Method for authenticating an image captured in an outdoor environment of a motor vehicle.

Abstract

Method for authenticating an image captured in the environment of a motor vehicle (100). Method for authenticating a candidate image acquired by an image acquisition means (2) mounted on a motor vehicle (100) to acquire images of an environment external to said motor vehicle (100), the authentication method being implemented when the vehicle (100) is stationary, and characterized in that it comprises the following steps: - a first acquisition step (E01), by the image acquisition means (2), of at least one reference image; followed by - a second acquisition step (E02) of at least one candidate image, by the image acquisition means (2), of an individual approaching the vehicle (100); Figure for the abstract: Fig. 7

Description

Title of the invention: Method for authenticating an image captured in an outdoor environment of a motor vehicle.

[0001] The invention relates to a method for authenticating an image captured in an external environment of a motor vehicle.

[0002] The automotive industry is now developing the use of facial recognition to activate various vehicle functionalities, the functionalities of which may require varying levels of security.

[0003] Facial recognition processing can be broken down into three steps: - an enrollment step corresponding to the capture of a portrait of a candidate person by an image sensor (in particular a camera) in order to construct a candidate facial biometric fingerprint, - an identification step in which the candidate facial biometric fingerprint is compared to a database of facial biometric fingerprints, the person being identified if one of the fingerprints in the database corresponds to the candidate fingerprint, and - an authentication step in which it is verified that the candidate is actually present in front of the camera. The invention relates to the authentication stage. More specifically, it focuses on detecting fraud involving the presentation of a decoy as input to facial recognition processing, in order to be falsely identified and exploit the rights of the impersonated individual. One of the most common and easiest methods of fraud involves presenting photos or videos of the impersonated individual to the sensor. It is therefore essential to protect against this type of fraud.

[0004] The object of the invention is to remedy the disadvantages described above by providing a vehicle capable of reliably detecting a family member or acquaintance walking towards said vehicle and capable of detecting an attempt at fraud by identity theft.

[0005] Depending on the type of sensor used to capture images of a person (for example, an RGB camera or an infrared camera), different technologies can be used to detect the presence of a decoy. For example, infrared camera technology can distinguish a decoy from a human face by analyzing the reflections present in the captured image. Alternatively, authentication using images from an RGB camera will use the detection of eye or facial movements, blood pulse, skin texture, or even facial geometry. These authentication solutions require the use of algorithms that are costly in terms of processing time and licensing fees. Furthermore, since these solutions are primarily developed for the banking or customs sectors, the processing methods they offer are generic and poorly suited to the specific requirements of facial recognition applied to the automotive industry. The aim of the invention is to provide a method for authenticating images that overcomes the above-mentioned drawbacks and improves upon known prior art methods for authenticating images. In particular, the invention enables a simple and reliable method that allows for fast, dependable, and computationally inexpensive authentication.

[0006] To this end, the invention relates to a method for authenticating a candidate image acquired by an image acquisition means arranged on a motor vehicle to acquire images of an environment external to said motor vehicle, the authentication method being implemented when the vehicle is stationary, the authentication method comprising the following steps: - a first acquisition step, using image acquisition methods, of at least one reference image; followed by - a second step of acquiring at least one candidate image, by means of image acquisition, of an individual approaching the vehicle; - a step to verify the authenticity of at least one candidate image; this verification step includes: - a recognition step, in at least one candidate image, of a face outline and / or a body outline of the individual approaching the vehicle; - a step of determining a set of probative zones of at least one candidate image, distributed over the contour of the face and / or the contour of the body of the individual approaching the vehicle in at least one candidate image, such that each zone of the set of probative zones includes both a portion of the face and / or a portion of the body of the individual approaching the vehicle, and a portion of the external environment, - a positioning step of a set of reference areas on the reference image, each reference area of ​​the reference image being associated with a probationary area of ​​at least one candidate image, the positions of the reference areas in the reference image being identical to the positions of the probationary areas in at least one candidate image, the size and shape of each reference area being respectively identical to the size and shape of the probationary area to which the reference area is associated; - a selection step, in each of the probationary areas of at least one candidate image, of a sub-area of ​​context excluding the portion of the face and / or body of the individual; - a step of comparing the content of each selected context sub-area in each of the probationary areas of at least one candidate image, with content from the reference area of ​​the reference image to which the probationary area is associated; - a step to calculate a correlation rate between each of the context sub-zones and a corresponding reference zone; - another step of comparing each correlation rate to a minimum threshold; - a concluding step regarding the authenticity of the content of at least one candidate image, based on the result of said comparison; - a step to confirm the absence of a decoy; or - a step to confirm the presence of a decoy, depending on the result of the verification step.

[0007] The authentication process may include a step of calculating a distance separating the individual from the motor vehicle, this calculation step being carried out after the second step of acquiring at least one candidate image.

[0008] The number of probationary areas distributed on the contour of the face and / or on the contour of the body of the individual in at least one candidate image may increase with a decrease in the distance between the vehicle and the individual approaching said vehicle.

[0009] The authentication process may include a substep of verifying the geometric consistency of the content of at least one candidate image, this verification substep being carried out before or during the verification step.

[0010] The authentication process may include a step of comparing a field of view of at least one candidate image with a field of view of the reference image, this comparison step being carried out before the verification step during the verification step.

[0011] The field of view comparison step may include a substep of comparing the content distortion of at least one candidate image with the content distortion of the reference image.

[0012] Said at least one candidate image may comprise a series of candidate images comprising at least two successive candidate images; the calculation step may comprise the calculation of a correlation rate between each of the context sub-areas of each candidate image in the series of candidate images and a corresponding reference area; the calculation step may also comprise the calculation of average correlation rate for each sub-area of ​​context, each average correlation rate being calculated by averaging the correlation rates corresponding to the same reference area, and said other comparison step may include comparing each average correlation rate to a minimum threshold.

[0013] The invention also relates to a device for authenticating a candidate image of an environment of a motor vehicle equipped with an image acquisition means, the authentication device comprising hardware and / or software elements implementing the authentication process as defined above.

[0014] The invention also relates to a motor vehicle comprising an authentication device as defined above or comprising hardware and / or software elements implementing the authentication process as defined above.

[0015] The invention also relates to a computer program product comprising program code instructions recorded on a computer-readable medium to implement the steps of the authentication process as defined above when said program is running on a computer.

[0016] The invention also relates to a computer-readable data recording medium on which is recorded a computer program comprising program code instructions for implementing the authentication method as defined above.

[0017] The invention also relates to a signal from a data carrier, carrying the computer program product as defined above.

[0018] It is understood that, in the invention, the image acquisition means is fixed relative to the vehicle, and associated with a field of vision that is also fixed relative to the vehicle.

[0019] These objects, features and advantages of the present invention will be described in detail in the following description of a particular embodiment, given by way of non-limiting example, with reference to the accompanying figures, among which:

[0020] Fig. 1 represents a motor vehicle equipped with an authentication device.

[0021] The [Fig.2] is a flowchart of an execution method of an image authentication process according to the invention.

[0022] Fig. 3 illustrates a reference image.

[0023] The [Fig.4] is a candidate image.

[0024] Fig. 5 is a detail view of the candidate image of Fig. 4 with several proof areas.

[0025] Fig. 6 illustrates the reference image of Fig. 3 with several reference areas.

[0026] Figure 7 illustrates an example of a probationary area placed on a candidate image containing a decoy.

[0027] Fig. 8 illustrates the reference image of Fig. 3 with a reference area.

[0028] Figure 9 is a graph that shows a first example of a rate evolution of correlation.

[0029] Fig. 10 is a graph that shows a second example of an evolution of the correlation rate.

[0030] An example of a motor vehicle 100 equipped with an embodiment of an image authentication device is described below with reference to [Fig.1].

[0031] The motor vehicle 100 can be any type of vehicle, for example, a passenger car, a commercial vehicle, or a public transport vehicle. The motor vehicle 100 includes an image acquisition means 2, for example, a camera, mounted on the motor vehicle 100 and configured to acquire images of an environment external to said motor vehicle. The image acquisition means 2 can, for example, be mounted on the roof of the motor vehicle 100, on the rearview mirrors, on the front of the vehicle, or on the rear, like a reversing camera.

[0032] The image acquisition means 2 can be configured to take several images, for example thirty images per second. However, an image acquisition means 2 with a higher acquisition frequency, for example sixty or even one hundred and twenty images per second, can also be considered.

[0033] Preferably, the image acquisition means 2 is intangible, i.e., its position and settings cannot be changed. A single motor vehicle 100 can be equipped with a multitude of image acquisition means 2 arranged at strategic locations on said vehicle 100.

[0034] The motor vehicle 100 also includes a facial recognition system 4 which enables the implementation of facial recognition processing used to authenticate images acquired by the image acquisition means 2. For this purpose, the facial recognition system 4 includes a memory 6 and a computing unit 7 described later in this document and implementing an authentication method according to the invention.

[0035] Memory 6 advantageously includes information relating to persons authorized to access the vehicle, or persons authorized to use one of the applications secured by facial recognition. The information contained in memory 6 may include the facial biometric scans of the persons authorized to access the vehicle. These scans will be used during identification. from the face of an individual approaching the vehicle from among those authorized to access said vehicle. These fingerprints can be stored in memory 6 following a learning procedure.

[0036] The computing unit 7 includes a microprocessor 71, a local electronic memory 72 and communication interfaces 73 enabling the microprocessor 71 to communicate with the image acquisition means 2, with the application systems set 5 and with the memory 6.

[0037] The vehicle also includes a set 5 of applications secured by facial recognition communicating with the facial recognition system 4. In one embodiment, the set of applications 5 may include, in particular, an application 51 for locking and / or unlocking the doors of the motor vehicle, specifically the door providing access to a driver's seat of the vehicle 100 and / or one or more doors providing access to a passenger compartment of the motor vehicle 100 and / or a tailgate of the motor vehicle 100; and a personalized welcome application 52 using facial recognition of the individual approaching the vehicle and allowing, for example, the configuration of vehicle settings according to the driver's preferences, including adjusting the driver's seat position. Other applications are conceivable within the framework of the set of applications 5.

[0038] The motor vehicle 100, in particular the facial recognition system 4, includes all the hardware and / or software elements configured to implement the method defined in the object of the invention or the method described later in the description.

[0039] An execution method for authenticating a candidate image acquired by the image acquisition means 2 mounted on the motor vehicle 100 is described below with reference to [Fig. 2]. The authentication method is implemented when the vehicle is stationary, in particular when the vehicle is parked in a parking space. As will be seen in more detail later, this provides at least a completely static background, which will be particularly useful for verifying whether a candidate person is actually present in front of the image acquisition means 2. The authentication method comprises a series of steps.

[0040] In a first acquisition step E01, the image acquisition means 2 acquires at least one reference image of an environment outside said motor vehicle 100. The image acquisition means 2 can acquire several reference images under different lighting conditions. An example of a reference image of an outdoor environment is illustrated in [Fig. 3].

[0041] The first acquisition step E01 is followed by a second acquisition step E02 of at least one candidate image of an individual approaching the vehicle. The candidate image(s) are acquired by the image acquisition means 2, under the same conditions as during the first acquisition step E01, with the possible exception of lighting conditions. Figure 4 illustrates an example of a candidate image.

[0042] In a preferred embodiment, the method includes a substep E011 of detecting an individual approaching the motor vehicle 100. In this embodiment, the method proceeds to the second acquisition step E02 of a candidate image when an individual is detected in the vicinity of the motor vehicle 100. In other words, as long as there has not been a detection of an individual approaching the vehicle 100, the method can loop back to the first acquisition step E01 and the image acquisition means 2 then acquires a multitude of reference images.

[0043] In a preferred embodiment, the second acquisition step E02 of a candidate image can be repeated several times, so that the image acquisition means 2 acquires a plurality of candidate images. In this embodiment, each candidate image differs from the reference image in that an individual is present in the environment outside the motor vehicle 100. At least some of the candidate images differ from one another in that the individual's position relative to the environment in which they are moving is different. In other words, the individual's position relative to the environment in which they are moving is different for each candidate image undergoing the image processing described below. These images differ from one another in the distance between the individual and the image acquisition means 2 mounted on the vehicle.

[0044] In this preferred embodiment, the method includes a calculation step E021 of the distance separating the individual from the vehicle 100. This calculation step E021 of the distance separating the individual from the vehicle allows verification of whether the individual present in the candidate images is indeed approaching the vehicle. In particular, if the calculated separation distance tends to decrease chronologically with the acquisition of candidate images, it can be concluded that the individual is approaching the vehicle. Thus, if a repetition of the calculation step E021 of the distance separating the individual from the vehicle 100 indicates a decrease in said distance with each acquisition of candidate images by the image acquisition means 2, this can serve as confirmation of an approach of the individual detected during the aforementioned detection substep E011.

[0045] The process then proceeds to a verification step E03 of the authenticity of the candidate image. This step corresponds to image processing, in particular image processing. Facial recognition. The E03 verification step includes a multitude of sub-steps described later in this description.

[0046] Overall, each candidate image acquired during a second acquisition step E02, or at least a part of these acquired candidate images, is intended to be subject to the verification step E03, in order to confirm or not the authenticity of this or these candidate images.

[0047] For the sake of understanding, the following steps of the process are described as a processing of an individual candidate image, knowing that these steps of the process are applicable to each candidate image acquired, or at least to a part of the candidate images acquired.

[0048] The method includes a recognition step E031, in the candidate image, of a face outline and / or a body outline of an individual approaching the vehicle. To do this, the recognition step E031 may include a substep E0311 of localizing the face and / or body of the individual approaching the vehicle in the candidate image to determine a face outline and a body outline of said individual, as illustrated in [Fig. 4].

[0049] When the face outline and / or body outline of the individual approaching the vehicle is recognized in the candidate image, the process proceeds to a determination step E032 of a set of probative zones ZP of the candidate image.

[0050] In a preferred embodiment, the process proceeds to this determination step E032 of a set of probative zones ZP of the candidate image when the distance separating the individual from the vehicle is less than or equal to a predetermined distance, for example a distance equal to five meters. The determination step E032 is illustrated more particularly in [Fig. 5].

[0051] The ZP probation zones are distributed around the contour of the face and / or the body of the individual approaching the vehicle. The ZP probation zones are distributed such that each zone in the set of ZP probation zones includes both a portion of the face and / or a portion of the body of the individual approaching the vehicle and a portion of the environment outside the motor vehicle. An example of the distribution of ZP probation zones around the contour of the face of the individual approaching the vehicle is illustrated in [Fig. 5].

[0052] Determining a probationary zone ZP includes determining its position, that is, its location in the candidate image, and its geometry, that is, its shape and size. In particular, a probationary zone ZP can be denoted ZP(x,y), where x and y denote the coordinates of the probationary zone ZP in the plane of the candidate image, for example, the coordinates of the center of the probationary zone ZP relative to the center of the candidate image. Thus, each probationary zone ZP is determined, on the one hand, by its position (xi, yi) in the plane of the candidate image, and on the other hand by its geometry. As an example, the geometry of a probationary zone ZP can be determined as being a quadrilateral, in particular a square or a rectangle, having a predetermined length and a predetermined width.

[0053] According to one embodiment, each probationary zone ZP may have its own unique geometry and position. More specifically, each probationary zone ZP may have a different size and shape. Furthermore, the positions and respective sizes of all the probationary zones ZP are such that the probationary zones ZP do not overlap.

[0054] Optionally, the number of probationary zones ZP distributed around the contour of the face and / or the body of the individual in the candidate image increases with a decrease in the distance separating the individual from the vehicle 100. In other words, the closer the individual is to the vehicle 100, the greater the number of probationary zones ZP, up to a predetermined maximum number of probationary zones ZP. The maximum number of probationary zones ZP distributed around the contour of the face is, for example, greater than ten probationary zones ZP, or even greater than fifteen probationary zones ZP.

[0055] According to one example, the number of probationary zones ZP distributed on the contour of the face and / or on the contour of the body of the individual in the candidate image increases linearly with the decrease in distance between the vehicle 100 and the individual approaching said vehicle 100.

[0056] According to another example, the number of probationary zones ZP doubles for each meter less of the distance between vehicle 100 and the individual approaching said vehicle 100.

[0057] Other examples of growth in the number of probationary zones ZP distributed on the contour of the face and / or on the contour of the body of the individual as a function of the decrease in the distance between the vehicle 100 and the individual approaching said vehicle 100 can be imagined.

[0058] Once the proof areas are determined for the candidate image being processed, the process proceeds to a positioning step E033 of a set of reference areas ZR on the reference image. This step is illustrated in [Fig. 6].

[0059] Each reference area ZR of the reference image is associated with a probationary area ZP of the candidate image. More precisely, each reference area ZR of the reference image is paired with a probationary area ZP of the candidate image; that is, for a pair of a reference area ZR paired with a probationary area ZP, the reference area ZR and the probationary area ZP share the same properties in terms of geometry and position in their respective image.

[0060] The reference image and the candidate image are associated with the same two-dimensional coordinate system. Thus, the coordinates xi and yi of a reference zone ZRi in the plane of the reference image are the same, in the plane of the candidate image, as the coordinates xi and yi of the probationary zone ZPi to which the reference zone ZRi is associated.

[0061] The number of reference areas ZR in the reference image is identical to the number of probationary areas ZP in the candidate image. The positions of the reference areas ZR in the reference image are identical to the positions of the probationary areas ZP in the candidate image. For each reference area ZR, its size and shape are respectively identical to the size and shape of the probationary area ZP to which said reference area ZR is associated. Thus, the set of reference areas ZR in the reference image is a reflection of the set of probationary areas ZP in the candidate image.

[0062] The process then proceeds to a selection step E034, in each of the probative zones ZP of the candidate image, of a context sub-zone excluding the portion of the individual's face and / or body. Therefore, it is not the portion of the individual's face and / or body that is subjected to comparison processing to authenticate the candidate image being analyzed, but rather the content of the context sub-zone that excludes the portion of the individual's face and / or body. The content of the context sub-zone represents the environment in which the individual is located.

[0063] Figure 5 illustrates a set of ten probationary zones ZPI to ZP10 defined for the authentication of the candidate image. Each of these ten probationary zones ZPI to ZP10 comprises a context sub-zone: - the ZPI zone includes a portion of a window frame of a building; - Zones ZP2, ZP3 and ZP4 each include a portion of the window frame as well as a portion of the glass of that same window; - Zones ZP5 and ZP6 each include a portion of the first facade of a building; - Zones ZP7, ZP8, ZP9 and ZP10 each include a portion of a first facade of a building and a portion of a second facade of the same building, the second facade being juxtaposed to the first facade of said building.

[0064] In this [Fig. 5], the window frame, the window pane, the first facade of the building, and the second facade of the same building correspond to context elements present in the different context sub-zones. These context elements are all located near the face and / or body of the individual approaching vehicle 100. The position and nature of the context elements in the candidate image are therefore dependent on the position of the individual approaching vehicle 100 in this candidate image. The context elements designate thus elements of the external environment which, in the candidate image, are located near the face and / or body of the individual approaching vehicle 100. In other words, in the candidate image, the contextual elements are contiguous, in other words directly adjacent, to the outline of the face and / or the outline of the body of the individual approaching vehicle 100.

[0065] Generally speaking, urban elements in the environment surrounding the motor vehicle can serve as contextual elements. Contextual elements include, for example, buildings, shop fronts, sidewalks, and so on. Contextual elements can also take the form of road markings, posts, such as signposts, etc. However, natural elements, such as vegetation, including trees, bushes, or floral arrangements, can also serve as contextual elements. The number and nature of the reference elements depend on the external environment in which the motor vehicle is located.

[0066] The process then proceeds to a comparison step E035 of the content of each selected context sub-area in the proof areas ZP of the candidate image with the content of each reference area ZR of the reference image. Thus, for all context sub-areas, the environment in which the individual is located in the candidate image is compared to the environment present in all the reference areas ZR in the reference image. More precisely, the content of the context sub-areas is compared to the content of the reference areas ZR in the reference image.In other words, the E035 comparison step allows us to check if the context elements located at the edge of the face and / or at the edge of the body of the individual in the candidate image are also present in the reference areas ZR in the reference image, and to check if these context elements are consistent with the content of the reference areas ZR in the reference image.

[0067] To achieve this, the process proceeds to a calculation step E036 of a correlation rate between each of the context sub-zones and a corresponding reference zone ZR. Advantageously, each reference zone ZR comprises a contour identical or substantially identical to the contour of the context sub-zone with which it is compared. For each candidate image, as many correlation rates are obtained as there are test zones.

[0068] The spatial correlation rate takes into account the level of similarity between each pixel in a context area and a corresponding pixel in a corresponding reference area (RA). Before this comparison, processing may be performed on the data of one or more of the candidate images to account for the fact that the different candidate images may have been acquired under different lighting conditions.

[0069] In the example of the candidate image in [Fig. 5] and the reference image in [Fig. 6], the contextual elements mentioned above, which are part of the content of the context sub-areas in the proof areas ZP in the candidate image, are also present in the reference areas ZR of the candidate image. The content of the context sub-areas in the candidate image is very similar, if not almost identical, to the content of the reference areas ZR in the reference image. The spatial correlation rates calculated for this example are therefore very high.

[0070] Figure 7 illustrates an example of a candidate image containing a decoy. The probative area placed on this candidate image, and more specifically the context area excluding the portion of the individual's face, certainly includes a portion of the external environment, in this case a portion of a shop window, but this probative area also includes a portion of a hand holding a display screen, as well as a portion of a white background corresponding to a background of an image displayed on said screen.

[0071] In the case of this example, during the comparison step E035, the content of the context sub-area of ​​the probationary area ZP placed on the candidate image illustrated in [Fig.7] is not consistent with the content of the corresponding reference area ZR in the reference image illustrated in [Fig.8], resulting in a low correlation rate during the calculation step E036.

[0072] The accuracy of the correlation rate depends on the size of the test areas ZP and their number. The calculation of each spatial correlation rate can be performed by a neural network, in particular a convolutional neural network.

[0073] The process then proceeds to another comparison step E037 of each correlation rate to a single minimum. To do this, a single CORMIN minimum of the correlation rate between a trial zone ZP and an associated reference zone ZR is defined, such that: - A candidate image will be authenticated if all correlation rates are above the minimum CORMIN threshold, - the candidate image will be rejected otherwise.

[0074] The minimum CORMIN threshold can, for example, be set at 97%, or even 98% or 99%, which corresponds to a very selective treatment. It can also be calibrated to lower values, for example 95% or 90%, if a less selective treatment than that described above is desired.

[0075] Finally, the process proceeds to a conclusion step E038 regarding the authenticity of the content of the candidate image, based on the result of said comparison performed during step E037. The candidate image is authenticated if all the correlation rates or at least most of the correlation rates are above the minimum CORMIN threshold. The candidate image is rejected otherwise.

[0076] As previously mentioned, when an individual approaches the stationary vehicle 100, the image acquisition system can acquire a multitude of candidate images. All of these candidate images, or at least a portion of them, are then subjected to verification step E03. The process is therefore iterative. Repeating the steps of the process in general, and in particular substeps E031 to E038 of verification step E03 for a large number of candidate images, allows for a greater number of correlation rate calculations, thereby increasing the accuracy of the correlation rate and thus its relevance. This increases the probability of detecting inconsistencies between the candidate images and the reference image(s); these inconsistencies are generally indicative of potential fraudulent behavior.

[0077] In particular, if most of the correlation rates calculated for the processed candidate images are high, this means that the environment in which the individual approaching vehicle 100 moves remains identical during the movement of said individual and that this environment corresponds well to that initially recorded at the time of the first acquisition step E01 of the reference image(s).

[0078] By applying the sub-steps of the authenticity verification step E03 to a large number of candidate images, it is possible to follow the evolution of the correlation rate as the individual moves relative to the vehicle 100. In particular, each iteration of the correlation rate calculation step E036, and of the following steps, makes it possible to refine the final result of the verification step E03.

[0079] The evolution of the correlation rate as a function of the distance separating the individual from vehicle 100 can be represented graphically, as shown in Figures 9 and 10. In these figures, the distance (in meters) separating vehicle 100 from the approaching individual is shown on the x-axis, while the y-axis indicates the correlation rate as a percentage. A horizontal dashed line indicates the predefined minimum CORMIN threshold.

[0080] Figure 9 illustrates a first example of the evolution of a correlation rate. In this first example, all calculated correlation rates are above the predefined minimum CORMIN threshold. Note that the calculated correlation rates increase as the distance between the individual and the vehicle decreases; the calculated correlation rates converge towards a maximum correlation rate. The candidate images processed in this first example are all authenticated in the conclusion step E038 of the process. The process then proceeds to a confirmation step E04 of the absence of a decoy, which means that the person nearby the vehicle is authentic, that is to say that it is not an attempt at fraud to access vehicle 100.

[0081] Conversely, [Fig. 10] illustrates a second example of the evolution of a correlation rate, in which all the calculated correlation rates are below the predefined minimum threshold. Note that the calculated correlation rates decrease as the distance between the individual and the vehicle 100 decreases: they converge towards a minimum correlation rate. The candidate images processed in this example are not authenticated but rejected during the conclusion step E038. The process then proceeds to a confirmation step E05 for the presence of a decoy; in other words, step E05 confirms that the person near the vehicle does not appear to be authentic, as in the case of [Fig. 7] where the decoy takes the form of a photograph of an individual's face displayed on a screen.

[0082] Optionally, the method may include a substep E022 for verifying the geometric consistency of the candidate image content. For example, given a candidate image in which the entire individual is shown from head to toe, a computer can estimate the individual's height and distance from vehicle 100. If the estimated height of the individual is not consistent with the individual's distance from vehicle 100, there is a geometric inconsistency in the candidate image, which may be a sign of attempted fraud. In this case, the method can proceed directly to the confirmation step E05 for the presence of a decoy, without necessarily going through the verification step E03 for the authenticity of the candidate image(s).

[0083] In general, the E022 substep for verifying the geometric consistency of the candidate image content allows for checking whether the dimensions of the individual approaching the vehicle 100 and / or their speed of movement and / or the position of said individual relative to their environment are consistent. This optional step allows for the efficient and rapid rejection of a fraudulent image because it can be performed before the E03 verification step of the process.

[0084] In the example of [Fig. 7], the height of the individual's face is inconsistent with their height, which may be pre-recorded in memory 6 of the facial recognition system 4, for example, or with their distance from the vehicle 100. Thanks to the substep E022, which verifies the geometric consistency of the candidate image content, the method can quickly detect a potential attempt at fraud and proceed directly to the confirmation step E05 for the presence of a decoy, without necessarily going through the verification step E03 for the authenticity of the candidate image(s). This optional verification step E022 allows for the efficient and rapid rejection of a fraudulent image because it can be performed before the verification step E03 of the method.

[0085] Alternatively, this optional verification step E022 could also be carried out during the verification step E03. In this case, the conclusion step E038 described above could take into account the result of this step: if a geometric inconsistency is detected in the candidate image then the candidate image would be rejected.

[0086] Optionally, the method may include a comparison step E023 of the field of view of the candidate image with a field of view of the reference image. This comparison step E023 verifies whether the angle of inclination at which the candidate image is obtained is identical to the angle of inclination of the image acquisition means on the vehicle 100 that takes the reference image. If the angles of inclination are identical or at least similar for the candidate image and the reference image, the probability of an attempted fraud is relatively low. If the angle of inclination for the candidate image is significantly different from the angle of inclination for the reference image, the probability of an attempted fraud is higher. In this case, the method can proceed directly to the confirmation step E05 for the presence of a decoy, without necessarily going through the verification step E03 for the authenticity of the candidate image(s).

[0087] Generally, the comparison step E023 of the field of view of the candidate image with a field of view of the reference image can also allow for the comparison of the distortion of said images. To this end, the comparison step E023 can include a substep E0231 for comparing the distortion of the content of the candidate image with the distortion of the content of the reference image.

[0088] If the distortion of the candidate image is identical or at least similar to the distortion of the reference image acquired with the image acquisition means 2 mounted on the vehicle 100, the probability of an attempted fraud is relatively low. If the distortion of the candidate image is significantly different from the distortion of the reference image, this may be because the candidate image was not acquired by the image acquisition means 2 mounted on the vehicle 100, which could be an indicator of an attempted fraud. In this case, the method can proceed directly to the confirmation step E05 for the presence of a decoy, without necessarily going through the verification step E03 for the authenticity of the candidate image(s).

[0089] This optional step makes it possible to reject a fraudulent image efficiently and quickly because it can be carried out upstream of the E03 verification step of the authenticity of the candidate image(s).

[0090] Alternatively, this optional verification step E023 could also be carried out during verification step E03. In this case, the conclusion step E038 described above could take into account the result of this step: if the distortion of the candidate image is significantly different from the distortion of the reference image then the candidate image would be rejected.

[0091] According to one embodiment, steps E031 to E036 can be repeated for a series of successive candidate images. Thus, for each reference area, a series of correlation rates is obtained, each correlation rate being calculated using a candidate image from the series of successive candidate images. Then, the correlation rates corresponding to the same reference area can be averaged. During the comparison step E037, the average of each correlation rate is compared to a minimum threshold. An authentication method based on a series of successive candidate images is more robust than a method based on a single candidate image.For example, if some candidate images were biased by the passage of an object or person in the background of the individual approaching the vehicle, it might still be possible to identify that individual provided there are a sufficient number of candidate images in which the background is not modified.

[0092] The authentication process thus makes it possible to detect decoys and therefore fraud attempts efficiently and cost-effectively. This solution is therefore difficult to circumvent by creating decoys. Furthermore, it is fast, which is beneficial for users when authentication requests are repetitive. In addition to reducing processing time, the invention eliminates the need to purchase licenses for commercially available algorithms.

Claims

1. Demands A method for authenticating a candidate image acquired by an image acquisition means (2) arranged on a motor vehicle (100) to acquire images of an environment external to said motor vehicle (100), the authentication method being implemented when the vehicle (100) is stationary, and characterized in that it comprises the following steps: - a first acquisition step (E01), by means of image acquisition (2), of at least one reference image; followed by - a second acquisition step (E02) of at least one candidate image, by means of image acquisition (2), of an individual approaching the vehicle (100); - a verification step (E03) of the authenticity of at least one candidate image; this verification step (E03) comprising: - a recognition step (E031), in at least one candidate image, of a face outline and / or a body outline of the individual approaching the vehicle (100); - a determination step (E032) of a set of probative zones (PZ) of at least one candidate image, distributed over the contour of the face and / or the contour of the body of the individual approaching the vehicle (100) in at least one candidate image, such that each zone of the set of probative zones (PZ) includes both a portion of the face and / or a portion of the body of the individual approaching the vehicle (100), and a portion of the external environment, - a positioning step (E033) of a set of reference zones (RZ) on the reference image, each reference zone (RZ) of the reference image being associated with a probationary zone (PZ) of at least one candidate image, the positions of the reference zones (RZ) in the reference image being identical to the positions of the probationary zones (PZ) in at least one candidate image, the size and shape of each reference zone (RZ) being respectively identical to the size and shape of the probationary zone (PZ) to which the reference zone (RZ) is associated; - a selection step (E034), in each of the probative areas of at least one candidate image, of a sub-area of ​​context excluding the portion of the face and / or body of the individual; - a comparison step (E035) of the content of each selected context sub-zone in each of the probative zones (PZ) of at least one candidate image, with content from the reference zone (RZ) of the reference image to which the probative zone (PZ) is associated; - a calculation step (E036) of a correlation rate between each of the context sub-zones and a corresponding reference zone (RZ); - another comparison step (E037) of each correlation rate to a minimum threshold; - a conclusion step (E038) regarding the authenticity of the content of at least one candidate image, depending on the result of said comparison; - a confirmation step (E04) of the absence of a decoy; or - a confirmation step (E05) of the presence of a decoy, depending on the result of the verification step (E03).

2. Authentication method according to the preceding claim, characterized in that it comprises a calculation step (E021) of a distance separating the individual from the motor vehicle (100), this calculation step (E021) being carried out after the second acquisition step (E02) of at least one candidate image.

3. Authentication method according to any one of the preceding claims, characterized in that the number of probative zones (PZs) distributed on the contour of the face and / or on the contour of the body of the individual in at least one candidate image increases with a decrease in distance between the vehicle (100) and the individual approaching said vehicle (100).

4. Authentication method according to any one of the preceding claims, characterized in that it comprises a verification substep (E022) of a geometric consistency of the content of at least one candidate image, this verification substep (E022) being carried out before the verification step (E03) or during the verification step (E03).

5. An authentication method according to any one of the preceding claims, characterized in that it comprises a step of comparing (E023) a field of view of at least one candidate image with a field of view of the reference image, this step comparison (E023) being carried out before the verification step (E03) or during the verification step (E03).

6. Authentication method according to the preceding claim, characterized in that the field of view comparison step (E023) includes a substep of comparison (E0231) of the content distortion of at least one candidate image with the content distortion of the reference image.

7. Authentication method according to any one of the preceding claims, characterized in that said at least one candidate image comprises a series of candidate images comprising at least two successive candidate images, in that the calculation step (E036) comprises the calculation of a correlation rate between each of the context sub-zones of each candidate image in the series of candidate images and a corresponding reference zone (RZ), in that the calculation step (E036) also comprises the calculation of average correlation rates for each context sub-zone, each average correlation rate being calculated by averaging the correlation rates corresponding to the same reference zone, and in that said other comparison step (E037) comprises the comparison of each average correlation rate to a minimum threshold.

8. Device for authenticating a candidate image of an environment of a motor vehicle (100) equipped with an image acquisition means (2), the authentication device comprising hardware and / or software elements (4, 5, 6, 7) implementing the authentication method according to one of the preceding claims.

9. Motor vehicle characterized in that it includes an authentication device according to claim 8 or in that it includes hardware and / or software elements (4, 5, 6, 7) implementing the authentication method according to any one of claims 1 to 7.

10. Product computer program comprising program code instructions recorded on a computer-readable medium to implement the steps of the authentication process according to any one of claims 1 to 7 when said program is running on a computer.

Images