS. 1 to 4. The same elements are identified with the same reference signs throughout the description which follows.

[0044] [Fig.l] schematically illustrates a vehicle 10, according to a particular and non-limiting exemplary embodiment of the present invention.

[0045] [Fig.l] shows a view of the interior of a vehicle 10, in particular of the front part of the passenger compartment of the vehicle 10 including in particular the steering wheel, the dashboard, the driver's seat (front left seat according to the example of [Fig.l]), the front passenger seat (front right seat according to the example of [Fig.l]) and the windshield of the vehicle 10.

[0046] The vehicle 10 corresponds for example to a vehicle with a thermal engine, with electric motor(s) or even a hybrid vehicle with a thermal engine and one or more electric motors. The vehicle 10 thus corresponds for example to a land vehicle, for example a car, a truck, a bus.

[0047] The vehicle 10 advantageously carries an imaging device which provides a set of images of the driver.

[0048] The imaging device comprises a camera 11 associated with an optical system for acquiring images and processing means such as a computer and storage means such as non-volatile memories of the EEPROM or flash type for storing the acquired images.

[0049] The lens of the camera 11 is directed towards the interior of the vehicle 10 for the acquisition of the set of images of the driver of the vehicle 10 seated on the driver's seat. This set of images is provided to the calculation means of the imaging device.

[0050] The camera 11 is for example arranged on or integrated into the dashboard of the vehicle 10, for example at the steering wheel. The shooting point of the camera 11 is directed towards the driver's seat to produce images of the driver of the vehicle 10.

[0051] The imaging device also comprises a front camera (located at the front of the vehicle 10) which provides images of a roadway on which the vehicle 10 is traveling.

[0052] The camera 11 and the front camera correspond, for example, to a monocular optical camera with a CCD sensor (Charge-Coupled Device) or with a CMOS sensor (Complementary Metal-Oxide Semiconductor).

[0053] The vehicle 10 also comprises a device 2 of [Fig. 3] which is configured to implement a process for determining a comfort speed of the vehicle 10 traveling on a roadway. The operations forming such a process are described below and illustrated in [Fig. 2].

[0054] Such a process is advantageously implemented by one or more computers of the vehicle 10.

[0055] In a first operation, the device 2 obtains a first data item DI representative of traffic conditions on the roadway on which the vehicle 10 is traveling from an image of the roadway obtained, for example by the front camera of the on-board imaging device of the vehicle.

[0056] According to an exemplary embodiment, the first data DI is obtained at the output of a trained model for predicting traffic conditions on the roadway when the image of the roadway is presented as the input of the trained model for predicting traffic conditions on the roadway.

[0057] According to a variant, the model for predicting traffic conditions on the roadway is implemented by a deep convolutional neural network.

[0058] According to a variant, the first data DI is representative of a segmentation mask of the image of the roadway presented as input to the trained model for predicting traffic conditions on the roadway, of a mapping of irregularities of the roadway on which the vehicle 10 is traveling, of a class identifier of the type of the roadway on which the vehicle 10 is traveling and of a class identifier of weather conditions in an environment of the vehicle.

[0059] The deep convolutional neural network comprises a set of layers of artificial neurons.

[0060] For example, each artificial neuron is a perceptron, that is to say a linear classifier generally comprising several inputs and a single output and characterized by an activation function, weights (or synaptic coefficients) and a bias (or threshold).

[0061] For example, a perceptron with n inputs ( Xj, .... X„ ) and a single output o is defined by n weights ( U]. ..., and a bias (or threshold) 0: 100621 (l, 1 V12' Of(z) = 1=:i 1 1 .0 otherwise.

[0063] The output o then results from the application of the Heaviside function to the post-synaptic potential z given by:

[0064]

[0065] with a nonlinear Heaviside activation function H(x) given for example by : 100661 VX6R. / / (x) = P If X > 1

[0067] The activation functions can be of other types such as ReLU, GeLU, Tanh or even sigmoid functions.

[0068] The internal parameters of the deep convolutional neural network are then these weights and biases for all the perceptrons.

[0069] The present invention is not limited to this definition of perceptron nor to the use of other basic elements forming a layer of the neural network. It is also not limited to the number of perceptrons (or other building blocks) used per layer nor by the number of layers. The internal parameters of the deep convolutional neural network are usually weights and biases regardless of the building blocks of the layers of the deep convolutional neural network.

[0070] The primary function of the deep convolutional neural network is to calculate the post-synaptic potential z which is usually defined by a vector in a feature space (called latent space in English). The vector z is formed of neural activations which we will subsequently call features. The feature vector z is then used to provide a segmentation mask to an input image of a roadway presented as input to the trained deep convolutional neural network and / or a mapping of roadway irregularities and / or to classify a roadway, i.e. to provide a roadway type class identifier and / or a weather condition class identifier. The feature space has a dimension lower than that of the input image space.For this reason, the deep convolutional neural network can be seen as comprising an encoder 12, two decoders 13 and 14 and two classifiers 15 and 16 ([Fig.2]).

[0071] The input image x presented as input to the deep convolutional neural network, i.e. the image of the roadway on which the vehicle 10 is traveling, is then the input image presented as input to the encoder 10. The output of the encoder 10 is then the feature vector z ([Fig.l]) obtained from the input image x by applying operations defined by the perceptrons (or other basic elements) of the different layers of the deep convolutional neural network.

[0072] The decoder 13 provides as output a segmentation mask MS of the input image x when a feature vector z is presented at its input.

[0073] For example, the segmentation mask defines regions in the input image x.

[0074] The decoder 14 provides as output a map of irregularities CI of the roadway on which the vehicle 10 circulates when a feature vector z is presented at its input. The irregularity map associates, for example, a level of road roughness with each region of the input image x defined by the segmentation mask.

[0075] The classifier 15 provides as output an identifier ID1 of the type class of the roadway on which the vehicle 10 is traveling when a vector of characteristics z is presented at its input.

[0076] For example, roadway type classes can define, for example, whether a roadway belongs to a mapped traffic lane (as part of a national road (class 1), departmental road (class 2), motorway (class 3), etc.) or not.

[0077] According to another example, road type classes can define different road surfaces (bitumen, earth, mixed, etc.).

[0078] The classifier 16 provides as output an identifier ID2 of a class of weather conditions in an environment of the vehicle 10 when a vector of characteristics z is presented at its input.

[0079] For example, weather condition classes can define whether it is raining, freezing, foggy, etc.

[0080] The environment of the vehicle 10 can be defined by a geographic area centered on the geographic coordinates of the vehicle 10.

[0081] This architecture of the deep convolutional neural network makes it possible to obtain a good compromise between a computational cost during the inference phase of the model for predicting traffic conditions on a roadway and the performance of the deep convolutional neural network in terms of segmentation.

[0082] According to one variant, the input images are adapted to optimize the calculations of the deep convolutional neural network.

[0083] For example, the input images are sized to have a resolution equal to 256 columns and 256 rows.

[0084] During the training phase of the deep convolutional neural network, the weights of the deep convolutional neural network are determined from a set of images suitable for segmentation and mapping of roadway irregularities and / or for classification of roadway type and / or classification of weather conditions. The set of images is for example formed from image databases such as ADE20K [Semantic Understanding of Scenes Through the ADE20K Dataset Bolei Zhou, Hang Zhao, Xavier Puig, Tete Xiao, Sanja Fidler, Adela Barriuso, Antonio Torralba https: / / arxiv.org / abs / 1608.05442], Open Surfaces [OpenSurfaces: A Richly Annotated Catalog of Surface Appearance http: / / opensurfaces.cs.cornell.edu] and / or Cityscapes [The Cityscapes Dataset for Semantic Urban Scene Understanding M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, R. Benenson, U. Franke, S. Roth, and B. Schiele https: / / www.cityscapes-dataset.com / ].

[0085] Image augmentation strategies such as horizontal image flipping, changes in contrast and brightness of images can be implemented to increase the variability of the set of images used for determining the weights of the deep convolutional neural network.

[0086] The neural network is trained from the set of images with the aim of obtaining a segmentation mask MS for each image of the set of images presented as input to the roadway traffic conditions prediction model, of obtaining a mapping of roadway irregularities CI, and of classifying the type of roadway and weather conditions, i.e. associating a roadway type class identifier ID1 and a weather conditions class identifier ID2 with each image of the set of images.

[0087] The classifiers 15 and 16 of the deep convolutional neural network are trained using, on the one hand, a metric measuring the performance of the classifiers 15 and 16 such as for example a metric based on a cross-entropy loss and, on the other hand, a metric measuring the performance of the decoders 13 and 14 such as for example a metric based on a cross-entropy loss per image pixel.

[0088] Learning can use a Stochastic Gradient Descent (SGD) method using an initial learning rate of 0.001 and momentum equal to 0.9 (An Improved Analysis of Stochastic Gradient Descent with Momentum Yanli Liu, Yuan Gao, Wotao Yin https: / / proceedings.neurips.cc / paper / 2020 / file / d3f5d4de09eal9461dab00590df91e4f-P aper.pdf).

[0089] The learning rate can then follow a cosine decay with a weight decay set to 0.0005.

[0090] Preferably, the neural network is a deep convolutional neural network of the SegNet type (SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation, V. Badrinarayanan, A. Kendall, and R. Cipoll https: / / ieeexplore.ieee.org / stamp / stamp.jsp?tp=&arnumber=7803544).

[0091] Alternatively, the neural network is a Deeplabv3 type deep convolutional neural network (https: / / www.imaios.com / fr / ressources / blog / deeplabv3-et-imagerie-medicale#:~:text= The architecture%20DeepLabV3%20propose%20d,r%C3%A9seau%20de%20neurones%20convolutionif%20profond) which is efficient for semantic segmentation of images. Its main strength lies in its ability to capture the semantic information of an image at different scales in order to segment this image.

[0092] Alternatively, the neural network is a deep convolutional neural network of type UNet (https: / / datascientest.com / u-net).

[0093] In a second operation, the device 2 obtains a second data item D2 representative of a pose of a driver of the vehicle 10 from an image c of the driver. The second data item D2 makes it possible to obtain an estimate of the comfort of the driver of the vehicle 10.

[0094] According to a variant, the image of the driver is captured by an on-board sensor of the vehicle 10, preferably housed on the dashboard of the vehicle 10. This sensor is, for example, the camera 11 of the on-board imaging device of the vehicle 10.

[0095] According to a variant, the second data D2 is obtained at the output of a trained model 17 for estimating the two-dimensional (2D) pose of a vehicle driver supplied as input by the image of the driver.

[0096] According to a variant, the 2D vehicle driver pose estimation model is implemented by a neural network such as that described in the article by Xiao Sun et al. entitled “Integral Human Pose Regression” https: / / arxiv.org / pdf / 1711.08229.pdf).

[0097] More specifically, the vehicle driver 2D pose estimation model is an image classification model implemented by a first deep convolutional neural network of type Resnet-50 to extract a feature vector representative of a 2D pose of a person skeleton for each person input image and a second shallow neural network used to estimate a 2D person pose from the feature vector.

[0098] A deep convolutional neural network of type ResNet-X is an X-layer convolutional neural network as defined in (Deep Residual Learning for Image Recognition aiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun https: / / arxiv.org / abs / 1512.03385). A deep convolutional neural network of type ResNet is different from traditional neural networks in that it takes residuals from each layer and uses them in subsequent connected layers.

[0099] The first deep convolutional neural network of type Resnet-50 is trained on the MPII benchmark dataset (2d Human Pose Estimation: New Benchmark and State of the Art Analysis Mykhaylo Andriluka, Leonid Pi- shchulin, Peter Gehler, and Bemt Schielel https: / / openaccess.thecvf.com / content_cvpr_2014 / papers / Andriluka_2D_Human_Pose_2014_CVPR_paper.pdf).

[0100] Image augmentation strategies such as scaling, rotation or even flipping of images with probability of 0.5 can be implemented to increase the variability of the set of images used for determining the weights of the deep convolutional neural network.

[0101] Training the first deep convolutional neural network may use a Stochastic Gradient Descent (SGD) method using an initial learning rate of 0.001 which may decrease, for example, to 0.00005 when the loss saturates on the data set used for training the first deep convolutional neural network.

[0102] In a third operation, the device 2 obtains a third piece of data D3 representative of a current operating state of the vehicle 10 from at least one sensor 18 on board the vehicle 10.

[0103] Each sensor monitors the state of vehicle components and records sensory information from the vehicle in the form of measurements that make it possible to evaluate the proper functioning of these components. These measurements make it possible to estimate a current state of the vehicle 10 and its behavior when the vehicle 10 is traveling on a roadway.

[0104] According to variants, said measurements are the current speed of the vehicle 10 and / or its acceleration and / or its jerk and / or its slippage and / or its axle angle.

[0105] According to a variant, the first, second and third data are normalized to make them compatible with each other at the input of a model 19 for determining the vehicle comfort speed.

[0106] In a fourth operation, the device 2 obtains a comfort speed v of the vehicle at the output of the trained vehicle comfort speed determination model 19 when the first, second and third data are presented at the input of the trained model 19.

[0107] The model 19 is trained to merge the first, second and third data so as to enable the model 19 to efficiently take advantage of the information from these three data.

[0108] According to one variant, the vehicle comfort speed determination model is implemented by a deep convolutional neural network which is composed, according to one variant, of a ResNet34 type architecture, a hidden layer and two dense linear layers.

[0109] The first data, i.e. the segments of the segmentation mask and the irregularity mapping, and the second data representative of a pose of the driver of the vehicle 10 are presented as input to the ResNet34 type architecture. which then provides first features. The third data is then passed through the hidden layer which then provides second features. The first and second features are then fused. The fused features are presented as input to the two dense linear layers and a prediction of the vehicle's comfort speed is provided as output from the second of the two dense layers.

[0110] According to a variant, the deep convolutional neural network which implements the vehicle comfort speed determination model is trained using the stochastic gradient descent optimizer with a momentum of 0.9 and a weight decay of 0.00005.

[0111] The training of the deep convolutional neural network implementing the vehicle comfort speed determination model may use a Stochastic Gradient Descent (SGD) method using an initial learning rate of 0.001 which may decrease according to a cosine curve for example down to 0.00005 when the loss saturates on the data set used for training the deep convolutional neural network. This decrease in the learning rate ensures that the network training converges efficiently to an optimum.

[0112] [Fig.3] schematically illustrates a device 2 configured to determine a comfort speed of a vehicle traveling on a roadway, according to a particular and non-limiting exemplary embodiment of the present invention.

[0113] The device 2 corresponds for example to a device on board the vehicle 10, for example a computer.

[0114] The device 2 is for example configured for the implementation of at least one operation described with regard to figures 1 and 2 and / or the steps of the method described with regard to [Fig.4]. Examples of such a device 2 include, but are not limited to, on-board electronic equipment such as an on-board computer of a vehicle, an electronic calculator such as an ECU (“Electronic Control Unit”), a smartphone, a tablet, a laptop. The elements of the device 2, individually or in combination, can be integrated into a single integrated circuit, into several integrated circuits, and / or into discrete components. The device 2 can be produced in the form of electronic circuits or software (or computer) modules or even a combination of electronic circuits and software modules.

[0115] The device 2 comprises one (or more) processor(s) 20 configured to execute instructions for carrying out at least one step of the method and / or for executing the instructions of the software(s) embedded in the device 2. The processor 20 may include integrated memory, an input / output interface, and different circuits known to those skilled in the art. The device 2 further comprises at least one memory 21 corresponding for example to a volatile and / or non-volatile memory and / or comprises a memory storage device which may comprise volatile and / or non-volatile memory, such as EEPROM, ROM, PROM, RAM, DRAM, SRAM, flash, magnetic or optical disk.

[0116] The computer code of the embedded software(s) comprising the instructions to be loaded and executed by the processor is for example stored in the memory 21.

[0117] According to various particular and non-limiting embodiments, the device 2 is coupled in communication with other similar devices or systems and / or with communication devices, for example a TCU (from the English “Telematic Control Unit” or in French “Telematic Control Unit”), for example via a communication bus or through dedicated input / output ports.

[0118] The vehicle comfort speed determination model, the roadway traffic conditions prediction model and / or the two-dimensional vehicle driver pose estimation model can be trained by a remote device, such as a server or a series of servers, in communication with the device 2 via a communication network. The internal parameters of the model trained by the remote device can then be transmitted by this remote device to the device 2 which receives them. The device 2 is then able to implement this model.

[0119] According to a particular and non-limiting exemplary embodiment, the device 2 comprises a block 22 of interface elements for communicating with external devices, for example a remote server or the “cloud”, other nodes of the ad hoc network. The interface elements of the block 22 comprise one or more of the following interfaces: - RF radio frequency interface, for example Wi-Fi® type (according to IEEE 802.11), for example in the 2.4 or 5 GHz frequency bands, or Bluetooth® type (according to IEEE 802.15.1), in the 2.4 GHz frequency band, or Sigfox type using UBN (Ultra Narrow Band) radio technology, or LoRa in the 868 MHz frequency band, LTE (Long-Term Evolution), LTE-Advanced; - USB interface (from the English “Universal Serial Bus” or “Universal Serial Bus” in French); - HDMI interface (from the English “High Definition Multimedia Interface” or “High Definition Multimedia Interface” in French); - LIN interface (from the English “Local Interconnect Network”).

[0120] Data, for example internal parameters of the vehicle comfort speed determination model and / or the roadway traffic conditions prediction model and / or the vehicle driver two-dimensional pose estimation model, are for example loaded to the device 2 via the interface of the block 22 using a Wi-Fi® network such as according to IEEE 802.11, an ITS G5 network based on IEEE 802.1 Ip or a mobile network such as a 4G (or 5G) network based on the LTE (Long Term Evolution) standard defined by the 3GPP consortium, in particular an LTE-V2X network.

[0121] According to another particular and non-limiting exemplary embodiment, the device 2 comprises a communication interface 23 which makes it possible to establish communication with other devices (such as other computers of the on-board system) via a communication channel 24. The communication interface 23 corresponds for example to a transmitter configured to transmit and receive information and / or data via the communication channel 24. The communication interface 23 corresponds for example to a wired network of the CAN (Controller Area Network) type, CAN FD (Controller Area Network Flexible Data-Rate), FlexRay (standardized by the ISO 17458 standard) or Ethernet (standardized by the ISO / IEC 802-3 standard).

[0122] Internal parameters of the vehicle comfort speed determination model and / or the roadway traffic condition prediction model and / or the two-dimensional vehicle driver pose estimation model may be received via the communication interface 23.

[0123] According to a particular and non-limiting exemplary embodiment, the device 2 can provide output signals to one or more external devices, such as a display screen 25, touch-sensitive or not, one or more speakers 26 and / or other peripherals 27 (projection system) via output interfaces 28, 29 and 30 respectively. According to a variant, one or other of the external devices is integrated into the device 2.

[0124] [Fig.4] illustrates a flowchart of the different steps of a method for determining a comfort speed of a vehicle traveling on a roadway, according to a particular and non-limiting exemplary embodiment of the present invention. The method is for example implemented by a device on board the vehicle 10 or by the device 2 of [Fig.3].

[0125] In a first step 41, a first piece of data is obtained. The first piece of data is representative of traffic conditions on the roadway on which the vehicle is traveling from an image of the roadway.

[0126] In a second step 42, a second piece of data is obtained. The second piece of data is representative of a pose of a driver of the vehicle from an image of the driver.

[0127] In a third step 43, a third piece of data is obtained. The third piece of data is representative of a current operating state of the vehicle from at least one on-board sensor of the vehicle.

[0128] In a fourth step 44, a vehicle speed is obtained as an output from a trained vehicle comfort speed determination model when the first, second and third data are presented as input to the trained vehicle comfort speed determination model.

[0129] According to a variant, the variants and examples of the operations described in relation to Figures 1 and 2 apply to the steps of the method of [Fig.4].

[0130] Of course, the present invention is not limited to the embodiments described above but extends to a method for determining a comfort speed of a vehicle traveling on a roadway which would include secondary steps without thereby departing from the scope of the present invention. The same would apply to a device configured for implementing such a method.

[0131] The present invention also relates to a vehicle, for example an automobile or more generally an autonomous land-based motor vehicle, comprising the device 2 of [Fig.3].

Claims

Claims

1. Method for determining a comfort speed of a vehicle traveling on a roadway, the method being implemented by at least one processor and comprising the following steps: - obtaining (41) a first data item representative of traffic conditions on the roadway from an image of the roadway; - obtaining (42) a second data item representative of a pose of a driver of the vehicle from an image of the driver; - obtaining (43) a third data item representative of a current operating state of the vehicle from at least one on-board sensor of the vehicle; - obtaining (44) the comfort speed of the vehicle as output from a trained model for determining the vehicle's comfort speed when the first, second and third data items are presented as input to the trained model for determining the vehicle's comfort speed.

2. Method according to claim 1, for which the first data is obtained as output from a trained model for predicting traffic conditions on the roadway when the image of the roadway is presented as input to the trained model for predicting traffic conditions on the roadway.

3. The method of claim 2, wherein the roadway traffic condition prediction model is implemented by a deep convolutional neural network.

4. Method according to one of claims 2 to 3, for which the first data is representative of a segmentation mask of the image of the roadway, of a mapping of irregularities of the roadway, of a class identifier of type of the roadway and of a class identifier of meteorological conditions in an environment of the vehicle.

5. Method according to one of the preceding claims, for which the second data is obtained at the output of a trained model for estimating the two-dimensional pose of a vehicle driver supplied as input by the image of the driver.

6. The method of claim 5, wherein the two-dimensional vehicle driver pose estimation model is implemented by a neural network.

7. Method according to one of the preceding claims, for which the vehicle comfort speed determination model is implemented by a deep convolutional neural network.

8. Computer program comprising instructions for implementing the method according to any one of the preceding claims, when these instructions are executed by a processor.

9. Device (2) for determining a comfort speed of a vehicle traveling on a roadway comprising a memory (21) associated with at least one processor (20) configured for implementing the steps of the method according to any one of claims 1 to 7.

10. Vehicle (10) comprising the device (2) according to claim 9.

Images