A method for selecting a person in an image of a scene

Abstract

A computer-implemented method for selecting a person in an image of a scene. The method takes as input an image of a scene. The method comprises the following steps: - Detect each person present in the image; - Measure, in the image, a test distance between two notable points of each person, said test distance corresponding to a reference biometric quantity; - Calculate the actual depth of each person, from a viewpoint of the image, based on the test distance, the corresponding reference biometric quantity, and at least one image acquisition parameter; - Select one or more persons from among the detected persons according to whether their actual depth in the scene satisfies a localization criterion in said scene.

Description

Title of the invention: Method for selecting a person in an image of a scene. Technical field

[0001] The present invention relates to a method for selecting a person in an image of a scene. It also relates to a method and a biometric identification system implementing such a method. Technical background

[0002] It is common to use individual identification protocols based on comparing certain biometric characteristics to allow access to remote services, verify identity during, for example, border control, or authorize access to a restricted area. For this purpose, the user generally presents themselves to an acquisition system that acquires an image from which certain relevant biometric characteristics are extracted during subsequent processing. The use of an identification protocol generally requires a preliminary enrollment step whereby a user registers with the entity implementing the protocol by sharing a certain amount of biometric and identity-related information.

[0003] Examples of biometric identification systems and methods using facial recognition are described in Xinyi, et al. "A survey of face recognition." arXiv preprint arXiv:2212.13038 (2022); US 2017 / 0330028 Al [MorphoTrack LLC] 16.11.2017; EP 3 285 209 A2 [Safran Identity and Security SAS] 21.02.2018.

[0004] Examples of a biometric identification system and method by pedestrian posture (“pose”) and / or gait (“gait”) recognition are described in Ye, Mang, et al. "Deep learning for person re-identification: A survey and outlook." IEEE transactions on pattern analysis and machine intelligence 44.6 (2021) 2872-2893; US 2017 / 0316255 Al [Panasonic Intellectual Property Management Co Ltd Wang] 02.11.2017; WO 2019 / 188111 Al [NEC CORP] 03.10.2019; US 2015 / 0193686 [Tata Consultancy Services Ltd] 09.07.2015.

[0005] In locations accommodating a multitude of individuals, such as airports or train stations, data acquisition systems for enrollment and identification systems, particularly facial recognition systems, are likely to acquire the biometric information of other people located near the user wishing to enroll or identify themselves. In particular, free-flow acquisition systems have a wide field of view so as to be able to acquire the biometric information of a user located in a given area. If several people are if they happen to be in this area at the same time as the user, their biometric characteristics may be acquired without their consent.

[0006] There are digital image processing methods that allow for the "removal" or "erasure" of third parties. However, these methods rely on prior, and not entirely optimal, detection of individuals within the image. All or part of their biometric characteristics may remain accessible. Therefore, the preservation of their anonymity is not guaranteed.

[0007] It is also possible to configure the acquisition systems so that their optical elements are focused solely on one or more areas dedicated to biometric identification. However, such a solution requires dedicated acquisition systems that cannot be used for other purposes. Furthermore, it does not allow the use of existing acquisition systems, such as monocular surveillance cameras, already installed in an access-controlled area. Moreover, over time, depending on the needs of the staff operating the access-controlled area, the size and geometry of the areas dedicated to identification may change. New adaptations and adjustments are then necessary for these acquisition systems.Finally, because the acquisition systems are spatially arranged in such a way as to be able to acquire an individual's biometric characteristics, they inevitably acquire images of their lateral and rear surroundings.

[0008] Another possible alternative is to evaluate the position of people in a scene from a depth map of the scene and to select the people located in a region of interest on said depth map. An example is described in EP 3 866 064 Al [IDEMIA IDENTITY & SECURITY FRANCE [FR]] 18.08.2021. However, a depth map of a scene requires either the use of specific acquisition systems such as stereoscopic cameras, time-of-flight cameras, or structured-light optical scanners, or the implementation of image processing based on convolutional neural networks, such as the one described by Eigen, D. et al. (2014). Depth map prediction from a single image using a multi-scale deep network. Advances in neural information processing Systems, 27, requiring prior training and significant computing resources.

[0009] There is therefore a need for a simple solution enabling the precise and real-time selection of individuals in a scene in order to allow the acquisition and subsequent processing of their biometric characteristics, provided they have given their consent. Advantageously, such a solution would preserve the confidentiality of the biometric characteristics as well as the anonymity of the individuals. who are likely not to consent to the acquisition and processing of their biometric characteristics. Summary of the invention

[0010] In a first aspect of the invention, a computer-implemented method 600 is provided for selecting a person in an image of a scene. The method takes, as input data, an image of a scene. The method comprises the following steps: - Detect 601 each person present in the image; -Measure 602, in the image, a test distance between two notable points of each person, said test distance corresponding to a biometric reference quantity; - Calculate 603 the actual depth of each person, from an image viewpoint, from the test distance, the corresponding biometric reference size, and at least one image acquisition parameter; - Select 604 one or more people from among the people detected depending on whether their actual depth in the scene satisfies a location criterion in said scene.

[0011] According to some embodiments, the reference biometric size is identical for all persons detected in step 601.

[0012] According to some embodiments, the reference biometric size is different between the persons detected and / or groups of persons detected in step 601.

[0013] According to some embodiments, the reference biometric size has a constant value for all persons detected in step 601.

[0014] According to certain embodiments, the value of the reference biometric size varies according to at least one morphological, biological and / or physiological criterion relating to each person detected.

[0015] According to some embodiments, the method 600 further includes, before step 604, a step 603a of determining the coordinates of the actual position of each person in the scene from its actual depth calculated for each person, and, in step 604, one or more persons among the detected persons are selected according to whether the coordinates of their actual position in the scene satisfy a localization criterion in said scene.

[0016] According to some embodiments, the process 600 further includes, before step 603, a step 602a of estimating the orientation of each person present in the image and a step 602b of correcting the test distance of each person from their estimated orientation.

[0017] According to some embodiments, the location criterion is a first defined area of ​​the scene, and the people located in said area are selected.

[0018] According to some embodiments, a second zone is defined around the first zone, and the people, among the detected people, located between the first zone and the second zone or outside the second zone are not selected.

[0019] According to some embodiments, the process takes as input data a plurality of images of the scene, each of the steps being executed on each of the images, and includes, in addition, before step 604, a step 603b of predicting the trajectory of each of the people detected through the plurality of images (400), and at step 604, the detected people are selected if their predicted trajectory crosses the boundary of the first zone.

[0020] According to some embodiments, the reference biometric size is chosen from the size of the face, the general morphological posture of the individual, the length of a limb, the interpupillary distance or the distance between the center of an eye and a corner of the mouth.

[0021] In a second aspect of the invention, a data processing device is provided comprising means for implementing a method 600 according to any embodiment of the first aspect of the invention.

[0022] In a third aspect of the invention, a computer program is provided comprising instructions which, when the program is executed by a data processing device, cause the latter to implement a method 600 according to any one embodiment of the first aspect of the invention.

[0023] In a fourth aspect of the invention, a biometric identification method is provided comprising the following steps: (a) Acquire an image of a scene using an acquisition device, preferably using a monocular camera; (b) Selecting one or more persons present in the scene using a method 600 according to any embodiment of the first aspect of the invention; (c) Acquire, for each selected person, a biometric characteristic; (d) Identify each selected person based on the acquired biometric characteristic.

[0024] In a fifth aspect of the invention, a biometric identification system is provided comprising: - an acquisition device, preferably a monocular camera, configured to acquire an image of a scene; - a data processing device according to the second aspect of the invention comprising, in addition, means for implementing a biometric identification process according to the fourth aspect of the invention.

[0025] A first notable advantage of the invention is that it makes it possible to estimate, in real time and with reasonable accuracy, the actual depth of each person present in a scene from a single image of said scene. The depth is estimated from a viewpoint of said image, in particular from the viewpoint of the acquisition device that acquired said image.

[0026] Another notable advantage of the invention is that it is suitable for images obtained with a monocular camera such as a surveillance monocular camera. It can therefore be directly used with surveillance devices already installed in a scene, for example in an access control area.

[0027] Another advantage of the invention is that the selection is effective even for people in the scene who do not, voluntarily or involuntarily, facilitate the acquisition of their biometric characteristics. For example, the method according to the invention is capable of selecting people who are not looking at the camera and / or are wearing eyeglasses and / or a face mask. Brief description of the drawings

[0028] [Fig.1] is a schematic representation of an example scene in the form of a border control zone.

[0029] [Fig.2] is a schematic representation of a biometric identification system for identifying people in a scene.

[0030] [Fig.3] is a schematic representation of a data processing device such as a biometric processing device.

[0031] [Fig.4] is a schematic representation of an image capable of being acquired by an acquisition device of a biometric identification system.

[0032] [Fig.5] is a schematic representation of an example of measuring a test distance between the centers of the pupils of a person's eyes.

[0033] [Fig.6] is a flowchart of a process for selecting people in a scene.

[0034] [Fig.7] is a schematic representation of the trajectories of people moving in and near an area in a scene. Detailed description of the implementation methods

[0035] In this disclosure, embodiments are described in the general context of one or more hardware or devices capable of executing preloaded instructions such as, for example, computer-executable instructions for executing program modules. The program modules can include one or more routines, programs, objects, variables, commands, scripts, functions, applications, components, data structures that can perform particular tasks or implement particular types of abstract data.

[0036] Certain embodiments can also be implemented in distributed computing environments where tasks are performed by remote data processing devices connected by a communication network. In a distributed computing environment, program modules may reside on local and / or remote computer storage media, including memory storage devices.

[0037] For the purposes of this disclosure, "scene" means a space in which one or more individuals are engaged in various activities. In particular, it includes places open to the public, such as airports, train stations, theaters, event venues, and businesses, where a biometric identification system may be implemented to control access.

[0038] With reference to [Fig. 1], by way of illustration, a scene 100 can be a border control zone 100, such as an airport is likely to have. In this control zone 100, travelers 101-1...101-6 generally have the opportunity to identify themselves using a free-flow biometric identification system to access the boarding gates 106, when departing from the territory, or the disembarking gates, when arriving in the territory.

[0039] The control zone 100 then includes a biometric identification consent zone 102 equipped with a device 103 for acquiring and / or recording tests of the anthropomorphic and / or anthropometric characteristics of the travelers 101 in order to analyze them. In the example of a control zone shown in [Fig. 1], the acquisition device 103 is a single aerial monocular camera oriented at a downward viewing angle above the consent zone 102. The single acquisition device 103 acquires one or more tests of the faces of the travelers 101-1, 101-2 who have previously completed their enrollment, for example during a prior registration step (“check-in”).The tests are then transmitted to a biometric processing device (not shown) which extracts biometric test templates to be compared to reference biometric templates in a database and to identify travelers based on this comparison.

[0040] With reference to [Fig. 2], a biometric identification system 200 that can be used for implementing free-flow identification in scene 100 of [Fig. 1] can be a facial and / or pedestrian recognition system. The System 200 comprises a biometric processing device 201 and a video and / or photographic recording device 103, such as a monocular camera. For security reasons, the biometric processing device 201 is generally located away from the video and / or photographic recording device 103, usually in a dedicated room. It communicates with the video recording device 103 via any suitable wired or electromagnetic telecommunication device.

[0041] A videographic and / or photographic recording device 103, such as a monocular camera, is defined by its optical characteristics, in particular its field of view (DOF), which includes a vertical and a horizontal component, and by its depth of field (DOF). The recording device 103 is configured to cover the entire consent area 102 and to record an image of an area 202 of interest of an individual 101-2, for example, their face, from which certain relevant anthropomorphic and / or anthropometric characteristics of the individual 101-2 can be extracted and converted into a biometric proof template by the biometric processing device 201. The biometric processing device 201 then compares the biometric proof template with one or more reference biometric templates from a database.If there is a match between the biometric template being tested and at least one reference biometric template associated with an individual identified in the database, individual 101-2 is considered identified by the identification system. Otherwise, individual 101-2 is not considered identified and must then undergo manual verification by a human agent.

[0042] The area of ​​interest 202 of individual 101-2 depends on the type of recognition implemented by the biometric identification system 200. In the case of facial recognition, the area of ​​interest 202 includes the face and / or one or more characteristic facial features, such as the eyes, the corners of the mouth, or the nose. In the case of pedestrian recognition, the area of ​​interest 202 includes at least the upper part of individual 101-2's body, or even their entire body, and the biometric processing device 201 extracts a biometric template from an analysis of individual 101-2's posture, build, and / or gait.

[0043] With reference to [Fig. 3], a biometric processing device 201 is generally a data processing device 300 comprising means for implementing a biometric identification method as described above. This device 300 comprises one or more central processing units (CPUs) 301 and / or one or more graphics processing units (GPUs) 302, a physical remote communication module 303, one or more physical input / output modules 304 for data exchange with external devices, a transient storage medium 305 such as random access memory (RAM), a non-transient recording medium 306, and communication buses (not shown) for data transfer between internal components of the device 300. It may also include a secure element 308 for the storage of cryptographic keys, the execution of encryption algorithms, and / or the storage and / or encryption of any other algorithm and / or data whose security and confidentiality must be preserved, for example, a database of reference biometric templates.

[0044] The biometric processing device 201 allows the execution of one or more program modules comprising instructions which, when the program module(s) are executed, cause the biometric processing device 201 to implement a biometric identification method as described above. The program module(s) may be written in any programming language, compiled or interpreted. They may be part of a software solution, i.e., a collection of executable instructions, code, scripts, or other components, and / or databases.

[0045] In this example, the consent zone 102 is an area delimited by floor markings 105 to facilitate its identification by travelers 101-1...101-6. This zone may also be delimited by any other appropriate means, for example, physical barriers such as retractable belt barriers. Depending on the configuration of the control zone 100, the consent zone 102 may be a transit area, a waiting area, or any other area identifiable for these purposes. Travelers 101-1...101-6 located within the consent zone 102 are considered to have consented to biometric identification. Travelers 101-1...101-6 located outside this zone 102 are considered to have not consented to biometric identification.

[0046] As explained previously, with reference to [Fig. 1] and [Fig. 2], the acquisition device 103, being configured to cover the entire consent zone 102, can also cover part of its immediate vicinity 104 in which a traveler 101-4 is likely to be located. In such a configuration, with reference to [Fig. 4], an image 400 acquired by an acquisition device 103 can then include one or more travelers 401-1, 401-2, 401-3 present in the consent zone 102 and one or more travelers present in the vicinity 104 of this zone. The anthropomorphic and / or anthropometric characteristics of this traveler 401-4, 401-5 are then likely to be subject to biometric processing even though they have not consented to it.

[0047] In a first aspect of the invention, with reference to [Fig.4], [Fig.5] & [Fig.6], a computer-implemented method 600 is provided for selecting a person in an image of a scene 100, the method 600 takes, as input data 1600, an image 400 of a scene 100, said method 600 comprises the following steps: - Detect 601 each person 401-1...401-5 present in the image 400; -Measure 602, in image 400, a test distance DE between two notable points PI, P2 of each person 401-1...401-5, said test distance DE corresponding to a reference biometric GBR quantity; - Calculate 603 the actual depth DP of each person 401-1...401-5 from a viewpoint of image 400 from the test distance DE, the corresponding biometric reference GBR quantity, and at least one image acquisition parameter 400; - Select 604 one or more persons 401-1...401-3 from among the persons detected 401-1...401-5 depending on whether their actual depth DP in scene 100 satisfies a criterion 102 of localization in said scene 100.

[0048] At step 601, each person 401-1...401-5 present in the image 400 is detected by any suitable means, in particular an object detection algorithm in an image. Examples of such algorithms are described in Carion, Nicolas, et al. (2020) "End-to-end object detection with transformers." European Conference on Computer Vision. Cham: Springer International Publishing, and Li, Chuyi, et al. (2022) "YOLOvô: A single-stage object detection framework for industrial applications." arXiv preprint arXiv:2209.02976.

[0049] In step 602, a test distance DE corresponding to a reference biometric quantity GBR is measured between two notable points PI, P2 on the image 400. The accuracy of the measurement of the test distance DE generally depends on the resolution of the image 400. Preferably, the image scale is at most 3 mm per pixel, or even at most 1 mm per pixel.

[0050] The reference biometric size GBR to which the test distance DE corresponds can be any morphological reference distance representative of a person or a category of persons.

[0051] By way of example, the reference biometric size GBR may be the size of the face, the general morphological posture of the person, the length of a limb such as an arm, leg or torso, the inter-pupillary distance (“inter eye distance” -IED), or the distance between the center of an eye and a corner of the mouth.

[0052] According to some embodiments, the reference GBR biometric value may be [R2] identical for all persons detected 401-1...401-5 in step 601. For example, if the chosen reference GRB biometric value is the interpupillary distance, it is used for all persons detected.

[0053] According to certain embodiments, the reference GRB biometric measurement differs between individuals detected 401-1...401-5 and / or groups of individuals detected 401-1...401-5 in step 601. In a first example, the reference GRB biometric measurement may be the interpupillary distance for one detected individual and the length of a limb for another detected individual. In a second example, the reference GRB biometric measurement may be the interpupillary distance for a first group of detected individuals and the length of a limb for a second group of detected individuals. In a third example, the reference GRB biometric measurement may be the interpupillary distance for one detected individual and the length of a limb for a group of other detected individuals.

[0054] According to some embodiments, the reference GBR biometric measurement has a constant value for all individuals detected in step 601. It has been found that even when the value of the reference GBR biometric measurement, for example, limb length, chosen for an individual is quite far from their actual value for that individual, the resulting error in the actual depth calculated in step 603 remains acceptable, or even negligible. In other words, the error in the calculated actual depth is negligible even if the value chosen for the reference GBR biometric measurement of the individual in question is not accurate for that individual.In practice, it has been found that a standard deviation of approximately 10% in the value of the reference biometric measurement is generally acceptable because it generates negligible uncertainty in the person's actual position; this uncertainty is estimated to be on the order of a few centimeters to a few tens of centimeters. The method according to the invention is therefore robust with respect to inaccuracies that may affect the chosen value for the reference GBR biometric measurement for individuals detected in step 601.

[0055] With reference to [Fig. 5], according to a first example, the two notable points P1, P2 are the pupil centers of the right and left eyes, and the measured DE test distance is the distance between the pupil centers. This DE test distance corresponds to the interpupillary distance. The pupil centers are notable points that can be detected using any suitable method. Examples of feature point detection algorithms are described in Zhu, X., Lei, Z., Liu, X., Shi, H., & Li, SZ (2016). Face alignment across large poses: A 3d solution. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 146-155), and Xu, Z., Li, B., Yuan, Y., & Geng, M. (2021). Anchorface: An anchor-based facial landmark detector across large poses. In Proceedings of the AAAI conference on artificial intelligence (Vol. 35, No. 4, pp. 3092-3100.

[0056] The reference GBR biometric measurement corresponding to this first example of the DE test distance between the centers of the two eyes is the interpupillary distance. This biometric measurement varies relatively little between individuals and can be considered constant without affecting the accuracy of the actual depth for each person calculated in step 603. The average interpupillary distance for the entire population is 62 mm.

[0057] According to certain embodiments, the value of the reference biometric size GBR varies according to at least one morphological, biological, and / or physiological criterion specific to each person, including the biological sex and / or age of the person in question. An example of a method for assessing the biological sex and / or age of a person is described in Kuprashevich, M., & Tolstykh, I. (2023). Mivolo: Multi-input transformer for age and gender estimation. In International Conference on Analysis of Images, Social Networks and Texts (pp. 212-226). Cham: Springer Nature Switzerland.

[0058] In the first example of the DE test distance in [Fig. 5], instead of a constant value for all individuals, the reference GBR biometric measurement can vary within a range of values ​​depending on the category to which the individual belongs. For example, when the morphological, biological, and / or physiological criterion corresponds to the biological sex of an adult population, the reference GBR biometric measurement value for the interpupillary distance can be chosen to be between 51 mm and 74.5 mm for the female biological sex and between 53 mm and 77 mm for the male biological sex.When the morphological, biological, and / or physiological criterion corresponds to age, the reference GBR biometric value for interpupillary distance can be chosen: an average value of 53 mm for a child aged between 5 and 10 years, 58 mm for an adolescent aged between 11 and 20 years, and 62 mm for an adult. It is possible to combine several morphological, biological, and / or physiological criteria, for example, age and gender.

[0059] According to a second example, the test distance DE is a measure of the person's height, and the reference biometric size GBR is a reference height for that person. When the population of people moving within scene 100 is relatively homogeneous in height, it is possible to use the same reference height value for all people in accordance with certain embodiments described above. For example, for a population of people who are predominantly European, a reference height could be 180 cm for males and 167 cm for females.

[0060] In step 603, the actual depth of each person 401-1...401-5 from a viewpoint of image 400 is calculated from the proof distance DE, of the corresponding biometric reference GBR size, and at least one image acquisition parameter 400.

[0061] The viewpoint of the image 400 is generally the viewpoint of the acquisition device 103 that acquired said image 400. The actual depth DP of each person 401-1...401-5 then corresponds to the actual distance DP between the objective lens of the acquisition device 103 and said person 401-1...401-5. In other words, with reference to Fig. 2, the actual depth DP is the distance, along the Zc axis, of a person 101-2 from the plane (Xo Yc) in the orthonormal coordinate system (O, Xc Yc Zc) of the acquisition device 103.

[0062] Each person 401-1...401-5 can be represented geometrically as a set of points (Xe yc, Zc) in the orthonormal frame (O, Xc, Yc Zc) of the acquisition device 103. In the context of the applications of the present invention, the persons 401-1...401-5 being generally standing in a vertical position, a real depth DP calculated between any one of the points of a person 401-1...401-5 is considered to be representative of the real depth of the whole person 401-1...401-5.

[0063] The image acquisition parameter 400 corresponds to the calibration parameters of the acquisition device 103 as a function of the model used to model it, for example the pinhole model.

[0064] According to an example, for an image 400 acquired by an acquisition device 103 in the form of a monocular camera whose optical arrangement is modeled according to a pinhole model, the actual depth DP of each person 401-1...401-5 can be calculated using the following formula: ii — ii — II where DE is the test distance measured on each | DE | person in image 400, GBR is the biometric reference quantity corresponding to the measured DE test distance and f is the focal length of the monocular camera according to the pinhole model.

[0065] In step 604, one or more persons 401-1...401-3 from among the detected persons 401-1...401-5 are selected according to whether their actual depth DP in scene 100 satisfies a location criterion 102 in said scene 100. The location criterion 102 depends on the configuration of scene 100 and the location in scene 100 where the persons are selected. In the embodiment shown in [Fig. 1], [Fig. 2] & [Fig. 4], the location criterion 102 is a first defined zone 102 of scene 100, called the "biometric identification consent zone," and the persons 401-1...401-3 located in said zone 102 are selected. This zone 102 is delimited by a marking, for example, a marking on the floor. It has usually a simple geometric shape such as a square, a rectangle, a circle or an oval.

[0066] Depending on the geometric configuration of the consent zone 102, the spatial arrangement of the acquisition device 103, and the spatial distribution of the persons 401-1... 401-5, knowledge of the depth DP for each person alone may be sufficient for the purposes of the invention. For example, when the consent zone 102 is in the form of a queue or access corridor, and the acquisition device 103 is facing the persons using it, its narrowness forces said persons to arrange themselves in rows one behind the other, generally in the form of a queue of one or two people per row.

[0067] In other circumstances, particularly where the density of people is higher and they move more freely, knowing the depth DP for each person alone may not be sufficient for the purposes of the invention. If the consent zone 102 is part of an open environment and the spatial distribution of people is more extensive, such as scene 100 illustrated in [Fig. 1], additional knowledge of each person's actual position in scene 100 may be advantageous for improving the accuracy of selecting people in said zone 102 and avoiding selecting people in its vicinity 104.

[0068] According to some embodiments, the method 600 further includes, before step 604, a step 603a of determining the coordinates of the actual position of each person 401-1...401-5 in the scene 100 from its actual depth DP calculated for each person 401-1...401-5, and, in step 604, one or more persons 401-1...401-3 from among the detected persons 401-1...401-5 are selected according to whether the coordinates of their actual position in the scene 100 satisfy a localization criterion 102 in said scene 100.

[0069] According to the preceding example, in which an image 400 is acquired by an acquisition device 103 in the form of a monocular camera whose optical arrangement is modeled according to a pinhole model, each person can be represented, with reference to [Fig. 2], by one or more real coordinates (X^J, Zc) in the orthonormal frame (O, Xc, Yc, Zc) of the acquisition device 103, and, with reference to [Fig. 4], by one or more image coordinates (¾, y) in the orthonormal frame (U, X, Yi) of the image 400. The real coordinates (x^Zc) and the image coordinates (x^Zc) are related by the following relation: / Xc\ IGBR i [X{\where DE is the test distance measured on For each person in image 400, GBR is the reference biometric size corresponding to the measured test distance DE and f is the focal length of the monocular camera according to the pinhole model.

[0070] Solving the preceding relation allows us to determine the actual coordinates (X^y) of each person 401-1...401-5 in the orthonormal frame (O, Xc, Yc, Zc) of the acquisition device 103, and thus the coordinate(s) (X^, y, Zc) of their actual position in the scene 100. The consent zone 102, generally having a simple geometric shape such as a square or a rectangle, can also be represented by a set of coordinates (Xc, y^c, Z^) in the orthonormal frame (O, Xc, Y^, Zc) of the acquisition device 103. By comparing the coordinates (X^y, Zc) of each person 401-1...401-5 with the set of coordinates (X^c, y^, Z^) of the consent zone 102, we can determine the actual position of each person relative to the consent zone 102. to know whether it is located inside or outside said area.

[0071] Alternatively and equivalently, it is possible to represent each person 401-1...401-5 and the consent zone 102 in an orthonormal frame of the scene 100 (not shown). A transformation matrix, called the extrinsic parameter matrix of the acquisition device 103, composed of a rotation matrix and a translation matrix, can then be determined to transform the coordinates in the orthonormal frame (O, Xc, Y^Z,. ) of the acquisition device 103 into coordinates in the orthonormal frame of the scene 100.

[0072] Depending on the arrangement of the acquisition device 103 in the scene 100, and the position and behavior of the persons 401-1...401-5 in the scene 100, the orientation of each person 401-1...401-5 relative to the acquisition device 103 may differ and lead to perspective effects that may be advantageous to consider when measuring the test distance. In particular, for facial recognition identification systems, facial orientation is taken into account for determining facial biometric characteristics. In the example of a test distance measurement corresponding to the interpupillary distance, the position of the head relative to the acquisition device 103, particularly through its yaw angle, may, due to perspective effects, cause this distance to appear reduced in the image 400.

[0073] Thus, according to advantageous embodiments, the process 600 further includes, before step 603, a step 602a of estimating the orientation of each person 401-1...401-5 present in the image 400 and a step 602b of correcting the proof distance DE of each person 401-1...401-5 from its estimated orientation.

[0074] Examples of algorithms for estimating the orientation of a face in an image are described in Hempel, T., et al. (2024) Toward Robust and Unconstrained Full Range of Rotation Head Pose Estimation. IEEE Transactions on Image Processing, 33, 2377-2387 and Albiero, V., et al. (2021). img2pose: Face alignment and detection via 6dof, face pose estimation. In Proceedings of the IEEE / CVF conference on computer vision and pattern recognition (pp. 7617-7627).

[0075] With reference to [Fig.1], [Fig.2] & [Fig.4], according to certain advantageous embodiments, particularly for applications requiring a very low level of uncertainty as to the persons selected, a second zone 107, called the "safety zone", is defined around the consent zone 102, and, at step 604, persons 401-4, 401-5, among the detected persons 401-1...401-5, located between the consent zone 102 and the safety zone 107 or outside the safety zone 107 are not selected.

[0076] In the example shown in [Fig. 1], [Fig. 2], and [Fig. 4], the safety zone 107 comprises the area 104 surrounding the consent zone 102. It is represented by a dashed line. In practice, the boundary of the safety zone 107 may be an imaginary line that is not physically marked by any floor markings or a set of barriers. Persons 101-1... 101-6 / 401-1... 401-5 moving within scene 100 are then unaware of its presence. In other situations, the security zone 107 may be marked to warn individuals (101-1...101-6 / 401-1...401-5) moving within scene 100 that they are approaching a consent zone 102 for biometric identification. However, anyone present, even inadvertently, between the consent zone 102 and the security zone 107 is not selected and is not subject to biometric identification.

[0077] One advantage of implementing a security zone 107 is to limit the risks of selecting people in the vicinity 104 of the consent zone 102, particularly when the reference biometric size GBR used for estimating their position is likely to suffer from significant variability between people.

[0078] Occasionally, some 401-5 individuals who intend to consent to biometric identification do not cross the boundary of the consent zone 102 to enter it. On the contrary, they remain, generally due to inattention, near the boundary in the vicinity 104 of the consent zone 102. Being able to predict individuals' intentions can therefore prove advantageous in order to discriminate between, on the one hand, those 401-5 individuals who consent to their biometric identification but who have not "truly" entered the consent zone 102 and, on the other hand, those 401-4 individuals who do not consent to their identification. biometric, but, circulating near the consent zone 102, risk, due to possible uncertainties in estimating their position, being subject to biometric identification against their will.

[0079] Also, according to certain embodiments, the process 600 takes, as input data, a plurality of images 400 of the scene 100, each of the steps 601, 602, 603 being executed on each of the images 400, the process 600 further includes, before step 604, a step 603b of predicting the trajectory of each of the persons detected through the plurality of images 400, and at step 604, the persons detected are selected if their predicted trajectory crosses the boundary of the first zone 102, called the "zone of consent".

[0080] According to an example, with reference to [Fig. 7], by performing each of steps 601, 602, 603 on each of the images 400 of a plurality of consecutive images, it is then possible, in step 603b, to predict the trajectories T1, T2 of two people moving towards and / or near the first consent zone 102. In step 603b, the trajectory prediction can be estimated using any suitable method from the positions determined after performing steps 601, 602, 603 on each of the images 400.

[0081] The predicted trajectory T1 of the first person leads towards the consent zone 102 and crosses its boundary. Their intention is considered to be consenting to the acquisition of their biometric characteristics, and they are therefore selected accordingly. In contrast, according to the predicted trajectory T2, the second person moves near the consent zone 102, without crossing its boundary. They remain in the vicinity 104, between the consent zone 102 and the security zone 107. The second person is then considered undecided and therefore not consenting to the acquisition of their biometric characteristics. They are therefore not selected.

[0082] In a second aspect of the invention, with reference to [Fig. 3], a data processing device 300 is provided, comprising means for implementing a method 600 according to any embodiment of the first aspect of the invention. Preferably, with reference to [Fig. 2], the data processing device 300 is in the form of a biometric processing device 201.

[0083] In a third aspect of the invention, a computer program is provided comprising instructions which, when the program is executed by a data processing device 300, 201, cause the latter to implement a method 600 according to any one embodiment of the first aspect of the invention.

[0084] In a fourth aspect of the invention, a biometric identification method is provided comprising the following steps: (a) Acquire an image 400 of a scene 100 using an acquisition device 103, preferably using a monocular camera; (b) Selecting one or more persons 401-1...401-3 present in scene 100 using a method 600 according to any embodiment of the first aspect of the invention; (c) Acquire, for each person selected 401-1...401-3, a biometric characteristic; (d) Identify each selected person 401-1...401-3 based on the acquired biometric characteristic.

[0085] In a fifth aspect of the invention, a biometric identification system 200 is provided, comprising: - an acquisition device 103, preferably a monocular camera, configured to acquire an image 400 of a scene 100; - a 201, 300 data processing device according to the second aspect of the invention comprising, in addition, means for implementing a biometric identification process according to the fourth aspect of the invention. References Literature patent

[0086] US 2015 / 0193686 [Tata Consultancy Services Ltd] 09.07.2015.

[0087] US 2017 / 0330028 Al [MorphoTrack LLC] 16.11.2017.

[0088] US 2017 / 0316255 Al [Panasonic Intellectual Property Management Co Ltd Wang] 02.11.2017.

[0089] EP 3 285 209 A2 [Safran Identity and Security SAS] 02.21.2018.

[0090] WO 2019 / 188111 Al [NECCORP] 03.10.2019.

[0091] EP 3 866 064 Al [IDEMIA IDENTITY & SECURITY FRANCE [FR]] 08.18.2021. Non-patent literature

[0092] Eigen, D. et al. (2014). Depth map prediction from a single image using a multi-scale deep network. Advances in neural information processing Systems, 27.

[0093] Zhu, X., Lei, Z., Liu, X., Shi, H., & Li, S. Z. (2016). Face alignment across large poses: A 3d solution. In Proceedings ofthe IEEE conférence on computer vision and pattern récognition (pp. 146-155).

[0094] Carion, Nicolas, et al. (2020) "End-to-end object détection with transformers." European conférence on computer vision. Cham: Springer International Publishing.

[0095] Xu, Z., Li, B., Yuan, Y., & Geng, M. (2021). AnchorFace: An anchor-based facial landmark detector across large poses. In Proceedings ofthe AAAI conférence on artificial intelligence (Vol. 35, No. 4, pp. 3092-3100).

[0096] Albiero, V., et al. (2021). img2pose: Face alignment and détection via 6dof, face pose estimation. In Proceedings ofthe IEEE / CVF conférence on computer vision and pattern récognition (pp. 7617-7627).

[0097] Ye, Mang, et al. "Deep learning for person re-identification: A survey and outlook." IEEE transactions on pattern analysis and machine intelligence 44.6 (2021) 2872-2893.

[0098] Li, Chuyi, et al. (2022) "YOLOvô: A single-stage object détection framework for industrial applications." arXivpreprint arXiv:2209.02976.

[0099] Xinyi, et al. "A survey of face récognition." arXiv preprint arXiv:2212.13038 (2022).

[0100] Kuprashevich, M., & Tolstykh, I. (2023). Mivolo: Multi-input transformer for âge and gender estimation. In International Conférence on Analysis of Images, Social Networks and Texts (pp. 212-226). Cham: Springer Nature Switzerland.

[0101] Hempel, T., et al. (2024) Toward Robust and Unconstrained Full Range of Rotation Head Pose Estimation. IEEE Transactions on Image Processing, 33, 2377-2387.

Claims

Demands

1. A computer-implemented method 600 for selecting a person in an image of a scene (100), the method (600) takes as input data (1600) an image (400) of a scene (100), said method (600) comprises the following steps: - Detect 601 each person (401-1...401-5) in the image (400); - Measure 602, in the image (400), a test distance DE between two notable points PI, P2 of each person (401-1...401-5), said test distance DE corresponding to a reference biometric GBR quantity; - Calculate 603 the actual depth DP of each person (401-1... 401-5) from an image viewpoint (400) based on the test distance DE, the corresponding biometric reference GBR quantity, and at least one image acquisition parameter (400); - Select 604 one or more people (401-1... 401-3G) from among the detected people (401-1...401-5) depending on whether its actual depth DP in the scene (100) satisfies a criterion (102) of localization in said scene (100).

2. Method 600 according to claim 1, wherein the reference biometric GBR size is identical for all persons detected in step 60).

3. Method 600 according to claim 1, wherein the reference biometric GBR size is different between individuals detected (401-1...401-5) and / or groups of individuals detected (401-1...401-5) at step (601).

4. Method 600 according to any one of claims 2 to 3, wherein the reference biometric GBR quantity has a constant value for all persons detected (401-1...401-5) at step (601).

5. Method 600 according to any one of claims 2 to 3, wherein the value of the reference biometric GBR varies according to at least one morphological, biological and / or physiological criterion relating to each person detected (401-1...401-5).

6. A method 600 according to any one of claims 1 to 5, further comprising, prior to step 604, a step 603a of determining the coordinates of the actual position of each person (401-1...401-5) in the scene (100) from its actual depth DP calculated for each person (401-1...401-5), and, at step 604, one or more persons (401-1...401-3) from among the detected persons (401-1...401-5) are selected according to whether the coordinates of their actual position in the scene (100) satisfy a criterion (102) of location in said scene (100).

7. Method 600 according to any one of claims 1 to 6, further comprising, prior to step 603, a step 602a of estimating the orientation of each person (401-1...401-5) present in the image (400) and a step 602b of correcting the proof distance DE of each person (401-1...401-5) from its estimated orientation.

8. Method 600 according to any one of claims 1 to 7, wherein the location criterion (102) is a first defined area (102) of the scene (100), and the persons (401-1...401-3) located in said area (102) are selected.

9. Method 600 according to claim 8, wherein a second zone (107) is defined around the first zone (102), and the persons (401-4, 401-5), among the detected persons (401-1...401-5), located between the first zone (102) and the second zone (107) or outside the second zone (107) are not selected.

10. A method 600 according to any one of claims 1 to 9, wherein it takes as input a plurality of images (400) of the scene (100), each of steps 601, 602, 603 being performed on each of the images (400), and further comprises, prior to step 604, a step 603b of predicting the trajectory of each of the people detected through the plurality of images (400), and at step 604, the detected people are selected if their predicted trajectory crosses the boundary of the first zone (102).

11. Method 600 according to any one of claims 1 to 10, wherein the reference biometric size GBR is chosen from face size, general morphological posture of the individual, limb length, interpupillary distance or distance between the center of an eye and a corner of the mouth.

12. Data processing device (201, 300) comprising means for implementing a method (600) according to any one of claims 1 to 11.

13. A computer program comprising instructions which, when the program is executed by a processing device (300, 201) data, lead him to implement a method (600) according to any one of claims 1 to 11.

14. A biometric identification method comprising the following steps: (a) Acquiring an image (400) of a scene (100) using an acquisition device (103), preferably using a monocular camera; (b) Selecting one or more persons (401-1...401-3) present in the scene (100) using a method 600 according to any one of claims 1 to 11; (c) Acquiring, for each selected person (401-1...401-3) a biometric characteristic; (d) Identifying each selected person (401-1...401-3) on the basis of the acquired biometric characteristic.

15. Biometric identification system (200) comprising: - an acquisition device (103), preferably a monocular camera, configured to acquire an image (400) of a scene (100); - a data processing device (201, 300) according to claim 12 further comprising means for implementing a biometric identification method according to claim 14.