FR2968812A1 - Method for rendering images in real-time from virtual three-dimensional scene, involves building data structure by merging all static objects of scene before making combinations of three-dimensional geometric data by spatial proximity - Google Patents

Abstract

The method involves building a model of a data structure of a virtual three-dimensional (3D) scene (Sc) to enable a fast search of intersections between rays and the 3D scene, where the data structure is used to group elements of the 3D scene by spatial proximity. The data structure is built by merging all static objects (OS) of the 3D scene before making combinations of 3D geometric data by spatial proximity, where the data structure is built with all the static objects and animated objects (OA) that are marked only when animation is activated. An independent claim is also included for a computer program product for rendering images in real-time from a virtual 3D scene.

Description

The present invention relates to a method and a system for rendering images of a virtual scene in three dimensions in real time. A real-time three-dimensional rendering method consists in generating an image of a scene, possibly comprising animated elements, in the instant preceding its display on a medium, for example a screen. Real-time rendering in three dimensions or 3D is different from a so-called "precalculated" rendering in that the rendering of the scene is immediate and the user can manipulate and interact directly with the scene, for example to manipulate and change the colors of an object, to move within a virtual universe, to configure the accessories of an automobile, to simulate emergency cases. The 3D rendering of the images is done directly on a display device in the case of a real-time rendering, unlike a pre-calculated 3D rendering in which the result of the rendering is a non-modifiable image or video. For the calculation time of a real-time 3D rendering method to be imperceptible to a user, it must be less than the retinal persistence. Since the rendering speed of the images is measured in frames per second (fps) or Hertz (Hz), it appears that the printing for the user of an interactive mode exists from a rendering speed of the order of 6 fps, and a rendering process can be considered real-time from 15 fps. Beyond 72 fps, the user can not notice any difference if the rendering speed is further increased.

To obtain such a rendering speed, hardware means for accelerating the graphical calculation dedicated to this function are commonly used, known as graphics processing unit or GPU (Graphic Processing Unit). Thus, the rendering method implemented by a 3D rendering engine uses optimized algorithms, but also many preprocessing on the 3D scenes operated by the GPU. The use of such real-time 3D rendering methods using hardware acceleration is hampered by the heterogeneity of the available machine parks which complicates the deployment of applications using such methods.

Thus, several versions of the same application must be developed and configured to adapt to different hardware configurations. The 3D models used by these applications are even frequently specific to the hardware configuration of the target machine. In addition, these applications require a minimum hardware configuration preventing their use on terminals or personal computers of low capacity such as "netbooks", office laptops, tablet PCs. It is therefore desirable to develop the most efficient rendering methods possible to not use the hardware acceleration means, or to use these means optionally. Among the real-time 3D rendering methods, it is known, for example from WO2009063319A2, to use a so-called radius casting method, consisting in calculating for a given point of the image the color attributed by calculating the intersections. between a ray passing through the point of the image considered and the objects of the scene. An advantage of the ray tracing method is to allow easy parallelization, particularly with regard to the calculations relating to the primary rays. Indeed, the calculation of the intersections of a ray does not affect that of a neighboring radius. It is therefore easy to take advantage of modern multi-core processors and SIMD (Single Instruction, Multiple Data) type instructions. In order to effectively calculate the intersections between the rays and the objects of the scene, it is known to organize the data concerning the objects of the scene in the form of a specific tree-type data structure grouping the elements of the scene. by proximity to limit the number of intersection calculations to be performed. In particular, data structures of the type Regular Grid, octree, BVH, QBVH, BIH, Kdtree are used. The use of a specific data structure of this type makes it possible to significantly accelerate the rendering of static scenes. In the case of a scene with animated objects, the use of such a structure is complex and more expensive in computing time because said structure must be updated or reconstructed when rendering each image. The object of the present invention is to solve all or some of the disadvantages mentioned above.

For this purpose, the present invention relates to a method for rendering real-time images from a three-dimensional virtual scene using a ray tracing technique, comprising the steps of building from the scene model a data structure for enabling an accelerated calculation of the intersections between the rays and the 3D scene, said data structure for grouping elements of the scene by spatial proximity, characterized in that the construction of the data structure comprises a first step of merging all the static objects of the scene before proceeding with the grouping of 3D geometric data by spatial proximity. Thanks to the provisions according to the invention, and to the creation of a data structure defining an organization of the objects in different bounding boxes on the basis of a merger of objects, the calculation performances of the intersection with the scene are significantly improved over a known solution in which the structure is built on the basis of independent objects. Indeed, it is thus possible to allow groupings between close 3D data initially belonging to two distinct objects. These arrangements make it possible to render a complex scene without using GPU-type graphics acceleration while maintaining acceptable performance. According to one aspect of the invention, the data structure is constructed with the set of 3D data of the static objects and, where appropriate, at least one animated object, the animated object (s) being specifically marked when their animation is activated. . These provisions make it possible on the one hand to treat animated objects as static objects when they are not animated and thus to benefit from performance gains, and on the other hand to eliminate these objects from the intersection calculations of the static part of the scene easily, without having to reconstruct the data structure containing the static objects. According to another aspect of the invention, a second specific data structure to be updated to each image including the animated objects being animated is constructed. According to one aspect of the invention, the animated object 3D data is not located in the second data structure, but remains in the first static data structure in nodes of this node dependent structure corresponding to the animated objects. According to another aspect of the invention, the rendering of a final image is achieved by: - calculating an intermediate image representing the static objects and comprising, for each point of the image or is shown a static object of the depth information of the intersection of a ray launched with said static object; and then - by updating in this intermediate image the pixels corresponding to animated objects being animated by comparing for the points of the image for which an intersection of a ray with an object being animated is detected. intersection depth information for the first intermediate image and for the moving object considered, the nearest intersection being maintained.

According to another aspect of the invention, the first intermediate image also includes animated objects that are not being animated. According to another aspect of the invention, the determination of the first image is performed using a path of the first data structure grouping the static objects; and the update corresponding to the objects being animated is performed using a path of the second data structure grouping the objects being animated. When the user's point of view stops moving relative to the scene during a determined period of time, the first intermediate image constitutes a background that remains identical for the entire duration and does not require a new calculation. In this case, it is sufficient for each image to copy this background and update the pixels with animated objects corresponding to the second intermediate image, if they correspond to points of intersection closer than those of the background.

According to another aspect of the invention, an intermediate result of the rendering method is produced in the form of a data buffer, comprising position information in the image and corresponding element identifiers of the geometry. According to another aspect of the invention, the lighting and / or shading steps and / or one or more reprocessing passes are performed using the data buffer.

These arrangements make it possible to carry out the lighting and / or shading steps and / or one or more reprocessing passes without interfering with the first ray launching steps. According to another aspect of the invention, an automatic detection of the configuration of the user machine is performed in order to choose the rendering mode and the available options, and in particular to allow the activation or deactivation of the realization of steps rendering, to decrease the resolution of the image when moving from the point of view of the user in the scene or to disable shading and / or reflection calculation and / or transparency features. According to another aspect of the invention, the data structure used can in particular be a structure of the type "bounding box hierarchy" BVH (Bounding Volume Hierarchy). According to another aspect of the invention, the 3D geometric data of the static objects are processed to apply transformations (or "baking") to express their coordinates in a frame of the scene common to all objects. According to another aspect of the invention, the 3D geometrical data corresponding to the animated objects also undergo a transformation processing (or "baking") so as to express their coordinates in the frame of the scene taking into account the original position of the objects animated in the animation. These provisions make it easier to calculate the intersections between the geometry of the scene and the launched rays.

According to another aspect of the invention, the data structure comprises a tree having a set of nodes, these nodes being stored in series in a table, each node of the tree containing a reference on a left child node or respectively of right, the right child node, respectively left, located in the next box of the table. According to another aspect of the invention, the data structure comprises a tree whose leaves include a pointer or a reference on the geometric data of the scene contained in the 3D model.

The invention also relates to a computer program product implementing the steps of the method described above, as well as a computer system comprising one or more computers or terminals in the possession of a user comprising one or more processors making it possible to execution of this program. The invention will be better understood from the detailed description which is set forth below with reference to the accompanying drawings in which: FIG. 1 is a flowchart of part of the method according to the invention; FIG. 2 schematically illustrates the principle of the step of melting the static objects of the method of FIG. 1; Figures 3 and 4 illustrate the step of constructing the data structure bringing together the elements of the scene by spatial proximity of the method of Figure 1; Figures 5 and 6 illustrate the data structure used by the method of Figure 1; Figure 7 illustrates the relationship between the static object data structure and the data structure of the objects being animated. Figure 8 illustrates the rendering step of the method of Figure 1. Figure 9 illustrates updating and rendering in the case of a scene including moving objects. Figure 10a illustrates a path of a data structure as part of the rendering step of the method of Figure 1 for a particular ray, for the static part of a scene including animated objects. FIG. 10b illustrates a path of a data structure as part of the rendering step of the method of FIG. 1 for a particular spoke, for the animated portion of a scene comprising animated objects. FIG. 11 illustrates the substeps of the rendering step of the process of FIG. 1. FIG. 12 illustrates a sub-step of taking into account the transparency in the rendering step of the process of FIG. 1 and FIG. FIG. 13 illustrates a sub-step of taking into account shadows carried in the rendering step of the process of FIG. 1 and FIG. 9. FIG. 14 is a table containing performance measurements for the rendering of a typical scene . FIG. 15 illustrates the separation of radius packets as part of a rendering step of the method of FIG.

As shown in FIG. 1, a method according to the invention makes it possible to input a 3D model representing a 3D virtual SC scene according to a standard type format and to apply to it in a first step E1 an analysis intended to identify the static OS objects from the Sc scene and OA animated objects. In a second step E2, a merger or a "collapse" of the static objects OS is performed, so as to obtain a single merged OSF object comprising all the 3D data relating to the static objects OS, as illustrated in FIG. 2 .

In a third step E3, a specific data structure S intended to allow an accelerated calculation of the intersections between the rays and the elements of the 3D scene is constructed. This data structure makes it possible to group the 3D data of the scene Sc, in particular constituted by triangles T, by spatial proximity.

The structure S comprises a tree with a set of nodes and leaves. A node contains BE bounding box information. The bounding box BE is aligned with the axes of the global coordinate system of the scene Sc, which is known in the known manner by the acronym "AABB" (Axis Aligned Bounding Box). When it is a sheet f, the node contains, in addition to the bounding box, a group of 3D data, in particular constituted by triangles. The data structure S used may in particular be a structure of the "bounding box hierarchy" or BVH (Bounding Volume Hierarchy) type. The BVH is used because it has a low memory footprint and is fast to build. Other types of data structures may, however, be used. By way of example, there may be mentioned structures comprising a regular grid, an octree, a BIH which is a variant of the BVH, a QBVH which is a variant of the BVH which aims to divide a node into 4 instead of 2, a KdTree. The two structures that are mainly used are KdTree and BVH because they are adaptive and can therefore effectively deal with scenes with significant differences in the density of 3D data, which can be illustrated by a scene including a teapot in a stadium, the teapot representing a very large local density of data in a comparatively larger size set. These two structures are relatively fast to build and cross compared to the other structures mentioned above. The tree structure S comprises a root R, a node NOSF corresponding to the static objects of the scene, NOA nodes corresponding to each of the animated objects, and intermediate nodes NI grouping two by two between the nodes NOSF and NOA until the root as shown in particular in Figure 6 and Figure 5. When setting up the structure S, the 3D data consisting of triangles belonging to the static object merged and accessible below the node NOSF are processed to apply them transformations (or "baking") so as to express their coordinates in the frame of the scene, and no longer in the local coordinate system of each object. This arrangement makes it possible to apply these transformations only once before rendering and thus to optimize the process speed. In the same way, the 3d data constituted by triangles of the NOA nodes corresponding to the animated objects undergo this pretreatment of transformation taking into account the original position of the object OA in the animation.

The construction of the lower levels of the Structure S tree, below the NOSF and NOA nodes, is done by grouping 3D data by proximity according to a heuristic called SAH. Other methods of spatial grouping can also be used. According to one embodiment, the first cut of the geometry of the scene is made along the longest axis thereof. By subsequently applying the SAH method, the bounding box B corresponding to a node will be cut into a predetermined number X of pre-cuts on each axis. A node n given not being a leaf of the tree 30 has two links to two child nodes nd and ng which will be respectively arranged to the right and left of a cut. Each triangle of the geometry included in the bounding box BE (n) of a node n at a given level of the tree will be inserted into the right or left nd child node corresponding to the right or left part of the cutoff. , depending on the position of its center of gravity. The position of the cutoff C among the X pre-cuts is determined as a function of a coefficient K weighted by the number of triangles T and the area of the bounding box BE of these triangles T being on each side of the cutoff C as this is illustrated in FIG. 4. The use of a discrete number X of re-interruptions makes it possible to increase the speed of calculation of these cuts. Two bounding boxes BE (nd) and BE (ng) are thus determined for the right nd and left ng respectively. The realization of the cut described above is repeated for the child nodes nd and ng until reaching a stop criterion and to constitute the leaves 10 of the tree. According to one embodiment, the stopping criterion is a number of triangles per leaf of the tree, for example equal to 5, this number being adjustable. According to one aspect of the invention, the nodes of the tree are stored serially in an array, each node n containing a reference on the left child node. The right child node is in the next box of the table. According to a variant, each node n containing a reference on the right child node. The left child node is in the next box of the table. The nodes f constituting the leaves of the tree structure for their part comprise a pointer to geometric data of the scene Sc contained in the 3D model M. In particular, the referenced data relate to: the 3D data constituted by the coordinates of the triangles T , information on the textures and the corresponding materials. Once the structure S as described above thus constituted, a rendering of the scene can be performed in a fourth step E4. The method according to the invention uses a ray casting rendering method. It is therefore necessary to calculate for a given point of the image the color allocated by calculating the intersections between a ray passing through the point of the image considered and the static objects OS or the animated objects OA of the scene Sc. In this way, this structure is traversed and the following tests are carried out on the nodes n of this structure: (1) Is the bounding box BE (n) of the node in front of or behind the camera? (2) Is this box in a pyramid of vision (or Frustum) delimited by the field of vision of the camera delimited between a close plane and a distant plane? (3) Is there an intersection between the box and a particular ray corresponding to a point in the image? In the case of a scene including animated objects OA, the structure S must take into account the movements of these animated objects from the moment when their animation is active. As long as these objects remain motionless, they can be treated as static objects. From the moment the animation is active, a specific treatment must be considered, since the bounding boxes calculated and the structure S must be modified.

For this purpose, when activating the animation of an object OA, the elements of this object are "blacklisted" or marked as animated in the data structure S. Moreover, the transformation processing ("baking" ) to express the coordinates of the geometry in the frame of the scene Sc are canceled. The NOA node corresponding to the animated object is inserted into a second "dynamic" data structure Sa corresponding to the animated objects being animated, for example of the BVH type. This second data structure Sa comprises, as shown in FIG. 7, a root R, animated nodes NOAa corresponding to the animated objects being animated, and intermediate nodes NI grouping two by two between the NOAa nodes. to the root. According to one aspect of the invention, the 3D data relating to the animated objects OA are not located in the data structure Sa, but remain in the static data structure Sa in the nodes n below the animated object node NOA . Thus, as shown in FIG. 7, during a traversal of the data structure of the objects being animated Sa, if we arrive at a node corresponding to an animated object, for example the reference node NOAa2 on In the figure, the part of the static data structure S under the node NOAa2 corresponding to the same object OAa2 is then traversed, as indicated by the transition arrow t. In the structure S, as indicated by the marking cross m, the node NOAa2 is blacklisted or marked as an animated object.

Thus geometric pre-transformations (or "baking") for this node have been canceled. The geometric transformations for the image considered, which correspond to a position in the animation, are applied during the course of the nodes n below the node NOAa2 to the sheets. These provisions avoid copying all the geometric data in two structures. According to one aspect of the invention, the rendering of a final image I is achieved by: calculating at first a first intermediate image Is comprising all the static objects OS and animated OAs not being animated and information of depth of the intersections corresponding to each point of the image Is; then - by updating in this intermediate image Is the pixels corresponding to animated objects being animated OAa by comparing for the points of the image I for which an intersection of a radius with an object being animated OAa is detected the intersection depth information d for the first intermediate image and for the animated object considered OAa, the closest intersection being maintained. The determination of the first intermediate image Is is carried out using a path of the first data structure S grouping the static objects OS and animated AOa not being animated; and updating corresponding objects being animated is performed using a path of the second data structure Sa grouping the objects being animated AOa. As is schematically represented in FIG. 9, it is therefore appropriate for each image to update the structure Sa in a first step ER1, then to proceed to a crossing of this structure Sa of the animated objects in a second step ER2 for each radius . If the point of view with respect to the scene has been modified, it is also necessary to proceed to a course of the structure S in a third step ER3. Finally, lighting ("shading") is achieved.

It should be noted that the steps ER2 and ER3 are not necessarily sequential but can be performed in parallel. In the same way, the steps ER2, ER3 and ER4 are not necessarily separated, as will be detailed later.

The intersection calculation used in step ER4 referenced in FIG. 9 for the determination of the first image Is based on the static part of the structure S is described in FIG. 10a for a given radius Ry. The path of the structure S starts from the root R as indicated in step E10 and uses a processing stack which is initially empty. In a step E11, it is determined whether the node n considered is a sheet f. If yes, in a step EI2, it is determined whether the ray Ry has an intersection with the bounding box BE of said sheet f. If so, the intersection of the radius Ry with the geometry corresponding to the sheet and constituted by triangles T is determined in a step EI3 and then, in a step EI4, the result is written, including in particular the depth d of the intersection with respect to the point of the image and a contribution to the color of the point of the image. Then, in a step EI5, it is checked whether the node processing stack n to be scanned is empty. If so, the tree path for the Ry ray is complete. Otherwise, return to step E11 for a new iteration. In the case where, in step EI2, it appears that there is no intersection between the ray Ry and the bounding box BE of the sheet f, the step EI5 is carried out. Now consider the case where the test result of step E11 is negative, ie node n is not a sheet. In this case, in a step EI7, it is determined whether the ray Ry intersects with the bounding box BE of the node n. If so, it is determined in a step EI8 if the child nodes nd and ng of the node n are marked as being animated (or "blacklisted"). If some of the wires are not marked ("blacklisted"), they are stacked on the processing stack in a step EI9. Then return to step EI5.

If in step EI8 it is found that all the wires are marked, step EI5 is executed as described above. The use of this traversal of the tree for the Ry rays corresponding to the points of the image makes it possible to constitute the first image 5 Is. The intersection calculation used in the step ER2 referenced in FIG. 9 which uses a path of the structure grouping the animated objects Sa in order to compute the image 1a, and described in the description in FIG. 10b, is similar to that described with reference to FIG. 10a. However, the marking test has been removed. Thus, in the case where the test of step E17a is positive, a stacking of the child nodes in the treatment stack is performed, without verification of any marking. As we have seen previously, the geometrical data in themselves are not contained in the structure Sa, but in the structure S The step ER4 of lighting ("shading") indicated in FIG. present in detail with reference to Figures 11 to 13. Figure 11 details in general the different stages of rendering. As indicated above, the lighting / shading steps can be performed in several steps, in combination with several ray tracing steps. Thus, after a first step of creating P-ray packets ERD1, the realization of a calculation for a primary ray Ry as detailed with reference to FIGS. 9, 10a, 10b is carried out in a second step ERD2. In a third step ERD3, lighting ("shading") is performed as well as casting of shadow rays as detailed in FIG. 13. In a fourth step ERD4, a calculation of the rays of transparency is carried out, which will be detailed below with reference to FIG. 12. Then, in a fifth step ERD5, lighting ("shading") 30 is performed as well as casting shadow rays. In a sixth step ERD6, a calculation of the transparency rays is performed. Then, in a seventh step ERD7, a lighting ("shading") is performed as well as a casting of shadow rays. In an eighth step ERD8, a calculation of the anti-aliasing rays is performed. Then, in a step ERD9, a lighting ("shading") is performed as well as a casting of shadow rays.

Referring now to Figure 12, consideration of transparent objects is detailed. In an ETO step, a rendering relative to the primary rays as described with reference to Figures 10a and 10b is made. During this rendering for the primary rays, a test is performed, not shown in FIGS. 10a and 10b but represented in step ET1 of FIG. 12, in which it is determined whether an intersection with a transparent object occurs. If no intersection with a transparent object is detected, the verification is completed in step ET5. If an intersection with a transparent object is detected, in a step ET2 subsequent to the rendering of the primary rays, a displacement of the origin of the ray is performed, to position the origin at the intersection with the transparent surface. Moreover, it is possible to mark or "blacklist" the transparent surface, that is to say the triangles T forming part of this surface, to avoid generating the detection of an intersection of the ray reflected on this surface. because of a lack of precision in the relative positioning of the new origin and the transparent surface. Then, in a step ET3, a new route of the data structures S, Sa is made to determine the new intersections, then in a step ET4, a lighting ("shading") is performed. As described in FIG. 13, the taking into account of the drop shadows is carried out as follows. As part of a rendering for the primary beams, it is determined in a step E01 the presence or absence of lights projecting shadows on the scene. In this case, the creation of a shadow packet is performed in a second step EO2, then in a third step E03, the rendering is performed taking into account these packets of shadow beams. It should be noted that the calculation of ray throws can be parallelized in the manner described in FIG. 15. Thus, by sharing the image I in portions each corresponding to a packet P of rays, for example comprising 16 × 16 rays, it is It is possible to process the calculations in parallel for the different ray packages. Single Instruction instructions, SIMD Multiple Data (MMX, 3DNow !, SSE, SSE2, SSE3, SSSE3 and SSE4) can be used. The calculations can be parallelized on the processors of the machine. The performance is almost linearly multiplied by the number of physical processors available. According to one aspect of the invention, when moving from the user's point of view in the scene, the resolution of the image may be lowered in order to improve the rendering performance. In addition, some parts of the illumination calculation can also be calculated at lower resolutions or disabled during travel, such as techniques known as Bump Mapping. In the same way, the reflection calculations can be calculated at lower resolutions or disabled if necessary. According to one aspect of the invention, an automatic detection of the configuration of the user machine is performed in order to choose the rendering mode and the available options allowing the best user experience to provide the best level of performance.

According to one aspect of the invention, in the case of a low-resolution rendering, the anti-aliasing is used to increase the quality of the rendering on the parts of the image corresponding to contours. According to one aspect of the invention, an intermediate result of the rendering method, before the final image I, is an intermediate data structure B, in the form of a data buffer, comprising positions in the image and identifiers of triangles T. These data are ordered in packets of radii P. In particular, the structure used may be of the type known as "FatBuffer".

So-called "Post-Process" effects can be applied to the result obtained, and in particular the following known effects: Depth of Field, Desaturate, RadialMask, LightDirection Mask, Fog, ColorCurve, Glow, MotionBlur, SSAO. It is also possible to take into account in the process according to the invention texturing effects, and in particular the following known effects, available for example in the 3DS MAX® software marketed by Autodesk: Bump, Diffuse, Specular Level, Self Ilium, Opacity, Reflection. According to one embodiment, lighting of the PHONG scene S is performed. Diffuse and specular lighting is particularly achieved. Of course, other types of lighting methods can be envisaged. By way of example, the following processors make it possible to implement software running the method according to the invention: Intel Core i7®, Atom®, Athlon X2® 64, Intel Core2 Quad®, Intel Pentium D®, Intel Pentium III®, Intel Centrino 2®. The table in Figure 14 represents various performance measurements, as a rough guide, for moving to the Crytek © Sponza 93 kTrl test scene with 93281 triangles and 62852 vertices with textures. The test protocol is as follows: 135 different camera positions are defined in the scene. The time required (in ms) is measured to calculate the images corresponding to each position. We then deduce in average fps all over the course. The data in the table indicates the type of processor used, the rendering resolution (in pixels), the use or not of textures, the time required to render a path on 135 images (in milliseconds) and the number of images per second (fps) average obtained on the defined path. Thanks to the provisions of the invention, an adaptive rendering pipeline is set up according to the calculation and storage possibilities of the materials used for visualization (workstation, laptop, PDA, mobile phone, etc.), allowing to visualize in real time scenes composed of a few million polygons and providing an image quality adapted to the material possibilities available. The arrangements according to the invention also make it possible not to adapt the 3D model at the input of the process to this one. Although the invention has been described in connection with particular embodiments, it is obvious that it is not limited thereto and that it comprises all the technical equivalents of the means described and their combinations if they are within the scope of the invention.

Claims (6)

REVENDICATIONS1. A real-time image rendering (I) method from a three-dimensional virtual scene (Sc) using a ray tracing technique, comprising the steps of constructing from the scene model a structure of data (S, Sa) for allowing an accelerated calculation of the intersections between the rays (Ry) and the 3D scene (Sc), said data structure (S, Sa) for grouping elements of the scene by spatial proximity, Characterized in that the construction of the data structure (S) comprises a step (E2) of merging the set of static objects (OS) of the scene before proceeding (E3) to the groupings of the 3D geometric data (T) by spatial proximity. The method of claim 1, wherein the data structure (S) is constructed with the set of 3D data of the static objects (OS) and animated objects (OA), the animated objects (OA) being specifically marked when their animation is enabled. 3. Method according to one of the preceding claims, wherein a second specific data structure (Sa) intended to be updated to each image (I) comprising the animated objects (OAa) being animated is constructed. 4. Method according to one of the preceding claims, wherein the rendering of a final image (I) is achieved by: - calculating in a first step a first intermediate image (Is) representing the static objects (OS) and comprising, for each point of the image or is shown a static object information depth (d) of the intersection of a ray launched with said static object (OS); then - by updating in this intermediate image (Is) the pixels corresponding to animated objects (OAa) being animated by comparing for the points of the image for which an intersection of a ray with an object in progress animation (OAa) is detected the intersection depth information (d) for the first intermediate image and for the moving object considered, the nearest intersection being preserved. The method of claim 4, wherein the first intermediate image (1a) also includes animated objects (OAs) that are not being animated. 6. Method according to one of claims 4 or 5, and according to claim 3, wherein the determination of the first image (Is) is performed using a path of the first data structure (S) grouping the static objects ( OS, OAs); and the update corresponding to the objects being animated (OAa) is performed using a path of the second data structure (Sa) grouping the objects being animated.7. Method according to one of the preceding claims, in which an intermediate result of the rendering method is produced in the form of a data buffer (B), comprising position information in the image (I) and element identifiers of the corresponding geometry (T). The method of claim 7, wherein the lighting and / or shading steps and / or one or more reprocessing passes are performed using the data buffer (B). 9. Method according to one of the preceding claims, wherein an automatic detection of the configuration of the user machine is performed in order to choose the rendering mode and available options, and in particular to enable activation, degradation or disabling rendering steps, decreasing the resolution of the image when moving from the viewpoint of the user to the scene, or disabling lighting and / or reflection and / or calculation of transparency. 10. Method according to one of the preceding claims, wherein the data structure (S) used may in particular be a structure of the type "BOH" (Bounding Volume Hierarchy). 11. Method according to one of the preceding claims, wherein the 3D geometrical data (T) of the static objects are processed to apply transformations (or "baking") so as to express their coordinates in a frame of the scene ( Sc) common to all the objects (OS) of this scene (Sc). 12. The method according to claim 11, wherein the 3D geometrical data (T) corresponding to the animated objects (OA) also undergo a transformation processing (or "baking") so as to express their coordinates in the frame of the scene (Sc). ) taking into account the original position of animated objects (OA) in the animation. Method according to one of the preceding claims, wherein the data structure (S) comprises a tree having a set of nodes (n), said nodes (n) being stored in series in an array, each node (n). ) not being a leaf of the tree containing a reference on the left child node (ng) respectively of right (nd), the node of right (nd), respectively of left (ng), lying in the next box of the table. 14. Method according to one of the preceding claims, wherein the data structure (S) comprises a tree whose leaves (f) comprise a pointer or a reference on geometric data (T) of the scene (Sc). ) contained in the 3D model (M). 15. Computer program product for implementing a method according to one of the preceding claims.