Selective filtering of virtual objects within digital three-dimensional scenes

Abstract

Method for modifying a three-dimensional scene (10) for its generation for an observer (20), comprising steps of extracting (S2) an object (11) from said three-dimensional scene (10), projecting (S3) said object onto a two-dimensional surface, according to a relative position and orientation of said object with respect to said observer, so as to obtain a two-dimensional image, applying (S4) at least one digital filter to said two-dimensional image; generating (S5) a new object whose texture comprises said two-dimensional image, and inserting (S6) said new object into said three-dimensional scene in place of said object. Figure for abstract: Fig. 1

Description

Title of the invention: Selective filtering of virtual objects within digital three-dimensional scenes FIELD OF THE INVENTION

[0001] The present invention relates to the generation of a three-dimensional scene for an observer, with a view to production on a screen or via a virtual or augmented reality device. It relates more particularly to the modification of a three-dimensional scene for its generation, in which a filter is applied to at least one object in the scene.

[0002] Many applications use three-dimensional scenes in which an observer is placed, and can possibly move, and can observe different objects which constitute this scene, respecting characteristics of distance and orientation just as in the physical world.

[0003] Some applications allow modeled people to be inserted into these three-dimensional scenes. These modeled people are realistic representations of physical people.

[0004] Depending on the applications, the observer can move freely in the three-dimensional scene. A projection of the scene is therefore calculated continuously, in real time or quasi-real time, in order to adapt the appearance of the scene to the movements of the observer.

[0005] These movements may be the result of actual movements of a user and observer who, for example, has put on an augmented reality headset, or of manipulation of a steering device (such as a joystick for example) for a user / observer facing a computer screen. Obviously, other use cases are also conceivable.

[0006] The observer can therefore approach the modeled person as closely as he wishes, and position himself at any angle in relation to it.

[0007] This situation can pose different types of problems.

[0008] First of all, in such a situation, the hyper-realistic virtual representation of existing people can pose a problem of “virtual” intimacy. Indeed, in real life, a person consciously or unconsciously has a social distance and will be uncomfortable if one approaches below this distance. Among other reasons, the intimacy of the body can be cited, which can notably depend on the clothing worn by the modeled person. Since this is a realistic representation, the problem naturally transposes to the virtual world, and the absence A brake preventing an observer from getting as close as he or she wishes to the representation can therefore pose a problem.

[0009] Furthermore, in a three-dimensional scene, depending on the logic of the application, the real person associated with the modeled person may not be informed that he or she is being observed in an unwanted manner, i.e. too closely (for example because he or she is not connected to the application managing the three-dimensional scene). In real life, he or she could react by moving away so as to keep his or her distance, but in such a situation, he or she cannot react.

[0010] “Forbidden zone” mechanisms have been proposed, such as the “guardian” zone of Oculus Quest headsets. Such a zone can be materialized by superimposing the three-dimensional scene viewed by the observer. However, such a mechanism is configured by the user / observer himself and nothing is provided to physically prevent him from crossing these limits.

[0011] Furthermore, in an augmented reality environment, for example, in which the modeled person corresponds to a determined location in the physical world, it is impossible to constrain the free movement of the observer using a headset or a smartphone. The latter can cross these limits as if they did not exist.

[0012] These proposals therefore do not respond to the problem previously mentioned.

[0013] Furthermore, another problem concerns defects in the modeled object (whether it whether it is a person or another object).

[0014] Among these defects, we can cite reconstruction defects.

[0015] The principle of reconstructing a three-dimensional model of an object, for example a real person, is based on taking multiple shots of the object. To do this, a set of cameras can be placed all around the object that one wants to reconstruct. The images from all the cameras at the same time instant then allow the reconstruction of a 3D representation, or model.

[0016] The shots from the multiple cameras are not necessarily taken at 360°. For technical reasons, the cameras may only be able to film over 180°, if the object is in front of a wall for example, or if the number of cameras available is not sufficient (at least thirty cameras are needed to cover 360°). In such a case, the capture of the real object is not complete, which can result in visual artifacts in the reconstructed model that are more or less significant depending on the geometric complexity of the object. Also, when navigating in a three-dimensional scene, these artifacts can appear and harm the quality of the rendering of the scene, due to the freedom of the observer to move around the objects and as close as he wishes.

[0017] Other defects concern resolution defects.

[0018] The resolution of the modeled object cannot be higher than the resolution of the source images, from the cameras used for the construction of the model. The resolution of the reconstructed three-dimensional model is therefore optimal when it is displayed with a point of view (of the observer) equivalent to that of the cameras used to take the source images, that is to say when the relative position of the observer corresponds to that of one of the cameras used to construct the model.

[0019] However, because the observer can move freely in the three-dimensional scene, to be displayed at a size corresponding to the relative position between the object and the observer, the object may have to be enlarged (zoomed) by a significant factor. These enlargements can cause visual artifacts to appear that impact the quality of the three-dimensional representation.

[0020] There is therefore a need to improve the current state-of-the-art proposals. Summary of the invention

[0021] The invention aims to avoid improving the situation compared to the proposals of the state of the art, and in particular, but not exclusively, to respond to the problems previously mentioned linked to excessive proximity and / or an inadequate observation angle(s).

[0022] For these purposes, according to a first aspect, the present invention can be implemented by a method of modifying a three-dimensional scene for its generation for an observer, comprising steps of - extraction of an object from said three-dimensional scene, - projection of said object onto a two-dimensional surface, according to a position and a relative orientation of said object with respect to said observer, so as to obtain a two-dimensional image, - applying at least one digital filter to said two-dimensional image; - generation of a new object whose texture includes said two-dimensional image, and insertion of said new object into said three-dimensional scene in place of said object.

[0023] Thus, the various problems described above, in particular, are resolved by managing a filtering of the appearance of the object as a function of the relative position and / or orientation of the object with respect to the observer. It is thus possible to manage the problem of “virtual” intimacy and visual artifacts caused by the (sometimes unavoidable) defects of three-dimensional models.

[0024] Furthermore, as will be seen later, the application of a digital filter to a bidirectional image resulting from a projection of the object, then the generation of a new textured object from this image and taking the place of the original object makes it possible to avoid costly three-dimensional calculations and thus offer a mechanism compatible with the constraints of real-time rendering of a user's navigation in a three-dimensional scene.

[0025] According to preferred embodiments, the invention comprises one or more of the following features which can be used separately or in partial combination with each other or in total combination with each other: - prior to the extraction step, said object is selected according to said position and / or said relative orientation. It is thus possible to process only part of the objects. - said digital filter is configured according to a parameter resulting from said relative position and / or orientation. - said at least one digital filter is configured according to said at least one parameter continuously within a determined range of values. This allows for progressive rendering, for example. - said at least one parameter comprises a distance between said object and said observer, and / or an angle resulting from said relative position and orientation, formed by a main direction of said object and a direction of observation of said observer. - said at least one digital filter comprises a filter acting on a level of detail of said object perceptible by said observer, such as blurring. Thus, the object can be more or less recognizable, which can be very interesting in the case of a model associated with a real person. - said at least one digital filter comprises a filter acting on the colorimetry of said object. This can, for example, make it possible to add artistic effects or highlight the object within a three-dimensional scene. - said object is a person modeled after a real person, said filter aiming to preserve the privacy of said real person

[0026] Another aspect of the invention relates to a computer program capable of being implemented on a web server, the program comprising code instructions which, when executed by a processor, carry out the steps of the method as previously defined.

[0027] Another aspect of the invention relates to a data medium on which at least one series of program code instructions has been stored for the execution of a method as previously defined.

[0028] Other characteristics and advantages of the invention will appear on reading the following description of a preferred embodiment of the invention, given by way of example and with reference to the appended drawings. BRIEF DESCRIPTION OF THE FIGURES

[0029] [Fig.l] illustrates an illustrative flowchart of a method according to one embodiment of the invention.

[0030] [Fig.2] represents an illustrative example of relative position and orientation of an object with respect to an observer.

[0031] [Fig.3] illustrates two examples of configuration of a filter as a function of a distance parameter, according to embodiments of the method described.

[0032] [Fig.4] represents two examples of configuration of a filter as a function of an angle parameter, according to embodiments of the method described.

[0033] [Fig.5] illustrates an example of a privacy or “forbidden” zone made possible by an implementation of the method.

[0034] [Fig.6] illustrates applications of two examples of filters on an object according to different intensities, according to embodiments of the method described.

[0035] DETAILED DESCRIPTION OF EMBODIMENTS OF THE INVENTION

[0036] In the following description, for the sake of clarity, we are interested in a single object within a three-dimensional scene. However, it is clear that a plurality or all of the objects present in a three-dimensional scene can be treated as described. Each object can be treated, or not, independently of each other. In the same way, the same scene can be generated for different observers. The description below is not exhaustive of the different use cases and, in particular, of the aspects which have just been mentioned.

[0037] These objects can be called "virtual objects" insofar as they are three-dimensional models that only exist in the three-dimensional scene. In particular, they can be modeled people, associated with real people. In such a case, the object (modeled person) can adopt an appearance that is as realistic and faithful as possible to the associated real person.

[0038] Such realistic rendering can be achieved by various state-of-the-art techniques. Among those used and accessible to those skilled in the art, we can cite the 3D reconstruction method set out in the article “3D Gaussian Splatting for Real-Time Radiance Field Rende ring” by Bernhard Kerbl et al, in ACM Trans. Graph, 2018.

[0039] The proposed method aims to modify a previously existing three-dimensional scene before its rendering, or generation, for an observer. To the extent that a three-dimensional scene must be periodically generated in order to account movements of the observer and / or the different objects present, this modification process must preferably be implemented at each iteration, that is to say before each generation.

[0040] The generation of the three-dimensional scene itself does not fall within the scope of the proposed method, and is part of the techniques accessible to those skilled in the art. In particular, different tools or platforms exist commercially and the proposed method can be applied to these existing proposals in order to improve them.

[0041] In [Fig.l], a three-dimensional scene 10 is thus modified by the different steps S1-S6 of the proposed method.

[0042] Step S7 consists of the generation (or rendering) of the three-dimensional scene for the observer. It can be considered as external and independent of the proposed method, and can be in accordance with the state of the art.

[0043] This generation may target various types of human-machine interface. In particular, it may be a display on a computer screen (including portable devices such as a smartphone-type mobile phone), digital tablets, etc.). It may also be a representation for a virtual or augmented reality headset.

[0044] The proposed method comprises modifying the rendering of the 3D model as a function of the relative position and / or orientation of the object with respect to the observer, with a view to rendering (or generating) the three-dimensional scene comprising the object for observation.

[0045] A preliminary step S1 consists of determining this relative position and / or orientation. They can be directly deduced from the three-dimensional scene 10 which includes the data of the different objects (fixed or mobile) making up the scene, as well as the parameters of the observer. This information is necessarily available to allow the generation of the three-dimensional scene 10 for the observer (step S7).

[0046] [Fig. 2] represents an illustrative example of relative position and orientation of an object 11 with respect to an observer 20.

[0047] In this figure, the arrows indicate the orientations of the observer and the object. The observer has an orientation (or “observation direction”) which corresponds to his gaze in the physical world and which is used to enable the generation of the three-dimensional scene 10. In particular, this orientation, or observation direction, creates a dichotomy between a visible part of the scene (located in front of him, i.e. in an area generally pointed by the arrow) and an invisible part (located behind him, i.e. opposite the visible part with respect to the observer).

[0048] The relative positions and orientations of an object 11 with respect to an observer can be expressed in different ways, as known to those skilled in the art. For example, it can be a 6-tuple with 3 components for a relative position and 3 components for a solid angle. This expression depends on the data representation mode used by the three-dimensional scene 10 and the software application used.

[0049] According to embodiments, the method described can be based on parameters derived from this data.

[0050] As examples, in [Fig.2], a distance d between the object 11 and the observer 20 is illustrated, as well as an angle a between their respective orientations.

[0051] These parameters may be sufficient to express different behaviors that can be implemented within the framework of the proposed method, but it is understood that obviously other parameters can be defined according to a desired behavior.

[0052] In a step S2, the object 11 is extracted from the three-dimensional scene 10.

[0053] A preliminary criterion can be set up before extraction, so that only a part of the objects are extracted and processed (application of a filter, etc.).

[0054] This criterion may be based on the position and / or relative orientation of the object 11 relative to the observer. In particular, the object may be selected when at least one parameter derived from the position and / or relative orientation(s) is within a value range.

[0055] The prior selection of the object for processing implies that it is not subject to processing (application of a filter) if it is not selected. This embodiment makes sense when a filter can only be applied in certain situations; in which case, it is advantageous not to extract the object if no filter will be applied, in order to avoid unnecessary digital processing (steps S2, S3, etc.).

[0056] According to another embodiment, the object is systematically extracted in step S2, without a prior selection step. This embodiment may make sense in the case where a processing (filter) is systematically applied to the object.

[0057] After its extraction in step S2, the object 11 is projected, in a step S3, onto a two-dimensional surface according to the relative position and orientation, in order to obtain a two-dimensional image

[0058] This projection step S3 may be conventional in itself, that is to say that it may correspond to the projection conventionally carried out in a generation phase S7 of an object of a three-dimensional scene. It consists of determining an appearance of a three-dimensional object from the point of view of an observer (who observes it according to a two-dimensional representation, typically on a screen).

[0059] This projection S3 therefore depends on the position and orientation (viewing angle) of the observer 20 and the position and orientation of the object.

[0060] In a step S4, a digital filter is applied to the two-dimensional image obtained by the projection of the object. Obviously, several filters can be applied, in combination or in succession.

[0061] These filters can be conventional digital image processing filters.

[0062] One of the advantages of the proposed method is that it is thus possible to use the numerous digital filters available for 2D imaging and to apply them, indirectly, to an object in a three-dimensional scene.

[0063] Another, even more important advantage is that it makes processing on the fly possible, due to the reduction in computational cost brought about by this method.

[0064] Indeed, 3D filters have been proposed in the scientific literature.

[0065] This model is a good compromise between quality and rendering performance for powerful computers with large computing and memory capacities. However, it cannot be applied to devices such as mobile phones or standalone augmented or virtual reality headsets.

[0066] Indeed, the static 3D display of an image whose point of view can be modified with the mouse or finger can be satisfied with a refresh rate of a few images per second.

[0067] In a context of simulated, augmented or totally virtual reality, rendering must be done continuously in order to reflect the movements, even minimal, of the user / observer. Maintaining a high refresh rate (or "frame rate" in English), for example between 30 and 60 images per second, is crucial to guarantee a good experience for the user.

[0068] Furthermore, a realistic 3D model, once decompressed for display, can correspond to a large volume of data, typically in excess of 10 MB. This large mass of data makes any real-time processing difficult, if not impossible, especially since the very nature of 3D filters does not facilitate, or even makes totally impossible, the implementation of conventional filters such as a simple blur.

[0069] This performance problem is obviously all the more crucial with a volumetric video, which will change the data to be displayed 30 times per second, for example, thus multiplying the mass of data to be transmitted and potentially leading to saturation of the computer's resources (smartphone, digital tablet, virtual reality headset, etc.), and in particular of the memory bus connecting the central memory, the CPU processor and the GPU graphics processor.

[0070] This problem of excessive computational cost incompatible with the real-time requirement on a computer with limited capabilities is solved by transforming the object into two dimensions to reduce computational complexity.

[0071] Once the digital filter has been applied, in a step S5, a new object is generated and then, in a step S6, inserted into the three-dimensional scene in place of the original object 11.

[0072] The term "instead" means that it replaces this original object which is therefore removed from the scene. The new object is positioned at the same position as the original object and in the same orientation.

[0073] The three-dimensional scene 10 is thus modified, before generation for an interface associated with the observer, in step S7. All of the steps S1-S6 are provided to allow ongoing processing that does not substantially delay this generation step, so that there is no perceptible impact for the user and observer.

[0074] Step S5 comprises the generation of an object whose texture comprises the two-dimensional image resulting from the application of the digital filter(s).

[0075] More specifically, according to one embodiment, this new object can be a rectangle oriented towards the observer. The texture is calculated from the projection of the original object according to the relative positions and orientations, so that the appearance of this textured object corresponds to that which the original object would have from the point of view of the observer, but with, in addition, the application of one or more digital filters.

[0076] This (or these) applied filter(s) may correspond to different use cases and choices of the developer of the virtual or augmented reality application. It may also vary according to different parameters, in nature or intensity.

[0077] According to one embodiment, the digital filter is configured as a function of a parameter resulting from the relative position and / or orientation of the object 11 with respect to the observer 20.

[0078] This parameter can be a distance d, or an angle a. The distance d can be the distance between the object 11 and the observer, and the angle can be the angle between their respective orientations, that is to say between an observation direction and a main direction, marking the front face, of the object 11. Other parameters can also be determined from the relative position and orientation. Combinations of parameters can also be used for the configuration of the filters.

[0079] [Fig.3] illustrates two examples of configuration of a filter as a function of a distance parameter d.

[0080] We see in these examples that the intensity of the filter, p, depends on the distance d, so that the closer the observer is to the object (d decreases), the stronger the filter is in order to impact the appearance of the object for the observer.

[0081] This dependency can be implemented in different ways.

[0082] The dotted line illustrates a first embodiment, in which simple thresholding is implemented. If the distance d is greater than a threshold dH then no filter is implemented (value of the filter intensity p zero). If the distance is less than the threshold dH then a value pm is used as the filter intensity.

[0083] [Fig.5] illustrates an example of such a situation. The distance dl forms a privacy or "forbidden" zone 12 around an object 11 which is here a realistic representation of a real person. When an observer approaches too close to the object 11, and enters this zone 12 (i.e. d <di), alors le filtre numérique est appliqué afin de modifier l’apparence de l’objet (par exemple de le rendre méconnaissable).

[0084] The solid line illustrates a second embodiment, proposing a progressive approach.

[0085] If the distance d is greater than a threshold d2, then no filter is implemented (value of the filter intensity p zero). If the distance is less than a threshold dh then a value pm is used as the filter intensity. In a range of values ​​[di; d2], the filter is configured continuously, that is to say that the intensity (or any other configuration parameter) of the digital filter evolves continuously, for example linearly, between the two extrema, here 0 and pm.

[0086] Thus, when approaching the object 11, the observer will see its appearance increasingly impacted up to a maximum impact when it enters its immediate vicinity (d <di).

[0087] For example, we can set the values ​​di=0.5 m and d2=1.5 m.

[0088] [Fig.4] illustrates two examples of configuration of a filter as a function of an angle parameter a. This angle can be the angle between the direction of observation and the main direction (front face) of the object 11. In the example, it is considered that an angle a=0 corresponds to the situation of the observer facing the main face of the object. Thus, an angle a=180° corresponds to an observer facing the rear face of the object.

[0089] We see in these examples that the intensity of the filter, p, depends on the angle a, so that the further the observer moves away from angle 0, the stronger the filter is in order to impact the appearance of the object for the observer.

[0090] This dependency can be implemented in different ways.

[0091] The dotted line illustrates a first embodiment, in which a simple thresholding is implemented. If the angle a is less than a threshold adi or greater than an angle agi then no filter is implemented (value of the intensity of the filter p zero). The thresholds adi and agi can have the same absolute value (in the case where the object is symmetrical, for example), but are of different signs since correspond to the two possible directions of rotation of the observer around the object.

[0092] If the angle a is less than the threshold agb or greater than the threshold adb then a value pm is used as the filter intensity. This case corresponds to the situation where the angle a is too far from the angle a=0 corresponding to a face observation.

[0093] The solid line illustrates a second embodiment, proposing a progressive approach.

[0094] If the angle a is greater than a threshold ag2, and if the angle a is less than a threshold ad2> then no filter is put in place (value of the intensity of the filter p zero).

[0095] If the angle a is greater than a threshold adb or if the angle a is less than a threshold agi then a value pm is used as the intensity of the filter.

[0096] In a range of values ​​[agi; ag2] and in a range of values ​​[ad2; adi], the filter is configured continuously, that is to say that the intensity (or any other configuration parameter) of the digital filter evolves continuously, for example linearly, between the two extrema, here 0 and pm.

[0097] The value of the thresholds and the intensity levels of the filters depend on the application cases, as do the parameter (distance, angle, etc.) and the nature of the filter itself (or filters, because several filters can be applied).

[0098] In a case of application to the management of the virtual intimacy of a modeled person corresponding to a real person, the filter can depend on the distance and be designed to act on a level of details perceptible by the observer.

[0099] The risk is, in fact, that an observer gets too close to observe a person at a level of detail not desired by that person. By applying a filter affecting the level of detail, such as for example a blurring filter, when the observer gets too close, this risk can be mitigated.

[0100] As seen previously, blurring can be triggered when the distance becomes lower than a given threshold, and it can also be progressive as the observer gets closer, up to a maximum level. At this maximum level, the object may no longer be recognizable.

[0101] Thus, it is possible to preserve the privacy of real people having a model (i.e. a modeled person) represented in a three-dimensional scene, in particular by playing on a level of details perceptible by the observer (this level of details being determined by a filter which depends on a distance between the model and the observer).

[0102] In the case of an imperfect object, for example due to incomplete reconstruction on certain orientations, in particular on a rear face, the filter may depend on the angle and be designed to play on a level of details perceptible by the observer. This situation may appear, as seen previously, in the case of an insufficient number of cameras, or poor positioning of these, during the construction of the 3D model by taking pictures of the real object (in particular of a real person).

[0103] Thus, the object can appear without modification when it is observed from an angle that does not reveal a defect resulting (in particular) from the reconstruction of the model, while a filter can be applied to impact its appearance when it is observed from an angle that would reveal a reconstruction defect. This filter can precisely be intended to mask the defect, by acting, for example, on a level of details perceptible by the observer (for example by applying a blur filter).

[0104] Another application case is the management of the resolution of the object. As explained previously, the modeling of an object is linked to the resolution of the digital images acquired by a set of cameras.

[0105] The filter can then depend on the distance and be designed to act on a level of detail perceptible by the observer. Thus, the filter can mask resolution defects when the observer is too close, in a more harmonious way. In particular, disharmonious or confusing artifacts in the clarity of the three-dimensional scene can be replaced by filtering on objects that are too close, so that their appearance is clearer and in line with the resolution of the 3D model.

[0106] These different application cases can be managed by the same application. Thus, the same object can be the subject of a first filter aimed at guaranteeing virtual privacy, a second filter aimed at managing an incompleteness of the model, a third filter aimed at managing the resolution of the model, etc.

[0107] Filters can be of any nature, as previously mentioned. Examples include: - a Gaussian blur, - a crystallization effect, - a pixelation effect, - a rendering based on black or white dots (or “dotscreen”) - a print-like rendering (or “halftone”) - an effect playing on saturation, contrast and / or intensity, - an effect playing on the exposure, - an effect playing on the vibrancy of colors, - noise reduction, - improved sharpness - a black and white photo effect, - a sepia recoloration, - a recoloration with a base color, etc.

[0108] These filters can be configured by at least one parameter, such as an intensity, in a manner known per se. In the context of the proposed method, the configuration can depend on one parameter or possibly several.

[0109] [Fig.6] illustrates the application of two examples of filters on an object according to different intensities.

[0110] In [Fig.6], Figure A illustrates an example of an object on which no filter is applied.

[0111] Figures B and C correspond to a “pixelation” type filter, with an average intensity for figure B and a strong intensity on figure C.

[0112] Figures D and E correspond to a “Gaussian blur” type filter, with an average intensity for figure D and a strong intensity in figure E.

[0113] Two intensity levels are shown for illustrative purposes only. In a real case, more levels may be implemented, including a continuity of levels as seen previously for certain embodiments of the method.

[0114] These two filters allow you to play on a level of perceptible details of the object. Possibly progressively, it is impossible to mask certain details or aspects of the object's appearance, or even to make it unrecognizable. In the case of a physical person, for example, it can be blurred more depending on the observer's proximity so that at too close a distance it becomes unrecognizable.

[0115] As an added benefit, it becomes impossible for a malicious observer to capture an accurate and recognizable image of a person modeled too closely.

[0116] It may be provided by the application in charge of the three-dimensional scene that each physical person can manage the behavior of the filters for their 3D model themselves. In particular, they can set the distance at which they do not wish to be recognizable, or sufficiently blurred so that no detail is perceptible (for example the threshold di in the examples previously described).

[0117] It may be provided that these elements provided by natural persons wishing to appear in a three-dimensional scene in a modeled form form an element of an image rights contract established between them and the scene manager. The method described therefore allows the establishment of new conventions and a more refined way of managing one's image in virtual worlds.

[0118] Other use cases are still possible. Examples include, but are not limited to: - Improving quality: by chaining color enhancement filters, modifying contrast, exposure or sharpness, it is possible to improve the quality of the rendering, to compensate for the poor conditions of capture of the images used for reconstruction, - Artistic effects: recoloring, black and white, sepia are examples of filters that can be used for artistic purposes, - Signaling effects: For example, it is possible to render in monochrome when the model is far away, then colorize it as the user approaches, thus attracting their attention,

[0119] It can be seen that in certain use cases, the object is subject to filtering processing regardless of its position and / or orientation.

[0120] Only the filter itself can be configured differently depending on these elements. Thus, in the case of signage, a black & white rendering filter can be systematically applied but with an increasing intensity value depending on the distance between the object and the observer.

[0121] In the case of an artistic filter, this may not even depend on the position or orientation of the object relative to the observer.

[0122] Other advantages of the method described and of the numerous embodiments include, among others, advantages which may be of different orders: - efficiency: apply filters conditionally and with arbitrary intensity to a data structure that does not lend itself to it either by the mass of data to be processed or by the type of data; - performance: applying a filter to a texture is a very efficient and GPU-efficient operation, unlike applying a filter directly to a 3D structure; - efficiency: the reduced data mass of a texture compared to a 3D frame makes it possible to filter volumetric videos in real time, - simplicity: the use of a new object, for example rectangular (called a “billboard”) allows natural integration into a 3D environment of the augmented reality type. - Flexibility: this system is not limited to masking and can be used for many other purposes.

[0123] Of course, the present invention is not limited to the examples and the embodiment described and shown. It is in particular susceptible of numerous variants accessible to those skilled in the art, some of which have been described previously or simply mentioned.

Claims

Claims

1. Method for modifying a three-dimensional scene (10) for its generation for an observer (20), comprising steps of - extracting (S2) an object (11) from said three-dimensional scene (10), - projecting (S3) said object onto a two-dimensional surface, according to a relative position and orientation of said object with respect to said observer, so as to obtain a two-dimensional image, - applying (S4) at least one digital filter to said two-dimensional image; - generating (S5) a new object whose texture comprises said two-dimensional image, and inserting (S6) said new object into said three-dimensional scene in place of said object.

2. Method according to the preceding claim, in which prior to the extraction step, said object is selected as a function of said position and / or said relative orientation.

3. Method according to one of the preceding claims, in which said digital filter is configured according to a parameter resulting from said relative position and / or orientation.

4. Method according to the preceding claim, in which said at least one digital filter is configured as a function of said at least one parameter continuously in a determined range of values.

5. Method according to one of claims 3 or 4, wherein said at least one parameter comprises a distance (d) between said object and said observer, and / or an angle arising from said relative position and orientation, formed by a main direction of said object (11) and a direction of observation of said observer (20).

6. Method according to one of the preceding claims, in which said at least one digital filter comprises a filter acting on a level of details of said object perceptible by said observer (20), such as blurring.

7. Method according to one of the preceding claims, in which said at least one digital filter comprises a filter acting on the colorimetry of said object (11).

8. Method according to one of the preceding claims, in which said object is a person modeled from a real person, said filter aiming to preserve the privacy of said real person.

9. A computer program capable of being implemented on a multimedia stream playback terminal, the program comprising code instructions which, when executed by a processor, performs the steps of the method defined in claims 1 to 8.

10. Computer-readable data carrier on which at least one series of program code instructions for executing a method according to one of claims 1 to 8 has been stored.

Images