Visual sealing device, vision system, vehicle and associated visual sealing method

Abstract

The invention relates to a device (16) comprising: a memory (20) storing a detection model (24) configured to:detect a part of interest in each individual in an image; andprovide a position, in a three-dimensional frame linked to the camera, of each detected part of interest;a processing unit (22) configured to:implement the detection model (24) on the basis of an image representative of a part of a passenger compartment (6) of a vehicle (2) to calculate, in a reference frame (26), a position of each detected part of interest; andfor each detected part of interest:compare the corresponding calculated position with delimitation data indicative of spatial limits of the passenger compartment (6); anddetermine that the corresponding individual (28) is a passenger of the vehicle (2) if the calculated position is included within the spatial limits of the passenger compartment. Figure 1

Description

Title of the invention: Visual sealing device, vision system, vehicle and associated visual sealing method Technical field

[0001] The present invention relates to a visual sealing device comprising a memory and a processing unit.

[0002] The invention also relates to a vision system, a vehicle carrying such a vision system, and an associated visual sealing method.

[0003] The invention applies to the field of transport, in particular rail transport. State of the art

[0004] It is known to equip a vehicle, in particular a railway vehicle, with a vision system, in particular an intelligent vision system.

[0005] Such an intelligent vision system comprises cameras arranged to image an area of ​​interest of the vehicle, as well as a computer configured to process the images acquired by the cameras, in particular by means of image analysis algorithms, and to provide, as output, information relating to the imaged area of ​​interest.

[0006] Such a vision system is, for example, implemented in applications for detecting and counting objects, or for detecting and counting actions.

[0007] However, conventional vision systems do not give complete satisfaction.

[0008] Indeed, when transparent surfaces of the vehicle (such as the bay windows of a railway vehicle) are present in the field of vision of a camera of the vision system, said camera is capable of imaging regions of space which do not belong to the area of ​​interest, but which are visible through the transparent surfaces of the vehicle.

[0009] As a result, the image analysis algorithms implemented by the computer are required to process visual data that is not representative of the area of ​​interest. Consequently, there is a risk that the information output by the computer may be erroneous.

[0010] For example, in the context of an automatic passenger counting application, an individual present on a platform, but detected through a vehicle window, is likely to be identified as a passenger present on board the vehicle. This is therefore detrimental to the reliability of the automatic passenger counting operation.

[0011] An aim of the present invention is to remedy at least one of the drawbacks of the state of the art.

[0012] Another object of the invention is to propose a device capable of carrying out a dis criminalization, in the images provided by a vehicle camera, between passengers present on board the vehicle and individuals not on board the vehicle.

[0013] Another aim of the invention is to propose a solution whose implementation does not require any structural alteration of the vehicle or its equipment. Statement of the invention

[0014] To this end, the invention relates to a visual sealing device of the aforementioned type, the memory being adapted to store a detection model configured to: • detect at least one predetermined part of the human body, forming a part of interest, in each individual represented on a two-dimensional image acquired by a camera; and • provide, as output, at least one position, in a predetermined three-dimensional reference frame linked to the camera, of each part of interest detected, the detection model having been previously trained on the basis of a labeled data set comprising a plurality of two-dimensional images and, for each image, and for each individual represented on said image, at least one position, in a three-dimensional reference frame linked to the camera having acquired said image, of the corresponding part of interest, the processing unit being configured so as to, for each two-dimensional image of at least one video stream acquired by a respective camera and representative of at least one part of a vehicle interior: • implementing the detection model on the basis of said image to calculate, for a part of interest represented on said image and detected by the detection model, a position of said part of interest in a predetermined reference frame; and • for each part of interest detected: • compare the corresponding calculated position with delimitation data indicative of spatial limits of the passenger compartment in the reference frame; and • determine that the individual associated with said party of interest is a passenger of the vehicle if a result of the comparison is indicative of the fact that the corresponding calculated position is included within the spatial limits of the passenger compartment, and that he is not on board the vehicle otherwise.

[0015] Indeed, thanks to the implementation of the detection model, a three-dimensional position of the detected part of interest is obtained, which allows easy discrimination of the imaged individuals, on the basis of simple prior knowledge of the spatial limits of the passenger compartment.

[0016] Furthermore, by using the detection model, determining the position of each part of interest in the three-dimensional reference frame requires only simple software modifications to the pre-existing vision systems on board the vehicle. Such software modifications are significantly less expensive than the hardware modifications required by installing rangefinders, or replacing existing on-board cameras with depth cameras.

[0017] As a result, the implementation of the visual sealing device according to the invention is favorable to a vision system that is more efficient and less expensive than existing vision systems.

[0018] Advantageously, the visual sealing device according to the invention has one or more of the following characteristics, taken in isolation or in any technically possible combination:

[0019] the part of interest is the face, and for each detected face, the processing unit being configured to determine that the individual associated with said face is not a passenger of the vehicle if the position calculated for said face corresponds to a position of a wall of the vehicle;

[0020] the processing unit is, furthermore, configured to write, in the memory, for each detected part of interest, a label determined according to the result of the comparison and representative of the fact that the corresponding individual is or is not on board the vehicle, associated with an identifier of the individual corresponding to said detected part of interest.

[0021] The device according to the invention can be any type of device such as a server, a computer, a tablet, a calculator, a processor, a computer chip, programmed to implement the method according to the invention, for example by executing the computer program according to the invention.

[0022] According to another aspect of the invention, there is provided a vision system comprising a visual sealing device as defined above and at least one camera, each camera being connected to the visual sealing device to transmit, to said visual sealing device, a video stream acquired by said camera.

[0023] According to another aspect of the invention, there is provided a vehicle, in particular a rail transport vehicle, comprising a vision system as defined above, each camera of the vision system being on board the vehicle and positioned so that at least part of a passenger compartment of the vehicle is in a field of vision of said camera.

[0024] Advantageously, the vehicle has the following characteristic:

[0025] at least one camera is mounted on an interior surface of a first wall side of the vehicle, an optical axis of the camera being perpendicular to an interior surface of a second side wall opposite the first side wall, the unit of processing being configured to determine that the individual associated with the detected part of interest is a passenger of the vehicle if the position of said part of interest along an axis coincident with the optical axis of the camera, and whose origin belongs to the interior surface of the first side wall, is less, preferably strictly less, than a predetermined width of the vehicle.

[0026] According to another aspect of the invention, a visual sealing method is proposed, the method being implemented by computer and comprising, for each two-dimensional image of at least one video stream acquired by a respective camera and representative of at least one part of a passenger compartment of a vehicle: • a detection step comprising an implementation of a detection model on the basis of said image to calculate, for a part of interest represented on said image and detected by the detection model, a position of said part of interest in a predetermined reference frame, the detection model being configured to: • detect at least one predetermined part of the human body, forming the part of interest, in each individual represented on a two-dimensional image acquired by a camera; and • provide, as output, at least one position, in a predetermined three-dimensional reference frame linked to the camera, of each part of interest detected, the detection model having been previously trained on the basis of a labeled data set comprising a plurality of two-dimensional images and, for each image, and for each individual represented on said image, at least one position, in a three-dimensional reference frame linked to the camera having acquired said image, of the corresponding part of interest; • for each detected part of interest, a step of comparing the corresponding calculated position with delimitation data indicative of spatial limits of the passenger compartment in the reference frame; and • for each detected part of interest, a decision step for determining that the individual associated with said part of interest is a passenger of the vehicle if a result of the comparison is indicative of the fact that the corresponding calculated position is included within the spatial limits of the passenger compartment, and that he is not on board the vehicle otherwise.

[0027] According to another aspect of the invention, there is provided a computer program comprising executable instructions which, when executed by computer, implement the steps of the method as defined above.

[0028] The computer program can be in any computer language, such as for example machine language, C, C++, JAVA, Python, etc. Brief description of the figures

[0029] The invention will be better understood on reading the following description, given solely by way of non-limiting example and made with reference to the appended drawings in which:

[0030] [Fig-1]: [Fig.l] is a schematic sectional view, along a horizontal plane, of a railway vehicle according to the invention; and

[0031] [Fig.2]: [Fig.2] is a flowchart of a visual sealing process implemented implemented by a visual sealing device of the vehicle of [Fig.l].

[0032] It is understood that the embodiments which will be described below are in no way limiting. In particular, it will be possible to imagine variants of the invention comprising only a selection of characteristics described below isolated from the other characteristics described, if this selection of characteristics is sufficient to confer a technical advantage or to differentiate the invention compared to the state of the prior art. This selection comprises at least one preferably functional characteristic without structural details, or with only a part of the structural details if it is this part which is only sufficient to confer a technical advantage or to differentiate the invention compared to the state of the prior art.

[0033] In particular, all the variants and all the embodiments described can be combined with each other if nothing prevents this combination from a technical point of view.

[0034] In the figures and in the remainder of the description, the elements common to several figures retain the same reference. Detailed description

[0035] A vehicle 2 according to the invention is illustrated by [Fig.l].

[0036] The vehicle 2 is, in particular, a passenger transport vehicle. preferably, the vehicle 2 is a rail transport vehicle such as a train, in particular a rail transport vehicle comprising at least one car 4 intended for the transport of passengers.

[0037] Conventionally, the car 4 comprises a passenger compartment 6 delimited by walls 8.

[0038] By "passenger compartment of a vehicle", it is understood, within the meaning of the present invention, a part of the vehicle 2 where passengers can sit. In the case of a vehicle comprising a plurality of cars (or wagons), each car (or wagon) has a corresponding passenger compartment of the vehicle.

[0039] In particular, the walls 8 comprise two substantially vertical side walls 10. The side walls 10 are arranged opposite one another and extend in a usual direction of movement of the vehicle 2.

[0040] Furthermore, at least one wall 8 of the car 4, in particular a side wall 10, comprises at least one transparent surface 11, such as a window, a bay window, or even a porthole.

[0041] For the purposes of the present invention, the term "transparent surface" means a surface which lets light pass through without blocking vision, or a translucent surface (i.e. a surface which lets light pass through but does not allow the contours or colours of objects to be clearly distinguished).

[0042] As a result, an external environment of the vehicle 2 is visible from the passenger compartment 6 through each of its transparent surfaces 11.

[0043] Furthermore, the vehicle 2 comprises a vision system 12 intended, for example, for monitoring the passenger compartment 6.

[0044] The vision system 12 comprises at least one camera 14 and a visual sealing device 16 connected to each other.

[0045] Camera 14

[0046] Each camera 14 is mounted on board the vehicle 2.

[0047] Conventionally, each camera 14 has a field of vision 17. Furthermore, each camera 14 is configured to acquire a video stream representative of a scene located in the corresponding field of vision 17, and to transmit the acquired video stream to the visual sealing device 16.

[0048] The video stream delivered by each camera 14 comprises a succession of two-dimensional images, simply referred to by the term “images” hereinafter.

[0049] Preferably, each camera 14 has a fixed position over time in the passenger compartment 6.

[0050] Furthermore, each camera 14 is positioned so that at least a portion of the passenger compartment 6 of the vehicle 2 is included in the corresponding field of vision 17.

[0051] In this case, at least one camera 14 is preferably positioned so that at least part of a transparent surface 11 of the vehicle 2 is included in the corresponding field of vision 17.

[0052] More preferably, at least one camera 14 is mounted on a side wall 10A of the vehicle 2, called the “first side wall”. More precisely, the camera 14 is mounted on an inner surface 18 of the first side wall 10A, oriented towards the side wall 10B (called the “second side wall”) opposite the first side wall 10A. In this case, an optical axis A of the camera 14 is advantageously orthogonal to the second side wall 10B.

[0053] Preferably, each camera 14 is an IP technology camera (also called an “IP camera”) connected to a port of a network switch (also called a “switch”) of a communication network of the vehicle 2, such as an Ethernet IP wired communication network having a ring topology. In this case, each camera 14 is configured to transmit the acquired video stream to the device visual sealing 16 via said communication network.

[0054] Visual sealing device 16

[0055] The visual sealing device 16 is intended to detect, for each camera 14, as a function of the video stream received from said camera 14, the individuals present in the corresponding field of vision. The visual sealing device 16 is also configured to determine, among the individuals detected, those who are actually on board the vehicle 2 and those who are not.

[0056] The visual sealing device 16 is, for example, on board the vehicle 2. Alternatively, the visual sealing device 16 is a ground-based device. In this case, the transmission, to the visual sealing device 16, of each video stream acquired from each camera 14, is likely to be carried out by means of any appropriate communication means, such as a wireless communication network.

[0057] The visual sealing device 16 comprises a memory 20 and a processing unit 22 connected to each other.

[0058] Memory 20

[0059] The memory 20 is configured to store an artificial intelligence model 24, hereinafter called a “detection model”.

[0060] Such a detection model 24 is configured to receive, as input, at least one two-dimensional image, and to provide, as output, for each image received, a position, in a predetermined three-dimensional reference frame, of each object detected in said image.

[0061] In particular, the detection model 24 is configured to detect individuals represented (at least in part) on said image, or even a predetermined part of the human body, called “part of interest”.

[0062] In particular, the part of interest is the face. The expression “part of interest” is also likely to designate the human body in its entirety.

[0063] The detection model 24 was previously trained on the basis of a labeled dataset.

[0064] In this case, the labeled dataset comprises a plurality of images (i.e., a plurality of two-dimensional images), each previously acquired using a corresponding camera. Furthermore, at least a portion of the images in the labeled dataset represents (totally or partially) one or more individuals. More specifically, for each individual, the corresponding portion of interest is visible in the image.

[0065] Each image of the labeled data set is associated, for each individual represented on said image, with a label indicative of the position of said individual (or of the corresponding part of interest) in a predetermined three-dimensional reference frame linked to the camera having acquired said image.

[0066] Preferably, for each image of the labeled data set, and for each individual represented on said image, the corresponding label is also indicative of at least one of: • data representing whether or not part of the individual leaves the camera's field of vision; • data representative of a degree of concealment of the individual; • data representative of an angle of observation of the individual (i.e. an angle formed between an optical axis of the camera and a straight line connecting the camera and the individual); • coordinates, in the image, of the corners of a two-dimensional bounding box associated with the individual; • data representative of the height, width and / or depth of the individual; and • data representative of a rotation of the individual, around a predetermined vertical axis, in relation to a predetermined reference orientation.

[0067] In the case where the detection concerns the part of interest mentioned above, the label data listed above relate to said part of interest (for example, position of the face in the three-dimensional frame linked to the camera, coordinates of the bounding box of the face, etc.).

[0068] Preferably, the detection model 24 is the YOLO3D model, as described by Arsalan Mousavian et al. in the digital preprint “3D Bounding Box Estimation Using Deep Learning and Geometry”, referenced arXiv: 1612.00496.

[0069] Furthermore, the memory 20 is configured to store, for each camera 14 of the vehicle 2, a respective position in a predetermined three-dimensional reference frame 26.

[0070] Preferably, as illustrated by [Fig.l], the reference mark 26 has a first axis included in a plane of the first side wall 10A, a second axis orthogonal to the second side wall 10B, and a third vertical axis. Furthermore, in this case, the first, second and third axes are orthogonal to each other.

[0071] More preferably, the memory 20 is configured to store, for each camera 14 of the vehicle 2, data for transforming coordinates in a frame linked to said camera into coordinates in the reference frame 26. Such transformation data comprises, for example, a transition matrix (also called a “base change matrix”) from the frame linked to the camera 14 to the reference frame 26.

[0072] The memory 20 is also configured to store boundary data indicative of spatial limits of the passenger compartment 6.

[0073] For example, the delimitation data comprises coordinates, in the reference frame 26, of all or part of the walls 8 of the passenger compartment 6.

[0074] According to another example, the delimitation data of the passenger compartment 6 comprise, for at least one axis of the reference frame 26, a range of limit coordinates such that an object whose position projected onto said axis does not belong to said range of coordinates is necessarily located outside the passenger compartment 6. In this case, for a given axis of the reference frame 26, the corresponding range of limit coordinates is defined by the extreme positions of the passenger compartment 6 along said axis.

[0075] Alternatively, or in addition, the delimitation data comprise dimensions of the passenger compartment 6, in particular a width of the passenger compartment 6 (i.e. a distance between the side walls 10).

[0076] Processing unit 22

[0077] The processing unit 22 is configured to receive the video stream delivered by each camera 14.

[0078] Furthermore, for each camera 14, the processing unit 22 is configured to implement the detection model 24 stored in the memory 20 on the basis of the respective video stream, in order to determine whether or not each individual 28 appearing in the video stream is a passenger of the vehicle (a “passenger” being understood as being a person on board the vehicle 2).

[0079] Preferably, the processing unit 22 has a compact shape guaranteeing a small footprint in the vehicle 2, and is, in particular, compliant with railway standards for electromagnetic compatibility, resistance to fire and smoke, and resistance to shock and vibrations.

[0080] In order to determine whether the individuals 28 appearing in each video stream are passengers of the vehicle 2, the processing unit 22 is configured to implement a visual sealing method 30 ([Fig.2]).

[0081] As illustrated by [Fig.2], the visual sealing method 30 comprises a detection step 32, a comparison step 34 and a decision step 36.

[0082] More precisely, for each camera 14, the processing unit 22 is configured to, during the detection step 32, implement the detection model 24 on the basis of at least one image of the respective video stream.

[0083] This results, at the output of the detection model 24, for each individual 28 represented on said image and detected by the detection model 24, in a position of said individual 28 detected in the reference frame linked to the camera 14.

[0084] Furthermore, for each individual 28 detected, the processing unit 22 is configured to convert, into the reference frame 26, the position of said individual 28 in the frame linked to the camera 14, calculated using the detection model 24.

[0085] In particular, the processing unit 22 is configured to carry out such a conversion from the position of said camera 14 in the reference frame 26, stored in the memory 20.

[0086] Preferably, the processing unit 22 is also configured to carry out such a conversion from the transformation data relating to said camera 14 and stored in the memory 20.

[0087] In the case where the detection model 24 is configured to detect a part of interest of the human body, the processing unit 22 is configured to, during the detection step 32, implement said detection model 24 in order to calculate a position of said detected part of interest in the reference frame linked to the camera 14, and to convert the calculated position of said part of interest into the reference frame 26.

[0088] Furthermore, for each individual 28 detected, the processing unit 22 is configured to, during the comparison step 34, compare the corresponding calculated position in the reference frame 26 with the delimitation data stored in the memory 20.

[0089] For example, for individual 28A, the corresponding calculated position is included in the volume delimited by the coordinates of the walls of the passenger compartment 6.

[0090] Conversely, the individual 28B is visible through the transparent surface 11, but is not present on board the vehicle 2, and his position is therefore outside the physical limits of the passenger compartment 6 represented by the delimitation data.

[0091] Advantageously, in the case where the camera 14 is mounted on a side wall 10A, its optical axis A being orthogonal to the opposite side wall 10B, the processing unit 22 is simply configured to compare a coordinate of the individual 28 detected along the second axis (or “longitudinal position”) to the width of the passenger compartment. Such a characteristic is advantageous, insofar as it allows simple discrimination, based on the comparison of only two values ​​(namely the longitudinal position of the detected individual and the width of the passenger compartment).

[0092] As illustrated in the figure, the longitudinal position is also the position along an axis coincident with the optical axis A of the camera 14, and whose origin belongs to the interior surface 18 of the first side wall 10.

[0093] In the case where the detection model 24 is configured to detect a part of interest of the human body, in particular a face, the processing unit 22 is advantageously configured to compare the corresponding position in the reference frame 26 with the delimitation data stored in the memory 20 to determine whether said position corresponds to a position of a wall 8 of the vehicle 2.

[0094] Such a characteristic is advantageous, insofar as it allows discrimination between a human face and a face represented on an illustration fixed on a wall 8 of the vehicle 2, such as an advertising insert.

[0095] Furthermore, for each individual 28 detected, the processing unit 22 is configured to, during the decision step 36, determine, based on a result of the comparison step 34, whether or not said individual 28 is a passenger of the vehicle 2.

[0096] More specifically, the processing unit 22 is configured to determine that the individual 28 is a passenger who is on board the vehicle 2 if a result of the comparison is indicative of the fact that the corresponding calculated position is included within the spatial limits of the passenger compartment 6.

[0097] Conversely, the processing unit 22 is configured to determine that the individual 28 is not on board the vehicle 2 if the result of the comparison is indicative of the fact that the corresponding calculated position is outside the spatial limits of the passenger compartment 6.

[0098] For example, during decision step 36, processing unit 22 determines that individual 28A is on board vehicle 2, but not individual 28B.

[0099] Preferably, in the case where the camera 14 is mounted on a side wall 10, its optical axis A being orthogonal to the opposite side wall 10, the processing unit 22 is simply configured to determine that the individual 28 is a passenger if his longitudinal position is less (preferably strictly less) than the width of the passenger compartment, and that he is not on board the vehicle 2 otherwise.

[0100] In the case where the detection model 24 is configured to detect a part of interest of the human body, in particular a face, the processing unit 22 is advantageously configured to determine that the individual corresponding to said part of interest is not a passenger of the vehicle 2 if the position calculated for said part of interest corresponds to a position of a wall 8 of the vehicle 2.

[0101] Advantageously, the processing unit 22 is also configured to, at the end of the decision step 36, write, in the memory 20, for each individual 28 detected, a label representative of the fact that said individual 28 is or is not on board the vehicle 2. In particular, said label is stored in association with an identifier of said individual.

[0102] This feature is advantageous, insofar as the label is likely to be used during a subsequent processing process, so as to discriminate the passengers of the vehicle 2 from individuals detected through the transparent surfaces 11 but not present on board the vehicle 2. In this way, errors in the subsequent processing process, linked to poor discrimination between passengers and persons not on board the vehicle, are likely to be avoided.

[0103] Operation

[0104] The operation of the vision system 12 will now be described with reference to FIGS. 1 and 2.

[0105] During a prior configuration step, the detection model 24 is trained based on the labeled dataset, then stored in memory 20.

[0106] Then, during operation of the vision system 12, each camera 14 transmits the respective acquired video stream to the visual sealing device 16.

[0107] In this case, during the detection step 32, for each camera 14, the processing unit 22 of the visual sealing device 16 implements the detection model 24 on the basis of at least one image of the respective video stream.

[0108] This results, at the output of the detection model 24, for each individual 28 represented on said image and detected by the detection model 24, a position, in the reference frame linked to the camera 14, of said detected individual 28.

[0109] Then, for each individual 28 detected, the processing unit 22 converts, into the reference frame 26, the position of said individual 28 initially calculated in the frame linked to the camera 14.

[0110] Then, during the comparison step 34, for each individual 28 detected, the processing unit 22 compares the corresponding calculated position in the reference frame 26 with the delimitation data stored in the memory 20.

[0111] Then, during the decision step 36, for each individual 28 detected, the processing unit 22 determines whether or not said individual 28 is a passenger of the vehicle 2, based on a result of the comparison step 34.

[0112] Advantageously, for each individual 28 detected, the processing unit 22 writes, in the memory 20, a label representing the fact that said individual 28 is or is not on board the vehicle 2.

[0113] Of course, the invention is not limited to the examples which have just been described.

Claims

Claims

1. Visual sealing device (16) comprising a memory (20) and a processing unit (22), the memory (20) being adapted to store a detection model (24) configured to: • detecting at least one predetermined part of the human body, forming a part of interest, in each individual represented on a two-dimensional image acquired by a camera (14); and • provide, as output, at least one position, in a predetermined three-dimensional reference frame linked to the camera, of each part of interest detected, the detection model (24) having been previously trained on the basis of a labeled data set comprising a plurality of two-dimensional images and, for each image, and for each individual represented on said image, at least one position, in a three-dimensional reference frame linked to the camera having acquired said image, of the corresponding part of interest, the processing unit (22) being configured so as to, for each two-dimensional image of at least one video stream acquired by a respective camera (14) and representative of at least one part of a passenger compartment (6) of a vehicle (2): • implementing the detection model (24) on the basis of said image to calculate, for a part of interest represented on said image and detected by the detection model (24), a position of said part of interest in a predetermined reference frame (26); and • for each part of interest detected: • compare the corresponding calculated position with delimitation data indicative of spatial limits of the passenger compartment (6) in the reference frame; and • determining that the individual (28) associated with said party of interest is a passenger of the vehicle (2) if a result of the comparison is indicative of the fact that the position corresponding calculated is included within the spatial limits of the passenger compartment, and that it is not on board the vehicle (2) otherwise.

2. A visual sealing device (16) according to claim 1, wherein the part of interest is the face, and for each detected face, the processing unit (22) being configured to determine that the individual (28) associated with said face is not a passenger of the vehicle (2) if the position calculated for said face corresponds to a position of a wall (8, 10) of the vehicle (2).

3. Visual sealing device (16) according to claim 1 or 2, wherein the processing unit (22) is further configured to write, in the memory (20), for each detected part of interest, a label determined according to the result of the comparison and representative of the fact that the corresponding individual (28) is or is not on board the vehicle (2), associated with an identifier of the individual corresponding to said detected part of interest.

4. A vision system (12) comprising a visual sealing device (16) according to any one of claims 1 to 3 and at least one camera (14), each camera (14) being connected to the visual sealing device (16) for transmitting, to said visual sealing device (16), a video stream acquired by said camera (14).

5. Vehicle (2), in particular a rail transport vehicle, comprising a vision system (12) according to claim 4, each camera (14) of the vision system (12) being on board the vehicle (2) and positioned so that at least a part of a passenger compartment (6) of the vehicle is located in a field of vision (17) of said camera (14).

6. Vehicle (2) according to claim 5, wherein at least one camera (14) is mounted on an inner surface (18) of a first side wall (10A) of the vehicle (2), an optical axis (A) of the camera (14) being perpendicular to an inner surface (18) of a second side wall (10B) opposite the first side wall (10A), the processing unit (22) being configured to determine that the individual (28) associated with the detected part of interest is a passenger of the vehicle (2) if the position of said part of interest along an axis coincident with the axis

7. optical (A) of the camera (14), and whose origin belongs to the inner surface (18) of the first side wall (10A), is less, preferably strictly less, than a predetermined width of the vehicle (2). Visual sealing method (30), the method being implemented by computer and comprising, for each two-dimensional image of at least one video stream acquired by a respective camera (14) and representative of at least one part of a passenger compartment (6) of a vehicle (2): • a detection step (32) comprising an implementation of a detection model (24) on the basis of said image to calculate, for a part of interest represented on said image and detected by the detection model (24), a position of said part of interest in a predetermined reference frame (26), the detection model (24) being configured to: • detect at least one predetermined part of the human body, forming the part of interest, in each individual represented on a two-dimensional image acquired by a camera; and • provide, as output, at least one position, in a predetermined three-dimensional reference frame linked to the camera, of each part of interest detected, the detection model (24) having been previously trained on the basis of a labeled data set comprising a plurality of two-dimensional images and, for each image, and for each individual represented on said image, at least one position, in a three-dimensional reference frame linked to the camera having acquired said image, of the corresponding part of interest; • for each detected part of interest, a step (34) of comparing the corresponding calculated position with delimitation data indicative of spatial limits of the passenger compartment (6) in the reference frame (26); and • for each detected party of interest, a decision step (36) for determining that the individual (28) associated with said party of interest is a passenger of the vehicle (2) if a result of the comparison is indicative of the fact that the corresponding calculated position is within the spatial limits of the passenger compartment, and that it is not on board the vehicle (2) otherwise.

8. A computer program comprising executable instructions which, when executed by a computer, implement the steps of the method according to claim 7.

Images