Method and device for modelling a real object

Abstract

The invention relates to a method for modelling a real object, the method comprising the steps of: • obtaining a set of two-dimensional and / or three-dimensional data linked to the object, wherein the dataset is associated with a first space; • generating a three-dimensional representation of the object in a second space from the dataset; wherein the three-dimensional representation comprises a plurality of parts, the surfaces of which are generated via a mathematical function describing a parametric surface, wherein the step of generating the three-dimensional representation of the object in the second space, referred to as the canonical space, comprises a sub-step of adjusting the shape and size of each of the parts to at least one element of the real object, and wherein the fitting is carried out while taking into account at least one structural constraint of the object.

Description

Method and device for modeling a real object TECHNICAL FIELD OF THE INVENTION

[0001] The field of invention is that of modeling an object.

[0002] More specifically, the invention relates to a method and device for modeling a real object.

[0003] The invention finds applications in particular in augmented reality to improve the rendering of a three-dimensional representation of the real object, for example in the context of a virtual fitting of the object by an individual. STATE OF THE ART

[0004] Techniques for modeling an object are known from the prior art, making it possible to obtain a three-dimensional representation of the object in a so-called modeling space. Conventionally, these techniques are based on image processing and / or a scan of the object obtained by varying the viewing angle of the object. Generally, the representation of the object obtained by these techniques is a model formed in a single step, in a discrete manner, without any segmentation between the different elements forming the object. For example, in the case of a pair of glasses, the three-dimensional representation obtained from a scan of the pair of glasses worn by a support does not differentiate the different elements such as the branches, the tenons, the circles, the lenses, the bridge, or even the support. All these elements are delimited by a cloud of points forming a single three-dimensional surface without any differentiation of the elements.This is also the case when scanning a pair of shoes where the laces are fused to the upper and / or tongue of each shoe. Similar difficulties are encountered when scanning a handbag with handles, a strap or a chain that tend to fall back and be fused to the outer lining of the bag without distinguishing the two pieces.

[0005] The discretization of the cloud points leads on the one hand to a heaviness of use of the file and on the other hand to inaccuracies at the level of the modeled surfaces, or even light artifacts in the rendering, especially when it is a translucent texture. Biases can also be introduced such as smoothing of details or the introduction of artifacts. In the example of the scanned pair of glasses, the smoothing of details is particularly visible at the level of the tenons where the different parts of the tenon such as the screws or the hinge are merged together and therefore unidentifiable. Artifacts are also introduced at the point of contact between the support and the temple of the pair of glasses which tend to be merged together, with in addition a burr which is added between the vertical support and the horizontal temple of glasses.

[0006] Furthermore, prior art techniques are often limited to opaque and matte objects. A paint providing a matte appearance is generally used when the object is shiny or translucent, to avoid misinterpretations regarding the position of the object's outer surface during image processing or scanning. Without this paint, holes in the mesh obtained during a scan can be formed due to reflections on the surface, particularly in the case of stereovision reconstruction with images taken from different angles (generally of the order of a few degrees of difference). In addition, the internal parts that would be visible on the real object are then masked by the matte paint in the images or scan obtained, resulting in a loss of information for the rendering of the modeled object.

[0007] It should be emphasized that the three-dimensional representation of an object, obtained by the techniques of the prior art being in a single block, without segmentation, is not suitable for a virtual fitting of the modeled object requiring an articulation to better fit a shape on which the object is positioned. For example, in the context of a virtual fitting of glasses, or a superposition of the three-dimensional representation of the object on an image of the object in a video stream, the articulation and the deformation of the branches relative to the face of the pair of glasses may be necessary to obtain a realistic adjustment. In the case of a superposition, alignment defects may further apply biases either in the case of a measurement being taken, or in the case of a masking of the actual pair of glasses visible in the image.These disadvantages are particularly significant in the context of augmented reality where the processing of images from the video stream is carried out at every moment.

[0008] On the other hand, the three-dimensional representation obtained being very heavy, with only the capture of the external envelope, that is to say without knowledge of the internal parts, it becomes complex for a computer operator to easily take it up in order to obtain a reliable segmented representation in terms of architecture of external or even internal parts – for example a metal core inside a translucent branch – or in terms of textures for each part. It thus frequently happens that the textures are imprecise, either in terms of texture type (solid or translucent material, plastic, glass, wood, etc.) or in terms of texture position, in particular at the level of existing details or inserts in the object to be modeled. In the same way, a simplification of the model is often necessary to convert it into a model usable in real time.Existing state-of-the-art meshing techniques are often resource-intensive and not optimized for the model to be reconstructed since they have no a priori on this model.

[0009] Since the resumption of the three-dimensional representation to obtain a satisfactory result is complex, it can hardly be carried out on a standard computer because it requires significant storage and computer memory capacities, or even a powerful graphics card to manipulate the computer file of the three-dimensional representation.

[0010] Furthermore, segmenting the resulting three-dimensional representation can be tedious, especially at the joints between multiple parts. It should be noted that joints are generally part of the smoothed details in the three-dimensional representation obtained by conventional processing of images and / or scans of the object.

[0011] Furthermore, with current modeling methods, the textures to be applied to the three-dimensional representation each have points grouped by proximity. It thus happens that points located on two distinct parts end up on the same texture, making the texture difficult to correct by an external operator who has difficulty seeing at first glance which part(s) the texture corresponds to.

[0012] Finally, it often happens that the object to be modeled is deformed at the time of image acquisition or scanning, either by the mechanical constraints of the support, or by gravity, as for example in the case of a handbag including a metal chain.

[0013] The three-dimensional representation then also includes the same deformations, which can lead to inaccuracies when the representation is used for taking measurements, for example.

[0014] None of the current systems can simultaneously meet all the required needs, namely to propose a technique for modeling an object from real data which makes it possible to obtain a three-dimensional representation of the object which is precise and reliable in order to obtain a very good rendering of the object for an application in augmented reality or for any other applications using a representation of a real object, in particular at the level of details and textures, and in particular for translucent parts of the object.

[0015] The present invention aims to remedy all or part of the drawbacks of the state of the art cited above.

[0016] To this end, the invention relates to a method for modeling a real object comprising a plurality of elements, the modeling method comprising steps of: Obtaining a set of two-dimensional and / or three-dimensional data, linked to the object, the data set being associated with a first space; Generating a three-dimensional representation of the object in a second space from the data set.

[0017] According to the invention, the three-dimensional representation comprises a plurality of parts whose surfaces are generated by a mathematical function describing a continuous parametric surface, the step of generating the three-dimensional representation of the object in the second space, called canonical, comprising a sub-step of adjusting the shape and dimension of each of the parts to at least one element of the real object, the adjustment being carried out taking into account at least one structural constraint of said object.

[0018] Advantageously, the three-dimensional representation is of the canonical type, that is to say without deformation, as if coming directly from computer-aided design software.

[0019] Furthermore, the parts of the three-dimensional representation are derived from a generic parametric three-dimensional model of the object, the constraints between said parts being inherited from said generic parametric three-dimensional model, a constraint being included in the list {positioning constraint, orientation constraint, animation constraint}. The adjustment is advantageously carried out by taking into account one of said constraints between parts.

[0020] Advantageously, the modeling method also comprises a step of automatically determining a geometric transformation between the object of the acquisition space and the three-dimensional representation of said object in the canonical space, the transformation comprising at least one of the following elements: Straightening of a part of the three-dimensional representation according to a pre-established constraint on said part; Adjustment of an angle between two parts of the three-dimensional representation according to a pre-established constraint between said two parts; Symmetrization of the three-dimensional representation.

[0021] Thus, the three-dimensional representation is of the canonical type, that is to say ideal in relation to a standard, because it is not deformed.

[0022] Furthermore, the rendering calculated on the three-dimensional representation of the real object is more accurate, less subject to rendering errors, because the surfaces of the parts composing the three-dimensional representation are continuous surfaces, calculated mathematically for example according to control points arranged in the second three-dimensional space. A mathematical function describing a mathematical surface via control points is for example a Bézier surface or non-uniform rational B-splines, also known as NURBS (English acronym for "Non-uniformrational B-spline"), composed of a plurality of Bézier or NURBS curves.Another example of a mathematical function describing a parametric surface is, for example, the concatenation of a plurality of elementary volumes, positioned relative to each other in the second three-dimensional space, each volume being parameterized by at least one value (for example, a radius for a sphere).

[0023] It should be emphasized that a mesh is a succession of points, each three-dimensional point being associated with a plurality of three-dimensional points, the set of points defining a discontinuous surface. The use in the modeling method according to the invention of a mesh to describe the parametric surface can be envisaged but corresponds to a degraded mode of modeling the real object. It can be envisaged to use both mathematical functions describing continuous parametric surfaces for the external or visible parts of the real object and mathematical descriptions defining mesh points delimiting a surface for internal or non-visible parts of the real object.

[0024] Furthermore, rendering accuracy is also improved by the fact that parts are modeled in a segmented manner, taking into account at least one constraint. A constraint can be an internal constraint to a part or an inter-part constraint, such as an alignment, an orientation of one part relative to another, an angle between the two parts, a connection between two parts, the presence of one part inside another, etc. Detail areas can thus be modeled more finely than in the prior art, significantly increasing the rendering quality on translucent parts of parts, for example comprising a metal core. Surface normals are also smoother because they come from mathematical models and not from a discrete mesh (generated above a point cloud), resulting in more continuous reflections.Texture seams can also be placed in less important areas if the semantics of the objects constituting the model are known to limit visual artifacts.

[0025] A constraint can be internal to a part, such as a radius of curvature, the presence of a concavity, surface continuity, a relationship between two control points, or even continuity of normals at a point. Constraints can be purely experimental or derived from professional knowledge of the manufacturing process (for example, since most eyeglass lenses are cut from a spherical cap, this shape constraint can be translated into a geometric constraint).

[0026] A constraint can also be of an inter-part type, such as an alignment relationship, an angular relationship, a connection between two parts, etc.

[0027] It should be noted that adjusting a part can cause it to deform by changing its dimensional parameters. Dimensional parameters can be global parameters – such as length, width, height, depth – or local parameters such as a radius of curvature or the position of a control point.

[0028] The three-dimensional representation being advantageously segmented, allows for more detailed modeling, which can be resumed more quickly by an operator who can effectively correct the defects in the initial data set.

[0029] The augmented reality process using three-dimensional representation can thus have a better rendering, for example for a virtual fitting of glasses, be more precise, in particular when taking measurements in relation to the object, positioning the virtual object, masking the object or a virtual fitting.

[0030] Two-dimensional data from the obtained dataset can be, for example, an image of the object, a plan of the object, an artistic work of the object, a map of the object, etc.

[0031] Preferably, two-dimensional data can be acquired by an imaging sensor, for example of the CMOS or CCD type. Two-dimensional data can be advantageously calibrated or not.

[0032] Three-dimensional data from the obtained dataset may be, for example, a three-dimensional model in a three-dimensional space, a 3D mesh, a point cloud in a three-dimensional space, etc. Three-dimensional data may be acquired by a three-dimensional imaging sensor, such as a scanner, an MRI or ultrasound imaging system, etc.

[0033] The initially obtained dataset may also be an elevation map, i.e., two-dimensional data forming an image, each pixel of the image being associated with an elevation value, generally corresponding to a distance of the pixel from the image acquisition device. Such data, subsequently called 2.5D data, may be, for example, a depth map, a topographic map, an architectural plan. A depth sensor may thus be used to acquire 2.5D data.

[0034] In particular embodiments of the invention, the adjustment of each part is constrained by the superposition of at least one surface curve of said part with a contour of a distinct element of the object, said contour being previously deduced from the data set.

[0035] In particular embodiments of the invention, a structural constraint is a predetermined geometric constraint on a part or between two parts.

[0036] Thus, a known architecture of the object can be followed, allowing to significantly improve the precision of the adjustment. For example, in the case of modeling a pair of glasses, it can be advantageous to know the composition of the tenon operating the hinge between the face of the pair of glasses and the branch to finely model this part which tends to be merged during a scan of the object.

[0037] Geometric constraints can also be used to straighten the three-dimensional representation in canonical space in order to compensate for possible deformations of the object during the acquisition of the data set. Such constraints are, for example, knowing that for a pair of glasses the arms are parallel in the open position and form a 90° angle with the front of the pair of glasses.

[0038] In particular embodiments of the invention, a structural constraint is a continuity constraint of a curve of a part or a continuity constraint of the tangent at at least one point of the curve of said part.

[0039] This makes it possible to automatically remove artifacts from a surface whose curvature is known. Such artifacts can come, for example, from a fusion between a part of the object and the support, generated by a scan of the object.

[0040] In particular embodiments of the invention, the modeling method also comprises a step of identifying an outline of the object in the object data set and a step of determining a surface curve of the three-dimensional representation, corresponding to said outline of the object.

[0041] In particular embodiments of the invention, the data set comprises data from an acquired image of the object, the modeling method also comprising a step of calculating a geometric deformation factor of at least one part of the object.

[0042] It should be emphasized that the geometric deformation can be a structural deformation, for example a branch of a pair of glasses adjusted by an optician to the morphology of an individual, or be an optical deformation linked to an optical parameter of a system for acquiring said image, for example linked to the focal length of an objective of the acquisition system.

[0043] In particular embodiments of the invention, the surface curve is an edge of the three-dimensional representation of the object.

[0044] In particular embodiments of the invention, the data set comprises at least two views of the object from distinct angles.

[0045] In particular embodiments of the invention, the dataset comprises a view of the object and a depth map obtained concurrently with the view of the object.

[0046] In particular embodiments of the invention, the dataset comprises at least two depth maps obtained from distinct angles or a scan of the object.

[0047] In particular embodiments of the invention, the modeling method also comprises steps of: Recognition of at least one type of shape of the object and / or of a part of the object in the data set; Automatic selection of the parametric three-dimensional model as a function of the type(s) of shape previously recognized.

[0048] It should be emphasized that the parametric three-dimensional model selected based on the previously recognized shape type(s) corresponds to a generic three-dimensional model whose parameters are then adapted specifically to the corresponding object or part of the object. The selected model is therefore a generic parametric three-dimensional model.

[0049] In particular embodiments of the invention, the object is a pair of glasses, the recognition of at least one type of shape comprises all or part of the following sub-steps: Recognition of a type of shape of a frame included in the pair of glasses; Recognition of a type of shape of lens included in the pair of glasses; Recognition of a type of shape of a circle included in the pair of glasses; Recognition of a type of shape of a tenon included in the pair of glasses; Recognition of a type of shape of a bridge included in the pair of glasses; Recognition of a type of shape of a hinge between the frame and a temple included in the pair of glasses; Recognition of a type of shape of temple included in the pair of glasses; Recognition of a type of shape of a sleeve included on the temple; Recognition of a type of shape of nose pad included in the pair of glasses.

[0050] In particular embodiments of the invention, the parts of the three-dimensional representation come from a generic three-dimensional model of the object, the relationships between said parts being inherited from said generic three-dimensional model.

[0051] Said relationships correspond to constraints between said parts.

[0052] For example, to model a sneaker, a generic model of a shoe is used. Such a generic model can include a sole, a heel, an upper, a vamp, a slide, a tongue, or even laces. The generic model is then adapted and adjusted to the initially obtained dataset.

[0053] The generic model thus includes a collection of standard parts of a generic of the object to be modeled.

[0054] Advantageously, the generic model has been previously positioned and oriented in a similar manner to the object in a three-dimensional reference frame of a virtual space representative of the object's environment.

[0055] For this purpose, a detection of characteristic points of the object may have been previously carried out in the data set, the characteristic points making it possible to orient and / or position the generic three-dimensional model in the three-dimensional reference frame.

[0056] Alternatively, the position and / or orientation of the object relative to a sensor that acquired the dataset is known or estimated. In this case, the generic three-dimensional model can be positioned and / or oriented in the three-dimensional frame relative to the sensor, the position of which may, for example, correspond to the origin of the frame.

[0057] In particular embodiments of the invention, the three-dimensional representation comprises at least one articulation between two parts.

[0058] In particular embodiments of the invention, the generation of a three-dimensional representation is performed within a time period of less than one second after obtaining a data set.

[0059] The invention also relates to a device for electronically storing the instructions of a modeling method according to any of the preceding implementation modes.

[0060] Finally, the invention also relates to an augmented reality method implementing the method of modeling a real object according to any of the preceding implementation modes, and the three-dimensional representation of said object obtained. BRIEF DESCRIPTION OF THE FIGURES

[0061] Other advantages, aims and particular characteristics of the present invention will emerge from the following non-limiting description of at least one particular embodiment of the devices and methods which are the subject of the present invention, with reference to the appended drawings, in which: is a block diagram of an example of an embodiment of the modeling method according to the invention; is a photo of an example of a real object to be modeled, here a pair of glasses; is an image of a three-dimensional representation of the object of the obtained according to the method of the; is an image of a three-dimensional representation of the object of the according to a technique of the prior art; is a photo of another example of a real object to be modeled, namely here a shoe with laces; is an image according to a partial view of a three-dimensional representation of the object of the according to a technique of the prior art;is an image of a three-dimensional representation of an object similar to that of the according to the method of the; is a block diagram of another example of a mode of implementation of the modeling method according to the invention.; DETAILED DESCRIPTION OF THE INVENTION

[0062] This description is given without limitation, each characteristic of an embodiment being able to be combined with any other characteristic of any other embodiment in an advantageous manner.

[0063] Please note, from now on, that the figures are not to scale. Example of a particular embodiment

[0064] This is a block diagram of an example of an implementation mode of a modeling method 100 according to the invention.

[0065] The actual object 200 to be modeled in the present non-limiting example of the invention, illustrated via an image obtained by an image acquisition device, is a pair of glasses comprising several elements including a translucent frame 210 revealing a metal core 220 articulated by a hinge 225, making it possible to connect the face 230 and the branch 240, through the tenon 250.

[0066] The method 100 comprises a first step 110 of obtaining a set of two-dimensional and / or three-dimensional data, linked to the object, the data set being associated with a first so-called acquisition space. For the sake of clarity of the illustration, the data set here is of the two-dimensional type, coming from photos of the real object to be modeled.

[0067] This data set is then processed during a second step 120 in order to generate a three-dimensional representation of the object in a second so-called modeling space.

[0068] The three-dimensional representation advantageously comprises a plurality of parts whose surfaces are generated by a mathematical function configured by control points. Such a mathematical function is for example a Bézier or NURBS curve. The notable advantage linked to the generation of surfaces by a mathematical function is to obtain a high-quality rendering regardless of the viewing angle or the zoom level used, which is not possible on a classic model obtained during a scan of an object because the classic modeling being discrete brings calculation errors at the rendering level.

[0069] The curves of the three-dimensional representation are adjusted during a sub-step 121 of step 120. This adjustment is advantageously carried out by taking into account the segmentation of the three-dimensional representation which has been previously chosen so that the representation is as faithful as possible. The generation of the three-dimensional representation is thus carried out by adjusting a part of the three-dimensional representation to an element of the object to be modeled. It should be emphasized that the adjustment of each part is constrained by the superposition of at least one surface curve of said part with a contour of a distinct element of the object, which has been previously deduced from the data set.

[0070] The structural constraint can be a predetermined geometric constraint on a part or between two parts – for example a radius of curvature, a concavity, an alignment relationship between two points, etc. – or a constraint on the continuity of a curve, or even its normal.

[0071] More precisely, the adjustment is carried out here by means of projection onto a plane, then by a superposition of contours, one from the data obtained initially and the other from the three-dimensional representation being calculated. An optimization calculation is thus implemented and makes it possible to define the parameters of the curves of the three-dimensional representation so that they fit as well as possible to the contours detected in the initial data set.

[0072] The modeling method 100 may also comprise a step 130 of automatically determining a geometric transformation between the object of the acquisition space and the three-dimensional representation in the second space so that the representation is of the canonical type, that is to say not deformed, or even perfectly symmetrical in the case of a pair of glasses. Step 130 may be carried out before, concomitantly with step 120, or even after to make a correction to the geometric representation generated in step 120.

[0073] The geometric transformation used during step 130 generally comprises at least one of the following elements: Straightening of a part of the three-dimensional representation according to a pre-established constraint on said part (different methods can be used for these transformations such as the skeletonization (also known by the English term "rigging") of an animation skeleton or the use of a deformation field for example); Adjustment of an angle between two parts of the three-dimensional representation according to a pre-established constraint between said two parts; Symmetrization of the three-dimensional representation.

[0074] The pre-established constraints can be, for example, angular constraints – a pair of glasses standing perpendicular to the face, a pantoscopic angle (orientation of the face of a pair of glasses relative to the vertical) which is generally defined between 5 and 10 degrees, etc. –, structural constraints – for example, straightening the outer lining of a bag which would have been deformed during the scan, placing a handle or a chain of a bag in the direct extension of the bag, i.e. upwards when the bag is placed vertically – or even a symmetrization constraint in the case of an object whose architecture is symmetrical, as is the case with a pair of glasses.

[0075] For this purpose, the three-dimensional representation can advantageously include an articulation between two parts.

[0076] In order to allow the adjustment of the curves of the three-dimensional representation, before or concomitantly with step 120, the modeling method 100 may also comprise a step 140 of identifying a contour of the object 200 in the data set (by conventional filtering or machine learning methods for example) and a step 150 of determining a surface curve of the three-dimensional representation.

[0077] It should be emphasized that the modeling method 100 may also comprise a step 155 of calculating a geometric deformation factor linked to an optical parameter of a system for acquiring an image of the real object. This deformation may for example be linked to the focal length of a lens of the acquisition system. The induced deformation may be calculated upstream by usual techniques, in order to be able to calibrate the camera and obtain optically non-distorted images.

[0078] Advantageously, the modeling method 100 may also comprise a step 160 of recognizing at least one type of shape of the object 200 or at least one type of shape of a part of the object 200 in the data set and a step 170 of automatically selecting the parametric three-dimensional model based on this recognition. The recognition of a type of shape may in particular be carried out by a conventional technique, based for example on a detection of landmarks or on deep learning techniques.

[0079] Steps 160 and 170 automate the modeling process from a data set.

[0080] The recognition step 160 may in particular, in the case of a pair of glasses, comprise recognition of a shape of a frame, that of a circle, that of a bridge, a type of hinge, a shape of branch, or even a shape of nose pad.

[0081] It should be noted that the parts of the three-dimensional representation can be derived from a generic three-dimensional model of the object, for example a model of a pair of glasses, shoes or any other type of object, which makes it possible to maintain an intrinsic consistency of the object to be modeled whose type has been previously recognized. The relationships, also called constraints, between the parts can then be advantageously inherited from the generic three-dimensional model. The parametric three-dimensional model is a specification of the generic three-dimensional model by adapting the shape of each part to the object 200 to be modeled. A hierarchy of models can thus be set up by gradually refining the shape of each part to arrive at a faithful three-dimensional representation of the real object 200. Statistical models such as a 3DMM (acronym for the English term "3DMorphableModel") can also be used for these purposes.

[0082] This is an image of the three-dimensional representation 300 of the object 200 obtained by the modeling method 100.

[0083] Compared with which is an image of a three-dimensional representation 400 of the same object 200 but obtained by a conventional technique of the prior art, namely a three-dimensional scan of the object 200, the three-dimensional representation 300 appears significantly more precise than the three-dimensional representation 400. In particular, unlike the three-dimensional representation 400, the three-dimensional representation 300 has a very smooth surface and very precise details. The part 310 representing the metal core 220 is thus well positioned in the three-dimensional representation 300, inside the envelope 320 of the three-dimensional representation 300, while the metal core 220 is formed in the three-dimensional representation 400 by a discontinuous texture 420 applied to the outer surface.

[0084] Furthermore, protrusions 410 linked to the support or bumps 430 are visible on the three-dimensional representation 400 obtained by the technique of the prior art, which is not the case of the three-dimensional representation 300 which is advantageously smoothed by the structural constraints implemented during the modeling method 100.

[0085] Once the modeling is done, the three-dimensional representation can advantageously be implemented as part of an augmented reality process, for example by embedding it in a real image of a video stream. A virtual fitting of an object can thus be implemented in a more realistic manner because the resulting rendering has fewer negative visual effects. In addition, the modeling process 100 makes it possible to provide segmentation and semantics to each part of the modeled object necessary for the correct positioning of these objects in an augmented reality context. For example, the pads will be easily identified to be positioned on the nose. Other benefits and optional features

[0086] An object to be modeled is included among the following non-exhaustive list: a pair of glasses; a pair of rimmed glasses; a pair of half-rimmed glasses; a pair of rimless glasses; a pair of multi-rimmed glasses; a pair of one-piece glasses; a pair of single-bridge rimmed glasses; a pair of double-bridge rimmed glasses; a pair of single-bridge half-rimmed glasses; a pair of double-bridge half-rimmed glasses; a pair of mask-type glasses; a bag; a shoe, or even a pair of shoes; a piece of jewelry; a watch; a helmet; an item of clothing; a piece of furniture.

[0087] Generally speaking, an object to be modeled is a manufactured object. However, the modeling process can also be applied to a part of a human or animal body, such as a head.

[0088] This is a photo of a 450 shoe with 455 laces to model.

[0089] This is an illustration of a portion of a three-dimensional representation 460 of the shoe 450 according to a prior art method, namely a scan. The laces 455 of the model 460 are merged with each other. It is furthermore difficult to delimit the laces 455, the tongue 465 and the two quarters 466 of the shoe 450 and therefore to easily change them.

[0090] By using the illustrated modeling method 100 on a similar shoe, the illustrated three-dimensional representation 470 is much more consistent, modifiable and usable by clearly separating each lace 475, the tongue 476 and the quarters 477 of the shoe 478. Another detailed example of implementation

[0091] Illustrates an example of implementation of a modeling method 500 according to the invention, in the form of a block diagram.

[0092] The modeling method 500 firstly comprises a step 510 of obtaining 2D or 3D input data, from photos, plans, 3D scans, etc. of the real object 200.

[0093] A general qualification of the real object 200 to be modeled is carried out initially during a step 515. This qualification notably includes recognition of the type of the object 200 by classification tools well known to those skilled in the art.

[0094] A geometric transformation is then evaluated during a step 520 to correct the deformations undergone by the real object 200 in order to straighten it in a perfect canonical space. It should be emphasized that the geometric transformation can be reversible.

[0095] The data are then positioned in whole or in part in the second three-dimensional space, called canonical, during a step 530, taking into account any known structural constraints of the object, for example positioning of a branch at 90° relative to a face of a pair of glasses.

[0096] A parametric model, forming all or part of the three-dimensional representation of the object 200, is then selected from a database during a step 540, to model at least part of the object 200. A particular order of the parts to be modeled can be followed depending on the architecture of the previously qualified object 200.

[0097] This selection is carried out by identifying characteristic points or characteristic landmarks in different views and then by matching them with a parametric model.

[0098] The parametric model is positioned and deformed according to the transformation determined in step 520 in the second space such that it is superimposed on the initial data during a step 550, in order to maintain the consistency of the object 200.

[0099] Advantageously, a hinge of the parametric model can be advantageously positioned relative to a hinge recognized in the initial data. Similarly, a bridge or plates can also be selected and positioned.

[0100] In the case where the initial data are images of the object 200, the image of the object may have undergone deformations due to the optics used. An optical deformation can also be calculated and taken into account in order to correct the images of the object 200 and limit errors in interpretation of the initial data.

[0101] It should be emphasized that the selection and default positioning of the parametric model are here performed automatically by the use of a general parametric representative model which is adjusted on different views of the object 200. For example, for a pair of glasses, the general parametric model is a statistical model which is on the one hand deformable and contains a hierarchy and a set of relationships – positioning constraints, orientation constraints, animation constraints – between the constituent elements each represented by its own parametric model.There may be several standard models corresponding to a subclass of objects: for example, the general model of a boot is different from that of a pair of sneakers while they are part of the same general class of shoes, the general class defining a more generic model consisting of elements having semantics relating to the construction of the manufactured object, namely a sole, a seam, a tongue, a closure, heels, etc.

[0102] An adjustment of the shape of the parametric model is then carried out during a step 560 by optimizing a superposition of curves projected onto a plane, curves resulting from the parametric model and the initial data.

[0103] The adjustment of the parametric model is carried out via intrinsic parameters (control points, size, etc.) and / or extrinsic parameters (position, orientation, twist directions, etc.). Positioning constraints can be advantageously taken into account.

[0104] A parametric model is a collection of curves and surfaces that have editing or adjustment constraints according to business parameters with a mathematical parameterization correspondence. This limitation of the space of possible deformations of an element of the parametric model ensures that this element is always represented with possible constraints with regard to the architecture of the object. The business a priori is thus directly integrated into the modeling. A parametric model also contains elements of positioning and interaction with the representation of the standard parametric model and the other constituent elements, which ensures its placement and orientation in the construction of the manufactured object. Thus, a hinge of glasses arms is always in the right place and always connects the arm with the frame of the pair of glasses.

[0105] Steps 540 to 560 may be repeated to model each part, element or piece of the object 200.

[0106] It should be noted that the adjustment performed during step 560 may be repeated for a plurality of images taken from potentially different viewing angles, in order to finalize the shape of the parametric model. Depending on the image (or more generally depending on the type of data), a geometric transformation may be applied to straighten the image of the object 200 so that it can be used as a model for the three-dimensional representation in the second canonical space.

[0107] Finally, an operator can manually adjust the three-dimensional representation of the object 200, formed by each parametric model, in order to perfect the three-dimensional representation during a step 570. It should be emphasized that this step 570 is optional.

Claims

Method for modeling a real object comprising a plurality of elements, said method comprising steps of:Obtaining a set of two-dimensional and / or three-dimensional data, linked to the object, the data set being associated with a first space;Generating a three-dimensional representation of the object in a second space from the data set;characterized in that the three-dimensional representation comprises a plurality of parts whose surfaces are generated by a mathematical function describing a parametric surface, the parts of the three-dimensional representation being derived from a generic parametric three-dimensional model of the object, the constraints between said parts being inherited from said generic parametric three-dimensional model, a constraint being included in the list {positioning constraint, orientation constraint, animation constraint}, the step of generating the three-dimensional representation of the object in the second space, called canonical, comprising a sub-step of adjustment in shape and dimension of each of the parts to at least one element of the real object, the adjustment being carried out taking into account one of said constraints between parts;the modeling method also comprising a step of automatically determining a geometric transformation between the object of the acquisition space and the three-dimensional representation of said object in the canonical space, the transformation comprising at least one of the following elements: Straightening of a part of the three-dimensional representation according to a pre-established constraint on said part; Adjustment of an angle between two parts of the three-dimensional representation according to a pre-established constraint between said two parts; Symmetrization of the three-dimensional representation.; Modeling method according to claim 1, wherein the adjustment of each part is constrained by the superposition of at least one surface curve of said part with an outline of a distinct element of the object, said outline being previously deduced from the data set. A modeling method according to any one of claims 1 to 2, wherein a structural constraint is a predetermined geometric constraint on a part or between two parts. Modeling method according to any one of claims 1 to 3, in which a structural constraint is a continuity constraint of a curve of a part or a continuity constraint of the tangent at at least one point of the curve of said part. Modeling method according to any one of claims 1 to 4, also comprising a step of identifying an outline of the object in the object data set and a step of determining a surface curve of the three-dimensional representation, corresponding to said outline of the object. The modeling method of claim 5, wherein the surface curve is an edge of the three-dimensional representation of the object. Modeling method according to any one of claims 1 to 6, wherein the data set comprises data from an acquired image of the object, the modeling method also comprising a step of calculating a geometric deformation factor of at least one part of the object. A modeling method according to any one of claims 1 to 7, wherein the data set comprises two views of the object from distinct angles. A modeling method according to any one of claims 1 to 8, wherein the data set comprises a view of the object and a depth map obtained concomitantly with the view of the object. A modeling method according to any one of claims 1 to 9, wherein the data set comprises at least two depth maps obtained from distinct angles or a scan of the object. Modeling method according to any one of claims 1 to 10, comprising steps of:Recognition of at least one type of shape of the object and / or of a part of the object in the data set;Automatic selection of the generic parametric three-dimensional model according to the type(s) of shape previously recognized. Modeling method according to claim 11, wherein when the object is a pair of glasses, the recognition of at least one type of shape comprises all or part of the following sub-steps: Recognition of a type of shape of a frame included in the pair of glasses; Recognition of a type of shape of a circle included in the pair of glasses; Recognition of a type of shape of a tenon included in the pair of glasses; Recognition of a type of shape of a bridge included in the pair of glasses; Recognition of a type of shape of a hinge between the frame and a temple included in the pair of glasses; Recognition of a type of shape of temple included in the pair of glasses; Recognition of a type of shape of a sleeve included on the temple; Recognition of a type of shape of nose pad included in the pair of glasses. Recognition of a type of shape of logo or inserts included in the pair of glasses Modeling method according to any one of claims 1 to 12, in which the three-dimensional representation comprises at least one articulation between two parts. Electronic storage device for instructions of a modeling method according to any one of claims 1 to 13. Augmented reality method implementing the method of modeling a real object according to any one of claims 1 to 14, and the three-dimensional representation of said object obtained.

Images