Method for monitoring the displacement of prefabricated blocks in a dike's outer layer

Abstract

Method for tracking the displacement of prefabricated blocks of a dike armor. Method for generating a 3D model of an armor (5) of prefabricated blocks (15) made of cementitious material, the method being implemented by computer and comprising: a) providing a point cloud (25) partially representing the armor (5), b) segmenting the point cloud into homogeneous regions (80) representing at least a portion of one of the blocks (15) discretized in the point cloud, c) aggregating at least two of the adjacent homogeneous regions (80) to form a candidate region (100), d) matching said candidate region (100) with a 3D model object (105) representing at least a portion of a block (15), and e) recording, in the 3D model of the armor, the position and orientation of the 3D model object. matched in step d). Figure for the abbreviation: Fig. 9

Description

Description of the invention: Method for tracking the movement of pre-blocks made from the shell of a dike technical field The invention relates to a method for modeling a shell of prefabricated blocks by processing data acquired on the shell. It further relates to a method for determining the displacement, that is to say the variation in position and orientation, of the blocks of the shell over time. Previous technique Sea-level embankments protect against cyclones, floods, and even tsunamis. They are also resistant to erosion caused by swells. To create such embankment breakwaters, it is common practice to revet the slope with a single-layer or multi-layered armor layer made of prefabricated concrete blocks. This armor layer mitigates the effect of swell by breaking the waves into a multitude of seawater streams that diffuse within the layer.Two families of blocks are known for forming shells. The first family consists of blocks with a simple geometric shape, either hollow or solid, such as cubic or prismatic. Blocks in this first family are generally intended to be stacked regularly. The second family consists of blocks with more complex shapes, such as branches extending in different directions. Examples of blocks in this second family include Accropodes®, Accropodes II®, Cubipod®, Core-loc II®, Ecopode®, Tetrapode™, Xbloc®, and Crablock®. Blocks in this second family are generally arranged in a staggered, row-by-row pattern according to a pre-established plan. However, they can be irregularly interlocked within the shell due to inaccuracies in block placement.The mechanical resistance of the shell to the stresses exerted by the swell comes in particular from the forces resulting from the contact between the branches of the blocks. The blocks can move relative to each other over time, particularly during storms, cyclones, or earthquakes, during which the stresses exerted by the swell on the blocks are amplified. Excessive movement of the blocks relative to each other can result, in extreme cases, in the breach of the dike and / or in damage to a structure supported by the dike, for example a road, a building or an engineering structure. However, the movement of the blocks is generally small compared to the size of the carapace, which can extend in bands several kilometers long and over a width that can reach several tens of meters. For example, following a cyclone, a block may have been displaced by a few centimeters, which is small considering its size, which is generally several meters. Furthermore, a carapace can consist of several hundred, or even several thousand, identically shaped blocks, arranged randomly, making it difficult to measure the slight displacement of each individual block. Identifying each block and equipping it with instruments to measure its displacement is also particularly expensive. There is therefore a need to facilitate the identification of prefabricated blocks within a shell and to easily track the evolution of the shell over time, particularly with a view to preventing damage described above. The invention proposes a method for generating a 3D model of a shell of prefabricated blocks made of cementitious material, the method being implemented by computer and comprising: a) the provision of a point cloud partially representing the carapace, b) the segmentation of the point cloud into homogeneous regions representing at least a part of one of the discretized blocks in the point cloud, c) the aggregation of at least two of the homogeneous regions that are adjacent to form a candidate region, d) the matching of said candidate region with a 3D model object representing at least part of a block, and e) the recording, in the 3D model of the carapace, of the position and orientation of the 3D model object matched in step d). The method according to the invention allows for the simple modeling of a block of the carapace by identifying the points corresponding to a part of the block in the point cloud and by assigning a 3D model object that represents the block. Once the block is identified, it is thus possible to track its evolution between different times within the carapace, independently of the other blocks of the carapace. Preferably, the process involves, after step e), repeating steps c) to c) with another candidate region in order to match a model object representing another of the carapace blocks. It is thus possible to model several blocks of the carapace, or even all the blocks represented in the point cloud of the carapace. Preferably, steps c) to e) are repeated until more than 30%, preferably more than 50%, preferably more than 70%, preferably more than 80%, preferably more than 90%, or even 100% of the blocks of the carapace are modeled, each using a corresponding model object. Preferably, in order to accelerate digital processing, the method should also include, after matching step e), the removal of the region candidate, possibly enlarged as will be described later, of the point cloud. Advantageously, the computer implementation of the process automates the generation of candidate regions, the matching process, and the registration of the model object, without requiring any user intervention to perform these steps. Specifically, the process only requires the 3D model object of the block and a point cloud of the shell. It does not require prior knowledge of the block placement plan for forming the shell. Shell and blocks The shell can represent a submerged part and an emerged part. The point cloud can contain coordinates of points from submerged blocks and coordinates of emerged blocks. The carapace may have an arrangement of blocks that is ordered or irregular. The shell can be made up of identical blocks or different blocks. The carapace can extend over a length exceeding 50 m, or even exceeding 100 m, or even exceeding 500 m, or even exceeding 10000 m The prefabricated blocks are made of cementitious material. Cementitious material is defined as a material containing a hydrated binder. Examples of hydrated binders include Portland cement or hydrated binders made from blast furnace slag. Prefabricated blocks can be made of unreinforced or reinforced concrete, and can be solid or hollow. The length of a block can range from 1 m to 5 m. The volume of a block, for example, ranges from 1 m³ to 30 m³. Prefabricated blocks can have a compact or slender shape. Prefabricated blocks can have a branching general shape. For example, they may have a core from which protrusions, such as branches, protrude and extend outwards. Alternatively, the prefabricated blocks can have a solid polyhedral shape, for example chosen from; - a prismatic shape, particularly cubic, parallelepiped or hexagonal, such as the prefabricated block known as "Seabee", - a truncated cone shape with a polygonal base, especially a square one, and whose lateral faces may be grooved. Alternatively, prefabricated blocks can take the form of a lattice element. For example, they can be formed of concrete bars connected to each other at their vertices, such as the "Cob" block, the "Shed" block or the "Diahitis" block. In particular, the prefabricated blocks can each be chosen from: - a block known as "tetrapod®", - a block as described in FR 2 449 164 Al or in FR 2 647 135 Al, marketed under the reference "Accropode®" by the company Artelia, - a block as described in FR 2 791 370 A1, marketed under the reference "Accropode II®" by the company Artelia, - a block as described in FR 2 748 044 Al, marketed under the reference "Ecopode®" by the company Artelia, - a block as described in EP 0 583 500 A1, known as "A-Jack®", - a block as described in WO 95 / 09279 A1, known as "Core loc®", - a block as described in WO 2004 / 009910 A2, known as "Xbloc®", - a block as described in US 2010 / 258704 A1, known as "Cubipod®", - a block as described in the article "First Tests on the Symmetrical Breakwater Armor Unit Crablock", M. Salaudin et al, Coastal Engineering Journal 59(04), doi: 10.1142 / S0578563417500206, known as "Crablock", and - a block known as "Core-loc 2". Scatter plot The point cloud partially represents the carapace. In particular, as will become clear later, only the part of the carapace that is visible to an acquisition device can be acquired. The point cloud can represent the submerged part of the shell and / or the emerged part of the shell. In one implementation variant, the process may include a preliminary step of acquiring the point cloud. The acquisition of the point cloud of the shell may involve: - an acquisition of at least one point cloud of the emerged part of the carapace and / or an acquisition of at least one point cloud of the submerged part of the carapace, - optionally, the merging of the point cloud of the emerged part and the point cloud of the submerged part. Such a point cloud may exhibit a heterogeneous spatial distribution of points, resulting from the implementation of different acquisition techniques. Furthermore, the measurement inaccuracy attributed to each point may also be spatially distributed heterogeneously. It varies, for example, from 1 cm to 10 cm. Moreover, the point cloud may contain measurement noise, meaning it may contain points that do not correspond to a material element of the shell. Measurement noise may be particularly high in the part of the point cloud that... This presents the waterline portion of the carapace. The "waterline of the carapace" corresponds to the portion of the carapace formed by partially submerged blocks, that is, blocks with a portion submerged in water and a portion exposed to the air. The method according to the invention has the advantage of eliminating these heterogeneities and measurement noise, which do not significantly impair its effectiveness. The point cloud of the submerged part of the carapace and the point cloud of the emerged part of the carapace may exhibit different point densities and / or different measurement accuracies of point positions. The point cloud can be acquired using at least one acquisition device comprising an optical remote sensing device, for example a lidar, and / or an acoustic remote sensing device and / or a photogrammetric acquisition device. The acquisition device can be a flying object, for example a satellite, an aircraft, or a drone, equipped with an optical remote sensing instrument and / or a photogrammetric acquisition instrument. The flying object can fly over the carapace, and the instrument can acquire a point cloud of the emerged portion of the carapace. Preferably, the optical remote sensing instrument is oriented vertically downwards to acquire the emerged portion of the carapace in a nadir direction. Where possible, the optical remote sensing instrument can be oriented obliquely to acquire a greater number of points in order to form a point cloud representing the carapace more accurately.The acquisition device can be a ground-based vehicle, for example a construction crane, and the remote sensing and / or photogrammetric acquisition device is mounted on the crane, for example on the crane's boom, so as to acquire the point cloud of the emerged part of the crust it overlooks. The ground-based vehicle can be static or mobile. The acquisition device may be a vessel comprising: - an optical remote sensing device, preferably a lidar, oriented to acquire a point cloud of the waterline portion of the carapace, and / or - an acoustic remote sensing device, for example a bathymetric sounder to acquire a point cloud of the submerged part of the carapace. More than five, or even more than ten, or more than fifty, or more than a thousand blocks, or even more than five thousand blocks can be represented in the point cloud. The point cloud can contain between one hundred thousand and ten million points to facilitate processing. Each point is assigned three-dimensional coordinates. The coordinates of the points are referenced in a local coordinate system or, preferably, in an absolute coordinate system. By "absolute coordinate system," we mean a geodetic reference system in which the location of an object on Earth can be defined. unambiguous manner. Its center is, for example, close to the Earth's center of gravity, its first two axes are in the plane of the equator, and its third axis is close to the Earth's axis of rotation. The absolute reference frame that can be used within the framework of the present invention is preferably chosen from the following: French Geodetic Network 1993 (RGF93), World Geodetic System (WGS84), International Terrestrial Rotational Service (ITRS), or European Terrestrial Reference System (ETRS). The point cloud can be saved on a computer-readable storage medium. A computer storage medium is a digital recording medium such as magnetic tape, optical disc, hard drive, SSD, SD card, or USB flash drive. Candidate regions Step b) involves segmenting the point cloud to form homogeneous regions. The homogeneous regions are formed from points in the cloud. "Segmentation" is a well-known point cloud processing technique. It consists of partitioning the cloud by grouping points into subsets, called homogeneous regions, which are characterized by spatial and / or geometric consistency. Such subsets may be referred to as "segments" in the literature. "Spatial coherence" means, for example, that the points of the homogeneous region are contained within a single, non-disjoint set. "Geometric coherence" means, for example, that the points of the region are arranged in a similar manner on the same plane or the same parameterized curved surface. Preferably, step b) includes: - voxelization of the point cloud, - the formation of non-disjoint and spatially coherent groups of voxels, for example by means of a "k-means" partitioning method, and - the generation of homogeneous regions from groups of voxels. "Non-disjoint" voxel groups are such that all the voxels they contain are related to each other, either directly by being adjacent to each other by at least one face or by at least one vertex, or indirectly through a sequence of voxels in the group that are adjacent to each other. Such a point cloud segmentation technique, including a voxelization step, is described, for example, in the article "Segmentation of 3D Point Cloud Data Based on Supervoxel Technique", RS Rampriya and R. Suganya, Procedia Computer Science, Volume 171, 2020, Pages 427-435, doi: 10.1016 / j.procs.2020.04.045, Preferably, voxelization includes: - the application of a three-dimensional voxel grid, that is, cubic cells of the same dimensions, onto the point cloud, and - assigning a characteristic to a voxel when the coordinates of at least one point in the cloud are included in the voxel. The voxel's characteristic is, for example, a binary label that is 0 when no point in the cloud is included in the voxel and 1 otherwise. Advantageously, only the coordinates of the points in the cloud are required to generate the 3D model at the end of step e). The process can thus do without other potential data from the point cloud, for example the intensity or color of the point. Furthermore, the voxel size can be adjusted automatically, or by the operator implementing the voxelization, depending on the point cloud density. For example, the side length of a voxel can be between 1 cm and 20 cm. The voxel size can be between 1 and 2 times the average distance between two points in the point cloud, for example, approximately 1.5 times. Preferably, point cloud segmentation involves generating homogeneous regions using the VCCS segmentation technique. According to the VCCS segmentation technique, homogeneous regions are called "super-voxels". The VCCS segmentation technique, short for "Voxel Cloud Connectivity Segmentation," is described in detail in the article "Voxel Cloud Connectivity Segmentation — Supervoxels for Point Clouds," by J. Papon, A. Abramov, M. Schoeler, and F. Wörgötter, published in the Proceedings of the 2013 IEEE Conference on Computer Vision and Pattern Recognition, June 23-28, 2013, doi: 10.1109 / CVPR.2013.264. VCCS segmentation has the advantage of being automatically adapted to the point density of the cloud. It is notably available in the PCL (Point Cloud Library), a C++ library dedicated to processing 3D point clouds. In step d), at least two of the homogeneous regions that are adjacent are aggregated to form a candidate region. Preferably, each candidate region includes a homogeneous region and preferably all the homogeneous regions adjacent to it. Two "adjacent" homogeneous regions share at least one common border. The formation of the candidate region is preferably carried out using an adjacency graph resulting from the implementation of the VCCS segmentation technique. The adjacency graph can include, for each homogeneous region, the center of the homogeneous region, the list of vertices of the homogeneous region that are shared with adjacent homogeneous regions, and the list of centers of the adjacent homogeneous regions. Matching In step d), the candidate region matching aims to determine the position and orientation, within the point cloud's coordinate system, of a model object representing at least part of a block. The model object can represent an entire block. Preferably, it represents a portion of the block, which improves the matching process. The matching process involves aligning the shape of the candidate region with the shape of the model object used as a reference. The 3D model object can be a point cloud or a mesh, or preferably, defined using vector geometric data. Preferably, the model object is created using computer-aided design software, for example, software such as AutoCAD®, Revit® (marketed by Autodesk®), SolidWorks®, and Catia® (marketed by Dassault Systèmes). The model object can thus represent all or part of a block of the shell simply and accurately. The model object can notably be located, in the point cloud reference frame, by means of the position of its barycenter and the orientation of its principal axes. Preferably, in step d), the candidate region and the model object are matched by means of a shape descriptor of the candidate region and a shape descriptor of the model object. Preferably, the matching of the candidate region with the model object is carried out by implementing a matching technique called "Fast Global Registration", described in Zhou, Q.-Ÿ et al, Computer Vision — ECCV 2016, Lecture Notes in Computer Science, Vol. 9906, 766-782 - September 2016, doi: 10.1007 / 978-3-319-46475-6_47. The local descriptor of the candidate region is a set of features that describe the model properties of the shape represented by the candidate region. It can be defined from points of interest in the candidate region, to which are associated attributes that describe the environment of the corresponding points of interest. Such points of interest are also known as "keypoints" in English terminology. Many shape descriptors of a set of points are known, for example from the article "Evaluation of Local 3-D Point Cloud Descriptors in Terms of Suitability for Object Classification", J. Garstka, G. Peters, ICINCO 2016, doi: 10.5220 / 0006011505400547, and methods for detecting points of interest are also known, for example described in the thesis of A. Shaiek, "Reconnaissance d'objets 3D par points d'intérêt", Ecole Nationale Supérieure des Mines de Paris, 2013. Points of interest can be detected using a point of interest detector chosen from an ISS detector, a SIFT detector, a Harris detector, a SUSAN detector and an AGAST detector. In particular, the point of interest detector can be an ISS detector, an acronym for "Intrinsic Shape Signature", as described in the article "Intrinsic Shape Signatures: A shape descriptor for 3D Object Recognition", Yu Zhong, International Conference on Computer Vision Workshops (ICCV Workshops), 2009, doi: 10.1109 / ICCVW.2009.5457637. A shape descriptor for the candidate region can be chosen from an FPFH descriptor, an English acronym for "Fast Point Feature Histograms", a PFH descriptor, an English acronym for "Point Feature Histogram", an LFSH descriptor, an English acronym for "Local Feature Statistic Histograms" and a VFP descriptor, an English acronym for "Viewpoint Feature Histogram". In particular, the shape descriptor of the candidate region can be an FPFH descriptor. The FPFH descriptor is described in the article "Fast Point Feature Histogram (FPFH) for 3D registration", RB Rusu et al., Proceedings of the IEEE International Conference on Robotics and Automation, 2009, pages 3212-3217, doi: 10.1109 / ROBOT.2009.5152473. Furthermore, the shape descriptor of the model object includes a set of parameters that locally discretize the shape of the object. In particular, the shape descriptor of the model object can be determined by generating a point cloud representing at least a part of the model object, called the discretized model object, by defining points of interest in the discretized model object, and then calculating point of interest descriptors for each point of interest, the shape descriptor of the model object comprising the point of interest descriptors, Furthermore, in step d) of the matching process, the mapping of the candidate region with the model object may involve minimizing the gap between the candidate region and the model object. Preferably, this minimization is performed by implementing an iterative closest point algorithm, also known as an ICP algorithm, an acronym for "Iterative Closest Point". The matching step may also include validation of the correspondence, notably by counting the number of points in the candidate region included in the volume of the model object and comparing this number of points to a predetermined threshold. If the number of points is less than this threshold, then the matching is considered successful. Otherwise, the matching is considered inconclusive. The candidate region is considered unidentifiable and no record is made in step d). Preferably, in order to improve the accuracy of the position and orientation of the matched block, the matching step (d) includes, preferably after validation of the matching, an optimization of the matching. of the 3D model object with the candidate region. Preferably, said optimization involves enlarging the area of ​​the candidate region by implementing at least one optimization cycle comprising: 1) the aggregation into the candidate region of at least one, or even several, homogeneous regions adjacent to it, ii) minimizing the gap between the enlarged candidate region and the model object, preferably using an ICP algorithm as described above, and iii) the calculation of a minimization residual of said difference, the optimization cycle being repeated with one or more other homogeneous regions and with the candidate region enlarged in step ii) as long as said residual calculated in step iii) is less than a predetermined threshold, for example between 1 mm and 15 mm. When the minimization residual is greater than the predetermined threshold, the homogeneous region aggregated in step i) is considered not to represent the same prefabricated block in the point cloud as the candidate region prior to step i). The matching optimization is then stopped. Conversely, when the minimization residual is less than the predetermined threshold, it is considered that the points of the adjacent region(s) aggregated in step i) represent, in the point cloud, the same prefabricated block as the candidate region prior to step i). In addition to identifying other homogeneous regions representing the same prefabricated block, minimizing the gap improves the accuracy of the position and orientation of the 3D model object. Furthermore, once the residual is greater than the predetermined threshold, the optimization may include the calculation of the block's visibility, the optimization being considered optimal when the visibility is greater than a predetermined visibility threshold, preferably between 10% and 25%. Recording in the 3D model In step e), the coordinates of at least one point and at least one orientation parameter of the model object in the point cloud reference frame are recorded. The 3D model can be saved on a computer-readable storage medium. For example, the coordinates of the center of gravity of the model object are recorded in said reference frame. The orientation parameter can be an axis-angle pair, that is, comprising an angle and the coordinates of a normalized vector around which the rotation by angle is performed to orient a reference frame attached to the model object in the point cloud's reference frame. It can be a set of angles, for example roll, pitch, and yaw, which can define the orientation of the reference frame attached to the model object relative to the axes of the point cloud reference frame. The orientation parameter can be an Euler angle triplet defining the orientation of a reference frame linked to the model object, for example formed by the principal axes of the model object, with respect to the reference frame of the point cloud. Furthermore, the process may also involve assigning an identifier, for example alphanumeric, such as a name, to the model object. This identifier facilitates the subsequent identification of a block of the shell. Preferably, steps c) to e) are implemented for several candidate regions, preferably for all candidate regions. The method according to the invention thus makes it possible to automatically match numerous candidate regions and to save the corresponding model objects when the corresponding matching tests are successful. The method is capable of identifying and modeling most, if not all, of the blocks acquired in the point cloud. The method can be such that the matching test (c) is successful for more than 80%, more than 90%, or even more than 95% of the candidate regions tested. In particular, the 3D model can include model objects representing more than 80%, more than 90%, or even more than 95% of the blocks represented in the point cloud. Furthermore, the method according to the invention is particularly robust, in that blocks partially represented by a small number of points in the point cloud can be fully represented by model objects in the 3D model of the carapace. In particular, blocks can be partially obscured by neighboring blocks from the view of an observer looking at the carapace, for example, nadirically from an observation point elevated relative to the carapace. At least a portion of the population of blocks represented in the point cloud can be fully represented by model objects in the 3D model of the carapace. In particular, the average visibility of the blocks represented in the point cloud can be between 25% and 35%. Preferably, blocks represented in the point cloud with a visibility of less than 20%, or even less than 15%, and ideally less than 10%, can be fully represented by model objects in the 3D model of the shell. The "visibility" of a block corresponds to the surface area of ​​the portion of the block represented in the point cloud, divided by the total surface area of ​​the block. The average visibility of the blocks is the arithmetic mean of the individual block visibilities. Preferably, at least 90%, or even at least 95%, better 100% of the number of blocks represented in the point cloud that have a visibility greater than or equal to 25%, or even greater than or equal to 20%, are represented in the 3D model of the carapace. Furthermore, the invention relates to a method for tracking the movement of blocks in a shell, comprising: - the implementation of the process according to the invention by providing in step a) a first point cloud representing the carapace at a first instant, in order to generate a first 3D model of the carapace, - the implementation of the method according to the invention by providing in step a) a second point cloud representing the carapace at a second time, in order to generate a second 3D model of the carapace, and -the calculation of the displacement of at least one, preferably of all the paired model objects common to the first and second 3D models between the first and second instants. The tracking method allows us to study the evolution of the movement of all the blocks of a carapace over time. For example, the tracking method allows us to observe the effects of a storm or cyclone on the carapace. Such observation is particularly useful for: - block designers, since it allows them to evaluate in situ the behavior of a set of blocks and to consider modifications to the shape of blocks where necessary to form future shells, - companies installing the blocks on site, as it allows them to assess the opportunity of modifying the way the blocks are arranged in relation to each other for future projects, - dike managers in order to assess the potential risks of damage to the dike. The "displacement" of a model object refers to the change in position of a characteristic point, for example, the centroid, and the change in orientation of the model object. In particular, in the case of "Accropodes II®" blocks, positional changes of less than 10 cm, or even less than 5 cm, and roll, pitch, and yaw changes of less than 5°, or even less than 2°, can be measured using the method according to the invention. The invention also relates to a method for generating a consolidated 3D model of a shell, comprising: - the implementation of the process according to the invention by providing in step a) a first point cloud representing a first portion of the carapace, in order to generate a first 3D model of the carapace, - the implementation of the method according to the invention by providing in step a) a second point cloud representing a second portion of the carapace connected to the first portion of the carapace, in order to generate a second 3D model of the carapace, and - the assembly of the first and second 3D models to form the consolidated 3D model. The first point cloud can represent a portion of the shell at an initial stage of its construction, and the second point cloud can represent at least another part of the shell at a second stage of construction subsequent to the first. The generation method thus allows the 3D model of the shell to be updated as the blocks are placed. The invention also relates to a computer for implementing the process according to the invention and / or the tracking method according to the invention and / or the generation method according to the invention, the computer preferably comprising: a. at least one computer storage unit in which the point cloud is recorded and a memory unit in which instructions for implementing the process are loaded, and b. a processor to execute the instructions. The invention also relates to a computer data storage module, for example a hard drive, in which the 3D model, obtained by the process according to the invention, is recorded. The invention further relates to a method of displaying a visual representation of the 3D model on a screen, for example, a computer screen. The method may include displaying a virtual shell formed by nesting paired model objects representing the corresponding blocks of the stack. It may also include displaying the movement of the blocks, for example, by means of indicators such as arrows whose length is proportional to the intensity of the movement and / or a color scale corresponding to the intensity of the movement. The invention will be better understood upon reading the detailed description that follows, the example that will be described, and the attached drawing in which: [Fig.1] is a photograph of part of a dike's armor layer formed from "Accropodes II®" blocks. [Fig.2] is a schematic vertical cross-sectional view of a point cloud representation illustrating the point cloud acquisition of the carapace, [Fig.3] represents the visualization of the point cloud of the submerged part of the carapace, [Fig.4] schematically illustrates the voxelization of a point cloud, [Fig. 5] schematically illustrates the segmentation of the point cloud and the formation of candidate regions, [Fig.6] is a view of a model object, [Fig.7] and [Fig.8] schematically illustrate the matching of a candidate region with a model object, [Fig. 9] illustrates the validation of the matching step, and [Fig.10] is a) a nadirically acquired photograph of part of the carapace of a dike and b) a visual representation of model objects representing the carapace blocks photographed in a). Figures 1 to 10 illustrate an example of implementation of the process according to the invention. Figure 1 is a photograph of the outer shell 5 of a sea wall 10, which is formed of prefabricated concrete blocks 15 known as "Accropodes II®". The blocks are interlocked, so that only parts of them are visible to an observer. Furthermore, some blocks, for example those referenced with the number 20, are partially obscured from view because they are partially covered by other blocks. In addition, some blocks are positioned above water while others are submerged in the ocean. The process requires as input a point cloud representing at least a portion of the carapace. As illustrated in [Fig. 2], various acquisition techniques can be used to acquire the point cloud 25 of the carapace. For example, a lidar can be mounted on a drone 30 to acquire a point cloud of the emerged portion 35 of the carapace along the acquisition cone 38. Any other vehicle that can provide an elevated view of the carapace can obviously be used, for example, a crane, a helicopter, or an airplane, and an optical remote sensing device other than a lidar can be used, for example, a photogrammetric acquisition device. Furthermore, a vessel 40 can carry a bathymetric sounder in order to acquire a point cloud of the submerged part 45 of the carapace, as indicated by the acquisition cone 48. Such a cloud is represented on [Fig.3], on which the corresponding prefabricated blocks are also shown for better visualization. Finally, a portion of the carapace blocks is partially submerged, subject to tidal variations and impacted by swell. This is referred to as the waterline portion 50 of the carapace. For example, a lidar can be mounted on the ship and directed towards the waterline portion of the carapace to acquire the corresponding point cloud, as indicated by cone 55. The point cloud 25, located in a local or absolute reference frame, can then be generated by aggregating the point cloud data from the emerged portion, the waterline portion, and the submerged portion of the carapace. As shown for example in [Fig. 3], blocks 2 of the carapace can be represented by a small number of points, for example because these blocks are Hidden from the bathymetric sonar by overlying blocks. Such blocks, which have a visibility of between 5% and 10% in the point cloud, can be modeled. Other blocks have high visibility, for example, around 30% for blocks 57. The point cloud can then be voxelized as schematically illustrated in [Fig. 4]. A 3D grid 60 composed of voxels 65 is generated in the same reference frame as the point cloud. It is checked whether one or more points in the cloud have coordinates included in the voxel. If so, the voxel 65 is considered filled; otherwise, it is considered empty. For example, a value of 1 is assigned to it if it is filled (filled voxel 70) and 0 otherwise (empty voxel 75). In the implementation example illustrated here, the side of the voxel chosen to discretize the point cloud is equal to 2 cm. The point cloud is then segmented using the generated voxels, as illustrated in [Fig. 5]. In the example shown, the VCCS segmentation technique is implemented. Using this technique, the point cloud is partitioned into homogeneous regions 80, based on VCCS-specific "super-voxels," each of which has a center 85 and vertices 90. The homogeneous regions represent, in particular, portions of the surface of a block. The homogeneous regions can be connected to adjacent homogeneous regions 95 by common vertices. The adjacent homogeneous regions represent other portions of the same block 15. A candidate region 100 can then be formed using an adjacency graph provided by the VCCS technique, which comprises a set of adjacent homogeneous regions. In addition, a shape descriptor of the candidate region, for example an FFPH descriptor, is calculated for the purpose of implementing the matching. The method according to the invention aims in particular to match, that is, to make the candidate region correspond with a model object that represents a prefabricated block. Figure 6 is a representation of a 3D model object representing an "Accropodes II®" block. The model object is, for example, created by computer-aided design. It is thus defined vectorially by means of vertices and segments joining the vertices. A shape descriptor of the model object is calculated, for example, a descriptor, for the purpose of the matching step. For this, the model object is discretized as a point cloud, and descriptors of points of interest are calculated beforehand in order to determine the descriptor of the discretized model object. Figure 7 represents, in the form of arrows, descriptors of points of interest of the model object. The matching step is then carried out by mapping at least a part of the model object 105 with the candidate region 100. In particular, the step The matching process involves minimizing the gap between the candidate region and the model object using an ICP algorithm. Figure 8 illustrates three matching results between three different candidate regions and a model object. For each result, the candidate regions 100, a portion of the model object 110 used for the matching, and the entire model object 105 are shown. A validation of the matching is then implemented and is illustrated in [Fig. 9]. The validation includes, in particular, counting the number of voxels from the candidate region that are included within the volume of the model object. The matching validation is considered successful, for example, when the ratio of the number of voxels contained within the model object to the total number of voxels in the candidate region is less than a predetermined threshold. The matching process is then optimized by aggregating other homogeneous regions with the candidate region, and then minimizing the gap between the enlarged candidate region and the model object. This optimization is called "Region Growing." It aims both to improve the determination of the orientation and position of the 3D model object and to identify all the homogeneous regions in the point cloud that represent the same prefabricated block. Once the optimization is performed, candidate region 100 can be extracted from point cloud 60. The process can be implemented again on another candidate region. The position and orientation of the model object are then recorded in the 3D model. Figure 10 illustrates a result of implementing the method. Figure 10a) is a photograph of the emerged portion of a carapace 5 from which a point cloud was acquired, and Figure 10b) represents the result of modeling this portion using the point cloud. In Figure 10b), each model object 105 of a block 15 is represented according to the corresponding position and orientation recorded in the 3D model. As is clear, the 3D model faithfully and precisely reproduces the interlocking of the prefabricated blocks of the carapace. Furthermore, the method is particularly robust and is suitable for modeling blocks with low visibility. This is the case, for example, of the block referenced as number 20 in Figure 10a), which is represented by the model object 120 in Figure 10b), whose visibility is less than 10%. Furthermore, implementing the method according to the invention, using point clouds representing the same blocks of a carapace at different times, makes it possible to track the evolution of the carapace between these two times. For example, the variation in the position of the center of gravity and the variation in the orientation of each block can thus be determined. The invention is obviously not limited to the implementation methods shown in the description by way of illustration and not limitation.

Claims

Demands

1. Method for generating a 3D model of a shell (5) of blocks (15) prefabricated from cementitious material, the process being implemented by computer and including: a) the provision of a point cloud (25) partially representing the carapace (5), b) the segmentation of the point cloud into homogeneous regions (80) re- presenting at least a part of one of the blocks (15) discretized in the scatter plot, (c) the aggregation of at least two of the homogeneous regions (80) which are adjacent to form a candidate region (100), d) the matching of said candidate region (100) with a model object 3D (105) representing at least part of a block (15), and e) the recording, in the 3D model of the carapace, of the position and of the orientation of the 3D model object matched in step d).

2. Method according to claim |, comprising subsequent to step e) repeating steps c) to e) with another candidate region in order to to match a model object representing another of the blocks of the shell,

3. A method according to claim 2, comprising the removal of the region candidate of the point cloud.

4. A method according to any one of claims 2 and 3, steps c) to (e) being repeated until more than 30%, preferably more than 50 %, preferably more than 70%, preferably more than 80%, of preferably more than 90%, or even 100% of the shell blocks are each modeled using a corresponding model object.

5. Method according to any one of the preceding claims, the step b) comprising: - voxelization of the point cloud (25), - the formation of non-disjoint and coherent groups of voxels spa- tially, for example by means of a "k-" partitioning method averages”, and - the generation of homogeneous regions (80) from the groups of voxels.

6. A method according to any one of the preceding claims, the seg- point cloud calculation including region generation homogeneous (80), also called "super-voxels", by putting in implements the VCCS segmentation technique.

7. A method according to any one of the preceding claims, the 3D model object being a mesh or a point cloud or, of preferably, being defined by means of vector geometric data.

8. Method according to the preceding claim, the model object being elaborated by computer-aided design.

9. A method according to any one of the preceding claims, the model object being located, in the point cloud reference frame of the carapace, by means of the position of its center of gravity and its orientation of its main axes.

10. A method according to any one of the preceding claims, the candidate region and the model object being matched to by means of a shape descriptor of the candidate region and one of the- shape decoder of the model object.

11. | Method according to the preceding claim, the shape descriptor of the candidate region being chosen from an FPFH descriptor, acronym English for "Fast Point Feature Histograms", a PFH descriptor, English acronym for "Point Feature Histogram", an LFSH descriptor, English acronym for "Local Feature Statistical Histograms" and one of the- VFP descriptor, English acronym for "Viewpoint Feature Histogram".

12. Method according to any one of claims 10 and 11, the des- the shape identifier of the model object being determined: - by generating a point cloud representing at least a part of the model object, also called the discretized model object, - by defining points of interest in the discretized model object, then - by calculating point of interest descriptors for each point of interest, the shape descriptor of the model object containing the des- Points of interest encoders.

13. Method according to any one of the preceding claims, the setting corresponding to the candidate region with the model object including minimizing the gap between the candidate region and the model object, preferably by implementing an algorithm of the nearest iterative point.

14. Method according to any one of the preceding claims, the step d) matching involving validation of the matching weighting, in particular by counting the number of points in the region candidate included in the volume of the model object and comparison of said number of points at a predetermined threshold.

15. Method according to any one of the preceding claims, the step d) of matching comprising, preferably after the va- ideation of the matching process, an optimization of the matching process correspondence of the 3D model object with the candidate region.

16. Method according to the preceding claim, the optimization comprising the expansion of the candidate region area by implementing at minus one optimization cycle comprising: 1) the aggregation into the candidate region of at least one, or even several homogeneous regions adjacent to it, 11) minimizing the gap between the enlarged candidate region and the model object, and iii) the calculation of a minimization residual of said difference, the optimization cycle being repeated with one or more other regions homogeneous and with the candidate region enlarged in step ii) as long as said residual calculated in step iii) is less than a predetermined threshold, by example between | mm and 15 mm.

17. | Method according to any one of the preceding claims, including the assignment of an identifier, for example alphanumeric, for example a name, to the 3D model object saved in step e).

18. Method according to any one of the preceding claims, according of which at least 90%, or even at least 95%, better 100% in number blocks represented in the point cloud that exhibit a vi- probability greater than or equal to 25%, or even greater than or equal to 20%, are represented in the 3D model of the shell.

19. A method according to any one of the preceding claims, comprising a preliminary step of acquiring the point cloud.

20. | Method according to the preceding claim, the acquisition of the point cloud points on the carapace containing: - an acquisition of at least one point cloud of the emerged part of the shell and / or an acquisition of at least one point cloud of the submerged part of the shell, - optionally, the merging of the point cloud of the emerged part and of the point cloud of the submerged part.

21. Method according to the preceding claim, the point cloud of the submerged part of the shell and the point cloud of the emerged part of the carapace exhibiting different point densities and / or different measurement precision for the positions of the points.

22. Method for tracking the movement of blocks in a carapace, including: - the implementation of the process according to any one of the claims previous instructions by providing in step a) a first cloud of points representing the shell at a first instant, in order to generate a first 3D model of the shell, - the implementation of the process according to any one of the claims previous instructions by providing in step a) a second cloud of points representing the shell at a second instant, in order to generate a second 3D model of the shell, and - the calculation of the displacement of at least one, preferably all of the common paired model objects of the first and second 3D models between the first and second moments.

23. Method for generating a consolidated 3D model of a carapace, including: - the implementation of the process according to any one of the claims instructions 1 to 21 by providing in step a) a first scatter plot representing a first portion of the carapace, in order to generate a first 3D model of the shell, - the implementation of the process according to any one of the claims instructions 1 to 21 by providing in step a) a second scatter plot representing a second portion of the carapace connected to the first portion of the carapace, in order to generate a second 3D model of the shell, and - the assembly of the first and second 3D models to form the Consolidated 3D model.

24. Computer to implement the process according to any one of the claims 1 to 21 and / or the monitoring method according to claim 22 or the generation method according to claim 23, the computer preferably including: a. at least one computer storage unit in which the cloud The point is recorded and a memory in which instructions those responsible for implementing the process are, and b. a processor to execute the instructions.

25. Computer data storage module, for example a disk hard, in which the 3D model obtained by the process according to one any of the claims | at 21, is registered.