Method for lightweighting regular building three-dimensional model data

Abstract

This invention discloses a method for lightweighting 3D model data of regular buildings, comprising the following steps: S1, reading data; S2, detecting planes that can be simplified; S3, simplifying the triangulation mesh; S4, detecting and adjusting the effect after simplification. The lightweighting method of this invention, while ensuring the structural integrity of the 3D model and without deleting pyramids, achieves a significant improvement in data compression ratio by planning and simplifying the triangulation mesh, utilizing the convergence and movement of vertices within the mesh. Furthermore, since this invention does not involve image reading and compression, the demand for GPU and image processing computing power is greatly reduced, effectively lowering costs.

Description

Technical Field

[0001] This invention relates to the field of lightweight data, specifically to a method for lightweighting data of a regular three-dimensional building model. Background Technology

[0002] With the deepening implementation of national strategies such as smart cities, digital twins, and disaster emergency management, high-precision 3D modeling technology has become a core support for the construction of new infrastructure. Buildings, as the main elements of urban space, directly impact the digital transformation of key scenarios such as urban planning and real estate registration through their modeling accuracy and efficiency. By using platforms such as drones equipped with multiple high-precision digital sensors, multi-angle synchronous data acquisition can be achieved. The generated irregular triangular meshes can not only completely record the high-resolution texture features of building roofs and facades, but also construct continuous surface models through multi-view image geometric correlation.

[0003] However, in actual production, the irregular triangular meshes are too complex, causing problems such as inability to load directly, long loading times, and lag when opening files. Therefore, optimizing the reconstructed triangular meshes to construct a complete surface model is a crucial step in image-based 3D reconstruction.

[0004] Currently, the main methods for lightweighting triangulation of regular buildings are: ① Based on directly deleting redundant pyramid levels and merging data root nodes, lightweighting is achieved through compression, texture removal, and deletion of redundant triangulations. ② Based on merging adjacent pyramid levels and extracting the corresponding number of layers upward to construct a new pyramid, lightweighting of the 3D model is achieved through data compression and algorithm simplification. Method ① involves simplification based on deleting or merging triangulations. While this effectively reduces the amount of data, changes to the pyramids can easily cause confusion in the 3D model structure, which inevitably amplifies the errors generated by later triangulation simplification. The lightweighting method proposed in Method ② achieves a compression ratio of 0.13–0.14 for the actual number of triangulations and a compression ratio of 0.2–0.24 for the total file size, indicating that its data lightweighting effect is not significant.

[0005] For example, Chinese invention patent CN114494641B discloses a method and apparatus for lightweighting a 3D model, relating to the field of image processing technology. The method includes: reconstructing a 3D model based on tile data to obtain a topologically continuous triangular grid model; segmenting surfaces based on the planar features of the triangular grid model to obtain flat, block-shaped surface regions after segmentation of the overall 3D model; calculating the planar difference between adjacent segmented surfaces based on the flat, block-shaped surface regions to obtain merged blocks; selecting geometric feature points from the boundary line nodes of the merged blocks, and using the spatial surface formed by these geometric feature points as the lightweight segmented surface of the 3D model; and generating a new texture using spatial orthophoto projection based on the spatial area covered by the lightweight segmented surface to obtain a lightweight 3D model. This invention can improve the simplification efficiency and quality of 3D models, achieving optimized storage and efficient loading of 3D models. While this technical solution has a wide range of applications, it also consumes a significant amount of resources. Furthermore, this technical solution does not address whether the lightweighted model will result in discrepancies with the original graphic. According to research, because it allows for incomplete planar merging, it is very easy for discrepancies in the graphics to occur. Summary of the Invention

[0006] One technical problem solved by this invention is that the existing technology does not effectively reduce the weight of 3D model data of regular buildings, and the weight reduction process can easily lead to confusion. This invention provides a method for reducing the weight of 3D model data of regular buildings.

[0007] The technical solution adopted in this invention is: a method for lightweighting 3D model data of regular buildings, comprising the following steps: S1, reading data, parsing the point, face, and texture coordinates of the 3D model, and constructing a 3D topological mesh triangulation; S2, detecting planes that can be simplified, calculating the normal vectors of each triangular face formed by the triangulation; clustering according to the principle of similarity, based on the rule that the directions of the normals of topologically continuous triangular mesh faces are consistent; after clustering, merging small isolated faces with slight differences in normal directions into the face sets of adjacent clusters to form the final face set; calculating the normal of each face set, and, based on the characteristics of building attributes, removing face sets whose z-values ​​on the normals are between 0.2 and 0.8 and less than -0.2; S3, simplifying the triangulation, locating the vertices inside each face set. Calculate the length of the internal edges formed by the vertices based on the located vertices; sort the edges according to their lengths and move the two vertices of each edge to the edge's endpoint; move the vertices of all edges sequentially, and merge the points in the same position after all moves are completed; S4, check and adjust the simplification effect. For each patch set, calculate the projection plane of the pre-simplification triangular mesh according to the average normal before simplification, mesh the projection plane, and calculate the depth value of each point, storing the depth value in the corresponding mesh point; similarly, calculate the projection plane and its depth value of the post-simplification patch set; compare the two sets of depth values ​​sequentially, and the part with the smaller difference in depth value passes the test; for the part with the larger difference in depth value, restore the original shape of the triangular mesh; after overall depth value comparison and restoration, the final simplified triangular mesh is obtained.

[0008] As a further improvement of the present invention, in step S2, small isolated patches with slight differences in the normal direction are merged into the patch set of adjacent clusters to form the final patch set. By setting a threshold for judging slight differences, the output can be controlled, thereby reducing the amount of computation and the computation time, and meeting practical needs.

[0009] As a further improvement of the present invention, in step S3, the vertices inside the patch set are the vertices used by the patch set; its internal edges do not include the triangulation boundary and the patch set boundary.

[0010] As a further improvement of the present invention, in step S3, the two endpoints of the moved edge are marked; in subsequent moves, the already moved vertices are not moved.

[0011] As a further improvement of the present invention, the difference in depth values ​​shall meet the following conditions:

[0012] δ = |z_original - z_simplified| (1)

[0013] δ_max = α×L + β (2)

[0014] Where δ: the absolute value of the depth difference before and after simplification (unit: centimeters);

[0015] z_original: Original model depth value;

[0016] z_simplified: The simplified model depth value;

[0017] δ_max: Maximum permissible depth difference (in centimeters)

[0018] L: Building characteristic dimensions (unit: meters), taking the maximum value of the following three: building height (H), longest facade length (L_facade), and key component dimensions (such as column spacing, window width, etc.); to set the strictest tolerance standards for the most visually or functionally important dimensions of the current building. Taking the maximum value of the three for L is firstly in line with the laws of human visual perception, that is, to make different adaptive tolerances for objects of different sizes, avoiding the visual "weakest link effect"; secondly, it is necessary in engineering practice, that is, if only the height is used, the allowable error will be overestimated for flat buildings (such as airports), and if only the building height is considered in industrial plants, while the accuracy of crane beams (key component dimensions) is ignored, the risk of equipment collision may increase.

[0019] α: Proportion factor (recommended value for buildings: 0.08 to 0.15, specific value depends on the type of building);

[0020] β: Basic tolerance (Recommended value for buildings: 2-5cm, specific value depends on the type of building).

[0021] As a further improvement to this invention, S5 is also included: data export, which exports the simplified triangulated mesh according to the OBJ format and the point-line-face topology. Exporting the topological relationships of points, lines, and faces aims to: 1. Ensure geometric integrity, avoiding problems such as cracks, isolated points, or non-manifold edges, ensuring the mesh is closed and usable. 2. Accelerate neighborhood lookups; topological relationships help us quickly and accurately find the target geometry. 3. Support simplified correspondences, recording the mapping history of vertices / faces, facilitating the calculation of simplification errors or implementing progressive LOD (Level of Detail) switching. 4. Ensure compatibility and scalability; the standard OBJ format has limited functionality, and adding topological information (such as adjacency lists, edge lists) or using formats like PLY / OFF can support more complex applications (such as CAE analysis, parametric analysis).

[0022] The beneficial effects of this invention are as follows: The lightweight method of this invention, while maintaining the integrity of the 3D model structure and without deleting pyramids, achieves a significant improvement in data compression ratio by planning and simplifying triangular meshes, utilizing the convergence and movement of vertices within the mesh. Furthermore, since this invention does not involve image reading and compression, the demand for GPU and image processing computing power is greatly reduced, effectively lowering costs. Attached Figure Description

[0023] Figure 1 This is a comparison diagram of the density of the triangular network before and after processing the top area of ​​a regular building using the method of the present invention. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] In this example, a lightweight data processing method is applied to a 3D model of a building complex in a certain district of a city in Anhui Province.

[0026] 1. Data Reading: The 3D model of the building complex is collected by drone. The irregular triangular mesh data obtained by the drone is read. The data is in raw .obj format. The point, face and texture coordinates of the 3D model are parsed line by line. Based on the face-point relationship, a 3D topological mesh triangular network is constructed.

[0027] 2. Simplified Plane Detection: First, the normal vectors of each triangular face are calculated. Following the principle of similarity, the topologically continuous triangular mesh faces are clustered based on the consistency of their normal directions. After clustering, a detection algorithm merges small, isolated faces with slight differences in normal direction into adjacent clusters, forming the final face set (isolated faces are processed together to prevent them from affecting the final result later; individual isolated faces do not affect overall accuracy). The normal vector of each face set is calculated. Based on the characteristics of the building, face sets with z-values ​​between 0.2 and 0.8 and less than -0.2 on the normal vector are removed (the normal vectors on the top surface of a building point upwards, while the normal vectors on the sides are almost horizontal; based on these two factors, non-building face sets are also removed).

[0028] 3. Simplify the triangulation: For each set of faces, find the vertices inside the set (vertices used only by that set); based on the found vertices, find the internal edges formed by vertices inside the set (ignoring cases where the edge is on the boundary of the triangulation or the boundary of the set), and calculate the length of the edge. Sort the edges according to their lengths, move the two endpoints of each edge to the endpoint position, and mark the edges involving the two points on that edge (in subsequent moves, vertices that have already been moved are not moved). Perform the move operation on all edges in turn; after all edges have been moved, merge the points in the same position (using the method of reconstructing the triangulation, first removing the original triangulation, and then reconstructing this part of the triangulation based on the relationship between the vertices of the source triangulation and the vertices in the same position).

[0029] 4. Detection Results: Check whether the simplified model maintains a similar shape to the original model. If the offset is large, restore the parts with the largest differences. Specific detection method: For each patch set, calculate the projection surface of the original triangular mesh using the average normal of the original mesh, and calculate the depth value of each point within the surface (here, the projection surface is meshed in a 1024*1024 format, and the corresponding depth value is stored within the grid). Similarly, calculate the projection surface and its depth value for the simplified patch set. Compare the two sets of depth values ​​sequentially by grid point. For parts with smaller depth value differences, the test passes; for parts with larger depth value differences, restore the original triangular mesh shape. After overall depth value comparison and restoration, the final simplified triangular mesh is obtained. To further understand the deformation of the triangular mesh before and after simplification...

[0030] Based on a comparison of the depth values ​​of the triangles before and after simplification, five evaluation levels were established through data analysis (Table 1). It can be found that the depth values ​​before and after simplification are generally within ±5 cm (the standard can be relaxed according to actual needs, but the threshold should not be relaxed too much), which can basically preserve all structural details, sharp edges without blunting, and smooth curved surfaces without step-like features.

[0031] Table 1. Evaluation of Depth Values ​​Before and After Simplification of the 3D Model

[0032]

[0033]

[0034] Note: Due to the randomness of the data statistics, the difference in depth values ​​before and after simplification is not continuous within the range of -25.8 to 25.4, and there are some gaps.

[0035] Based on the statistical analysis of the above data, the inventor creatively summarized the following patterns:

[0036] δ = |z_original - z_simplified| (1)

[0037] δ_max = α×L + β (2)

[0038] δ: Absolute value of the depth difference before and after simplification (unit: m)

[0039] z_original: Original model depth value

[0040] z_simplified: The simplified model depth value

[0041] δ_max: The maximum allowable depth difference (unit: centimeters, in accordance with construction industry practice).

[0042] L: Characteristic dimension of the building (unit: meters), taking the maximum value of the following three:

[0043] Building height (H)

[0044] Longest facade length (L_facade)

[0045] Dimensions of key components (such as column spacing, window width, etc.)

[0046] α: Proportionality coefficient (Recommended value for buildings: 0.08–0.15, specific value depends on building type)

[0047] β: Basic tolerance (Recommended value for buildings: 2-5cm, specific value depends on the type of building).

[0048] Detailed explanations of each parameter are shown in Tables 2 and 3 below.

[0049] Table 2 Parameter Details

[0050]

[0051]

[0052] Table 3 Comparison of Different Building Precision Levels

[0053]

[0054] For different building types, different scaling factors (α) and basic tolerances (β) are selected. The maximum allowable depth difference (δ_max) is established by combining the building's characteristic dimensions (L) and the required accuracy (evaluation level). If the depth difference before and after simplification is within the maximum allowable depth difference range, the simplification detection is passed; otherwise, the original triangle is retained.

[0055] The depth difference before and after the simplification has the following characteristics:

[0056] (1) Dual tolerance mechanism

[0057] α×L addresses the scale effect (a 200m high-rise building allows for a larger absolute error than a 20m villa);

[0058] +β ensures basic accuracy (even small components have minimum requirements);

[0059] Specifically, it is recommended to use α ≤ 0.05 for historical buildings and α = 0.1 for new buildings to balance efficiency and accuracy.

[0060] (2) Conforms to the practices of the construction industry

[0061] Depth units are in centimeters (non-industrial millimeters).

[0062] Feature dimensions are taken from the dominant visual dimensions (human eyes primarily perceive height and facade).

[0063] (3) Corresponding experience evaluation criteria

[0064] When L = 50m and is directly opposite a historical building:

[0065] "Very good" rating: δ_max≤0.05×50+2=4.5cm, which is close to the 5.8cm corresponding to the "Very good" effect in Table 1.

[0066] "Good" rating: δ_max≤0.08×50+3=7cm, which is close to the 7.9cm corresponding to the "Good" effect in Table 1.

[0067] 5. Data Export: Export the simplified triangular mesh according to the obj format and its own point, line and surface topology.

[0068] Table 4 Comparison of data volume before and after lightweighting treatment

[0069]

[0070] Table 4 and Figure 1 To compare the data processing volume in Example 1, the density of the triangular mesh before and after processing the top area of ​​a regular building is compared. Figure 1 It can be observed that the density of the triangular mesh after weight reduction (green box) is significantly smaller than that before weight reduction (red box).

[0071] From an overall regional perspective, this invention significantly improves the compression ratio after lightweighting without resorting to traditional methods such as pyramid reduction, while maintaining the structural integrity of the 3D model (see Table 1 for comparison). Specifically, the compression ratio of the number of triangular meshes is nearly doubled compared to previous studies (from 0.137 in the prior art to 0.24). This is of great significance for enhancing the visualization effect of 3D building models, improving the processing efficiency of 3D building models, and reducing data storage and transmission costs.

[0072] Those skilled in the art should understand that the protection scheme of the present invention is not limited to the above embodiments, and various arrangements, combinations and transformations can be made on the basis of the above embodiments. Without departing from the spirit of the present invention, all transformations made to the present invention fall within the protection scope of the present invention.

Claims

1. A method for lightweighting 3D model data of regular buildings, characterized by: Includes the following steps: S1: Read data, parse the point, surface and texture coordinates of the 3D model, and construct a 3D topological mesh triangulation. S2, detects a simplified plane and calculates the normal vectors of each triangular face formed by the triangular mesh; Based on the principle of similarity, clustering is performed according to the rule that the normal directions of topologically continuous triangular mesh patches are consistent. After clustering, small isolated patches with slight differences in normal directions are merged into the patch set of adjacent clusters to form the final patch set. The output is controlled by setting a threshold for judging slight differences. Calculate the normal of each patch set, and based on the characteristics of the building properties, remove patch sets whose z-values ​​on the normal are between 0.2 and 0.8 or less than -0.2; S3, a simplified triangulation, locates the vertices inside each face set, calculates the length of the internal edge formed by the vertices based on the located vertices, sorts the edges according to their lengths, moves the two endpoints of the edges to the end positions of the edges, moves the vertices of all edges in turn, and merges the points in the same position after all the moves are completed. S4. Detect and adjust the simplification effect. For each patch set, calculate the projection surface of the pre-simplification triangular mesh according to the average normal before simplification. Mesh the projection surface and calculate the depth value of each point. The depth value is stored in the corresponding mesh point. Similarly, calculate the projection surface and its depth value of the simplified patch set. The two sets of depth values ​​before and after simplification are compared sequentially. For different building types, different scaling factors α and basic tolerance β are selected. Combined with the building feature size L and the required accuracy, the maximum allowable depth difference δ_max is established. If the depth difference before and after simplification is within the maximum allowable depth difference range, simplification is passed; otherwise, the original triangle is retained. After overall depth value comparison and recovery, the final simplified triangulation is obtained. The difference in the depth values ​​should meet the following conditions: δ = |z_original - z_simplified| (1) δ_max = α×L + β (2) Where δ: the absolute value of the depth difference before and after simplification, in centimeters; z_original: Original model depth value; z_simplified: The simplified model depth value; δ_max: The maximum allowable depth difference, in centimeters; L: Characteristic dimension of the building, in meters; Take the maximum value of the following three: building height H, longest facade length L_facade, and key component dimensions; α: Proportionality coefficient, the specific value of which depends on the type of building; β: Basic tolerance, the specific value depends on the type of building.

2. The method for lightweighting regular building 3D model data according to claim 1, characterized in that: In step S3, the vertices inside the patch set are the vertices used by the patch set; Its internal edges do not include the boundaries of the triangular mesh and the boundaries of the patch set.

3. The method for lightweighting regular building 3D model data according to claim 1, characterized in that: In step S3, the two endpoints of the moved edge are marked; in subsequent moves, the already moved vertices are not moved.

4. The method for lightweighting regular building 3D model data according to any one of claims 1 to 3, characterized in that: It also includes S5, data export, which exports the simplified triangulation network in obj format and with point, line and surface topology.

Images