Projection of a light pattern to improve multiple image processing functions of a vehicle

Abstract

The invention relates to a method of processing an image representative of a scene facing a vehicle. A light intensity map is generated (501) from a generation model configured to receive as input an image representative of a scene facing the vehicle. The determined light intensity map is transmitted (502) to the light module, for projection (503) of a pixelated light beam according to the light pattern corresponding to the generated light intensity map. A first image representative of the scene facing the vehicle is obtained (503), following the projection of the at least one pixelated light beam according to the light pattern. The method includes determining (505; 506) at least a first image processing result by applying a first image processing function to the first image and a second image processing result by applying a second image processing function to the first image.

Description

[0001] Description

[0002] Title: Projecting a light pattern to improve several image processing functions of a vehicle

[0003] The present invention relates to the field of image processing, particularly of images representing a scene facing a vehicle. More specifically, the invention relates to a system and a method for applying several image processing functions to at least one image representing a scene facing a vehicle, such as a motor vehicle for example, after projecting a luminous pattern onto the scene, in particular several image processing functions to several images representing a scene facing a vehicle, such as a motor vehicle for example, after projecting luminous patterns onto the scene.

[0004] Advanced Driver Assistance Systems (ADAS) are becoming increasingly common in vehicles, such as automobiles. These systems assist the driver in piloting the vehicle, and in the case of autonomous vehicles, they can even completely control certain driving parameters without driver intervention.

[0005] ADAS functions rely on sensor information from the vehicle, particularly images acquired by a camera, to which various image processing functions can be applied. The results of these image processing functions are then used by the ADAS functions. Image processing functions may include, for example:

[0006] - a function for determining depth information of the scene represented by an image. Such information can be used to confirm depth data acquired by another sensor on the vehicle, for example by a lidar. Such redundancy thus ensures robustness in estimating the distance to an object, and ensures greater safety in the execution of the driver assistance function based on depth information;

[0007] - a semantic segmentation function designed to segment each image captured by the camera into several pixel regions, each region being labeled with a class from a set of predefined classes. In automotive applications, the following classes might be used: car, pedestrian, sign, road, etc. A semantic segmentation score can be associated with the semantic segmentation of the image, or with each segment identified within the image, the score representing the degree of certainty associated with the semantic segmentation; and / or

[0008] - an object detection function designed to detect one or more objects in the scene represented by each image, to identify the category of each detected object from among several predefined categories, and to determine the position of each detected object (or a part of the image in which the object is located). Each detected object is further associated with a detection score representing the certainty associated with the detection;

[0009] - an instance segmentation function aimed at segmenting each image captured by the camera into several pixel regions for each object in the scene, as with semantic segmentation, but without labeling each region with a class from a set of predefined classes. An instance segmentation score can be associated with the instance segmentation of the image, or with each segment determined in the image, the score being representative of the degree of certainty associated with the instance segmentation;

[0010] - a panoptic segmentation function, which implements both semantic segmentation and instance segmentation, thus aiming to segment each image captured by the camera into several pixel regions for the different objects in the scene, each object being labeled with a class from a set of predefined classes. A panoptic segmentation score can be associated with the panoptic segmentation of the image, or with each segment determined in the image, the score being representative of the degree of certainty associated with the panoptic segmentation;

[0011] - an object tracking function capable of determining a motion vector for at least one identified object in a processed image. However, such processing is less effective in low or very low ambient light, such as when the camera captures images representative of a night scene. In such situations, the performance of these image processing functions is significantly lower than in daytime driving conditions, which can prevent the implementation of certain driver assistance features and even lead to safety issues.

[0012] There is therefore a need to improve the accuracy associated with several image processing functions applied to at least one image representative of a scene facing a motor vehicle, or even to images representative of a scene facing a motor vehicle, when the ambient light is low.

[0013] The present invention improves the situation.

[0014] A first aspect of the invention relates to an image processing method in a vehicle, the method comprising, during a typical phase, the following steps:

[0015] - obtaining a representative image of a scene facing the vehicle;

[0016] - generation of at least one first light intensity map, from a light intensity map generation model and the image obtained, the light intensity map generation model being configured to receive as input the image representing the scene and to generate as output the first light intensity map, the first light intensity map corresponding to a light pattern and indicating light intensity values ​​to control light elements of at least one pixelated light source of at least one light module of the vehicle;

[0017] - transmission of at least the first determined light intensity map to at least one light module, for projection of at least a first pixelated light beam according to the light pattern corresponding to the first generated light intensity map, in the scene facing the vehicle;

[0018] - obtaining at least one first representative image of the scene facing the vehicle, following the projection of at least one first pixelated light beam according to the light pattern;

[0019] - determination of at least one first image processing result by applying a first image processing function to at least one first image and / or of a second image processing result by applying a second image processing function to at least one first image.

[0020] Thus, the invention enables the generation of a light intensity map to project a light pattern onto the scene, thereby improving several image processing functions (at least the first and second image processing functions), particularly in night driving situations or when ambient light is low. Adapting the projected light pattern to the scene facing the vehicle (based on the image of the scene obtained before the light pattern was projected) allows for the adaptive improvement of the accuracy of the image processing functions.To this end, the light intensity map generation model can be designed, for example, using machine learning, particularly supervised learning, specifically to improve at least the first and second image processing results (and optionally other image processing results from at least a third image processing function implemented in the vehicle), compared to a situation in which the light pattern is not projected onto the scene facing the vehicle. Furthermore, high responsiveness is possible for lighting control (represented by the time between image acquisition and the first image), since the light intensity map generation model is capable of directly outputting a command that can be used to control the pixelated light source of the lighting module.

[0021] Another aspect of the invention relates to a first aspect of the invention concerning a method for processing images in a vehicle, the method comprising, during a typical phase, the following steps:

[0022] - obtaining a representative image of a scene facing the vehicle;

[0023] - generation of light intensity maps comprising a first light intensity map and a second light intensity map, from a light intensity map generation model and the image obtained, the light intensity map generation model being configured to receive as input the image representing the scene and to generate as output light intensity maps, each light intensity map corresponding to a light pattern and indicating light intensity values ​​to control light elements of at least one pixelated light source of at least one light module of the vehicle;

[0024] - transmission of the generated light intensity maps to at least one light module, for projection of at least a first pixelated light beam according to a first light pattern corresponding to the first generated light intensity map, in a first wavelength range, and of a second pixelated light beam according to a second light pattern corresponding to the second generated light intensity map, in a second wavelength range, in the scene facing the vehicle;

[0025] - obtaining at least a first image representative of the scene facing the vehicle, in which the first light beam pixelated according to the first light pattern is projected in the first wavelength range and a second image representative of the scene facing the vehicle, in which the second light beam pixelated according to the second light pattern is projected in the second wavelength range;

[0026] - determination of at least one first image processing result by applying a first image processing function to the first image and / or the second image, and of a second image processing result by applying a second image processing function to the first image and / or the second image.

[0027] Thus, the invention enables the generation of light intensity maps to project light patterns onto the scene, across several wavelength ranges, thereby improving several image processing functions (at least the first and second image processing functions), particularly in night driving situations or when ambient light is low. Adapting the projected light patterns to the scene facing the vehicle (based on the image of the scene obtained before the light pattern is projected) allows for the adaptive improvement of the accuracy of the image processing functions.To this end, the light intensity map generation model can be designed, for example trained by machine learning, particularly supervised learning, specifically to improve at least the first and second image processing results (and optionally other image processing results from at least a third image processing function implemented in the vehicle), compared to a situation in which light patterns are not projected onto the scene facing the vehicle. Furthermore, high responsiveness is possible for lighting control (represented by the time between image acquisition and the first and second images), since the light intensity map generation model is capable of directly producing usable commands to control at least one pixelated light source in at least one lighting module.

[0028] According to some embodiments, the first image processing function can be a function for determining depth information of the scene represented in at least one first image.

[0029] Such a first function is advantageous in that it can provide redundant information with data captured by a depth camera or lidar of the vehicle, which enhances the reliability of depth information, and therefore the reliability of vehicle functions that are based on such depth information.

[0030] If necessary, the first image processing function can be a function for determining depth information of the scene represented in the first and second images.

[0031] In some embodiments, the second image processing function can be a semantic perception function. Semantic perception functions make it possible to identify and classify segments or areas of the scene facing the vehicle, and such information is used by many functions implemented in current vehicles to enhance safety or improve driving comfort.

[0032] According to some embodiments, the method may further include transmitting the first image processing result and / or the second image processing result to a vehicle driver assistance module, for implementation of at least one driver assistance function based on the first image processing result and / or the second image processing result.

[0033] Thus, improving the accuracy associated with image processing functions makes it possible to improve at least one driver assistance function, and therefore safety and / or driving comfort.

[0034] According to some embodiments, the light intensity map generation model can have one of the following structures:

[0035] - a convolutional neural network;

[0036] - an artificial neural network of the auto-encoder or variational auto-encoder type;

[0037] - a self-aware or transformative model; or

[0038] - a network generating a system of antagonistic generative networks.

[0039] Such model structures are particularly well-suited for receiving an image as input and generating an output command in the form of a light intensity map. Furthermore, such structures can be used for machine learning.

[0040] According to some embodiments, the representative image of the scene facing the vehicle can be obtained in a predetermined wavelength range, and the first representative image of the scene facing the vehicle can be obtained in a first predetermined wavelength range, identical to or distinct from the predetermined wavelength range, following the projection of a light beam pixelated according to the light pattern in the first wavelength range. Thus, the invention can be implemented in one or more wavelength ranges, which may include visible or non-visible wavelengths.The wavelength range(s) can be determined according to various criteria, such as the accuracy in determining the first and second image processing results, the cost and availability of a camera capable of obtaining the image and / or the first image, the wavelengths in which the pixelated light source of the light module is able to form a light pattern, etc.

[0041] In one embodiment of the invention, the first aspect of the invention:

[0042] - the generation step includes the generation of a second light intensity map, from said light intensity map generation model and the image obtained, the light intensity map generation model being configured to receive as input the image representing the scene and to generate as output light intensity maps, each light intensity map corresponding to a light pattern and indicating light intensity values ​​to control light elements of at least one pixelated light source of at least one light module of the vehicle;

[0043] - the transmission step includes the transmission of the second generated light intensity map to at least one light module, for projection of a second pixelated light beam according to a second light pattern corresponding to the second generated light intensity map, in a second wavelength range, in the scene facing the vehicle;

[0044] - the acquisition step involves obtaining a second image representative of the scene facing the vehicle, into which the second pixelated light beam is projected according to the second light pattern in the second wavelength range;

[0045] - the determination step includes determining the first image processing result by applying a first image processing function to the first image and / or the second image, and a second image processing result by applying a second image processing function to the first image and / or the second image.

[0046] Where appropriate, in this embodiment of the first aspect of the invention or according to the second aspect of the invention, the representative image of the scene facing the vehicle can be obtained in a predetermined wavelength range, the predetermined wavelength range being identical to the first wavelength range, identical to the second wavelength range, or distinct from the first and second wavelength ranges.

[0047] Thus, the invention can be implemented in at least two wavelength ranges, which may include visible or non-visible wavelengths. The wavelength ranges can be determined according to various criteria, such as the accuracy in determining the first and second image processing results, the cost and availability of a camera capable of obtaining the image and / or the first image and / or the second image, the wavelengths in which the pixelated light source of at least one light module is capable of forming light patterns, etc.

[0048] According to a first embodiment, the first and second light intensity maps generated can be transmitted to a first light module of the vehicle, and the first light beam and the second light beam can be projected by the first light module.

[0049] Thus, the first light module is capable of projecting light in at least two wavelength ranges. A single control unit for the first light module can receive both the first and second light intensity maps, which facilitates control of the pixelated light source for projecting the first and second light beams.

[0050] In addition, the first and second generated light intensity maps can be transmitted to a second light module on the vehicle, and the first and second light beams can be projected by this second light module. Thus, the first beam projected by the first light module can combine with the first beam projected by the second light module to form the first light pattern. Similarly, the second beam projected by the first light module can combine with the second beam projected by the second light module to form the second light pattern.

[0051] Alternatively, according to a second embodiment, the first light intensity map can be transmitted to a first light module of the vehicle and the second light intensity map can be transmitted to a second light module of the vehicle, the first light beam can be projected by the first light module and the second light beam can be projected by the second light module.

[0052] Thus, each light module can be dedicated to projecting a light beam in a given wavelength range, which simplifies the control of the pixelated light source in each light module.

[0053] In another embodiment of the invention, the first light intensity map can be transmitted to the light module for projection of the first pixelated light beam, based on the light pattern corresponding to the first generated light intensity map, in the first wavelength range, and for projection of a second pixelated light beam, also based on the light pattern corresponding to the first generated light intensity map, in a second wavelength range distinct from the first. The method may further include obtaining a second image representing the scene facing the vehicle, following the projection of the second pixelated light beam, based on the light pattern, in the second wavelength range.The first image processing result can be determined by applying the first image processing function to the first image obtained and / or the second image obtained, and the second image processing result can be determined by applying the second image processing function to the first image obtained and / or the second image obtained.

[0054] Thus, the first and second image processing results can be determined from two images acquired from a scene in which an adaptive light pattern is projected into two distinct wavelength ranges, simultaneously or alternately during the same period. The accuracy associated with the first and second image processing results is thereby improved. In some embodiments, the light pattern is projected into strictly more than two distinct wavelength ranges, and the image processing functions are capable of obtaining image processing results from strictly more than two images (the first image in the first wavelength range, the second image in the second wavelength range, and at least one other image in a different wavelength range).

[0055] According to some embodiments, the method may further include a training phase for the light intensity map generation model, comprising a modification of at least one parameter of the generation model as a function of an overall loss, the overall loss being evaluated from:

[0056] - of an initial loss assessed based on a first training image processing result determined by a first training image processing function, following the generation of a training light intensity map by the generation model during the training phase; and

[0057] - a second loss evaluated based on a second training image processing result determined by a second training image processing function, following the generation of the training light intensity map by the generation model during the training phase.

[0058] Thus, the light intensity map generation model is specifically trained to produce a light intensity map for projecting a light pattern that improves image processing results. Where appropriate, the method may include a training phase for the light intensity map generation model, comprising modifying at least one parameter of the generation model based on an overall loss, the overall loss being evaluated from:

[0059] - of a first loss evaluated according to a first result of training image processing determined by a first training image processing function, following the generation of training light intensity maps by the generation model during the training phase;

[0060] - a second loss evaluated based on a second training image processing result determined by a second training image processing function, following the generation of training light intensity maps by the generation model during the training phase.

[0061] Thus, the light intensity map generation model is specifically trained to produce light intensity maps to project light patterns that improve image processing results.

[0062] According to embodiments, the first image processing function of the training phase may be identical to the first image processing function of the current phase, and / or the second image processing function of the training phase may be identical to the second image processing function of the current phase.

[0063] Thus, the generation model is specifically trained to improve the performance of at least the first and second image processing functions implemented during the current phase of the process, which allows for optimization of the accuracy of the first and second image processing results.

[0064] In addition, the training phase may include the following steps:

[0065] - obtaining an association of training data, the association comprising a representative training image of a scene, a first reference image processing result and a second reference image processing result;

[0066] - application of the light intensity map generation model to the training image, to obtain a training light intensity map;

[0067] - obtaining at least one synthetic image representative of the scene of the training image in which at least one pixelated light beam is projected according to a light pattern corresponding to the training light intensity map;

[0068] - application of the first training image processing function to at least one synthetic image, to determine the first training image processing result;

[0069] - application of the second training image processing function to at least one synthetic image, to determine the second training image processing result;

[0070] - evaluation of the first loss by comparing the first training image processing result with the first reference image processing result;

[0071] - evaluation of the second loss by comparing the second training image processing result with the second reference image processing result;

[0072] - assessment of the overall loss based on the first loss and the second loss;

[0073] - modification of at least one parameter of the map generation model - of light intensity as a function of the overall loss assessed.

[0074] Such training steps enable supervised learning of the light intensity map generation model when these steps are repeated until a predefined convergence criterion is reached. The resulting light intensity map generation model is optimal for improving the results of the first and second image processing functions.

[0075] If applicable, the training phase may include the following steps:

[0076] - obtaining an association of training data, the association comprising a representative training image of a scene, a first reference image processing result and a second reference image processing result;

[0077] - application of the light intensity map generation model to the training image, to obtain training light intensity maps comprising a first training light intensity map and a second training light intensity map;

[0078] - obtaining at least a first synthetic image representative of the scene of the training image in which is projected a first light training pattern corresponding to the first light intensity map of the training, in the first wavelength range, and a second synthetic image representative of the scene of the training image in which is projected a second light training pattern corresponding to the second light intensity map, in the second wavelength range;

[0079] - application of the first training image processing function to the first synthetic image and / or the second synthetic image, to determine the first training image processing result;

[0080] - application of the second training image processing function to the first synthetic image and / or the second synthetic image, to determine the second training image processing result;

[0081] - evaluation of the first loss by comparing the first training image processing result with the first reference image processing result;

[0082] - evaluation of the second loss by comparing the second training image processing result with the second reference image processing result;

[0083] - assessment of the overall loss based on the first loss and the second loss;

[0084] - modification of at least one parameter of the light intensity map generation model based on the overall assessed loss.

[0085] Such training steps enable supervised learning of the light intensity map generation model when these steps are repeated until a predefined convergence criterion is reached. The resulting light intensity map generation model is optimal for improving the results of the first and second image processing functions.

[0086] According to some embodiments, the pixelated light source of at least one light module may comprise electroluminescent semiconductor elements of submillimeter dimensions, epitaxially mounted directly on a common substrate.

[0087] Such a lighting module allows the projection of a pixelated light beam according to a high-resolution light pattern, which improves the ability of the light intensity map generation model to improve the first and second image processing results.

[0088] A second aspect of the invention relates to an assembly in a vehicle comprising:

[0089] - at least one camera arranged to obtain a representative image of a scene facing the vehicle;

[0090] - at least one light module comprising a pixelated light source comprising a plurality of individually controllable light elements;

[0091] - a control device configured to determine at least one first light intensity map, from a light intensity map generation model and the image obtained by at least one camera, the light intensity map generation model being configured to receive as input the image representative of the scene and to generate as output the first light intensity map, the first light intensity map corresponding to a light pattern and indicating light intensity values ​​to control light elements of the pixelated light source of at least one light module of the vehicle.

[0092] The control device is capable of transmitting the determined light intensity map to at least one light module, for projection of at least one first pixelated light beam, according to the light pattern corresponding to the first generated light intensity map, onto the scene facing the vehicle. At least one camera is capable of obtaining at least one first representative image of the scene facing the vehicle, following the projection of at least one first pixelated light beam according to the light pattern.The set further includes at least one first image processing module configured to determine a first image processing result by applying a first image processing function to at least one first image obtained, and / or a second image processing module configured to determine a second image processing result by applying a second image processing function to at least one first image obtained.

[0093] A second aspect of the invention relates to an assembly in a vehicle comprising:

[0094] - at least one camera arranged to obtain a representative image of the scene facing the vehicle;

[0095] - at least one light module comprising a pixelated light source comprising a plurality of individually controllable light elements;

[0096] - a control device configured to determine light intensity maps comprising a first light intensity map and a second light intensity map, from a light intensity map generation model and the image obtained by at least one camera, the light intensity map generation model being configured to receive as input the image representing the scene and to generate as output light intensity maps, each light intensity map corresponding to a light pattern and indicating light intensity values ​​to control light elements of the pixelated light source of at least one light module of the vehicle;in which the control device is capable of transmitting the generated light intensity maps to at least one light module, for projection of at least a first pixelated light beam according to a first light pattern corresponding to the first generated light intensity map, in a first wavelength range and a second pixelated light beam according to a second light pattern corresponding to the second generated light intensity map, in the second wavelength range, in the scene facing the vehicle.

[0097] At least one camera is capable of obtaining at least a first representative image of the scene facing the vehicle, following the projection of the first pixelated light beam according to the first light pattern in the first wavelength range, and a second representative image of the scene facing the vehicle, following the projection of the second pixelated light beam according to the second light pattern in the second wavelength range. The system further includes at least one first image processing module configured to determine a first image processing result by applying a first image processing function to the first image obtained and / or to the second image obtained, and a second image processing module configured to determine a second image processing result by applying a second image processing function to the first image obtained and / or to the second image obtained.

[0098] A third aspect of the invention relates to a vehicle comprising an assembly according to the second aspect of the invention and further comprising a driver assistance module. The driver assistance module is configured to implement at least one driver assistance function based on the first image processing result and / or based on the second image processing result.

[0099] Other features and advantages of the invention will become apparent upon examination of the detailed description below, and the accompanying drawings in which:

[0100] [Fig. 1a] illustrates an image processing system in a vehicle, according to embodiments of the invention;

[0101] [Fig. 1 b] illustrates a light module of a vehicle image processing system, according to embodiments of the invention;

[0102] [Fig. 2a] illustrates a light pattern projected by a light module of a vehicle image processing system, according to embodiments of the invention; [Fig. 2b] illustrates an image acquired by a camera following the projection of a light pattern in a scene facing a vehicle, according to embodiments of the invention;

[0103] [Fig. 3a] illustrates the structure of a control device, according to embodiments of the invention;

[0104] [Fig. 3b] illustrates the structure of an image processing module according to embodiments of the invention;

[0105] [Fig. 4a] illustrates a pixelated light source from a light module of a vehicle image processing system, according to one embodiment;

[0106] [Fig. 4b] illustrates a pixelated light source from a light module of a vehicle image processing system, according to another embodiment;

[0107] [Fig. 5] is a diagram illustrating the steps of a common phase of an image processing method in a vehicle, according to embodiments of the invention;

[0108] [Fig. 6] is a drive system for a light intensity map generation model, according to embodiments of the invention;

[0109] [Fig. 7] is a diagram illustrating the steps of a training phase of a light intensity map generation model, according to embodiments of the invention;

[0110] [Fig. 8a] illustrates an image processing system in a vehicle, according to embodiments of the invention;

[0111] [Fig. 8b] illustrates a light module of a vehicle image processing system, according to embodiments of the invention;

[0112] [Fig. 9a] illustrates a first light pattern projected by a light module of a vehicle image processing system, according to embodiments of the invention; [Fig. 9b] illustrates a first image acquired by a camera following the projection of a first light pattern in a scene facing a vehicle, according to embodiments of the invention;

[0113] [Fig. 10a] illustrates the structure of a control device, according to embodiments of the invention;

[0114] [Fig. 10b] illustrates the structure of an image processing module, according to embodiments of the invention;

[0115] [Fig. 11a] illustrates a pixelated light source of a light module of a vehicle image processing system, according to one embodiment;

[0116] [Fig. 11b] illustrates a pixelated light source of a light module of a vehicle image processing system, according to another embodiment;

[0117] [Fig. 12] is a diagram illustrating the steps of a common phase of an image processing method in a vehicle, according to embodiments of the invention;

[0118] [Fig. 13] is a drive system for a light intensity map generation model, according to embodiments of the invention;

[0119] [Fig. 14] is a diagram illustrating the steps of a training phase of a light intensity map generation model, according to embodiments of the invention.

[0120] The description focuses on the features that distinguish the system and process from those known in the state of the art.

[0121] In the description that follows, identical elements or steps, by structure or by function, appearing on different figures retain, unless otherwise specified, the same references.

[0122] Figure 1a illustrates an image processing system for a scene 10 facing a vehicle 100, according to embodiments of the invention. The vehicle 100 comprises a right front headlight 105.1 and a left front headlight 105.2. The right front headlight 105.1 comprises at least one first right-hand light module 101.1, and the left front headlight comprises at least one light module 101.2.

[0123] According to the invention, at least one of the light modules 101.1 and 101.2 is capable of projecting at least one light beam forming a luminous pattern in the scene 10, within a first wavelength range, which may be a range of visible wavelengths or a range of wavelengths not including any visible wavelengths. The two light modules 101.1 and 101.2 may be capable of projecting light beams forming the same luminous pattern by superimposing their respective beams.

[0124] A straight light beam 120.1 is projected by the straight light module 101.1. The straight light beam 120.1 can comprise several light beams in several distinct wavelength ranges, including the first wavelength range, as will be better understood upon reading the following. Similarly, a left light beam 120.2 is projected by the left light module 101.2. The left light beam 120.2 can comprise several light beams in several distinct wavelength ranges, including the first wavelength range, as will be better understood upon reading the following.

[0125] No restrictions are attached to the first wavelength range according to the invention, which may be a visible range, or an infrared or ultraviolet range, for example. By way of example, the first wavelength range may be included in a near-infrared range, also called NIR, for “Near Infrared”, in a short-wave infrared range, also called SWIR, for “Short Wave Infrared”, in a medium-wave infrared range, also called MWIR, for “Medium Wave Infrared”, or in a long-wave infrared range, also called LWIR, for “Long Wave Infrared”. Note that the MWIR and LWIR ranges are also called the thermal range. In certain embodiments, and as explained below, the light module 101.1 and / or 101.2 may also be capable of projecting at least a second light beam in at least a second wavelength range, in addition to the first light beam in the first wavelength range, to project the same light pattern as in the first wavelength range or to perform a lighting or signaling function. For example, the light module 101.1 and / or 101.2 may project, alternately at a given frequency, or simultaneously, the first light beam in the first wavelength range and the second light beam in the second wavelength range (and optionally at least a third light beam in a third wavelength range).

[0126] Thus, the light module 101.1 and / or 101.2 can be capable of projecting during the same time interval (simultaneously or by alternating beams at a frequency corresponding to a period smaller than the time interval):

[0127] - a light beam according to a light pattern generated according to the invention, in the first wavelength range; and

[0128] - a light beam in the second wavelength range, according to the same light pattern generated according to the invention, or performing a lighting or signaling function when the second wavelength range is the visible range.

[0129] Note that when the first wavelength range does not include any visible wavelengths, it is avoided to disturb the driver while driving the vehicle when determining the first and second image processing results described below.

[0130] According to the invention, the light beams projected by the light module 101.1 and / or 101.2 are pixelated, which allows the light pattern to be created with a resolution dependent on the number of pixels permitted by a pixelated light source of the light module 101.1 and / or 101.2. Figure 1b illustrates the structure of a light module 101 with a pixelated light source 111, for example, matrix-based, according to embodiments of the invention. The light module 101 can be the front right module 101.1 and / or the front left module 101.2 described above.

[0131] The 101 light module includes:

[0132] - a control unit 110 of the pixelated light source 111;

[0133] - the pixelated light source 111;

[0134] - a projection optic for the light coming from the pixelated light source

[0135] 111 to produce a pixelated beam of light projected in front of the vehicle towards scene 10. No restrictions are attached to the projection optics, which may include any set of optical elements.

[0136] No restrictions are placed on the number of pixels of the pixelated light source 111. Preferably, the pixelated light source 111 is a high-definition light source, that is, one capable of projecting a light beam comprising more than one hundred pixels, preferably more than 1000 pixels. The pixelated light source 111 may also be capable of projecting a light beam comprising more than 10,000 pixels according to embodiments of the invention.

[0137] Furthermore, there are no restrictions attached to the technology associated with the pixelated light source 111, which can be:

[0138] - according to a first example, a matrix of individually controllable light elements, of which at least a first set of light elements is capable of emitting in the first wavelength range. As detailed below, according to embodiments of the invention, the same matrix of light elements may comprise a first set of light elements capable of emitting in the first wavelength range and a second set of light elements capable of emitting in a second wavelength range;

[0139] - according to a second example, a first light source capable of emitting in the first wavelength range and a micromirror array, also called DMD for Digital Micromirror Devices, individually activatable to reflect the light from the first light source towards the projection optics 112. In addition, the pixelated light source 111 may include a second light source capable of emitting in the second wavelength range, as described later. In this case, the micromirror array is alternately controlled, at a given frequency, to produce a first beam when the first light source is activated, and to produce a second beam when the second light source is activated;

[0140] - According to a third example, a first laser light source capable of emitting in the first wavelength range, and a controllable mirror for scanning a predetermined set of positions. Such technology is called laser scanning. The control unit 110 is capable of synchronously controlling the movement of the mirror and the activation of the first laser light source. Furthermore, the pixelated light source 111 may include a second laser light source capable of emitting in the second wavelength range, as described later.

[0141] Note that the pixelated light source 111, according to each of the three examples above, can also emit light in at least one other wavelength range, in addition to the first and second wavelength ranges. Thus, more generally, the pixelated light source 111 is capable of emitting a light beam according to a light pattern in at least one wavelength range, including at least the first wavelength range.

[0142] The three examples listed above have the advantage of allowing the realization of a light pattern in a pixelated beam with high resolution, from a light intensity map generated and transmitted by a control device 103 described below.

[0143] In the first example described above, the light elements can be electroluminescent elements, individually controlled by a voltage applied to the terminals of each light element by the control unit 110. Each electroluminescent light element can be mounted on its own substrate. Alternatively, the electroluminescent light elements can be on the same substrate, in which case the pixelated light source is said to be monolithic.

[0144] A so-called "monolithic" light source can exhibit a particularly high density of light-emitting elements, making it especially attractive for a wide range of applications. A monolithic source consists of multiple submillimeter-sized electroluminescent semiconductor elements epitaxially bonded directly to a common substrate, typically silicon. Unlike conventional LED arrays, where each individual light-emitting element is an individually manufactured electronic component mounted on a substrate such as a printed circuit board (PCB), a monolithic source is considered a single electronic component. During its production, multiple arrays of electroluminescent semiconductor junctions are generated on a common substrate, forming a matrix.This production technique allows for the creation of closely spaced electroluminescent areas, each acting as a basic light element. The gaps between these light elements can be submillimeter in size. One advantage of this production technique is the high pixel density that can be achieved on a single substrate.

[0145] The first example has the advantage, when the pixelated light source comprises a first set of light elements and a second set of light elements, of projecting two light beams simultaneously in the first wavelength range and in the second wavelength range. This is because the two sets of light elements can be controlled separately by the control unit 110.

[0146] In the second example described above, the control unit 110 controls the micromirrors of the array to produce the light beam containing the light pattern in the first wavelength range, based on a first received light intensity map, when the first light source is active. The resolution of the light pattern then depends on the number of micromirrors in the array. Optionally, when the pixelated light source 111 includes at least one second light source and when the second light source is active, the control unit 110 controls the micromirrors of the array to produce the second light beam in the second wavelength range, based on the received light intensity map (to project the same light pattern as in the first wavelength range) or to perform a lighting function.

[0147] In the third example described above, the control unit 110 controls the mirror and the first laser light source to scan the predetermined set of positions and produce the light beam comprising the light pattern in the first wavelength range, based on the first received light intensity map. The resolution of the light pattern then depends on the number of positions scanned by the mirror. Optionally, when the pixelated light source 111 includes at least one second laser light source, the control unit 110 controls the mirror and the second laser light source to produce the light beam in the second wavelength range, based on the first received light intensity map (to project the same light pattern as in the first wavelength range) or based on a second received light intensity map to perform a lighting function.

[0148] Referring again to Figure 1a, the vehicle 100 further comprises at least one camera 102 including a sensor capable of acquiring a first image, or a series of first images, of the scene 10 in the first wavelength range. When the first wavelength range is infrared, the at least one camera 102 includes a camera comprising at least one infrared sensor, for example, a thermal camera. When the first wavelength range is visible, the at least one camera 102 includes a color camera, for example, an RGB (Red Green Blue) camera.According to advantageous embodiments, the camera 102 may comprise several sensors, including a first sensor capable of acquiring a first image, or a first series of images, in the first wavelength range, and a second sensor capable of acquiring a second image, or a second series of images, in the second wavelength range. When the pixelated light source is also capable of emitting in at least a third wavelength range, the camera 102 further comprises at least one other sensor dedicated to at least a third wavelength range.

[0149] Alternatively, the vehicle 100 comprises several cameras 102: a first camera 102 capable of acquiring a first image, or a first series of images, in the first wavelength range, and a second camera 102 capable of acquiring a second image, or a second series of images, in the second wavelength range. When the pixelated light source is also capable of emitting in at least a third wavelength range, the vehicle 100 further comprises at least one other camera 102 dedicated to at least a third wavelength range.

[0150] Furthermore, as described below, at least one camera is capable of acquiring an image, or a series of images, in a predetermined wavelength range, which may be identical to the first wavelength range, identical to the second wavelength range (or identical to the third wavelength range), or distinct from the first and second wavelength ranges.

[0151] According to the invention, when a light pattern is projected into several different wavelength ranges, at least one camera 102 is capable of acquiring several images of the same scene, one image corresponding to each wavelength range into which the light pattern is projected.

[0152] When the same light module is capable of projecting a light pattern in several wavelength ranges with the same matrix light source, and when the matrix light source is according to the second or third example, the alternation frequency between the first beam and the second beam is preferably greater than 100 Hz, for example equal to 700 Hz, so as to correspond to a period much shorter than the exposure time of at least one camera 102, which allows simultaneous capture of several images in the several wavelength ranges.

[0153] Vehicle 100 also includes at least:

[0154] - a first image processing module 104.1 configured to apply a first image processing function to at least one image acquired by at least one camera (at least in the first wavelength range) to obtain a first image processing result. The first image processing module 104.1 may, in particular, be capable of executing a first image processing model 106.1; and

[0155] - a second image processing module 104.2 configured to apply a second image processing function to at least one image acquired by at least one camera (at least in the first wavelength range), to obtain a second image processing result. The second image processing module 104.2 may, in particular, be capable of executing a second image processing model 106.2.

[0156] According to undescribed embodiments, the vehicle may further comprise at least one third image processing module configured to apply a third image processing function to at least one image acquired by at least one camera to obtain a third image processing result. Thus, the vehicle 100 more generally comprises at least two image processing modules.

[0157] No restrictions are attached to the first image processing function, nor to the second image processing function, nor optionally to the third image processing function and any other possible image processing functions. For example, each of the first and second image processing functions can be one of the following image processing functions:

[0158] - a function for determining depth information of the scene represented by a first image. Such depth information (which thus forms the result of image processing) can be used to confirm depth data acquired by another sensor on the vehicle, for example by a lidar. Such redundancy thus ensures robustness in estimating the distance to an object, and ensures greater safety in the execution of the driver assistance function based on depth information;

[0159] - a semantic segmentation function designed to segment each initial image captured by the camera into several pixel regions, each region being labeled with a class from a set of predefined classes, the respective regions and classes forming the output of the semantic segmentation function. In automotive applications, the following classes might be used: car, pedestrian, sign, road, etc. A semantic segmentation score can be associated with the semantic segmentation of the initial image, or with each segment determined within the initial image, the score representing the degree of certainty associated with the semantic segmentation; and / or

[0160] - an object detection function designed to detect one or more objects in the scene represented by each first image, to identify the category of each detected object from among several predefined categories, and to determine the position of each detected object (or a part of the first image in which the object is located). Each detected object is further associated with a detection score representing the certainty associated with the detection. The objects, positions, categories, and scores thus form the result of the object detection;

[0161] - an instance segmentation function aimed at segmenting each first image captured by the camera into several pixel regions for each object in the scene, as with semantic segmentation, but without labeling each region with a class from a set of predefined classes. An instance segmentation score can be associated with the instance segmentation of the first image, or with each segment determined in the first image, the score being representative of the degree of certainty associated with the instance segmentation;

[0162] - a panoptic segmentation function, which implements both semantic segmentation and instance segmentation, aiming to segment each first image captured by the camera into several pixel regions for each object in the scene, each object being labeled with a class from a set of predefined classes, the respective objects and classes thus forming the result of the instance segmentation. A panoptic segmentation score can be associated with the panoptic segmentation of the first image, or with each segment determined in the first image, the score being representative of the degree of certainty associated with the panoptic segmentation;

[0163] - an object tracking function, also called "tracking" in English, capable of determining a motion vector for at least one object identified in a first processed image, the motion vector of the object thus forming the result of the object tracking function.

[0164] Object detection, semantic segmentation, panoptic segmentation, and instance segmentation functions are all functions of semantic perception.

[0165] According to one embodiment of the invention, described in detail below by way of illustration, the first image processing function is a depth information determination function, and the second image processing function is a semantic perception function (for example, one of the semantic perception functions described above). In a variant not described below, the first and second image processing functions are different semantic perception functions. According to other variants not described, the vehicle further comprises at least one third image processing module configured to apply a third image processing function, for example, a semantic perception function.

[0166] According to the invention, the projection of a light pattern in the first wavelength range in the scene 10, by controlling the light module 101 by a light intensity map generated according to the invention from an image in the predetermined wavelength range, makes it possible to optimize the first result of the first image processing function and the second result of the second image processing function, particularly in night driving situations when the results of image processing functions are degraded.

[0167] Furthermore, for the first image processing function, which is a depth information determination function in the described embodiment, the projection of a light pattern eliminates the need for a two-camera system capable of acquiring images in the same wavelength range, according to the principle of stereovision, to measure the disparity of each object or pixel in scene 10 and to deduce depth information from the image. Indeed, according to a known stereovision method, two cameras, whose relative positions are predefined and known, can each acquire an image of the same scene in the same wavelength range. The offset between pixels corresponding to the same object is called disparity, and allows, geometrically, and from the known separation between the two cameras, the object's distance to be determined.This makes it possible to determine a depth map for a pair of images captured by the two cameras. The depth map indicates the distance of each pixel in the captured scene. Projecting a light pattern onto the scene eliminates the need for a two-camera system, as the distortions of the light pattern by the scene allow access to the depth information of each pixel in an image acquired by a single camera, as will be better understood from the description of Figures 2a and 2b below.

[0168] The system also includes a driver assistance module 109, also called ADAS module 109 (for “Advanced Driver-Assistance Systems”), capable of implementing at least one driver assistance function based on the first result of the first image processing function and the second result of the second image processing function. As mentioned previously, the ADAS module can receive information about the distance to objects (and therefore the depth of the scene), which constitute the first result of the first image processing function, redundantly from at least one other sensor on the vehicle, for example, from a lidar, thus improving the safety associated with the driver assistance function. The ADAS module can also receive the second result of the second semantic perception function.

[0169] The system according to the invention further comprises a control device 103 capable of obtaining an image in the predetermined wavelength range from at least one camera 102, and of applying to said image a generation model 108 to generate a light intensity map. The light intensity map indicates a light intensity for each light element (or for a set of light elements when the source comprises several sets emitting in different wavelength ranges) of the matrix source of the light module 101. Upon receiving a light intensity map, the control unit 110 can thus control the matrix source 111 so as to project a light beam according to a light pattern corresponding to the light intensity map, in the first wavelength range.In addition, in certain embodiments, the control unit 110 can control the matrix source 111 so as to project the light beam according to the same light pattern defined by the light intensity map, in the second wavelength range, and optionally in at least a third wavelength range.

[0170] As previously stated, the predetermined wavelength range can be the first wavelength range, the second wavelength range, or another wavelength range distinct from the first and second ranges. When the predetermined wavelength range is the other wavelength range, at least one camera 102 includes a sensor or camera capable of acquiring images in the other wavelength range.

[0171] The generation model 108 is thus capable of generating a light intensity map corresponding to a light pattern, from an image received as input in the predetermined wavelength range. The control device 103 can transmit the light intensity map to the control unit 110, for projection of a first pixelated light beam according to the light pattern in the first wavelength range, and optionally for projection of a second pixelated beam according to the same light pattern in the second wavelength range (and optionally in at least a third wavelength range).

[0172] Thus, the light intensity map is interpreted by the control unit 110 to control the light intensity of each light element of the matrix source 111, or for a set of light elements of the matrix source 111 when the matrix source comprises several sets emitting in different wavelength ranges. For example, the light intensity map includes a light intensity value for each light element of a set (when the light intensity map and the set of light elements of the matrix source 111 have the same resolution) or to control the light intensity of a subset of light elements of the set of light elements (when the matrix source 111 has a higher resolution than the light intensity map).

[0173] Figure 2a presents an example of a light beam 120 according to a light pattern corresponding to a light intensity map generated according to the invention, projected by a light module 101 as previously described at least in the first wavelength range.

[0174] In particular, Figure 2a represents a beam of light projected onto a screen equipped with an orthonormal coordinate system and positioned 25 meters from the projector. Therefore, Figure 2a does not represent the projection of the light pattern onto a real scene, which will be described with reference to Figure 2b.

[0175] According to the invention, the generation model 108 is capable of generating a light intensity map corresponding to a discontinuous light pattern, composed, for example, of dark and illuminated areas. The distortions induced by the scene 10 in the light pattern, particularly at the discontinuities that are the boundaries between dark and illuminated areas, allow the determination of depth information by the first image processing function, and further enhance the second semantic perception function. Advantageously, the light intensity map is generated from the image acquired by at least one camera 102 in the predetermined wavelength range, describing the scene facing the vehicle. This makes it possible to improve the overall performance associated with the first image processing function in relation to the second image processing function, as described below.

[0176] In the example in Figure 2a, the lighting pattern is a checkerboard pattern generated by the 108 generation module for a given scene, i.e., from a descriptive image of the scene facing the vehicle. The checkerboard pattern exhibits a regular alternation of dark areas 201 and illuminated areas 202, with a strong contrast between these areas.

[0177] Each zone 201 or 202 corresponds to a set of at least one pixel, and preferably to a plurality of pixels, for example several tens or hundreds of pixels.

[0178] Thus, a dark area is created by deactivating the pixels in the area, while a lit area is obtained by activating at least some of the pixels in the lit area. The activation or deactivation of pixels is implemented by the control unit 110, which is capable of controlling the pixelated light source 111. In the first example of a pixelated light source described previously, the control unit 110 creates the light pattern by applying a voltage to the light elements corresponding to the pixels of the lit areas 202, and applies no voltage to the light elements corresponding to the pixels of the dark areas 201, according to the light intensity map received from the control device 103.

[0179] A high-definition pixelated light source 111 thus enables the projection of a light pattern with a large number of dark and illuminated areas, allowing for high precision in the implementation of the first and second image processing functions. The checkerboard pattern shown in Figure 2a is provided for illustrative purposes. It comprises 9 columns and 8 rows of areas, each area containing a plurality of pixels. There are no restrictions on the shape of the areas, nor on the distribution of illuminated and dark areas.

[0180] Thus, according to the invention, if the scene facing the vehicle varies, the image submitted as input to the 108 generation model varies, which can induce a variation in the light intensity map at the output of the 108 generation model, such a variation improving the overall performance of the first and second image processing functions.

[0181] Figure 2b presents a first image 210 acquired by the camera 102 following the projection of a light beam according to a light pattern in the scene 10 facing the vehicle 100, according to embodiments of the invention.

[0182] The scene shown in figure 2b is intentionally simplified, with few objects, in order to simplify the understanding of the invention.

[0183] The scene in which the light pattern is projected includes another vehicle, as well as a vertical wall located behind the other vehicle. The scene could thus correspond to the interior of a parking lot, a situation given for illustrative purposes only. The invention can advantageously be implemented in an outdoor driving scene, particularly during a night scene.

[0184] As previously explained, the light pattern projected into the scene can be formed in the first wavelength range, and the first image 210 captured by the camera 102 is thus an image in the first wavelength range.

[0185] Thus, at least one 102 camera can acquire:

[0186] - in one embodiment, a single first image 210 of the scene in the first wavelength range during the projection of a light beam in the first wavelength range according to a light pattern corresponding to the generated light intensity map; or

[0187] - in a variant, the first image 210 described above, as well as a second image 210 of the scene in the second wavelength range, during the projection of a second light beam in the second wavelength range, according to the same light pattern as that projected in the first wavelength range.

[0188] In the above variant in which first and second images are acquired, the first and second images are acquired simultaneously (i.e., the exposure time periods of the camera 102 for each of its sensors, or of the cameras 102, overlap or are identical).

[0189] Other variants are possible according to the invention, including the acquisition by at least one camera 102 of at least a third image in the third wavelength range, when a third light beam projects the light pattern in the third wavelength range.

[0190] Thus, depending on the embodiment, the camera 102 can acquire as many images as there are wavelength ranges in which the light pattern is projected. In the examples described above, one or two wavelength ranges are used for projecting light patterns, but the invention also applies to strictly more than two wavelength ranges in which the light pattern is projected.

[0191] In what follows, for the sake of simplification, it is assumed that the first image 210 is in the first wavelength range and follows the projection of the light pattern in the first wavelength range by the two light modules 101.1 and 101.2, from a light intensity map received from the control device 103.

[0192] It can be observed in figure 2b:

[0193] - that the dark and lit areas 211 projected onto a substantially horizontal plane such as the ground have an elongated shape and are thus lengthened;

[0194] - that the dark and lit areas 212 projected onto a substantially vertical and close plane, such as the rear of the other vehicle, retain a square format and have a first given size;

[0195] - that the dark and lit areas 213 projected onto a substantially vertical plane and further away than the areas 212, such as the wall behind the other vehicle, also retain a square format but have a second size greater than the first size;

[0196] - that the dark and light areas 214 projected onto irregular objects, with curves, such as the top of the other vehicle, are strongly distorted.

[0197] Thus, the light pattern in Figures 2a and 2b can correspond to a light intensity map generated by the control device, particularly when the first image processing function is a depth information determination function. Indeed, projecting a light pattern onto a scene provides, through analysis of the deformation of the projected pattern and after image processing by the first image processing module 104.1, scene disparity information that allows for the estimation of scene depth information. Therefore, a light pattern comprising a regular repetition of contrasting sub-patterns, such as a repetition of square dark and light areas to form a checkerboard, is particularly well-suited for estimating scene depth information.The 108 generation model can adapt the characteristics of such a regular repetition (size and shape of dark and light areas for example) according to the image in the predetermined wavelength range received as input, and the parameters of the 108 generation model are trained to improve both the first result of the first depth information determination function (which can be a depth map), but also the second result of the second semantic perception function, compared to the acquisition of a first image without prior projection of the light pattern (and optionally other image processing results from other image processing functions).

[0198] Alternatively, particularly when the first and second image processing functions are both semantic perception functions (so no image processing to determine depth information is implemented), the generation model 108 may be able to generate a light intensity map corresponding to a light pattern without such repetition of contrasting sub-patterns, depending on the image in the predetermined wavelength range received as input.

[0199] Figure 3a shows a structure of a control device 103 according to embodiments of the invention.

[0200] The control device 103 includes an input interface 303 for receiving the image captured by at least one camera 102, in the predetermined wavelength range, the image being representative of the scene 10 facing the vehicle.

[0201] The control device 103 further includes a computing unit 301, such as a processor, configured to generate a light intensity map from the image received on the input interface 313, by application of the generation model 108.

[0202] The processor 301 is configured to communicate unidirectionally or bidirectionally, via one or more buses or via a direct wired connection, with a memory 302 such as Random Access Memory (RAM), Read Only Memory (ROM), or any other type of memory (Flash, EEPROM, etc.). Alternatively, the memory 302 may contain several of the aforementioned types. The memory can, for example, temporarily store the image received on the input interface 303.

[0203] The 302 memory can also permanently store an algorithm running the 108 generation model capable of receiving as input an image in the predetermined wavelength range, and providing as output a light intensity map.

[0204] Memory 302 can, for example, store instructions enabling the execution of the generation model 108, which can be defined by a set of parameters and capable of generating a light intensity map from at least one image representative of a scene facing the vehicle.

[0205] Memory 302 can also store a previous brightness map (the last generated brightness map), while the generation 108 model also takes the previous brightness map as input as described below.

[0206] According to embodiments of the invention, the generation model 108 is trained in a preliminary phase by machine learning, the machine learning enabling the training of the parameters defining the model, the model having a predefined structure. The model may, for example, have one of the following structures:

[0207] - a convolutional neural network, such as an ll-Net type network for example;

[0208] - an artificial neural network of the auto-encoder or variational auto-encoder type, also called VAE in English;

[0209] - a self-aware or transformative model; or

[0210] - any other model structure capable of receiving as input an image representing a scene, and of generating a light intensity map (or two light intensity maps).

[0211] There are no restrictions on the machine learning applied to the generation 108 model to optimize its parameters. For example, the machine learning can be supervised, as described later with reference to Figure 7, shown below.

[0212] The control device 103 further includes an output interface 304 capable of transmitting the generated light intensity map to the light module 101.

[0213] The control device 103 may further include another output interface 305 suitable for transmitting the generated light intensity map to at least one of the image processing modules 104.1 and 104.2.

[0214] Figure 3b shows a structure of an image processing module 104 according to embodiments of the invention.

[0215] The image processing module 104 can be the first image processing module 104.1 or the second image processing module 104.2 (or any other image processing module of the vehicle 100). The image processing module 104 includes an input interface 313 for receiving the first image, or the first and second images, captured by at least one camera 102 and representative of the projection of the light pattern in the scene 10 facing the vehicle 100, the pattern corresponding to the light intensity map generated by the control device 103.

[0216] The image processing module 104 further includes a computing unit 311, such as a processor, configured to determine an image processing result from at least one first image received on the input interface 313. The processor 311 can, for example, be a graphics processor, also called GPU for “Graphical Processing Unit”.

[0217] The processor 311 is configured to communicate unidirectionally or bidirectionally, via one or more buses or via a direct wired connection, with a memory 312 such as Random Access Memory (RAM), Read Only Memory (ROM), or any other type of memory (Flash, EEPROM, etc.). Alternatively, the memory 312 may contain several of the aforementioned types. The memory can, for example, temporarily store the first image, or the first and second images, received on the input interface 313.

[0218] Memory 312 can also permanently store an algorithm running image processing model 106.1 or 106.2 capable of receiving as input a first image in the first wavelength domain, or first and second images in the first and second wavelength domains, and providing as output an image processing result (which can be depth information, or a semantic perception result, as previously described).

[0219] Memory 312 can, for example, store instructions for executing image processing model 106.1 or 106.2, which can be defined by a set of parameters optimized to generate an image processing result from at least one first image representing the projection of the light pattern onto the scene. In some embodiments, the first image processing model 106.1 is capable of determining depth information from at least one first image representing the projection of the light pattern onto the scene, as well as from the light intensity map corresponding to the projected pattern. To this end, when the control device 103 transmits the light intensity map to the light module 101, the control device 103 also transmits the light intensity map to the first image processing module 104.1.

[0220] According to embodiments of the invention, the image processing model 106.1 or 106.2 is trained in a preliminary phase by machine learning, the machine learning enabling the training of the parameters defining the model, the model having a predefined structure. The image processing model 106.1 or 106.2 may, for example, have one of the following structures:

[0221] - a convolutional neural network, such as an ll-Net type network for example;

[0222] - an artificial neural network of the auto-encoder or variational auto-encoder type, also called VAE in English;

[0223] - a self-aware or transformative model; or

[0224] - any other model structure capable of receiving as input a first image, or first and second images, representing the projection of a light pattern in the scene, and optionally the light intensity map (or light intensity maps), and of determining an image processing result.

[0225] There are no restrictions on the machine learning applied to the model to optimize its parameters. The machine learning can, for example, be supervised.

[0226] Alternatively, the image processing model 106.1 or 106.2 is not derived from machine learning, but is a software comprising a set of rules capable of determining an image processing result by applying the set of rules to at least a first image representing the projection of the light pattern in a scene, and optionally from the light intensity map corresponding to the projected light pattern.

[0227] The image processing module 104 can apply pre-processing to each first received image (or first and second received images) before applying the image processing model 106.1 or 106.2. The pre-processing can consist of cropping the first received image to retain only the part of the first received image corresponding to the projection of the light pattern in the scene 10. Other pre-processings can be applied according to the invention.

[0228] The image processing module 104 further includes an output interface 314 capable of transmitting the image processing result to a separate module. For example, the separate module to which the image processing result is transmitted could be the ADAS module 109 described previously.

[0229] The image processing module 104 may further include another input interface 315 suitable for receiving the light intensity map generated by the control device 103, as described previously.

[0230] Figure 4a illustrates a pixelated light source 111 according to the first example above, capable of forming two light beams in two distinct wavelength ranges.

[0231] The pixelated light source 111 according to the first example includes:

[0232] - a first set of 401 light elements, individually controllable, capable of emitting light in the first range of wavelengths;

[0233] - a second set of 402 light elements, individually controllable, capable of emitting in the second wavelength range.

[0234] In the example in Figure 4a, the first wavelength range is an infrared range and the second wavelength range is a visible range: for this purpose, the second set of light elements 402 includes subsets of blue, red and green light elements, which makes it possible to project a colored light beam into scene 10 facing the vehicle:

[0235] - to project the same light pattern as the first set of light elements according to the light intensity map generated by the control device 103; or

[0236] - to perform a lighting or signaling function, as previously described.

[0237] Alternatively, each element of the second 402 set can be capable of emitting white light.

[0238] Thus, the light elements of the first set 401 can be controlled to form a first light beam according to the light pattern corresponding to the received light intensity map, with a resolution depending on the number of light elements in the first set 401.

[0239] Simultaneously, the light elements of the second set 402 can be controlled to form a second light beam according to the light pattern corresponding to the light intensity map or performing the lighting or signaling function.

[0240] According to undescribed variants, at least a third set of light elements can be controlled to form a third light beam according to the light pattern corresponding to the light intensity map, in the third wavelength range.

[0241] Figure 4b illustrates a pixelated light source 111 according to the second example, comprising two light sources 411.1 and 411.2 and a DMD micromirror array.

[0242] The first light source 411.1 is capable of emitting light rays in the first wavelength range. The second light source 411.2 is capable of emitting light rays in the second wavelength range. According to undescribed variants, at least a third light source is capable of emitting light rays in the third wavelength range.

[0243] The light sources 411.1 and 411.2 can be arranged on a single support 412, as shown in Figure 4b, or on two separate supports. Only one micromirror 410 is shown in Figure 4b for clarity. However, in practice, the array comprises a large number of micromirrors, in particular more than 100, or even more than 1000 or more than 10,000 micromirrors individually controllable by the control unit 110 described previously.

[0244] Each micromirror 410 can be controlled to switch between at least one first position 420.1 and a second position 420.2. In the first position 420.1, the micromirror is arranged to reflect a light beam from a source, for example a light beam 413 emitted by the first light source 411.1, towards the projection system 112 so as to contribute to the formation of a light beam outside the vehicle 100. Thus, by controlling the positions of the micromirrors of the matrix by the control unit 110, the light module projects a first light beam according to the light pattern corresponding to the received light intensity map, in the first wavelength range, when the first light source 411.1 is activated.Similarly, by controlling the positions of the micro-mirrors of the matrix by the control unit 110, the light module projects a second light beam according to the same pattern corresponding to the first light intensity map received or to perform a lighting or signaling function (for the visible range).

[0245] The control unit 110 can thus control the same micromirror array, alternately at a given frequency, to project at least the first and second beams alternately. The switching frequency between the at least two beams can be greater than 700 Hz, so that the flicker of the second beam is not perceptible to the human eye and therefore does not disturb the driver and other road users, and is not perceptible to the camera 102 or cameras 102, having an exposure time greater than the period associated with the switching frequency. Figure 5 is a diagram illustrating the steps of a typical phase of an image processing method representing a scene facing a vehicle, according to embodiments of the invention.

[0246] At a stage 500, at least one camera 102 obtains a representative image of the scene facing the vehicle in the predetermined wavelength range, which can be, as explained previously, the first wavelength range, the second wavelength range, or some other wavelength range distinct from the first and second wavelength ranges.

[0247] At a step 501, the control device 103 generates a light intensity map from the image obtained at step 500, by applying the generation model 108. The generation model can further take as input the previous light intensity map, which is the light intensity map generated during a previous iteration of step 501.

[0248] At step 502, the control device 103 transmits the generated light intensity map to the light module 101 (or to the light modules 101.1 and 101.2). Furthermore, the generated light intensity map can be transmitted to at least one of the first and second image processing modules 104.1 and 104.2, for example, to the first image processing module 104.1 capable of determining depth information.

[0249] At step 503, the control unit 110 controls the matrix source 111 according to the received light intensity map, to project:

[0250] - a first beam of light according to the light pattern corresponding to the light intensity map in the first wavelength range; or

[0251] - a first light beam according to the light pattern corresponding to the light intensity map in the first wavelength range and a second light beam according to the light pattern corresponding to the light intensity map in the second wavelength range (and optionally in at least the third wavelength range); or

[0252] - a first light beam according to the light pattern corresponding to the first light intensity map in the first wavelength range and a second light beam according to an undescribed lighting or signaling instruction.

[0253] Following step 503, the process can return to step 500 to obtain a new image in the predetermined wavelength range, which allows continuous and real-time adaptation of the projected light beam, or projected light beams.

[0254] Furthermore, following step 503, at least one camera 102 obtains, at a step 504, at least one first image representing the projection of the light pattern in the scene, in the first wavelength range. Optionally, at least one camera also obtains a second image representing the projection of the same light pattern (corresponding to the light intensity map) in the scene in the second wavelength range (and optionally in at least the third wavelength range).

[0255] Note that when the predetermined wavelength range is the first wavelength range, obtaining the new image in the predetermined wavelength range, at the new step 500, and obtaining the first image in the first wavelength range at step 504 are one and the same step.

[0256] Similarly, when the predetermined wavelength range is the second wavelength range, obtaining the new image in the predetermined wavelength range at the new step 500, and obtaining the second image in the second wavelength range at step 504 are one and the same step.

[0257] At a step 505, the first processing module 104.1 determines the first image processing result (the depth map in the described embodiment), by applying the first image processing model 106.1, to the first image in the first wavelength range obtained at step 504, and optionally to the second image in the second wavelength range obtained at step 504, and optionally to the light intensity map generated at step 501 and transmitted by the control device 103.

[0258] At a step 506, implemented independently of step 505, the second processing module 104.2 determines the second image processing result (the semantic perception result in the described embodiment), by applying the second image processing model 106.2, to the first image in the first wavelength domain obtained in step 504, and optionally to the second image in the second wavelength domain obtained in step 504, and optionally to the light intensity map generated in step 501 and transmitted by the control device 103.

[0259] At a step 507, the first image processing module 104.1 can transmit the first image processing result to at least one other module, for example to the ADAS module 109. The ADAS module can thus implement at least one driver assistance function based on the first image processing result.

[0260] At step 508, the second image processing module 104.2 can transmit the second image processing result to at least one other module, for example to the ADAS module 109. The ADAS module can thus implement at least one driver assistance function based on the second image processing result.

[0261] According to some embodiments, the ADAS 109 module implements at least one driver assistance function based on the first image processing result and the second image processing result.

[0262] As described previously, at least a third image processing module can determine a third image processing result by applying the third image processing function to the first image, and optionally to the second image, the third result being able to be transmitted to another module such as the ADAS 109 module. Figure 6 is a 600 drive system of the generation 108 model of light intensity card, according to embodiments of the invention.

[0263] Such a drive system 600 is external to the vehicle 100 and is capable of implementing the drive phase of the process according to the invention, which is described later with reference to Figure 7. The drive phase precedes the current phase described with reference to Figure 5, which takes place during a driving situation of the vehicle 100. The drive phase may, in particular, be part of the design and manufacturing process of the control device 103, before its integration into the vehicle 100.

[0264] The 600 training system includes a 601 training database, capable of storing training data associations, each training data association comprising:

[0265] - a representative training image of a scene in the predetermined wavelength range;

[0266] - ground truth which includes first reference data corresponding to a first result of reference image processing (reference depth information in the described embodiment) and second reference data corresponding to a second result of reference image processing (a semantic perception result in the described embodiment).

[0267] Preferably, the associations stored in the training database are varied, that is to say, they were obtained in scenes varying according to several criteria: scene composition, scene brightness level, weather conditions, etc.

[0268] In addition, the 601 training database includes more than one hundred training data associations, and preferably several thousand or even tens of thousands of training data associations.

[0269] No restrictions are attached to the manner in which the 601 training database was constructed. The 601 training database can be obtained from images captured by a fleet of vehicles in real-world night driving situations (preferably varied according to several criteria as described previously), with one or more sensors, or a human operator, capable of determining the first and second reference image processing results. For example, a depth camera and / or lidar is capable of obtaining reference depth information (first reference data in the considered embodiment), and a human operator, or complementary sensors, determines a semantic perception result (second reference data in the considered embodiment) for each training image captured in the predetermined wavelength range.

[0270] The training system 600 further includes the generation model 108 of the light intensity map to be trained. The generation model 108 of the light intensity map is structurally capable of generating a light intensity map of a given resolution, determined from the matrix source 111 with which the control device 103 is associated in the vehicle 100 during the current phase, from a representative image of a scene in the predetermined wavelength range, an image which may be a training image during the training phase, and which is an image captured in real time by at least one camera 102 during step 500 of the current phase.

[0271] For this purpose, as described previously, the generation 108 light intensity map model can have one of the following structures:

[0272] - a convolutional neural network, such as a U-Net type network for example;

[0273] - an artificial neural network of the auto-encoder or variational auto-encoder type, also called VAE in English;

[0274] - a generating network of a system of generative adversarial networks, also called GAN, for “Generative Adversarial Networks” in English;

[0275] - a self-aware or transformative model; or

[0276] - any other generative model structure capable of receiving as input an image representing a scene, and of producing as output a light intensity map in a format capable of controlling the light module described above.

[0277] The 600 training system further includes a 602 synthesis module, capable of producing at least one representative synthetic image of the scene in the training image, illuminated by a light pattern corresponding to the light intensity map (called the training light intensity map during the training phase) produced as output by the generation model 108, in the first wavelength range and optionally in the second wavelength range. Thus, the 602 synthesis module allows for the simulation of a camera capturing the scene in the training image, onto which a light pattern corresponding to the training light intensity map output by the generation model 108 would be projected, in the first wavelength range and optionally in the second wavelength range. For this purpose, the synthesis module receives the following input:

[0278] - the training image;

[0279] - the training light intensity map generated by the generation 108 model from the training image.

[0280] According to embodiments, the synthesis module 602 can generate a first synthetic image representative of the scene of the training image, illuminated by a light pattern corresponding to a training light intensity map produced at the output of the generation model 108, in the first wavelength range, and a second synthetic image representative of the same scene illuminated by the light pattern corresponding to the training light intensity map, in the second wavelength range.

[0281] The training system 600 further comprises at least one first training image processing module 603.1 implementing a first training image processing function and a second training image processing module 603.2 implementing a second training image processing function. In the described embodiment, the first training image processing module 603.1 is capable of determining depth information forming a first training image processing result, and the second training image processing module 603.2 is capable of determining a semantic perception result forming a second training image processing result.

[0282] The first drive image processing module 603.1 is preferably identical to the first image processing module 104.1 integrated into the vehicle 100 during the current phase of the process according to the invention. Similarly, the second drive image processing module 603.2 is preferably identical to the second image processing module 104.2 integrated into the vehicle 100 during the current phase of the process according to the invention. The generation module 108 is thus specifically driven to improve the performance of at least the first and second image processing functions implemented in the vehicle 100 during the current phase (and optionally at least a third image processing function).

[0283] Thus, the first training image processing module 603.1 can implement the first image processing model 106.1, and the training image processing module 603.2 can implement the second image processing model 106.2. Image processing models 106.1 and 106.2 may have been obtained, for example, through machine learning during dedicated training phases not described herein. Alternatively, at least one image processing model 106.1 and 106.2 may implement a software function not derived from machine learning.

[0284] Alternatively, at least one of the first and second image processing models 106.1 and 106.2 is trained at the same time as the generation model 108, for example by supervised learning as described with reference to Figure 7.

[0285] The first 603.1 training image processing module is capable of receiving:

[0286] - at least one synthetic image from the synthesis module 602; and - optionally, the training light intensity map generated by the generation model 108 after receiving the training image.

[0287] Based on the information received, the first training image processing module 603.1 is able to generate a first training image processing result.

[0288] In addition, the second 603.2 training image processing module is capable of receiving:

[0289] - at least one computer-generated image from the 602 synthesis module; and

[0290] - optionally, the training light intensity map generated by the generation 108 model after receiving the training image.

[0291] Based on the information received, the second training image processing module 603.2 is capable of generating a second training image processing result.

[0292] The 600 drive system also includes at least:

[0293] - a first loss assessment module 604.1 capable of assessing a first loss by comparison between the first ground truth reference data from the training database 601 and the first training image processing result from the first training image processing module 603.1 (the training depth information in the described embodiment); and

[0294] - a second loss assessment module 604.2 capable of assessing a second loss by comparison between the second ground truth reference data from the training database 601 and the second training image processing result from the second training image processing module 603.2 (the semantic perception result in the described embodiment).

[0295] The 600 system also includes a 605 fusion module capable of:

[0296] - to determine an overall loss from at least the first loss from the first loss evaluation module 604.1 and the second loss from the second loss evaluation module 604.2. There are no restrictions on the function applied to obtain the overall loss from at least the first and second losses. It may, for example, be an average, a sum, or any other function capable of taking two input values; and - to modify one or more parameters of the generation model 108 based on the evaluated overall loss, and according to a predefined training strategy to reduce the overall loss when steps 700 to 707 of the training phase are iterated for multiple training images.According to embodiments in which at least one of the first and second image processing models is trained at the same time as the generation model, the fusion module is further capable of modifying one or more parameters of this image processing model as a function of the overall loss evaluated.

[0297] No restrictions are attached to the loss function used to determine the first loss or the loss function used to determine the second loss. The loss functions can be based on a distance according to a given standard.

[0298] Such a 600 training system is thus capable of implementing supervised learning of the 108 generation model.

[0299] Alternatively, the 108 generation light intensity map model can be derived from a machine learning method other than supervised learning. For example, the 108 generation light intensity map model can be optimized by reinforcement learning, with a reward determined based on at least one first loss evaluated from the first training image processing result from the first training image processing module 603.1 and a second loss evaluated from a second training image processing result from the second training image processing module 603.2.

[0300] Thus, in general, the 108 generation light intensity map model can be trained by machine learning in such a way as to improve the overall performance (to decrease the respective losses) of at least the first image processing function (depth information determination function) and the second image processing function (semantic perception function), by projecting a light pattern corresponding to a light intensity map generated by the 108 generation model.

[0301] Figure 7 is a diagram illustrating the steps of a training phase of a generation model 108 of a light intensity map, in an image processing method in a vehicle, according to embodiments of the invention.

[0302] As previously explained, the training phase of the generation 108 light intensity map model can be implemented in the 600 drive system described with reference to Figure 6.

[0303] At a step 700, the training database 601 obtains a training data association as previously described, comprising a training image of a scene facing a vehicle, a first reference image processing result and a second reference image processing result.

[0304] At a step 701, the training image obtained is submitted as input to the generation model 108 which generates a training light intensity map as a function of the training image obtained.

[0305] The training light intensity map is transmitted from the generation model 108 to the synthesis module 602, which also receives the training image from the training database 601. The generation model 108 can further transmit the generated training light intensity map to the first training image processing module 603.1 and / or the second training image processing module 603.2.

[0306] At step 702, the synthesis module 602 determines at least one synthetic image representative of the scene in the training image, illuminated by the light pattern corresponding to the light intensity map produced as output by the generation model 108, in the first wavelength range. As described previously, the synthesis module can produce a first synthetic image (for the projection of the light pattern in the first wavelength range) and a second synthetic image (for the projection of the same pattern in the second wavelength range), and optionally at least a third synthetic image (for the projection of the same pattern in the third wavelength range).

[0307] At least one synthetic image is passed to the first training image processing module 603.1, which generates the first training image processing result in a step 703.1.

[0308] At least one synthetic image is also passed to the second training image processing module 603.2, which generates the second training image processing result in a step 703.2.

[0309] The first training image processing result generated in step 703.1 is passed from the first training image processing module 603.1 to the first loss evaluation module 604.1. At a step 704.1, the first loss evaluation module 604.1 evaluates a first loss by comparison between the first training image processing result and the first reference image processing result.

[0310] The second training image processing result generated in step 703.2 is passed from the second training image processing module 603.2 to the second loss evaluation module 604.2. At a step 704.2, the second loss evaluation module 604.2 evaluates a second loss by comparison between the second training image processing result and the second reference image processing result.

[0311] Optionally, at least a third loss can be evaluated when the generation module 108 is trained to optimize the third image processing result of at least a third image processing module.

[0312] At step 705, the fusion module 605 determines an overall loss based on at least the first and second losses. At step 706, the fusion module 605 can further determine whether a predefined convergence criterion is met, based in particular on the overall loss evaluated at step 705, and optionally on losses evaluated during previous iterations of steps 700 to 705.

[0313] If the convergence criterion is not met, the fusion module 605 modifies at least one parameter of the generation model 108 at a step 707, depending on the evaluated loss, according to the predefined training strategy, and may modify one of the image processing models 106.1 and 106.2, when this image processing model is trained concurrently with the generation model. Such parameter optimization training strategies during supervised learning are well known and are not described further herein.

[0314] Following step 707, the process training phase returns to step 700 to repeat steps 700 to 705 based on a new association of training data from training database 601.

[0315] The 108 generation model can thus be trained until convergence.

[0316] At step 708, when the fusion module 605 determines that the convergence criterion is met, the training phase is completed and the generation model 108 can be implemented in the control device 103 for implementation of step 501 of the current phase of the process according to the invention.

[0317] According to embodiments in which the generation model 108 is further capable of taking as input the previous light intensity map, the supervised learning described with reference to Figure 7 can be adapted in one of the following ways:

[0318] - in step 700, a random light intensity map is generated, and the synthesis module 602 simulates the projection of a light pattern corresponding to the random light intensity map onto the training image to obtain a synthetic image, and the synthetic image as well as the random light intensity map are submitted as input to the generation model in step 701; or - the associations of the training base 601 comprise training images forming a sequence, the training images thus being submitted sequentially during the iterations of steps 700 to 707; or

[0319] - during step 700, the synthesis module 602 simulates the projection of a light pattern corresponding to the previous light intensity map into the newly obtained training image to obtain a synthesis image, and the synthesis image as well as the previous light intensity map are submitted as input to the generation model during step 701.

[0320] Figure 8a illustrates an image processing system of a scene 10 facing a vehicle 100N, according to embodiments of the invention.

[0321] The vehicle 100N includes a right front headlight 105.1 N and a left front headlight 105.2N. The right front headlight 105.1 N includes at least one first light module 101.1 N right and the left front headlight includes at least one light module 101.2N.

[0322] According to the invention, the light modules 101.1 N and 101.2 N are capable of projecting at least a first light beam producing a first light pattern in the scene 10N, in at least a first range of wavelengths, and a second light beam producing a second light pattern in a second range of wavelengths.

[0323] More generally, the 101.1N and 101.2N light modules are capable of projecting N light beams forming N light patterns in M ​​wavelength ranges, each light pattern among the N light patterns being projected into at least one wavelength range, and each wavelength range containing one projected pattern among the N light patterns. Thus, M is an integer greater than or equal to N.

[0324] In what follows, in order to simplify the description of the invention, it is considered, without limitation, that M and N are both equal to 2.

[0325] Each of the first and second wavelength ranges can be a range of visible wavelengths or a range containing no visible wavelengths. The two light modules 101.1 N and 101.2 N can be capable of projecting light beams that combine into a light pattern within a given wavelength range, and this for several wavelength ranges, by superimposing the respective beams.

[0326] A straight 120.1 N beam of light is projected by the straight 101.1 N light module. The straight 120.1 N beam of light may comprise several light beams in several distinct wavelength ranges, including the first and / or second wavelength ranges, as will be better understood from the following. Alternatively, a left-handed 120.2 N beam of light is projected by the left-hand 101.2 N light module. The left-handed 120.2 N beam of light may comprise several light beams in several distinct wavelength ranges, including the first and / or second wavelength ranges, as will be better understood from the following.

[0327] No restrictions are attached to the first wavelength range according to the invention, which may be a visible range, or an infrared or ultraviolet range, for example. By way of example, the first wavelength range may be within a near-infrared (NIR) range, a short-wave infrared (SWIR) range, a medium-wave infrared (MWIR) range, or a long-wave infrared (LWIR) range. Note that the MWIR and LWIR ranges are also called the thermal range.

[0328] Similarly, no restrictions are attached to the second wavelength range according to the invention, which may be a visible range, or an infrared or ultraviolet range, for example, but which is distinct from the first wavelength range. According to a first embodiment, the light module 101.1N or 101.2N can project, alternately at a given frequency, or simultaneously, the first light beam according to the first light pattern in the first wavelength range and the second light beam according to the second light pattern in the second wavelength range (and optionally other light beams in other wavelength ranges, according to the first light pattern, according to the second light pattern, or according to other light patterns). When several light modules 101.1N and 101.2N are capable of projecting in the same wavelength ranges, the light modules 101.1N and 101.2N can project light beams that overlap and combine to create a light pattern for each wavelength range, specifically a first light pattern in the first wavelength range and a second light pattern in the second wavelength range. According to another example of the first embodiment, the vehicle comprises a single light module 101.1N or 101.2N capable of projecting several light beams according to several light patterns.

[0329] In a second alternative embodiment, the light module 101.1 N projects at least the first light beam according to the first light pattern in the first wavelength range and the light module 101.2N projects at least the second light beam according to the second light pattern in the second wavelength range, each of the first and second light modules 101.1 N and 101.2N thus illuminating in distinct wavelength ranges.

[0330] Thus, more generally, at least one 101.1 N and / or 101.2 N light module can be capable of projecting during the same time interval (simultaneously or by alternating beams at a frequency corresponding to a period smaller than the time interval):

[0331] - the first light beam according to a first light pattern corresponding to a first light intensity map generated according to the invention, in the first wavelength range; and

[0332] - the second light beam according to a second light pattern corresponding to a second light intensity map generated according to the invention, in the second wavelength range.

[0333] In addition, at least one of the 101.1 N and / or 101.2 N light modules can project a beam of lighting and / or signaling in the visible range to perform a lighting or signaling function.

[0334] Note that when the first and second wavelength ranges do not include any visible wavelengths, it is avoided to disturb the driver while driving the vehicle when determining the image processing results described below.

[0335] According to the invention, the light beams projected by the light module 101.1 N and / or 101.2N are pixelated, which allows the creation of light patterns with a resolution depending on the number of pixels allowed by a pixelated light source of the light module 101.1 N and / or 101.2N.

[0336] Figure 8b illustrates the structure of a light module 101 N with a pixelated light source 111 N, for example matrix-based, according to embodiments of the invention. The light module 101 N can be the front right module 101.1 N and / or the front left module 101.2 N described previously.

[0337] The 101 N light module includes:

[0338] - a control unit 110 of the pixelated light source 111 N;

[0339] - the pixelated light source 111 N;

[0340] - a projection optic for the light coming from the pixelated light source

[0341] 111 N to produce a pixelated light beam projected in front of the vehicle towards the scene 10N. No restrictions are attached to the projection optics, which can include any set of optical elements.

[0342] No restrictions are placed on the number of pixels of the pixelated light source 111 N. Preferably, the pixelated light source 111 N is a high-definition light source, that is, one capable of projecting a light beam comprising more than one hundred pixels, preferably more than 1000 pixels. The pixelated light source 111 N can also project a light beam comprising more than 10,000 pixels according to embodiments of the invention.

[0343] Furthermore, there are no restrictions attached to the technology associated with the 111 N pixelated light source, which can be:

[0344] - according to a first example, a matrix of individually controllable light elements, of which at least one set of light elements is capable of emitting in the first wavelength range or in the second wavelength range. According to the first embodiment of the invention, the same matrix of light elements may comprise a first set of light elements capable of emitting in the first wavelength range and a second set of light elements capable of emitting in a second wavelength range;

[0345] - according to a second example, at least one light source capable of emitting in the first wavelength range or in the second wavelength range and a micromirror array, also called DMD for Digital Micromirror Devices, individually activatable to reflect the light from at least one light source towards the projection optics 112N. According to the first embodiment, the pixelated light source 111N may comprise a first light source capable of emitting in the first wavelength range and a second light source capable of emitting in the second wavelength range, as described later. In this case, the micromirror array is alternately controlled, at a given frequency, to produce the first beam when the first light source is activated, and to produce the second beam when the second light source is activated;

[0346] - According to a third example, at least one laser light source capable of emitting in the first or second wavelength range, and a controllable mirror for scanning a predetermined set of positions. Such technology is called laser scanning. The control unit 110N is capable of synchronously controlling the movement of the mirror and the activation of at least one laser light source. According to the first embodiment, the pixelated light source 111N may comprise a first laser light source capable of emitting in the first wavelength range and a second laser light source capable of emitting in the second wavelength range, as described later.

[0347] It should be noted that the pixelated light source 111 N, according to each of the three examples above, can also emit light in at least one other wavelength range, in addition to the first and second wavelength ranges. Thus, more generally, the vehicle 100N includes at least one light module capable of emitting at least two light beams in two distinct wavelength ranges.

[0348] The three examples listed above have the advantage of enabling the creation of light patterns in pixelated beams with high resolution, from light intensity maps generated and transmitted by a 103N control device described below.

[0349] Moreover, these three examples of pixelated light sources 111 N enable the first embodiment in which a single light module 101.1 N or 101.2 N is capable of projecting at least two light beams in at least two distinct wavelength ranges. In what follows, the first embodiment is considered, with a single light module 101.1 N or 101.2 N capable of projecting several light beams in several wavelength ranges, the first light beam following the first light pattern and the second light beam following the second light pattern.

[0350] In the first example of a pixelated light source 111 N described above, the light elements can be electroluminescent elements, individually controlled by a voltage applied to the terminals of each light element by the control unit 110 N. Each electroluminescent light element can be mounted on its own substrate. Alternatively, the electroluminescent light elements can be on a single substrate, in which case the pixelated light source is called monolithic. A so-called "monolithic" source can have a particularly high density of light elements, making it especially attractive for a wide range of applications. A monolithic source involves a plurality of submillimeter-sized electroluminescent semiconductor elements epitaxially mounted directly onto a common substrate, which is usually silicon.Unlike conventional LED arrays, where each individual light-emitting element is a separate electronic component mounted on a substrate such as a printed circuit board (PCB), a monolithic LED source is considered a single electronic component. During its production, multiple areas of light-emitting semiconductor junctions are generated on a common substrate, forming an array. This manufacturing technique allows for the creation of closely spaced areas of light-emitting elements, each acting as a single light-emitting element. The gaps between these elements can be submillimeter in size. One advantage of this manufacturing technique is the high pixel density that can be achieved on a single substrate.

[0351] The first example of a pixelated light source, 111 N, has the advantage, when the pixelated light source 111 N comprises a first set of light elements and a second set of light elements according to the first embodiment, of projecting two light beams simultaneously in the first wavelength range and in the second wavelength range, to project a first light pattern corresponding to the first received light intensity map and a second light pattern corresponding to the second received light intensity map. Indeed, the two sets of light elements can be controlled separately by the control unit 110N.

[0352] In the second example of a pixelated light source 111N described above, the control unit 110N controls the micromirrors of the array to produce the first light beam containing the first light pattern in the first wavelength range from a first received light intensity map, when the first light source is active. The resolution of the light pattern then depends on the number of micromirrors in the array. Furthermore, in the first embodiment described, the pixelated light source 111N includes at least a second light source (according to the first embodiment), and when the second light source is active, the control unit 110N controls the micromirrors of the array to produce the second light beam in the second wavelength range, based on the second received light intensity map.

[0353] In the third example of a pixelated light source 111N described above, the control unit 110N controls the mirror and the first laser light source to scan the predetermined set of positions and produce the first light beam comprising the first light pattern in the first wavelength range, based on the first received light intensity map. The resolution of the light pattern then depends on the number of positions scanned by the mirror. In the first embodiment described, the pixelated light source 111N includes at least a second laser light source; the control unit 110N controls the mirror and the second laser light source to produce the second light beam in the second wavelength range, based on the second received light intensity map.

[0354] Referring again to Figure 8a, the vehicle 100N further comprises at least one camera 102N including a sensor capable of acquiring at least one first image, or a series of first images, of the scene 10N in the first wavelength range and a second image, or a series of second images, of the scene 10N in the second wavelength range. Where the first or second wavelength range is infrared, the at least one camera 102N comprises a camera including at least one infrared sensor, for example, a thermal camera. Where the first or second wavelength range is visible, the at least one camera 102N comprises a color camera, for example, an RGB (Red Green Blue) camera.According to advantageous embodiments, the 102N camera may include several sensors, including a first sensor capable of acquiring the first image, or series of first images, in the first wavelength range, and a second sensor capable of acquiring the second image, or series of second images, in the second wavelength range. When the pixelated light source of at least one light module is also capable of emitting a third light beam in at least a third wavelength range (to produce the first pattern, the second pattern, or a third light pattern different from the first and second patterns), the 102N camera further includes at least one other sensor dedicated to at least a third wavelength range.

[0355] Alternatively, the 100N vehicle comprises several 102N cameras: at least one first 102N camera capable of acquiring the first image, or series of first images, in the first wavelength range, and a second 102N camera capable of acquiring the second image, or series of second images, in the second wavelength range. When the pixelated light source is also capable of emitting in at least a third wavelength range (to produce the first pattern, the second pattern, or a third light pattern different from the first and second patterns), the 100N vehicle further comprises at least one third 102N camera dedicated to at least a third wavelength range.

[0356] Furthermore, as described below, at least one 102N camera is capable of acquiring an image, or a series of images, in a predetermined wavelength range, which may be identical to the first wavelength range, identical to the second wavelength range (or identical to the third wavelength range), or distinct from the first and second wavelength ranges.

[0357] Thus, according to the invention, at least one 102N camera is capable of acquiring several images of the same scene, each image corresponding to a range of wavelengths in which one of the light patterns corresponding to one of the light intensity maps is projected.

[0358] According to the first embodiment, a single light module is capable of projecting several light patterns in several wavelength ranges with the same matrix light source. When the matrix light source is according to the second or third example, the alternation frequency between the first and second beams is preferably greater than 100 Hz, for example equal to 700 Hz, so as to correspond to a period much shorter than the exposure time of at least one 102N camera, which allows simultaneous capture of several images in the several wavelength ranges.

[0359] The 100N vehicle also includes at least:

[0360] - a first image processing module 104.1 N configured to apply a first image processing function to at least one first and / or second image acquired by at least one camera 102N (the first image, the second image, or the first and second images) to obtain a first image processing result. The first image processing module 104.1 N may, in particular, be capable of executing a first image processing model 106.1 N; and

[0361] - a second 104.2N image processing module configured to apply a second image processing function to at least one first and / or second image acquired by at least one 102N camera (the first image, the second image, or the first and second images), to obtain a second image processing result. The second 104.2N image processing module may, in particular, be capable of executing a second 106.2N image processing model.

[0362] According to undescribed embodiments, the vehicle may further comprise at least one third image processing module configured to apply a third image processing function to at least one first and / or second image acquired by at least one camera (the first image, the second image, or the first and second images) to obtain a third image processing result. Thus, the 100N vehicle more generally comprises at least two image processing modules.

[0363] No restrictions are attached to the first image processing function, nor to the second image processing function, nor optionally to the third image processing function and any other possible image processing functions. For example, each of the first and second image processing functions can be one of the following image processing functions:

[0364] - a function for determining depth information of the scene represented by at least one first and / or second image received as input to the function. Such depth information (which thus forms the result of image processing) can be used to confirm depth data acquired by another sensor on the vehicle, for example by a lidar. Such redundancy thus ensures robustness in estimating the distance to an object, and ensures greater safety in the execution of the driver assistance function based on depth information;

[0365] - a semantic segmentation function designed to segment the first and / or second image received as input into several pixel regions, each region being labeled with a class from a set of predefined classes, the respective regions and classes forming the output of the semantic segmentation function. In automotive applications, the following classes might be used: car, pedestrian, sign, road, etc. A semantic segmentation score can be associated with the semantic segmentation of the first and / or second image, or with each segment determined within the first and / or second image, the score being representative of the degree of certainty associated with the semantic segmentation; and / or

[0366] - an object detection function designed to detect one or more objects in the scene represented by the first and / or second image, to identify the category of each detected object from among several predefined categories, and to determine the position of each detected object (or a part of the first and / or second image in which the object is located). Each detected object is further associated with a detection score representing the certainty associated with the detection. The objects, positions, categories, and scores thus form the result of the object detection;

[0367] - an instance segmentation function aimed at segmenting the first and / or second image received as input to the function into several pixel regions for each object in the scene, as with semantic segmentation, but without labeling each region with a class from a set of predefined classes. An instance segmentation score can be associated with the instance segmentation of the first and / or image, or with each segment determined in the first and / or second image, the score being representative of the degree of certainty associated with the instance segmentation;

[0368] - a panoptic segmentation function, which implements both semantic segmentation and instance segmentation, aiming to segment the first and / or second input image into several pixel regions for each object in the scene, each object being labeled with a class from a set of predefined classes, the respective objects and classes thus forming the result of the instance segmentation. A panoptic segmentation score can be associated with the panoptic segmentation of the first and / or second image, or with each segment determined in the first and / or second image, the score being representative of the degree of certainty associated with the panoptic segmentation;

[0369] - an object tracking function, also called "tracking" in English, capable of determining a motion vector for at least one object identified in the first and / or second image received as input to the function, the motion vector of the object thus forming the result of the object tracking function.

[0370] Object detection, semantic segmentation, panoptic segmentation, and instance segmentation functions are all functions of semantic perception.

[0371] According to one embodiment of the invention, described in detail below by way of illustration, the first image processing function is a depth information determination function, and the second image processing function is a semantic perception function (for example, one of the semantic perception functions described above). In a variant not described below, both the first and second image processing functions are semantic perception functions. According to other variants not described, the vehicle further comprises at least one third image processing module configured to apply a third image processing function, for example, a semantic perception function.

[0372] More generally, the 100N vehicle comprises K image processing modules implementing K distinct image processing functions, K being an integer greater than or equal to 2. In what follows, to simplify the disclosure of the invention, it is considered, for illustrative purposes, that K is equal to 2.

[0373] Furthermore, several embodiments can be provided, depending on the number of images received as input by each image processing function. For example:

[0374] - according to one embodiment, each image processing function takes as input only one image from among at least the first and second images, and at least two image processing functions take different images as input. For example, in the embodiment described with K=N=M=2, the first image processing function can receive the first image as input and the second image processing function can receive the second image as input;

[0375] Alternatively, at least one image processing function takes as input several images from among at least the first and second images. For example, in the embodiment described with K=N=M=2, the first image processing function can receive the first and second images as input, and / or the second image processing function can receive the first and second images as input. According to another example, in the embodiment described with K=N=M=2, the first image processing function can receive the first image as input, and the second image processing function can receive both the first and second images as input.According to the invention, the projection of several light patterns in at least the first and second wavelength domains in the scene 10N, by controlling at least one light module 101N by at least two light intensity maps generated according to the invention from an image in the predetermined wavelength domain, makes it possible to optimize the first result of the first image processing function and the second result of the second image processing function, particularly in night driving situations when the results of image processing functions are degraded.

[0376] Furthermore, for the first image processing function, which in the described example determines depth information, the projection of light patterns eliminates the need for a two-camera system capable of acquiring images in the same wavelength range, according to the principle of stereovision, to measure the disparity of each object or pixel in the scene 10N and deduce depth information from the image. Indeed, according to a known stereovision method, two cameras, whose relative positions are predefined and known, can each acquire an image of the same scene in the same wavelength range. The offset between pixels corresponding to the same object is called disparity, and it allows, geometrically, and from the known separation between the two cameras, the object's distance to be determined.This makes it possible to determine a depth map for a pair of images captured by the two cameras, the depth map indicating the distance of each pixel in the captured scene. Projecting light patterns onto the scene (one or more of which are taken as input by the first image processing function) eliminates the need for a two-camera system. The distortions of the light pattern(s) by the scene allow access to the depth information of each pixel in the first or second image acquired by one camera of at least one 102N camera, as will be better understood from the description of Figures 9a and 9b below.The system also includes a 109N driver assistance module, also known as the 109N ADAS module (for "Advanced Driver-Assistance Systems"), capable of implementing at least one driver assistance function based on the first image processing result of the first image processing function and the second image processing result of the second image processing function. As mentioned previously, the ADAS module can receive information about the distance to objects (and therefore the depth of the scene), which constitute the first result of the first image processing function, redundantly from at least one other sensor on the vehicle, for example, a lidar, thus improving the safety associated with the driver assistance function. The ADAS module can also receive the second result of the second semantic perception function.

[0377] The system according to the invention further comprises a control device 103N capable of obtaining an image in the predetermined wavelength range from at least one camera 102N, and of applying to said image a generation model 108N to generate at least two light intensity maps. Each light intensity map indicates a light intensity for each light element (or for a set of light elements in the first embodiment where the pixelated light source of a light module comprises several sets emitting in different wavelength ranges) of the pixelated light source 111N of at least one light module 101N.Upon receiving a light intensity map, the control unit 11 ON can thus control the pixelated light source 111 N so as to project at least one light beam according to at least one light pattern corresponding to at least one light intensity map, in at least one wavelength range.

[0378] In the first embodiment, upon receipt of at least the first and second light intensity maps, the control unit 110N of a light module 101.1 N and / or 101.2N can thus control the pixelated light source 111 N so as to project the first light beam according to the first light pattern corresponding to the first light intensity map, in the first wavelength range, and so as to project a second light beam according to a second light pattern corresponding to the second light intensity map, in the second wavelength range.

[0379] In the second embodiment, the light module 101.1 N receives at least the first light intensity map, and the light module 101.2 N receives at least the second light intensity map. The control unit 110N of the light module 101.1 N controls the pixelated light source 111 N of the light module 101.1 N so as to project the first light beam according to the first light pattern corresponding to the first light intensity map, in the first wavelength range. The control unit 110N of the light module 101.2 N controls the pixelated light source 111 of the light module 101.2 N so as to project the second light beam according to the second light pattern corresponding to the second light intensity map, in the second wavelength range.

[0380] As previously stated, the predetermined wavelength range (that of the input image to the 108N generation model) can be the first wavelength range, the second wavelength range, or another wavelength range distinct from the first and second ranges. When the predetermined wavelength range is another wavelength range, at least one 102N camera includes a sensor or camera capable of acquiring images in that other wavelength range.

[0381] The 108N generation model is thus capable of generating at least first and second light intensity maps corresponding respectively to first and second light patterns, from an image received as input in the predetermined wavelength range.

[0382] The control device 103N can transmit the light intensity maps to the control unit 11 ON of at least one light module 101 N, for projection of a first pixelated light beam according to the light pattern in the first wavelength range, and for projection of a second pixelated beam according to the second light pattern in the second wavelength range (and optionally for projection of the first pattern, the second pattern, or a third pattern corresponding to a third generated light intensity map, in at least a third wavelength range).

[0383] Thus, each light intensity map is interpreted by the 110N control unit to control the light intensity of each light element in the 111N matrix source, or of a set of light elements in the 111N matrix source when the matrix source comprises several sets emitting in different wavelength ranges. For example, each light intensity map includes a light intensity value for each light element in a set (when the light intensity map and the set of light elements in the 111N matrix source have the same resolution) or to control the light intensity of a subset of light elements in the set of light elements (when the 111N matrix source has a higher resolution than the light intensity map).

[0384] Figure 9a presents an example of a 120N light beam according to a first light pattern corresponding to a first light intensity map generated according to the invention, projected by a 101N light module as previously described at least in the first wavelength range, in particular when the first image processing function is a depth information determination function and the first image processing function takes at least as input the first image including the first projected light pattern.

[0385] In particular, Figure 9a represents a beam of light projected onto a screen equipped with an orthonormal coordinate system and positioned 25 meters from the projector. Therefore, Figure 9a does not represent the projection of the first light pattern onto a real scene, which will be described with reference to Figure 9b.

[0386] According to the invention, the 108N generation model is capable of generating light intensity maps corresponding to discontinuous light patterns, composed for example of dark areas and lit areas, the deformations induced by the 10N scene in the first light pattern, in particular at the discontinuities which are the boundaries between dark areas and lit areas, allowing the determination of depth information by the first image processing function, and further promoting the second semantic perception function.

[0387] Advantageously, the first and second light intensity maps are generated from image acquired by at least one camera 102 in the predetermined wavelength range, descriptive of the scene facing the vehicle: it is thus possible to improve the overall performance associated with the first image processing function to the second image processing function, as described below.

[0388] In the example in Figure 9a, the first light pattern is a checkerboard pattern generated by the 108N generation module for a given scene, thus from a descriptive image of the scene facing the vehicle in the predetermined wavelength range. The checkerboard pattern exhibits a regular alternation of dark areas (201N) and bright areas (202N), with a strong contrast between these areas.

[0389] Each 201N or 202N zone corresponds to a set of at least one pixel, and preferably to a plurality of pixels, for example several tens or hundreds of pixels.

[0390] Thus, a dark area is created by deactivating the pixels in the area, while a lit area is obtained by activating at least some of the pixels in the lit area. The activation or deactivation of pixels is implemented by the control unit 110N, which is capable of controlling the pixelated light source 111N. In the first example of the pixelated light source 111N described previously, the control unit 110N creates the first light pattern by applying a voltage to the light elements corresponding to the pixels of the lit areas 202N, and applies no voltage to the light elements corresponding to the pixels of the dark areas 201N, based on the first light intensity map received from the control device 103N.A high-definition pixelated 111 N light source thus allows the projection of the first light pattern with a large number of dark and lit areas, which then allows high precision in the implementation of the image processing function(s) taking as input at least the first image including the first light pattern.

[0391] The checkerboard pattern shown in Figure 9a is for illustrative purposes only. It comprises 9 columns and 8 rows of zones, each zone containing a plurality of pixels. There are no restrictions on the shape of the zones, nor on the distribution of light and dark areas.

[0392] Thus, according to the invention, if the scene facing the vehicle varies, the image submitted as input to the 108N generation model varies, which can induce a variation in the light intensity maps at the output of the 108N generation model, such a variation improving the overall performance of the image processing functions implemented in the 100N vehicle.

[0393] Figure 9b presents a first image 21 ON acquired by at least one camera 102N following the projection of the first light beam according to the first light pattern in the scene 10N facing the vehicle 100N, according to embodiments of the invention.

[0394] The scene shown in figure 9b is intentionally simplified, with few objects, in order to simplify the understanding of the invention.

[0395] The scene in which the first light pattern is projected includes another vehicle, as well as a vertical wall located behind the other vehicle. The scene could thus represent the interior of a parking lot, a situation given for illustrative purposes only. The invention can advantageously be implemented in an outdoor driving scene, particularly during a night scene.

[0396] As previously explained, the first light pattern projected onto the scene is in the first wavelength range, and the first image 21 ON captured by the camera 102N is thus acquired in the first wavelength range. According to the invention, at least one camera 102N acquires the first image 21 ON in the first wavelength range during the projection of the first light beam in the first wavelength range according to the first light pattern corresponding to the first generated light intensity map, as well as a second, unillustrated image of the scene 10N, in the second wavelength range, during the projection of the second light beam in the second wavelength range, according to the second light pattern.

[0397] The first and second images can be acquired simultaneously (i.e., the exposure time periods of the 102N camera for each of its sensors, or of the 102N cameras, overlap or are identical).

[0398] Other variants are possible according to the invention, including the acquisition by at least one camera 102N of at least a third image in the third wavelength range, when a third light beam is projected by at least one light module 101 N according to the first light pattern, the second light pattern or a third light pattern corresponding to a third generated light intensity map, in the third wavelength range.

[0399] Thus, according to the embodiments, at least one 102N camera can acquire as many images as there are wavelength ranges in which light patterns are projected. In the examples described above, two wavelength ranges are used for projecting two light patterns, but the invention also applies to strictly more than two wavelength ranges in which at least two light patterns are projected.

[0400] It can be seen in figure 9b:

[0401] - that the dark and lit areas 211 N projected onto a substantially horizontal plane such as the ground have an elongated shape and are thus lengthened;

[0402] - that the dark and lit areas 212N projected onto a substantially vertical and close plane, such as the rear of the other vehicle, retain a square format and have a given first size; - that the dark and lit areas 213N projected onto a substantially vertical plane further away than the areas 212N, such as the wall behind the other vehicle, also retain a square format but have a second size greater than the first size;

[0403] - that dark and illuminated areas 214N projected onto irregular objects, with curves, such as the top of the other vehicle, are strongly distorted.

[0404] Thus, the first light pattern in Figures 9a and 9b can correspond to a first light intensity map generated by the 103N control device, particularly when the first image processing function is a depth information determination function that takes at least the first image as input. Indeed, projecting the first light pattern onto a scene provides, through analysis of the deformation of the first projected light pattern and after image processing by the first image processing module 104.1 N, scene disparity information that allows for the estimation of scene depth information. Therefore, a first light pattern comprising a regular repetition of contrasting sub-patterns, such as a repetition of square dark and light areas to form a checkerboard, is particularly well-suited for estimating scene depth information.The 108N generation model can adapt the characteristics of such a regular repetition (size and shape of dark and light areas for example) according to the image in the predetermined wavelength range received as input, and the parameters of the 108N generation model are trained to improve both the first result of the first depth information determination function (which can be a depth map), but also the second result of the second semantic perception function, compared to the acquisition of a first image without prior projection of the light patterns (and optionally other image processing results from other image processing functions).

[0405] The second light pattern may be, like the first light pattern in the first image 21 ON described above, a pattern with a repetition of contrasting sub-patterns, for example, a checkerboard pattern, but with different or identical characteristics to the first light pattern described previously, for example, with a different size and / or geometry of light and / or dark areas. Alternatively, the second light pattern may not include such a repetition of contrasting sub-patterns.

[0406] According to another variant, particularly when the first and second image processing functions are both semantic perception functions (so no image processing function determines depth information in the 100N vehicle), the 108N generation model may be able to generate two light intensity maps corresponding to distinct first and second light patterns not including such repetition of contrasting sub-patterns, depending on the image in the predetermined wavelength range received at input.

[0407] Figure 10a shows a structure of a control device 103N according to embodiments of the invention.

[0408] The control device 103N includes an input interface 303N to receive the image captured by at least one camera 102N, in the predetermined wavelength range, the image being representative of the scene 10N facing the vehicle.

[0409] The control device 103N further includes a computing unit 301N, such as a processor, configured to generate at least two light intensity maps, comprising the first light intensity map and the second light intensity map, from the image received on the input interface 313N, by application of the generation pattern 108N.

[0410] The 301N processor is configured to communicate unidirectionally or bidirectionally, via one or more buses or via a direct wired connection, with a 302N memory such as Random Access Memory (RAM), Read Only Memory (ROM), or any other type of memory (Flash, EEPROM, etc.). Alternatively, the 302N memory may contain several of the aforementioned types. For example, the memory can temporarily store the image received on the 303N input interface. The 302N memory can also permanently store an algorithm executing the 108N generation model, which is capable of receiving an image in the predetermined wavelength range as input and outputting light intensity maps.

[0411] The 302N memory can, for example, store instructions enabling the execution of the 108N generation model, which can be defined by a set of parameters and capable of generating light intensity maps from at least one representative image of a scene facing the vehicle.

[0412] The 302N memory can also store previous brightness maps (the latest brightness maps generated by the 108N generation model), when the 108N generation model also takes previous brightness maps as input as described below.

[0413] According to embodiments of the invention, the 108N generation model is trained in a preliminary phase by machine learning, the machine learning enabling the training of the parameters defining the model, the model having a predefined structure. The model may, for example, have one of the following structures:

[0414] - a convolutional neural network, such as a U-Net type network for example;

[0415] - an artificial neural network of the auto-encoder or variational auto-encoder type, also called VAE in English;

[0416] - a self-aware or transformative model; or

[0417] - any other model structure capable of receiving as input an image representing a scene, and of generating several light intensity maps (at least two light intensity maps).

[0418] No restrictions are attached to the machine learning applied to the 108N generation model to optimize its parameters. The machine learning can, for example, be supervised, as described later with reference to Figure 14, shown below. The 103N control device also includes a 304N output interface capable of transmitting the generated light intensity maps to the 101N light module.

[0419] The control device 103N may further include another output interface 305N capable of transmitting the generated light intensity maps to at least one of the image processing modules 104.1 N and 104.2 N.

[0420] Figure 10b shows a structure of the 104N image processing module according to embodiments of the invention.

[0421] The 104N image processing module can be the first 104.1 N image processing module or the second 104.2N image processing module (or any other 100N vehicle image processing module).

[0422] The image processing module 104N includes an input interface 313N for receiving the first image, the second image, or the first and second images (and optionally at least a third image, when more than two light beams are projected into strictly more than two wavelength ranges), captured by at least one camera 102N and representative of the projection of the first light pattern and / or the second light pattern in the scene 10N facing the vehicle 100N, the patterns corresponding to the light intensity maps generated by the control device 103N.

[0423] The 104N image processing module further includes a 311N computing unit, such as a processor, configured to determine an image processing result from the first and / or second image received on the 313N input interface. The 311N processor can, for example, be a graphics processor, also called a GPU for “Graphical Processing Unit”.

[0424] The 311N processor is configured to communicate unidirectionally or bidirectionally, via one or more buses or via a direct wired connection, with a 312N memory such as Random Access Memory (RAM), Read Only Memory (ROM), or any other type of memory (Flash, EEPROM, etc.). Alternatively, the 312N memory may contain multiple memories of the aforementioned types. For example, the memory can temporarily store the first and / or second image received on the 313N input interface.

[0425] The 312N memory can also permanently store an algorithm running the 106.1 N or 106.2 N image processing model capable of receiving as input the first image in the first wavelength domain and / or the second image in the second wavelength domain, and providing as output an image processing result (which can be depth information, or a semantic perception result, as previously described).

[0426] The 312N memory can, for example, store instructions enabling the execution of the 106.1 N or 106.2 N image processing model, which can be defined by a set of parameters optimized to generate an image processing result from at least one first image representative of the projection of the light pattern in the scene.

[0427] According to some embodiments, the first image processing model 106.1 N is capable of determining depth information from the first and / or second image representing the projection of the first and / or second light pattern in the scene, but also from the first and / or second light intensity map corresponding to the first and / or second projected light pattern. To this end, when the control device 103N transmits the first and / or second light intensity map to the light module 101 N, the control device 103N also transmits the first and / or second light intensity map to the first image processing module 104.1 N.

[0428] According to embodiments of the invention, the image processing model 106.1 N or 106.2 N is trained in a preliminary phase by machine learning, the machine learning enabling the training of the parameters defining the model, the model having a predefined structure. The image processing model 106.1 N or 106.2 N may, for example, have one of the following structures:

[0429] - a convolutional neural network, such as an ll-Net type network for example;

[0430] - an artificial neural network of the auto-encoder or variational auto-encoder type, also called VAE in English;

[0431] - a self-aware or transformative model; or

[0432] - any other model structure capable of receiving as input the first and / or second image, representing the projection of the first light pattern and / or the second light pattern in the scene, and optionally the first and / or second light intensity map, and of determining an image processing result.

[0433] There are no restrictions on machine learning applied to the image processing model to optimize its parameters. Machine learning can, for example, be supervised.

[0434] Alternatively, the 106.1 N or 106.2N image processing model is not derived from machine learning, but is software executing a set of rules capable of determining an image processing result by applying the set of rules to the first and / or second image representing the projection of the first and / or second light pattern in the scene, and optionally from the first and / or second light intensity map corresponding to the first and / or second projected light pattern.

[0435] The 104N image processing module can apply pre-processing to each first and / or second received image before applying the 106.1N or 106.2N image processing model. This pre-processing can consist of cropping the first and / or second received image to retain only the portion corresponding to the projection of the first and / or second light pattern in the 10N scene. Other pre-processing techniques can be applied according to the invention. The 104N image processing module further includes a 314N output interface capable of transmitting the image processing result to a separate module. For example, the separate module to which the image processing result is transmitted could be the 109N ADAS module described previously.

[0436] The image processing module 104N may further include another input interface 315N suitable for receiving the first and / or second light intensity card generated by the control device 103N, as described previously.

[0437] Figure 11a illustrates a pixelated light source 111 N according to the first example above, and in the first embodiment, capable of forming the first and second light beams in the first and second distinct wavelength ranges, so as to project the first light pattern and the second light pattern.

[0438] In the first embodiment, the pixelated light source 111 N according to the first example comprises:

[0439] - a first set of individually controllable 401 N luminous elements, capable of emitting light in the first wavelength range; and

[0440] - a second set of individually controllable 402N light elements, capable of emitting in the second wavelength range.

[0441] In the example in Figure 11a, the first wavelength range is an infrared range and the second wavelength range is a visible range: for this purpose, the second set of light elements 402N includes subsets of blue, red and green light elements, which allows a second colored light beam to be projected into the scene 10N facing the vehicle, to project the second light pattern according to the second light intensity map generated by the control device 103N.

[0442] Alternatively, each element of the second 402N set can be capable of emitting white light for the projection of the second light pattern in the 10N scene. Thus, the light elements of the first 401N set can be controlled to form the first light beam according to the first light pattern corresponding to the first received light intensity map, with a resolution depending on the number of light elements in the first 401N set.

[0443] Simultaneously, the light elements of the second 402N set can be controlled to form the second light beam according to the second light pattern corresponding to the second light intensity map or performing the lighting or signaling function.

[0444] According to undescribed variants of the first embodiment, at least a third set of light elements can be controlled to form a third light beam according to the first light pattern, according to the second light pattern or according to a third light pattern corresponding to a third light intensity map, in the third wavelength range.

[0445] Figure 11b illustrates a pixelated light source 111 N according to the second example, and in the first embodiment, comprising two light sources 411.1 N and 411.2 N and a DMD micromirror array.

[0446] The first light source, 411.1 N, is capable of emitting light rays in the first wavelength range. The second light source, 411.2 N, is capable of emitting light rays in the second wavelength range. According to undescribed variants, at least a third light source is capable of emitting light rays in the third wavelength range.

[0447] The light sources 411.1 N and 411.2 N can be arranged on the same support 412 N, as shown in Figure 11 b, or on two separate supports.

[0448] Only one micromirror 41 ON is shown in Figure 11b to facilitate understanding of the figure. However, in practice, the matrix comprises a large number of micromirrors, in particular more than 100, or even more than 1000 or more than 10,000 micromirrors individually controllable by the control unit 110N described previously.

[0449] Each micromirror 41 ON can be controlled to switch between at least one first position 420.1 N and a second position 420.2 N. In the first position 420.1 N, the micromirror is arranged to reflect a light beam from a source, for example a light beam 413 N emitted by the first light source 411.1 N, towards the projection system 112 N so as to contribute to the formation of the first light beam outside the vehicle 100 N. Thus, by controlling the positions of the micromirrors in the array with the control unit 110 N, the light module projects the first light beam according to the first light pattern corresponding to the first received light intensity map, in the first wavelength range, when the first light source 411.1 N is activated.Similarly, by controlling the positions of the matrix micro-mirrors with the 110N control unit, the light module projects the second light beam according to the second light pattern corresponding to the second received light intensity map.

[0450] The 110N control unit can thus control the same micromirror array, alternating at a given frequency to alternately project at least the first and second beams. The switching frequency between the at least two light beams can be greater than 700 Hz, so that the flickering of the first and second light beams is not perceptible to the human eye and therefore does not disturb the driver and other road users (when the first or second wavelength range is visible), and is not perceptible to the 102N camera(s), having an exposure time longer than the period associated with the switching frequency.

[0451] Figure 12 is a diagram illustrating the steps of a typical phase of a method for processing representative images of a scene facing a vehicle, according to embodiments of the invention. At a step 500N, at least one camera 102N obtains a representative image of the scene facing the vehicle in the predetermined wavelength range, which can be, as previously stated, the first wavelength range, the second wavelength range, or another wavelength range distinct from the first and second wavelength ranges.

[0452] At a step 501N, the control device 103N generates light intensity maps (comprising at least the first light intensity map and the second light intensity map) from the image obtained at step 500N, by applying the generation model 108N. The generation model can further take as input the previous light intensity maps, which include at least the first and second light intensity maps generated during a previous iteration of step 501N.

[0453] At a step 502N, the control device 103N transmits the generated light intensity maps to at least one light module 101.1N or 101.2N. According to the first embodiment, the control device 103N can transmit the light intensity maps to a single light module 101N for projection of the first and second light beams, or to both light modules 101.1N and 101.2N for superimposition of their respective first and second light beams in the first and second wavelength ranges. In the second embodiment, the control device 103N transmits the first light intensity map to the light module 101.1N and the second light intensity map to the light module 101.2N. In addition, one or more of the generated light intensity maps can be transmitted to at least one of the first and second image processing modules 104.1 N and 104.2 N, for example to the first image processing module 104.1 N capable of determining depth information.

[0454] At step 503N, in the first embodiment, the control unit 11 ON of each light module 101.1 N and 101.2 N, having received the light intensity cards, controls the matrix source 111 N according to the received light intensity cards, to project:

[0455] - the first light beam according to the first light pattern corresponding to the first light intensity map, in the first wavelength range; and

[0456] - the second light beam according to the second light pattern corresponding to the second light intensity map, in the second wavelength range (and optionally a third light beam in at least the third wavelength range, to project the first pattern, the second pattern or a third light pattern corresponding to a third light intensity map).

[0457] In the second embodiment:

[0458] - the control unit 11 ON of the light module 101.1 N, having received the first light intensity map, controls the matrix source 111 N of the light module 101.1 N according to the first received light intensity map, to project the first light beam according to the first light pattern corresponding to the first light intensity map, in the first wavelength range; and

[0459] - the control unit 11 ON of the light module 101.2N having received the second light intensity map, controls the matrix source 111 N of the light module 101.2N according to the second light intensity map received, to project the second light beam according to the second light pattern corresponding to the second light intensity map, in the second wavelength range.

[0460] Following step 503N, the process can return to step 500N to obtain a new image in the predetermined wavelength range, allowing continuous and real-time adaptation of the light patterns of the projected light beams.

[0461] In addition, following step 503N, at least one camera 102N obtains, at a step 504N, at least one first image representative of the projection of the first light pattern in the scene, in the first wavelength range, and a second image representative of the projection of the second light pattern in the scene, in the second wavelength range (and optionally a third image of the scene in at least the third wavelength range).

[0462] Note that when the predetermined wavelength range is the first wavelength range, obtaining the new image in the predetermined wavelength range, at the new step 500N, and obtaining the first image in the first wavelength range at the step 504N are one and the same step.

[0463] Similarly, when the predetermined wavelength range is the second wavelength range, obtaining the new image in the predetermined wavelength range at the new 500N step and obtaining the second image in the second wavelength range at the 504N step are one and the same step.

[0464] At a step 505N, the first processing module 104.1 N determines the first image processing result (the depth map for example), by applying the first image processing model 106.1 N, to the first image in the first wavelength range obtained at step 504N, and / or to the second image in the second wavelength range obtained at step 504N, and optionally to the first and / or second light intensity map generated at step 501 N and transmitted by the control device 103N.

[0465] At a step 506N, implemented independently of step 505N, the second processing module 104.2N determines the second image processing result (the semantic perception result, for example) by applying the second image processing model 106.2N to the first image in the first wavelength range obtained in step 504N, and / or to the second image in the second wavelength range obtained in step 504N, and optionally to the first and / or second light intensity map generated in step 501N and transmitted by the control device 103N. At a step 507N, the first image processing module 104.1N can transmit the first image processing result to at least one other module, for example, to the ADAS module 109N. The ADAS module can therefore implement at least one driver assistance function based on the first image processing result.

[0466] At step 508N, the second image processing module 104.2N can transmit the second image processing result to at least one other module, for example to the ADAS module 109N. The ADAS module can thus implement at least one driver assistance function based on the second image processing result.

[0467] According to some embodiments, the ADAS 109N module implements at least one driver assistance function based on the first image processing result and the second image processing result.

[0468] As described previously, at least one third image processing module can determine a third image processing result by applying the third image processing function to the first image and / or the second image, the third result being able to be passed on to at least one other module such as the ADAS 109N module.

[0469] Figure 13 is a 600N drive system of the 108N generation model of light intensity cards, according to embodiments of the invention.

[0470] Such a 600N drive system is external to the 100N vehicle and is capable of implementing the drive phase of the process according to the invention, which is described later with reference to Figure 14. The drive phase precedes the current phase described with reference to Figure 12, which takes place during a driving situation of the 100N vehicle. The drive phase can, in particular, be part of the design and manufacturing process of the 103N control device, before its integration into the 100N vehicle.

[0471] The 600N training system includes a 601N training database, capable of storing training data associations, each training data association comprising:

[0472] - a training image representative of a scene in the predetermined wavelength range;

[0473] - ground truth which includes first reference data corresponding to a first result of reference image processing (reference depth information in the example described) and second reference data corresponding to a second result of reference image processing (a semantic perception result in the example described).

[0474] Preferably, the associations stored in the training database are varied, that is to say, they were obtained in scenes varying according to several criteria: scene composition, scene brightness level, weather conditions, etc.

[0475] In addition, the 601 N training database includes more than one hundred training data associations, and preferably several thousand or even tens of thousands of training data associations.

[0476] There are no restrictions on how the 601 N training database was constructed. The 601 N training database can be obtained from images captured by a fleet of vehicles in real-world night driving situations (preferably varied according to several criteria as described previously), with one or more sensors, or a human operator, capable of determining the first and second reference image processing outcomes. For example, a depth camera and / or lidar can obtain reference depth information (as the first reference data), and a human operator, or complementary sensors, can determine a semantic perception outcome (as the second reference data), for each training image captured in the predetermined wavelength range.

[0477] The 600N training system further includes the 108N generation model of light intensity maps to be trained. As previously described, the 108N generation model of light intensity maps is structurally capable of generating several light intensity maps of a given resolution, including the first and second light intensity maps, from a representative image of a scene in the predetermined wavelength range. This image can be a training image during the training phase, and is an image captured in real time by at least one 102N camera during the 500N step of the current phase.

[0478] For this purpose, as described previously, the 108N generation model of light intensity cards can have one of the following structures:

[0479] - a convolutional neural network, such as an ll-Net type network for example;

[0480] - an artificial neural network of the auto-encoder or variational auto-encoder type, also called VAE in English;

[0481] - a generating network of a system of generative adversarial networks, also called GAN, for “Generative Adversarial Networks” in English;

[0482] - a self-aware or transformative model; or

[0483] - any other generative model structure capable of receiving as input an image representing a scene, and of producing as output light intensity maps in a format capable of controlling at least one previously described 101 N light module.

[0484] The 600N drive system also includes a 602N synthesis module, capable of producing several synthetic images, including:

[0485] - a first synthetic image representative of the scene in the training image, illuminated by the first light pattern corresponding to the first light intensity map (called the first training light intensity map during the training phase) produced at the output of the 108N generation model, in the first wavelength range, and

[0486] - a second synthetic image representative of the scene in the training image, lit by the second light pattern corresponding to the second light intensity map (called second training light intensity map during the training phase) produced at the output of the 108N generation model, in the second wavelength domain.

[0487] Thus, the 602N synthesis module allows for the simulation of a camera capturing the training image scene, into which at least the following would be projected:

[0488] - the first light pattern corresponding to the first training light intensity map, in the first wavelength range, and

[0489] - the second light pattern corresponding to the second training light intensity map, in the second wavelength range.

[0490] For this purpose, the 602N synthesis module receives the following input:

[0491] - the training image;

[0492] - the training light intensity maps generated by the 108N generation model from the training image.

[0493] The 600N training system further includes at least one first training image processing module 603.1 N implementing a first training image processing function and a second training image processing module 603.2N implementing a second training image processing function. In the example described, the first training image processing module 603.1 N is capable of determining depth information, forming a first training image processing result, and the second training image processing module 603.2N is capable of determining a semantic perception result, forming a second training image processing result.

[0494] The first drive image processing module 603.1 N is preferably identical to the first drive image processing module 104.1 N integrated into the vehicle 100N during the current phase of the process according to the invention. Similarly, the second drive image processing module 603.2 N is preferably identical to the second drive image processing module 104.2 N integrated into the vehicle 100N during the current phase of the process according to the invention. The generation module 108N is thus specifically driven to improve the performance of at least the first and second image processing functions implemented in the vehicle 100N during the current phase (and optionally at least a third image processing function).

[0495] Thus, the first training image processing module 603.1 N can implement the first image processing model 106.1 N, and the second training image processing module 603.2 N can implement the second image processing model 106.2 N. The image processing models 106.1 N and 106.2 N may have been obtained, for example, through machine learning during dedicated training phases not described herein. Alternatively, at least one image processing model 106.1 N and 106.2 N may implement a software function not derived from machine learning.

[0496] Alternatively, at least one of the first and second image processing models 106.1 N and 106.2N is trained at the same time as the generation model 108N, for example by supervised learning as described with reference to Figure 14.

[0497] The first 603.1 N training image processing module is capable of receiving:

[0498] - the first and / or second computer-generated image from the 602 N synthesis module; and

[0499] - optionally, the first and / or second training light intensity map generated by the 108N generation model after receiving the training image.

[0500] Based on the information received, the first 603.1 N training image processing module is capable of generating a first training image processing result.

[0501] In addition, the second 603.2N training image processing module is capable of receiving:

[0502] - the first and / or second synthesis image from the 602 N synthesis module; and - optionally, the first and / or second training light intensity map generated by the 108N generation model after receiving the training image.

[0503] Based on the information received, the second 603.2N training image processing module is capable of generating a second training image processing result.

[0504] The 600N drive system also includes at least:

[0505] - a first loss assessment module 604.1 N capable of assessing a first loss by comparing the first ground truth reference data from the 601 N training database with the first training image processing result from the first training image processing module 603.1 N (the training depth information in the described example); and

[0506] - a second loss evaluation module 604.2N capable of evaluating a second loss by comparison between the second ground truth reference data from the training database 601 N and the second training image processing result from the second training image processing module 603.2N (the semantic perception result in the example described).

[0507] The 600N system also includes a 605N fusion module capable of:

[0508] - determine an overall loss from at least the first loss from the first loss assessment module 604.1 N and the second loss from the second loss assessment module 604.2 N. There are no restrictions on the function used to obtain the overall loss from at least the first and second losses. For example, it could be an average, a sum, or any other function capable of taking two input values; and

[0509] - to modify one or more parameters of the generation model 108N based on the evaluated overall loss, and according to a predefined training strategy that reduces the overall loss when steps 700N to 707N of the training phase are iterated for several training images. In embodiments where at least one of the first and second image processing models is trained simultaneously with the generation model, the fusion module is further capable of modifying one or more parameters of this image processing model based on the evaluated overall loss.

[0510] No restrictions are attached to the loss function used to determine the first loss or the loss function used to determine the second loss. The loss functions can be based on a distance according to a given standard.

[0511] Such a 600N training system is thus capable of implementing supervised learning of the 108N generation model.

[0512] Alternatively, the 108N generation model of light intensity maps can be derived from machine learning other than supervised learning. For example, the 108N generation model of light intensity maps can be optimized by reinforcement learning, with a reward determined based on at least one first loss evaluated from the first training image processing result from the first training image processing module 603.1 N and a second loss evaluated from a second training image processing result from the second training image processing module 603.2 N.

[0513] Thus, in general, the 108N generation model of light intensity maps can be trained by machine learning in such a way as to improve the overall performance (to decrease the respective losses) of at least the first image processing function (depth information determination function) and the second image processing function (semantic perception function), by projecting several light patterns corresponding to several light intensity maps generated by the 108N generation model.

[0514] Figure 14 is a diagram illustrating the steps of a training phase of a 108N generation model of light intensity maps, in an image processing method in a vehicle, according to embodiments of the invention.

[0515] As previously explained, the training phase of the 108N generation model of light intensity maps can be implemented in the 600N training system described with reference to Figure 13.

[0516] At a 700N step, the 601N training database obtains a training data association as previously described, comprising a training image of a scene facing a vehicle, a first reference image processing result, and a second reference image processing result.

[0517] At a step 701 N, the obtained training image is submitted as input to the generation model 108N which generates N training light intensity maps based on the obtained training image, N being an integer greater than or equal to 2, comprising the first light intensity map and the second light intensity map.

[0518] The training light intensity maps are transmitted from the 108N generation model to the 602N synthesis module, which also receives the training image from the 601 N training database. The 108N generation model can further transmit the first and / or second generated training light intensity map to the first training image processing module 603.1 N and / or can transmit the first and / or second generated training light intensity map to the second training image processing module 603.2N.

[0519] At a 702N step, the 602N synthesis module determines synthetic images respectively representative of the scene of the training image, illuminated by N training light patterns corresponding to the N training light intensity maps produced at the output of the generation model 108N, in M ​​wavelength ranges (M being greater than or equal to N, e.g. equal to N).As described previously, the synthesis module can produce at least the first synthesis image (for the projection of a first training light pattern in the first wavelength range) and the second synthesis image (for the projection of the second training light pattern in the second wavelength range), and optionally at least a third synthesis image (for the projection of the first training light pattern, the second training light pattern or a third training light pattern corresponding to a third training light intensity map, in the third wavelength range).

[0520] At least one of the generated synthetic images is passed to the first training image processing module 603.1 N, which generates the first one-step training image processing result 703.1 N.

[0521] At least one of the synthetic images is passed to the second training image processing module 603.2N, which generates the second one-step training image processing result 703.2N.

[0522] The first training image processing result generated in step 703.1 N is transmitted from the first training image processing module 603.1 N to the first loss evaluation module 604.1 N. At a step 704.1 N, the first loss evaluation module 604.1 N evaluates a first loss by comparison between the first training image processing result and the first reference image processing result.

[0523] The second training image processing result generated in step 703.2N is passed from the second training image processing module 603.2N to the second loss evaluation module 604.2N. At a step 704.2N, the second loss evaluation module 604.2N evaluates a second loss by comparing the second training image processing result with the second reference image processing result.

[0524] Optionally, at least a third loss can be evaluated when the 108N generation module is trained to optimize the third image processing result of at least one third image processing module. At a 705N step, the 605N fusion module determines an overall loss based on at least the first and second losses.

[0525] At a step 706N, the fusion module 605N can further determine whether a predefined convergence criterion is met or not, based in particular on the overall loss evaluated at step 705N, and optionally on losses evaluated during previous iterations of steps 700N to 705N.

[0526] If the convergence criterion is not met, the fusion module 605N modifies at least one parameter of the generation model 108N at a step 707N, depending on the evaluated loss, according to the predefined training strategy, and may modify one of the image processing models 106.1N and 106.2N, when this image processing model is trained concurrently with the generation model. Such parameter optimization training strategies during supervised learning are well known and are not described further herein.

[0527] Following step 707N, the process training phase returns to step 700N to repeat steps 700N to 705N based on a new association of training data from training database 601.

[0528] The 108N generation model can thus be trained until convergence.

[0529] At step 708N, when the fusion module 605N determines that the convergence criterion is reached, the training phase is completed and the generation model 108N can be implemented in the control device 103N for implementation of step 501 N of the current phase of the process according to the invention.

[0530] According to embodiments in which the generation model 108 is also capable of taking as input the previous light intensity maps, the supervised learning described with reference to Figure 14 can be adapted in one of the following ways:

[0531] - during step 700N, N random light intensity maps are generated, and the synthesis module 602N simulates the projection of training light patterns corresponding respectively to the N random light intensity maps into the training image to obtain synthetic images, and the synthetic training images as well as the random light intensity maps are submitted as input to the generation model during step 701N; or

[0532] - the associations of the 601N training base include training images forming a sequence, the training images being submitted sequentially during the iterations of steps 700N to 707N; or

[0533] - during step 700N, the synthesis module 602N simulates the projection of light patterns corresponding to the previous light intensity maps into the newly obtained training image to obtain synthetic images, and the synthetic images as well as the previous light intensity map are submitted as input to the generation model during step 701N.

[0534] The present invention is not limited to the embodiments described above by way of example; it extends to other variants.

Claims

Demands 1. Image processing method in a vehicle (100), the method comprising, during a typical phase, the following steps: - obtaining (500) a representative image of a scene facing the vehicle; - generation (501) of at least one first light intensity map, from a light intensity map generation model (108) and the image obtained, the light intensity map generation model being configured to receive as input the image representing the scene and to generate as output the first light intensity map, the first light intensity map corresponding to a light pattern and indicating light intensity values ​​to control light elements of at least one pixelated light source of at least one light module of the vehicle; - transmission (502) of at least the first determined light intensity map to at least one light module, for projection (503) of at least a first pixelated light beam (120) according to the light pattern corresponding to the first generated light intensity map, in the scene facing the vehicle; - obtaining (504) at least one first representative image of the scene facing the vehicle, following the projection of at least one first pixelated light beam according to the light pattern; - determination (505; 506) at least of a first image processing result by applying a first image processing function to at least a first image and / or of a second image processing result by applying a second image processing function to at least a first image.

2. Method according to claim 1, wherein the first image processing function is a function for determining depth information of the scene (10) represented in at least one first image.

3. A method according to claim 1 or 2, wherein the second image processing function is a semantic perception function.

4. A method according to any one of the preceding claims, further comprising the transmission (507; 508) of the first image processing result and / or the second image processing result to a driver assistance module (109) of the vehicle (100), for implementing at least one driver assistance function based on the first image processing result and / or the second image processing result.

5. A method according to any one of the preceding claims, wherein the light intensity map generation model (108) has a structure among the following structures: - a convolutional neural network; - an artificial neural network of the auto-encoder or variational auto-encoder type; - a self-aware or transformative model; or - a network generating a system of antagonistic generative networks.

6. A method according to any one of the preceding claims, wherein the representative image of the scene (10) facing the vehicle (100) is obtained (500) in a predetermined wavelength range, and wherein the first representative image of the scene facing the vehicle is obtained (504) in a first predetermined wavelength range, identical to the predetermined wavelength range, or distinct from the predetermined wavelength range, following the projection (503) of a pixelated light beam (120) according to the light pattern in the first wavelength range.

7. A method according to any one of the preceding claims, wherein: - the generation step includes the generation of a second light intensity map, from said light intensity map generation model (108) and the image obtained, the light intensity map generation model being configured to receive as input the image representing the scene and to generate as output light intensity maps, each light intensity map corresponding to a first light pattern and indicating light intensity values ​​to control light elements of at least one pixelated light source of at least one light module of the vehicle; - the transmission step includes the transmission (502) of the second generated light intensity map to at least one light module, for projection (503) of a second pixelated light beam according to a second light pattern corresponding to the second generated light intensity map, in a second wavelength range, into the scene facing the vehicle; - the acquisition step involves obtaining (504) a second image representative of the scene facing the vehicle, into which the second pixelated light beam is projected according to the second light pattern in the second wavelength range; - the determination step includes the determination (505; 506) of the first image processing result by applying a first image processing function to the first image and / or the second image, and of a second image processing result by applying a second image processing function to the first image and / or the second image.

8. A method according to any one of claims 1 to 6, wherein the first light intensity map is transmitted to the light module (101) for projecting the first pixelated light beam (120) according to the light pattern corresponding to the first generated light intensity map, in the first wavelength range, and for projecting a second pixelated light beam according to the light pattern corresponding to the first generated light intensity map, in a second wavelength range distinct from the first wavelength range; the method further comprising obtaining (504) a second image representative of the scene facing the vehicle, following the projection (503) of the second light beam pixelated according to the light pattern, in the second wavelength range; and wherein the first image processing result is determined (505) by applying the first image processing function to the first image obtained and / or to the second image obtained; and wherein the second image processing result is determined (506) by applying the second image processing function to the first image obtained and / or to the second image obtained.

9. A method according to any one of the preceding claims, further comprising a training phase for the light intensity map generation model, comprising a modification (707) of at least one parameter of the generation model (108) as a function of an overall loss, the overall loss being evaluated (705) from: - of a first estimated loss (704.1) as a function of a first training image processing result determined by a first training image processing function, following the generation (701) of a training light intensity map by the generation model (108) during the training phase; - of a second loss evaluated (704.2) as a function of a second training image processing result determined by a second training image processing function, following the generation of the training light intensity map by the generation model during the training phase.

10. A method according to claim 8, wherein the first image processing function of the training phase is identical to the the first image processing function of the current phase, and / or the second image processing function of the training phase is identical to the second image processing function of the current phase.

11. A method according to claim 8 or 9, wherein the training phase comprises the following steps: - obtaining (700) an association of training data, the association comprising a representative training image of a scene, a first reference image processing result and a second reference image processing result; - application (701) of the light intensity map generation model (108) to the training image, to obtain a training light intensity map; - obtaining (702) at least one synthetic image representative of the scene of the training image in which at least one pixelated light beam is projected according to a light pattern corresponding to the training light intensity map; - application of the first training image processing function to at least one synthetic image, to determine (703.1) the first training image processing result; - application of the second training image processing function to at least one synthetic image, to determine (703.2) the second training image processing result; - evaluation (704.1) of the first loss by comparison of the first training image processing result with the first reference image processing result; - evaluation (704.2) of the second loss by comparison of the second training image processing result with the second reference image processing result; - assessment (705) of the overall loss from the first loss and the second loss; - modification (707) of at least one parameter of the generation model (108) of light intensity map as a function of the overall loss evaluated.

12. A method according to any one of the preceding claims, wherein the pixelated light source (111) of at least one light module (101) comprises electroluminescent semiconductor elements of submillimeter dimensions, epitaxially mounted directly on a common substrate.

13. Set in a vehicle (100) comprising: - at least one camera (102) arranged to obtain a representative image of the scene (10) facing the vehicle; - at least one light module (101; 101.1; 101.2) comprising a pixelated light source (111) comprising a plurality of individually controllable light elements; - a control device (103) configured to determine at least one first light intensity map, from a light intensity map generation model (108) and the image obtained by at least one camera, the light intensity map generation model being configured to receive as input the image representing the scene and to generate as output the first light intensity map, the first light intensity map corresponding to a light pattern and indicating light intensity values ​​to control light elements of the pixelated light source of at least one light module of the vehicle; wherein the control device is capable of transmitting the determined light intensity map to at least one light module, for projection of at least one first pixelated light beam according to the light pattern corresponding to the first generated light intensity map, in the scene facing the vehicle;in which at least one camera is capable of obtaining at least one first representative image of the scene facing the vehicle, following the projection of; at least a first pixelated light beam according to the light pattern; and the assembly further comprising at least a first image processing module (104.1) configured to determine a first image processing result by applying a first image processing function to at least a first image obtained, and / or a second image processing module (104.2) configured to determine a second image processing result by applying a second image processing function to at least a first image obtained.

14. Vehicle comprising an assembly according to claim 12, further comprising a driver assistance module (109), wherein the driver assistance module is configured to implement at least one driver assistance function based on the first image processing result and / or based on the second image processing result.

Images