Method, device, equipment, medium and program product for structural reconstruction of building complex

By performing voxel spatial semantic analysis and structural reconstruction of the point cloud data of the building complex, a polyhedral grid of building monomers was generated, which solved the problem of difficulty in defining the boundary of building monomers within the building complex in urban scenes, and achieved high-precision Structural reconstruction of the building complex.

CN118570395BActive Publication Date: 2024-11-26SHENZHEN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411017473.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-11-26
Estimated Expiration
2044-07-29

AI Technical Summary

Technical Problem

In urban market scenarios, the interval between building instances inside the building complex is small, which makes it difficult to accurately define the boundaries of building single units based on point cloud data, and is unable to achieve structured reconstruction.

Method used

By obtaining the original point cloud data of the building complex, semantic analysis of voxel space and globally consistent structured reconstruction, the target normal vector is obtained, and monolithically reconstructed based on these data to generate each building monolith in the building complex. target polyhedral mesh.

Benefits of technology

Accurate structured reconstruction of building monomers in building complexes, improve generalization and accuracy of semantic analysis, reduce the robustness to noise, reduce dependence on precise normal vectors, and improve monomer reconstruction accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118570395B_ABST
    Figure CN118570395B_ABST
Patent Text Reader

Abstract

The present application relates to a method, device, equipment, medium and program product for structural reconstruction of a building complex. The method comprises: obtaining the original point cloud data of the building complex; performing semantic analysis of the original point cloud data in voxel space to obtain semantically aligned target point cloud data; performing globally consistent structural reconstruction on the original point cloud data to obtain target normal vectors of polyhedrons in the building complex; performing monomer reconstruction based on the target point cloud data and the target normal vector to obtain the target polyhedron mesh of each building monomer in the building complex. The present method can accurately perform structural reconstruction of building monomers in the building complex.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of computer graphics, and in particular to a method, apparatus, computer equipment, computer-readable storage medium and computer program product for structural reconstruction of building complexes. Background Art

[0002] Semantic perception and reconstruction of city-scale point clouds are an important data foundation for smart city applications, and they play an irreplaceable role in spatial computing needs such as traffic management, engineering monitoring, crowd flow prediction, and geographic information analysis. In order to build a city digital twin and meet various application needs of smart city management, it is necessary to rely on remote sensing satellites, vehicle-mounted radars, drones and other equipment to complete scene-oriented data acquisition. Among them, multi-view acquisition and laser point cloud scanning based on drones are currently the mainstream solutions for generating high-precision data for urban scenes. By using drones carrying multi-angle photography equipment to obtain multiple high-precision images of an area, multi-view Figure 3 Reconstruction.

[0003] However, for building clusters in urban scenes, the internal building instances are closely spaced, and the data representation based on point clouds cannot define the precise instance boundaries, making individual reconstruction very difficult. There is also the problem of being unable to accurately perform structural reconstruction of building units in the building cluster. Summary of the invention

[0004] Based on this, it is necessary to provide a method, device, computer equipment, computer-readable storage medium and computer program product for structural reconstruction of a building complex, which can accurately reconstruct the building units in the building complex in order to solve the above-mentioned technical problems.

[0005] In a first aspect, the present application provides a method for structural reconstruction of a building complex, comprising:

[0006] Obtain the original point cloud data of the building complex;

[0007] Performing semantic analysis in voxel space on the original point cloud data to obtain semantically aligned target point cloud data;

[0008] Performing globally consistent structural reconstruction on the original point cloud data to obtain target normal vectors of polyhedrons in the building complex;

[0009] A monomer reconstruction is performed based on the target point cloud data and the target normal vector to obtain a target polyhedral mesh of each building monomer in the building complex.

[0010] In one embodiment, performing semantic analysis in voxel space on the original point cloud data to obtain semantically aligned target point cloud data includes:

[0011] Normalizing the original point cloud data to obtain normalized point cloud data;

[0012] The normalized point cloud data is semantically aligned in voxel space to obtain semantically aligned target point cloud data.

[0013] In one embodiment, the normalizing the original point cloud data to obtain normalized point cloud data includes:

[0014] For each voxel region in the original point cloud data, a first voxel subset and a second voxel subset in the voxel region are obtained; the voxel cells in the first voxel subset contain point cloud data, and the second voxel subset includes voxels at the bottom layer of the voxelized point cloud data;

[0015] Determine the spatial occupancy of the voxel region based on the first voxel subset and the second voxel subset, and determine a target voxel region from each voxel region based on a specified spatial occupancy threshold and the spatial occupancy of each voxel region;

[0016] The original point cloud data is normalized using the target voxel region to obtain normalized point cloud data.

[0017] In one embodiment, the globally consistent structural reconstruction of the original point cloud data to obtain the target normal vectors of the polyhedrons in the building complex includes:

[0018] Performing facet estimation and convex hull polyhedron generation on the original point cloud data to construct a polyhedron space of the building complex; the polyhedron space includes a polyhedron, and the polyhedron includes faces;

[0019] Performing mesh reconstruction and normal vector assistance on the original point cloud data to obtain an initial normal vector of the face patch;

[0020] Structural reconstruction is performed based on the polyhedron space and the initial normal vector to obtain a target normal vector of the polyhedron.

[0021] In one embodiment, the monomeric reconstruction based on the target point cloud data and the target normal vector to obtain a target polyhedral mesh of each building unit in the building complex includes:

[0022] Extracting building calibration points from the target point cloud data;

[0023] Performing nearest neighbor clustering on the building calibration points to obtain different initial monomerized point clouds;

[0024] Based on the target normal vector, the initial monomerized point cloud is de-adhesion processed to obtain a target polyhedral mesh of each building monomer in the building complex.

[0025] In one embodiment, the de-adhesion processing is performed on the initial monomerized point cloud based on the target normal vector to obtain a target polyhedral mesh of each building monomer in the building complex, including:

[0026] Performing structured reconstruction on the initial monomerized point cloud to obtain an initial polyhedral mesh;

[0027] Determine a non-manifold patch from the initial polyhedron mesh, and determine an adjacent polyhedron of the non-manifold patch; the non-manifold patch is a patch that is not contained in any closed polyhedron;

[0028] Based on the target normal vector and the influence of the non-manifold patch on the winding number of the adjacent polyhedron, a target polyhedron mesh of each building unit in the building complex is generated.

[0029] In a second aspect, the present application also provides a structural reconstruction device for a building complex, comprising:

[0030] The original point cloud data acquisition module is used to acquire the original point cloud data of the building complex;

[0031] A semantic analysis module, used for performing semantic analysis in voxel space on the original point cloud data to obtain semantically aligned target point cloud data;

[0032] A structured reconstruction module, used for performing globally consistent structured reconstruction on the original point cloud data to obtain target normal vectors of the polyhedrons in the building complex;

[0033] The monomer reconstruction module is used to perform monomer reconstruction based on the target point cloud data and the target normal vector to obtain a target polyhedral mesh of each building monomer in the building complex.

[0034] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0035] Obtain the original point cloud data of the building complex;

[0036] Performing semantic analysis in voxel space on the original point cloud data to obtain semantically aligned target point cloud data;

[0037] Performing globally consistent structural reconstruction on the original point cloud data to obtain target normal vectors of polyhedrons in the building complex;

[0038] A monomer reconstruction is performed based on the target point cloud data and the target normal vector to obtain a target polyhedral mesh of each building monomer in the building complex.

[0039] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0040] Obtain the original point cloud data of the building complex;

[0041] Performing semantic analysis in voxel space on the original point cloud data to obtain semantically aligned target point cloud data;

[0042] Performing globally consistent structural reconstruction on the original point cloud data to obtain target normal vectors of polyhedrons in the building complex;

[0043] A monomer reconstruction is performed based on the target point cloud data and the target normal vector to obtain a target polyhedral mesh of each building monomer in the building complex.

[0044] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0045] Obtain the original point cloud data of the building complex;

[0046] Performing semantic analysis in voxel space on the original point cloud data to obtain semantically aligned target point cloud data;

[0047] Performing globally consistent structural reconstruction on the original point cloud data to obtain target normal vectors of polyhedrons in the building complex;

[0048] A monomer reconstruction is performed based on the target point cloud data and the target normal vector to obtain a target polyhedral mesh of each building monomer in the building complex.

[0049] The above-mentioned method, apparatus, computer equipment, computer-readable storage medium and computer program product for structured reconstruction of a building complex obtain the original point cloud data of the building complex, perform semantic analysis in voxel space on the original point cloud data, obtain semantically aligned target point cloud data, perform globally consistent structured reconstruction on the original point cloud data, and obtain the target normal vectors of the polyhedrons in the building complex. Then, based on the target point cloud data and the target normal vectors, a monomeric reconstruction is performed to accurately obtain the target polyhedron mesh of each building monomer in the building complex. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 It is a schematic diagram of a flow chart of a method for structural reconstruction of a building complex in one embodiment;

[0052] Figure 2 A schematic diagram of structured reconstruction of a semantically consistent urban building complex in one embodiment;

[0053] Figure 3 A schematic diagram of visualizing space occupancy in one embodiment;

[0054] Figure 4 A schematic diagram for visualizing space occupancy in another embodiment;

[0055] Figure 5 A schematic diagram of correcting semantic classification based on semantic alignment of spatial analysis in one embodiment;

[0056] Figure 6 A schematic diagram of globally consistent structured reconstruction in one embodiment;

[0057] Figure 7 A schematic diagram of a dough sheet de-adhesion process in one embodiment;

[0058] Figure 8 It is a structural block diagram of a building complex structural reconstruction device in one embodiment;

[0059] Fig. 9 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0061] In one embodiment, Figure 1As shown, a method for structural reconstruction of a building complex is provided. This embodiment uses the method applied to a computer device as an example for illustration. The computer device can be a terminal or a server. It can be understood that the method can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. Among them, the terminal can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car devices, projection devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0062] In this embodiment, the structured reconstruction method of the building complex includes the following steps:

[0063] Step S102, obtaining original point cloud data of the building complex.

[0064] The point cloud data refers to the point set of the target surface characteristics, and the original point cloud data refers to the point set of the original surface characteristics of the building complex.

[0065] Optionally, the original point cloud data of the building complex can be collected by remote sensing satellites, lidars, drones and other equipment.

[0066] Step S104, performing semantic analysis in voxel space on the original point cloud data to obtain semantically aligned target point cloud data.

[0067] Among them, the target point cloud data is the point cloud data obtained after the original point cloud data is semantically aligned.

[0068] Optionally, the computer device performs scene data scale perception and spatial information-assisted semantic analysis in voxel space on the original point cloud data to obtain semantically aligned target point cloud data. Voxelization can be regarded as a set of regular cubic space division methods, and a voxel is a cubic unit space under the division.

[0069] Step S106, performing globally consistent structural reconstruction on the original point cloud data to obtain target normal vectors of the polyhedrons in the building complex.

[0070] Normal vector is a concept in spatial analytic geometry. The vector represented by a line perpendicular to a plane is the normal vector of the plane. The target normal vector is the normal vector of the polyhedron target in the building complex.

[0071] Optionally, the computer device performs globally consistent structural reconstruction on the original point cloud data, obtains the initial normal vector of each polyhedron in the building complex, and adjusts the initial normal vector to obtain the target normal vector of each polyhedron in the building complex. The initial normal vector of each polyhedron can be estimated by the IPSR (Iterative Poisson Surface Reconstruction) algorithm.

[0072] Step S108, performing monomer reconstruction based on the target point cloud data and the target normal vector to obtain a target polyhedral mesh of each building unit in the building complex.

[0073] Optionally, the computer device performs monomer reconstruction based on the point cloud neighborhood data and the target normal vector in the target point cloud data, removes the adhesion patches in the building complex, and obtains the target polyhedral mesh of each building monomer in the building complex.

[0074] Optionally, the building complex may include one or more building units.

[0075] The above-mentioned structured reconstruction method of the building complex obtains the original point cloud data of the building complex, performs semantic analysis on the original point cloud data in voxel space, obtains the target point cloud data after semantic alignment, performs globally consistent structured reconstruction on the original point cloud data, and obtains the target normal vectors of the polyhedrons in the building complex. Then, based on the target point cloud data and the target normal vector, a monomer reconstruction is performed to accurately obtain the target polyhedron mesh of each building monomer in the building complex.

[0076] In addition, by effectively utilizing the spatial information expressed by voxels, the generalization accuracy of semantic analysis is improved; a globally consistent structured grid reconstruction method is established to improve the robustness to noise; and grid de-adhesion combined with structured information improves the accuracy of singulation.

[0077] Traditional semantic analysis methods obtain semantic analysis results through neural network training and feature extraction. The generalization of feature analysis is poor, especially for complex urban scenes, the probability of misidentification is high. The above-mentioned building complex structured reconstruction method effectively improves the generalization and accuracy of semantic analysis by utilizing the spatial information expressed by voxels.

[0078] Traditional structured reconstruction methods often rely on high-quality normal vectors to support mesh generation schemes and are sensitive to noise. The above-mentioned building complex structured reconstruction method uses globally consistent structured mesh reconstruction, combines facet estimation with the faceted space generation strategy, and uses winding number optimization to reduce the dependence on precise normal vectors and improve the robustness of the mesh reconstruction process to the original data. Mesh de-adhesion combined with structured information uses the manifold distribution characteristics of the main facet and the winding number distribution of the adjacent polyhedron points to determine whether to delete them to obtain accurate monomer reconstruction results.

[0079] In one embodiment, Figure 2 As shown, the computer device inputs the urban scene point cloud data, performs voxel space-based semantic analysis on the urban scene point cloud data, and obtains semantically aligned target point cloud data; performs globally consistent structured reconstruction on the urban scene point cloud data, and obtains the target normal vectors of the polyhedrons in the building complex of the urban scene; performs monomer reconstruction based on the target point cloud data and the target normal vector, and obtains the target polyhedron mesh of each building monomer in the building complex, so as to complete the structured reconstruction of the building complex.

[0080] In one embodiment, semantic analysis is performed in voxel space on the original point cloud data to obtain semantically aligned target point cloud data, including: normalizing the original point cloud data to obtain normalized point cloud data; and semantically aligning the normalized point cloud data in voxel space to obtain semantically aligned target point cloud data.

[0081] Optionally, the computer device may use the spatial ratio to perform normalization calculations based on the scale perception of the voxel space, or may adopt other types of voxel scale normalization methods, which are not limited here.

[0082] Optionally, the original point cloud data is normalized to obtain normalized point cloud data, including: for each voxel area in the original point cloud data, obtaining a first voxel subset and a second voxel subset in the voxel area; the voxel cells in the first voxel subset contain point cloud data, and the second voxel subset includes voxels at the bottom layer of the voxelized point cloud data; based on the first voxel subset and the second voxel subset, determining the spatial occupancy of the voxel area, based on a specified spatial occupancy threshold and the spatial occupancy of each voxel area, determining a target voxel area from each voxel area; using the target voxel area, the original point cloud data is normalized to obtain normalized point cloud data.

[0083] Optionally, the computer device voxelizes the original point cloud data to obtain voxelized point cloud data; and segments the voxelized point cloud data to obtain various voxel regions.

[0084] Optionally, for each voxel region in the original point cloud data, the computer device determines a first number of a first voxel subset and a second number of a second voxel subset; and uses a ratio between the first number and the second number as the spatial occupancy rate of the original point cloud data.

[0085] Optionally, the computer device determines the spatial occupancy of the voxel region using the following formula (1):

[0086] (1)

[0087] in, It represents the point cloud data after voxelization of the original point cloud data. The diameter of the voxel is one fifth of the highest altitude value in the original point cloud data. After voxelization, some voxel cells contain point cloud data, which is recorded as The first voxel subset of ; The bottom voxel in the voxelized point cloud data is used to construct another The second voxel subset of ;based on and The number of voxels included and , determine the spatial occupancy of the voxel region , is explicitly calculated, such as Figure 3 and Figure 4 shown.

[0088] It is understandable that one of the difficulties in semantic analysis of urban scene point cloud data is the prior knowledge of its corresponding scale, which makes it impossible to perform stable and unified normalization calculations and achieve point cloud density optimization, which to a certain extent restricts the generalization of semantic analysis.

[0089] Therefore, the computer device obtains the spatial occupancy of each voxel area and more accurately implements normalization calculation based on the specified spatial occupancy threshold and the spatial occupancy of each voxel area to guide the efficient preprocessing of the scene point cloud data.

[0090] Optionally, for each voxel region, the computer device determines the difference between a specified spatial occupancy threshold and the spatial occupancy corresponding to the voxel region; based on the difference corresponding to each voxel region, determines a target voxel region closest to the spatial occupancy threshold from each voxel region.

[0091] The computer device may use the voxel region corresponding to the smallest difference as the target voxel region.

[0092] Optionally, each voxel region satisfies a relatively uniform range constraint in terms of spatial occupancy, as shown in formula (2):

[0093] (2)

[0094] in, Represents a The voxel segmentation method corresponds to a voxel area; is a specified space occupancy threshold, obtained empirically; For the corresponding A set of spatial occupancy rates; calculated by iteration The average value and The difference is calculated to obtain an adaptive target voxel region that is closest to the specified space occupancy threshold. Based on the target voxel region, a unified sampling calculation can be driven to determine the voxelization method and guide the number of sampling points retained in each voxel, so as to achieve global normalization of the point cloud density and realize efficient data preprocessing.

[0095] Optionally, the computer device adopts spatial information assistance technology, which can be based on the longitudinal height of the voxel to assist in semantic classification, or can adopt other information analysis methods based on the voxel space to assist in semantic classification, which is not limited here.

[0096] Optionally, semantic alignment is performed in voxel space on the normalized point cloud data to obtain semantically aligned target point cloud data, including: obtaining a first voxel and a second voxel from the normalized point cloud data; the longitudinal height of the first voxel is greater than a height threshold, and the second voxel is a voxel other than the first voxel in the voxelized normalized point cloud data; updating the semantic classification of the first voxel to a building to obtain an updated first voxel; using the updated first voxel as a calibration, updating the semantic classification of the second voxel to obtain an updated second voxel; and obtaining semantically aligned target point cloud data based on the updated first voxel and the updated second voxel.

[0097] The height threshold can be set as needed, such as half of the highest altitude position in the normalized point cloud data.

[0098] Optionally, the computer device voxelizes the normalized point cloud data to obtain voxelized point cloud data, counts the longitudinal height of each voxel in the voxelized point cloud data, and obtains the first voxel and the second voxel based on the longitudinal height of each voxel.

[0099] It is understandable that deep learning methods are widely used for urban scene semantic analysis tasks, and compared with traditional methods, they have achieved significant improvements in recognition accuracy and robustness. However, for complex scene point clouds, there are still cases of misclassification, especially at the bottom, top and side facades of high-rise buildings. To solve this problem, this embodiment introduces semantic alignment based on spatial analysis to improve the performance of semantic analysis.

[0100] The computer device counts the longitudinal heights of all voxels based on the voxelized voxel structure, selects the first voxel greater than the height threshold, and obtains the semantic classification of the first voxel; if the semantic classification of the first voxel is a building, no update is required; if the semantic classification of the first voxel is a non-building, the semantic classification of the first voxel is updated to a building.

[0101] For example, in the voxelized point cloud data, if the first voxel whose vertical height is greater than the height threshold is semantically classified as ground or vehicle, the semantic classification of the first voxel is updated to building. At the same time, the location of the updated (corrected) first voxel is recorded, and the adjacent voxels are searched downward in the vertical direction of the updated first voxel locked at the position, and the semantic classification of the adjacent voxels is updated to building to complete the semantic alignment.

[0102] It is understandable that the execution of the semantic alignment method is based on an assumption that in large-scale urban scenes, there are high-rise buildings, which are used as calibration to check the classification of other points. Since the altitude of high-rise buildings is significantly higher than that of other semantic objects, the result of semantic classification is corrected based on the height threshold, which effectively considers the spatial distribution law of urban semantic objects, that is, semantic alignment is achieved.

[0103] like Figure 5 As shown in the figure, after semantic alignment, the erroneous points in the semantic annotation based on the deep network are corrected, which improves the accuracy of the semantic annotation of high-rise buildings and realizes the semantic alignment of point cloud data.

[0104] In this embodiment, the computer device normalizes the original point cloud data to obtain normalized point cloud data, and then performs semantic alignment in voxel space on the normalized point cloud data to obtain semantically aligned target point cloud data, which can improve the scale estimation of point cloud data and enhance the ability to recognize semantic objects in complex environments.

[0105] In one embodiment, a globally consistent structured reconstruction is performed on the original point cloud data to obtain target normal vectors of the polyhedrons in the building complex, including: performing facet estimation and convex hull polyhedron generation on the original point cloud data to construct a polyhedron space of the building complex; the polyhedron space includes polyhedrons, and the polyhedrons include faces; meshing and normal vector assistance are performed on the original point cloud data to obtain initial normal vectors of the faces; and structured reconstruction is performed based on the polyhedron space and the initial normal vector to obtain the target normal vector of the polyhedron.

[0106] The polyhedron space is the space where each polyhedron of the building complex is located. A polyhedron is an object formed by a building and has multiple facets, and a facet is one of the faces of the polyhedron. The computer device obtains the center of the polyhedron, the center of the facets constituting the polyhedron, and the area of ​​the facets from the polyhedron space.

[0107] Optionally, the computer device uses a patch estimation algorithm to perform patch estimation on the original point cloud data, and extracts initial structured information except the original point cloud data; uses a convex hull polyhedron generation algorithm to generate a convex hull polyhedron for the initial structured information, and constructs a polyhedron space to complete the perception of the internal space of the point cloud and effectively compensate for the missing areas of the original point cloud.

[0108] Optionally, the computer equipment performs globally consistent structured reconstruction of the original point cloud data, which can be combined with IPSR normal assistance and winding number global optimization orientation scheme, or other normal vector-based global orientation methods for polyhedral mesh reconstruction and output, which are not limited here.

[0109] Optionally, the computer device uses the IPSR grid to perform mesh reconstruction and normal vector assistance on the original point cloud data to obtain the initial normal vector of the facet to drive global consistent structured reconstruction.

[0110] Optionally, the computer device performs a globally consistent polyhedron mesh reconstruction with optimized winding number based on the polyhedron space and the initial normal vectors of the facets to obtain target normal vectors of the polyhedrons in the building complex.

[0111] Optionally, structured reconstruction is performed based on the polyhedron space and the initial normal vector to obtain the target normal vector of the polyhedron, including: determining the initial winding number of the polyhedron based on the polyhedron space and the initial normal vector; determining the target orientation of the center of the polyhedron according to the initial winding number and the winding threshold; determining the target normal vector of the polyhedron based on the target orientation of the center of the polyhedron and the initial normal vector.

[0112] The wrapping threshold can be set as needed. For example, the wrapping threshold 𝜖=0.5.

[0113] Optionally, for each polyhedron in the polyhedron space, the computer device determines the initial winding number of the polyhedron based on the center of the polyhedron, the centers of each facet constituting the polyhedron, the area of ​​the facet, and the initial normal vector of the facet; if the initial winding number is greater than the winding threshold, the target orientation of the center of the polyhedron is determined to be an inner point; if the initial winding number is less than or equal to the winding threshold, the target orientation of the center of the polyhedron is determined to be an outer point.

[0114] Based on the target orientation of the center of the polyhedron, the patch reconstruction calculation for the adjacent polyhedrons can be derived, and finally the polyhedron mesh can be derived.

[0115] Optionally, the computer device is based on a globally consistent polyhedral mesh reconstruction method based on winding number optimization, and its optimization term is expressed as:

[0116]

[0117] Among them, the parameters represents the center of the polyhedron, Represents the facets of a polyhedron The center of express The area of is based on The normal vector of has been preliminarily corrected by the initial normal vector estimated by IPSR. Based on this formula, the center points inside all polyhedrons generated in the previous stage will have a corresponding winding w value. According to the winding number optimization theory, when When the value approaches 1, A point is defined as an interior point; when When the value approaches 0, The point is defined as an outlier. Here we set the wrapping threshold ,when , Points are considered as interior points, otherwise they are considered as exterior points. The orientation of the points enables the calculation of the reconstruction of the patches of the adjacent polyhedrons and ultimately the polyhedral mesh.

[0118] like Figure 6 As shown, the computer equipment performs face estimation and convex hull polyhedron generation on the original point cloud data to construct the polyhedron space of the building complex; performs IPSR mesh reconstruction and normal vector assistance on the original point cloud data to obtain the initial normal vector of the face; performs polyhedron orientation and structured reconstruction based on the polyhedron space and the initial normal vector to obtain the target normal vector of the polyhedron.

[0119] In this embodiment, the computer device uses a winding number optimization strategy to achieve global orientation of the polyhedral mesh data, so that the simplified polyhedral mesh data can be accurately exported.

[0120] In one embodiment, monomeric reconstruction is performed based on the target point cloud data and the target normal vector to obtain a target polyhedral mesh of each building unit in the building complex, including: extracting building calibration points from the target point cloud data; performing nearest neighbor clustering on the building calibration points to obtain different initial monomeric point clouds; based on the target normal vector, de-adhesion processing is performed on the initial monomeric point cloud to obtain a target polyhedral mesh of each building unit in the building complex.

[0121] The building calibration points are points that are semantically classified as buildings. The target point cloud data includes the semantic classification (semantic annotation) of each point cloud data, and the computer device can extract the building calibration points that are semantically classified as buildings from the target point cloud data.

[0122] Optionally, the computer device uses a K-nearest neighbor algorithm to perform nearest neighbor clustering on the building calibration points to obtain different initial individualized point clouds. The initial individualized point cloud is a point cloud obtained by the computer device by roughly classifying each building monomer.

[0123] Optionally, based on the target normal vector, the initial monomerized point cloud is de-adhesion processed to obtain a target polyhedral mesh for each building unit in the building complex, including: performing structured reconstruction on the initial monomerized point cloud to obtain an initial polyhedral mesh; determining non-manifold patches from the initial polyhedral mesh, and determining adjacent polyhedrons of the non-manifold patches; non-manifold patches are patches that are not included in any closed polyhedron; based on the target normal vector and the influence of the non-manifold patches on the winding number of adjacent polyhedrons, a target polyhedral mesh for each building unit in the building complex is generated.

[0124] Optionally, the computer device performs globally consistent structured reconstruction on the initial monomerized point cloud to obtain an initial polyhedral mesh.

[0125] It is understandable that the computer device can re-determine the new winding number of the polyhedron based on the target normal vector to determine whether the non-manifold face needs to be de-adhesion processed.

[0126] Optionally, the computer device determines the center of the adjacent polyhedron of the non-manifold patch, and determines the first winding number of the adjacent polyhedron based on the target normal vector; if the non-manifold patch is added to the adjacent polyhedron, a new adjacent polyhedron is obtained, and the second winding number of the new adjacent polyhedron is determined based on the target normal vector; the initial winding number corresponding to the adjacent polyhedron is obtained, and based on the initial winding number, the first winding number and the second winding number, a target polyhedron mesh for each building unit in the building complex is generated.

[0127] The computer device compares the first winding number and the second winding number with the initial winding number respectively, and selects a target winding number from the first winding number and the second winding number, where the target winding number is one of the first winding number and the second winding number that is closer to the initial winding number; if the target winding number is the first winding number, the non-manifold face is deleted to obtain a target polyhedral mesh of the building unit; if the target winding number is the second winding number, the non-manifold face is retained to obtain a target polyhedral mesh of the building unit.

[0128] It is understandable that in actual urban 3D reconstruction application scenarios, the distance between models in a building complex is relatively close, which affects the accuracy of single-unit calibration and reduces the interaction efficiency of scene data. In this embodiment, the computer device combines the single-unit reconstruction method of target point cloud data and structured information analysis to solve the grid "sticking" problem in the building reconstruction process and more accurately generate the target polyhedral grid of each building unit. Figure 7As shown in the figure, after the patch de-adhesion process, the incorrectly connected patches between dense building models will be deleted to improve the accuracy of building monomerization.

[0129] By using various open-source algorithms to conduct experiments on the above data, including data sets such as BuildingNet, Sensat and UrbanBIS, the results show that the above building complex structured reconstruction method has significant performance improvements in urban scene semantic analysis and structured reconstruction accuracy, and for the structured reconstruction of dense building complexes, through monomer reconstruction, it can effectively remove the adhesion of patches and obtain a clearer expression of building monomers.

[0130] In one embodiment, a method for structural reconstruction of a building complex is also provided, which is applied to a computer device. The method for structural reconstruction of a building complex comprises the following steps:

[0131] Step A1, obtaining the original point cloud data of the building complex.

[0132] Step A2, for each voxel region in the original point cloud data, obtain a first voxel subset and a second voxel subset in the voxel region; the voxel cells in the first voxel subset contain point cloud data, and the second voxel subset includes the voxels at the bottom layer of the voxelized point cloud data; based on the first voxel subset and the second voxel subset, determine the spatial occupancy of the voxel region, based on the specified spatial occupancy threshold and the spatial occupancy of each voxel region, determine the target voxel region from each voxel region; with the target voxel region, normalize the original point cloud data to obtain normalized point cloud data.

[0133] Step A3, obtaining a first voxel and a second voxel from the normalized point cloud data; the longitudinal height of the first voxel is greater than a height threshold, and the second voxel is a voxel other than the first voxel in the voxelized normalized point cloud data; updating the semantic classification of the first voxel to a building to obtain an updated first voxel; using the updated first voxel as a calibration, updating the semantic classification of the second voxel to obtain an updated second voxel; based on the updated first voxel and the updated second voxel, obtaining the semantically aligned target point cloud data.

[0134] Step A4, performing face estimation and convex hull polyhedron generation on the original point cloud data to construct a polyhedron space of the building complex; the polyhedron space includes polyhedrons, and the polyhedrons include faces; performing mesh reconstruction and normal vector assistance on the original point cloud data to obtain the initial normal vector of the face; based on the polyhedron space and the initial normal vector, determining the initial winding number of the polyhedron; based on the initial winding number and the winding threshold, determining the target orientation of the center of the polyhedron; based on the target orientation of the center of the polyhedron and the initial normal vector, determining the target normal vector of the polyhedron.

[0135] Step A5, extracting building calibration points from the target point cloud data; performing nearest neighbor clustering on the building calibration points to obtain different initial monomerized point clouds; performing structured reconstruction on the initial monomerized point cloud to obtain an initial polyhedron mesh; determining non-manifold patches from the initial polyhedron mesh, and determining the adjacent polyhedrons of the non-manifold patches; the non-manifold patches are patches that are not included in any closed polyhedron; based on the target normal vector and the influence of the non-manifold patches on the winding number of the adjacent polyhedrons, generate a target polyhedron mesh for each building unit in the building complex.

[0136] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0137] Based on the same inventive concept, the embodiment of the present application also provides a building complex structured reconstruction device for implementing the above-mentioned building complex structured reconstruction method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more embodiments of the building complex structured reconstruction device provided below can refer to the limitations of the building complex structured reconstruction method above, and will not be repeated here.

[0138] In an exemplary embodiment, Figure 8 As shown, a building complex structure reconstruction device is provided, comprising: an original point cloud data acquisition module 802, a semantic analysis module 804, a structure reconstruction module 806 and a monomer reconstruction module 808, wherein:

[0139] The original point cloud data acquisition module 802 is used to acquire the original point cloud data of the building complex.

[0140] The semantic analysis module 804 is used to perform semantic analysis on the original point cloud data in voxel space to obtain semantically aligned target point cloud data.

[0141] The structured reconstruction module 806 is used to perform globally consistent structured reconstruction on the original point cloud data to obtain target normal vectors of the polyhedrons in the building complex.

[0142] The monomer reconstruction module 808 is used to perform monomer reconstruction based on the target point cloud data and the target normal vector to obtain a target polyhedral mesh of each building unit in the building complex.

[0143] The above-mentioned building complex structured reconstruction device obtains the original point cloud data of the building complex, performs semantic analysis on the original point cloud data in voxel space, obtains the target point cloud data after semantic alignment, performs globally consistent structured reconstruction on the original point cloud data, and obtains the target normal vectors of the polyhedrons in the building complex. Then, based on the target point cloud data and the target normal vector, a monomer reconstruction is performed to accurately obtain the target polyhedron mesh of each building monomer in the building complex.

[0144] In one embodiment, the semantic analysis module 804 is further used to normalize the original point cloud data to obtain normalized point cloud data; and perform semantic alignment in voxel space on the normalized point cloud data to obtain semantically aligned target point cloud data.

[0145] In one embodiment, the semantic analysis module 804 is also used to obtain a first voxel subset and a second voxel subset in the voxel region for each voxel region in the original point cloud data; the voxel cells in the first voxel subset contain point cloud data, and the second voxel subset includes the voxels at the bottom layer of the voxelized point cloud data; based on the first voxel subset and the second voxel subset, the spatial occupancy of the voxel region is determined, and based on the specified spatial occupancy threshold and the spatial occupancy of each voxel region, the target voxel region is determined from each voxel region; the original point cloud data is normalized with the target voxel region to obtain normalized point cloud data.

[0146] In one embodiment, the semantic analysis module 804 is further used to obtain a first voxel and a second voxel from normalized point cloud data; the longitudinal height of the first voxel is greater than a height threshold, and the second voxel is a voxel other than the first voxel in the voxelized normalized point cloud data; the semantic classification of the first voxel is updated to a building to obtain an updated first voxel; the semantic classification of the second voxel is updated using the updated first voxel as a calibration to obtain an updated second voxel; based on the updated first voxel and the updated second voxel, the semantically aligned target point cloud data is obtained.

[0147] In one embodiment, the above-mentioned structured reconstruction module 806 is also used to perform facet estimation and convex hull polyhedron generation on the original point cloud data to construct a polyhedron space of the building complex; the polyhedron space includes polyhedrons, and the polyhedrons include faces; the original point cloud data is meshed and normal vector-assisted to obtain the initial normal vector of the facet; structured reconstruction is performed based on the polyhedron space and the initial normal vector to obtain the target normal vector of the polyhedron.

[0148] In one embodiment, the above-mentioned structured reconstruction module 806 is also used to determine the initial winding number of the polyhedron based on the polyhedron space and the initial normal vector; determine the target orientation of the center of the polyhedron according to the initial winding number and the winding threshold; and determine the target normal vector of the polyhedron based on the target orientation of the center of the polyhedron and the initial normal vector.

[0149] In one embodiment, the above-mentioned monomer reconstruction module 808 is also used to extract building calibration points from the target point cloud data; perform nearest neighbor clustering on the building calibration points to obtain different initial monomer point clouds; based on the target normal vector, de-adhesion process is performed on the initial monomer point cloud to obtain the target polyhedral mesh of each building unit in the building complex.

[0150] In one embodiment, the above-mentioned monomerized reconstruction module 808 is also used to perform structured reconstruction on the initial monomerized point cloud to obtain an initial polyhedral mesh; determine non-manifold patches from the initial polyhedral mesh, and determine the adjacent polyhedrons of the non-manifold patches; non-manifold patches are patches that are not included in any closed polyhedron; based on the target normal vector and the influence of the non-manifold patches on the winding number of adjacent polyhedrons, generate a target polyhedral mesh for each building unit in the building complex.

[0151] Each module in the above-mentioned building complex structural reconstruction device can be implemented in whole or in part by software, hardware and their combination. Each module can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0152] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Fig. 9As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be realized through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a method for structural reconstruction of a building complex is realized. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0153] Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0154] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0155] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0156] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0157] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0158] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0159] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0160] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for structural reconstruction of a building complex, characterized in that: The method comprises: Obtain the original point cloud data of the building complex; For each voxel region in the original point cloud data, a first voxel subset and a second voxel subset in the voxel region are obtained; the voxel cells in the first voxel subset contain point cloud data, and the second voxel subset includes voxels at the bottom layer of the voxelized point cloud data; Determine a first number of the first voxel subset and a second number of the second voxel subset; and use a ratio between the first number and the second number as a spatial occupancy of the voxel region; Based on a specified space occupancy threshold and the space occupancy of each voxel region, determining a target voxel region from each voxel region; Normalizing the original point cloud data using the target voxel region to obtain normalized point cloud data; Performing semantic alignment in voxel space on the normalized point cloud data to obtain semantically aligned target point cloud data; Performing globally consistent structural reconstruction on the original point cloud data to obtain target normal vectors of polyhedrons in the building complex; A monomeric reconstruction is performed based on the target point cloud data and the target normal vector to obtain a target polyhedral mesh of each building monomer in the building complex.

2. The method according to claim 1, characterized in that: The globally consistent structural reconstruction of the original point cloud data to obtain the target normal vectors of the polyhedrons in the building complex includes: Performing facet estimation and convex hull polyhedron generation on the original point cloud data to construct a polyhedron space of the building complex; the polyhedron space includes a polyhedron, and the polyhedron includes faces; Performing mesh reconstruction and normal vector assistance on the original point cloud data to obtain an initial normal vector of the face patch; Structural reconstruction is performed based on the polyhedron space and the initial normal vector to obtain a target normal vector of the polyhedron.

3. The method according to claim 1, characterized in that The monomeric reconstruction based on the target point cloud data and the target normal vector to obtain a target polyhedral mesh of each building monomer in the building complex includes: Extracting building calibration points from the target point cloud data; Performing nearest neighbor clustering on the building calibration points to obtain different initial monomerized point clouds; Based on the target normal vector, the initial monomerized point cloud is de-adhesion processed to obtain a target polyhedral mesh of each building monomer in the building complex.

4. The method according to claim 3, characterized in that The de-adhesion processing is performed on the initial monomerized point cloud based on the target normal vector to obtain a target polyhedral mesh of each building monomer in the building complex, including: Performing structured reconstruction on the initial monomerized point cloud to obtain an initial polyhedral mesh; Determine a non-manifold patch from the initial polyhedron mesh, and determine an adjacent polyhedron of the non-manifold patch; the non-manifold patch is a patch that is not contained in any closed polyhedron; Based on the target normal vector and the influence of the non-manifold patch on the winding number of the adjacent polyhedron, a target polyhedron mesh of each building unit in the building complex is generated.

5. A structural reconstruction device for a building complex, characterized in that: The device comprises: The original point cloud data acquisition module is used to acquire the original point cloud data of the building complex; A semantic analysis module, for obtaining, for each voxel region in the original point cloud data, a first voxel subset and a second voxel subset in the voxel region; the voxel cells in the first voxel subset contain point cloud data, and the second voxel subset includes voxels at the bottom layer of the voxelized point cloud data; determining a first number of the first voxel subset and a second number of the second voxel subset; taking the ratio between the first number and the second number as the spatial occupancy of the voxel region; determining a target voxel region from each voxel region based on a specified spatial occupancy threshold and the spatial occupancy of each voxel region; normalizing the original point cloud data with the target voxel region to obtain normalized point cloud data; performing semantic alignment in voxel space on the normalized point cloud data to obtain semantically aligned target point cloud data; A structured reconstruction module, used for performing globally consistent structured reconstruction on the original point cloud data to obtain target normal vectors of the polyhedrons in the building complex; The monomer reconstruction module is used to perform monomer reconstruction based on the target point cloud data and the target normal vector to obtain a target polyhedral mesh of each building monomer in the building complex.

6. The device according to claim 5, characterized in that The structured reconstruction module is also used to perform facet estimation and convex hull polyhedron generation on the original point cloud data to construct the polyhedron space of the building complex; the polyhedron space includes polyhedrons, and the polyhedrons include facets; the original point cloud data is meshed and normal vector assisted to obtain the initial normal vector of the facet; Structural reconstruction is performed based on the polyhedron space and the initial normal vector to obtain a target normal vector of the polyhedron.

7. The device according to claim 5, characterized in that The monomer reconstruction module is also used to extract building calibration points from the target point cloud data; perform nearest neighbor clustering on the building calibration points to obtain different initial monomer point clouds; Based on the target normal vector, the initial monomerized point cloud is de-adhesion processed to obtain a target polyhedral mesh of each building monomer in the building complex.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.