A three-dimensional scene tile model optimization method, device and medium

Abstract

The application discloses a three-dimensional scene tile model optimization method, equipment and medium, the method comprises the following steps: acquiring a three-dimensional scene and constructing the global triangle net of the three-dimensional scene; the node division is carried out to the triangle net and generates octree index and the hash file of the corresponding sub-scene of triangle net, and the octree index includes parent node and the corresponding child node of the parent node; according to the characteristics and attributes of LOD model, the LOD model is grouped, and at least one group of LOD model is obtained; the node alignment operation of the same level is carried out to each group of LOD model; the main model and the secondary model in each group of LOD model are determined, and the model is combined according to the depth difference between the main model and the secondary model; according to all hash files of the three-dimensional scene and the model after combination, the model file of 3D Tiles format of the three-dimensional scene and the metadata of tile set are generated. It is optimized through tile, greatly reduces the problem of large number of tile models under complex scene, effectively improves tile analysis speed and rendering efficiency.

Description

Technical Field

[0001] This invention belongs to the field of 3D scene tile model optimization technology, specifically relating to a 3D scene tile model optimization method, device and medium. Background Technology

[0002] With the development of technologies such as oblique photogrammetry, geographic information systems, and virtual reality, applications based on 3D scenes have played a significant role in fields such as metaverse, digital twins, smart cities, industrial management, gaming, and tourism. To improve the convenience and user experience of these applications, various fields tend to make them lightweight and cross-platform, and web applications are an excellent carrier for this. Meanwhile, in 3D visualization technology, the development of Web Graphics Library (WebGL) technology has enabled browsers to render 3D models with increasingly mature functions and excellent effects. However, as the scale of 3D scenes continues to expand, and the accuracy and complexity of scene models continue to increase, web applications, when rendering large-scale 3D scenes, are often limited by various computer hardware devices, such as memory, CPU, and GPU, resulting in less than smooth rendering, with stuttering and frame drops. This seriously affects the user experience and work efficiency of applications that require loading large 3D model data. Two important processes are involved in 3D visualization: model data parsing and scheduling, and model rendering.

[0003] The parsing and scheduling of model data here refers to the application parsing the component content data inherent in the 3D model to obtain relevant information, such as tile type, tile position, and tile layer. At the same time, the components are scheduled accordingly based on factors such as camera viewpoint and zoom level to select appropriate components for display.

[0004] 3D tiles are currently the mainstream specification for visualizing and analyzing 3D data on the web for geographic information systems. They are a 3D model data format developed for streaming and rendering of 3D geospatial data, such as photogrammetry, 3D architecture, BIM / CAD, instantiated features, and point clouds. The 3D tile construction process can be divided into two steps: scene partitioning and tile transformation. Each tile is essentially a triangular mesh constructed from the points within its area and the vertices of the tile. Tiles segment 3D data, allowing for further subdivision. Each tile serves as the basic unit for transmission and rendering, representing a portion of a large-scale 3D scene. However, complex scenes have a vast number of tile models, and the logical relationships between these models are more complex. Tile parsing is time-consuming, and organization is more prone to errors. Different partitioning and organization methods affect the complexity of the model file. Rendering requires parsing and scheduling tile data, thus impacting rendering efficiency.

[0005] Therefore, how to optimize the model to improve rendering efficiency is a topic worth studying. Summary of the Invention

[0006] To address the issue of long tile parsing times caused by the large number of tile models in complex scenes, this invention provides a method, device, and medium for optimizing 3D scene tile models. By optimizing the tiles, it significantly reduces the number of tile models in complex scenes, effectively improving tile parsing speed and rendering efficiency.

[0007] The objective of this invention is achieved through the following technical solution: The first aspect of this invention provides a method for optimizing a three-dimensional scene tile model, comprising the following steps: acquiring a three-dimensional scene and constructing a global triangular mesh of the three-dimensional scene; The triangular mesh is divided into nodes to generate an octree index and a hash file of the corresponding sub-scenes of the triangular mesh. The octree index includes a parent node and child nodes corresponding to the parent node. Each parent node contains a set of LOD models of tree space index. The LOD models are grouped according to their features and attributes to obtain at least one group of LOD models; Perform the same level of node alignment operation on each group of LOD models; Identify the primary and secondary models in each group of LOD models, and merge the models based on the depth difference between the primary and secondary models; Generate a 3D Tiles format model file and tile set metadata for the 3D scene based on all hash files and the merged model of the 3D scene.

[0008] A second aspect of the present invention provides a three-dimensional scene tile model optimization device, comprising a memory and a controller connected in sequence, wherein the memory stores a computer program, and the controller is used to read the computer program and execute a three-dimensional scene tile model optimization method as described in the first aspect.

[0009] A third aspect of the present invention provides a computer-readable storage medium storing instructions, wherein when the instructions are executed on a computer, the controller is configured to read the computer program and execute the three-dimensional scene tile model optimization method described in the first aspect.

[0010] Compared with the prior art, the present invention has at least the following advantages and beneficial effects: This invention optimizes 3D tile data by combining LOD trees and octree indexes, greatly reducing the problem of a large number of tile models in complex scenes and effectively improving tile parsing speed. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 is a flowchart of the method of the present invention; Figure 2 is a schematic diagram of three LOD models; Figure 3 is a schematic diagram of the LOD model in Figure 2 after node alignment; Figure 4 is a schematic diagram of the fusion of the two LOD models; Figure 5 shows the result of fusing the LOD model shown in Figure 3. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0014] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0015] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.

[0016] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0017] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use, or the orientation or positional relationship commonly understood by those skilled in the art. They are only used for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. In addition, the terms "first," "second," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0018] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0019] Large-scale 3D scenes often contain many independent 3D models, which is not conducive to indexing the models during rendering. A logically clear and structurally sound tile organization can greatly reduce data redundancy and improve data transmission speed. Therefore, the first aspect of this invention discloses a method for optimizing 3D scene tile models. This method can be, but is not limited to, executed by a 3D scene tile model optimization device. The 3D scene tile model optimization device can be software, or a combination of software and hardware, and can be integrated into smart devices such as mobile terminals, tablets, and computers. Specifically, the 3D scene tile model optimization method includes the following steps S01 to S06.

[0020] Step S01: Obtain the 3D scene and construct the global triangular mesh of the 3D scene.

[0021] In this step, the 3D scene is read and parsed, and a global triangulation is constructed based on its plane in the 2D scene.

[0022] Step S02: Divide the triangular mesh into nodes to generate an octree index and a hash file of the corresponding sub-scenes of the triangular mesh. The octree index includes a parent node and child nodes corresponding to the parent node. Each parent node contains a set of LOD models of tree space index.

[0023] This step is the process of generating the LOD model.

[0024] Octree indexing is a method for organizing and arranging spatial objects. It can divide three-dimensional space and allocate spatial elements based on the density of spatial features. The introduction of this indexing method makes the processing, optimization, and rendering of spatial objects more convenient, and can utilize memory more effectively, thereby improving computational performance.

[0025] By combining the LOD model with the octree spatial index, the 3D scene tile model is efficiently optimized. Specifically, this step includes steps S021 to S023.

[0026] Step S021: Traverse each triangulation in the triangulation network and determine whether the triangulation network meets the partitioning condition. The partitioning condition is that the node where the triangulation network is located has reached the maximum recursion depth or the number of triangles in the node where the triangulation network is located is greater than a first threshold.

[0027] Step S022: If the triangulation does not meet the partitioning conditions, that is, the node where the triangulation is located has not reached the maximum recursion depth or the number of triangles in the node where the triangulation is located is less than or equal to a first threshold, generate a hash file of the sub-scene corresponding to the current node.

[0028] Step S023: Conversely, if the triangular mesh meets the partitioning conditions, the current node is partitioned using the octree indexing method to obtain the octree index. First, the bounding box of the triangular mesh is built according to the maximum size of the scene, and this bounding box is used as the parent node; then, the child nodes of the parent node are constructed according to the parent node, that is, it is divided into eight non-overlapping child nodes. The spatial data contained in the eight child nodes are added together and are consistent with the spatial data contained in the parent node. The index of the triangular mesh is recorded, and the index records the relationship between the parent node and the corresponding child node. The number of triangles in each child node is determined according to the bounding box of the child node, and the index of all triangles is stored. Each parent node contains a set of LOD models of tree spatial indexes, which are recorded as an LOD tree.

[0029] Step S024: Recursively traverse all child nodes of the parent node, determine whether the child node meets the partitioning condition, repeat steps S022 and S023 until all nodes no longer meet the partitioning condition, and finally generate a hash file of the global scene.

[0030] Step S03: Group the LOD models according to their features and attributes to obtain at least one group of LOD models.

[0031] Models meeting the prerequisites for LOD model fusion are grouped according to their features and attributes. Models with minimal differences in features and attributes are grouped together. Current rendering and scheduling often employ Levels of Detail (LOD) technology to construct tile pyramids, reducing CPU computational load on the same scene. The LOD model, based on LOD, assumes that when an object covers a small area of ​​the screen, a coarser model can be used to describe that object, providing a geometric hierarchy model for visibility judgment algorithms to facilitate rapid rendering of complex scenes. Within the same image, objects that are closer, move slower, and are closer to the center of the viewer's field of vision are displayed with more detail; that is, the higher the LOD level, the greater the detail. Generally, the more LOD levels, the more detail can be displayed.

[0032] This step groups LOD models that meet the prerequisites for LOD model fusion. These prerequisites mean that the feature layers must have a close spatial distribution and the same or related feature types. Specifically, the spatial distribution between feature layers must be less than a second threshold. The choice of the second threshold depends on the specific application scenario. Specifically, 3D Tiles stipulate that holes are not allowed inside 3D tiles. If all geographic features in two layers are to be merged, the spatial range covered by the tiles cannot be less than the bounding volume formed by all geographic features. The greater the spatial distance between geographic features, the larger the spatial range covered by the tiles, and the higher the data volume and complexity of the tiles. Therefore, layers with large spatial distances should not be merged to prevent the creation of giant tiles. Furthermore, from a logical perspective, to facilitate understanding of the definitions and concepts of each model, models should be grouped according to their characteristics and attributes. Therefore, models with significant differences in type should not be merged to prevent misinterpretation of the models.

[0033] For example: Taking the Lujiazui scene in Shanghai as an example, Lujiazui is a large-scale 3D scene composed of ArcGIS Earth imagery, Cesium 3D topography, and three independent building clusters near the Huangpu River. The three independent building clusters belong to Building Information Modeling (BIM), have close spatial relationships and the same spatial index, which meet the necessary conditions for model fusion and can be merged. Therefore, they can be divided into the same group.

[0034] Step S04: Perform node alignment operation at the same level for each group of LOD models.

[0035] This step analyzes the LOD tree nodes of models that meet the necessary conditions for LOD model fusion, and achieves node alignment.

[0036] The necessary condition for model fusion is that feature layers must have the same 3D spatial index and their LOD levels must overlap. Specifically, 3D Tiles stipulate that a single 3D tile can only have one 3D spatial index; all geographic features within the tile will be divided by that spatial index, therefore feature layers with different spatial indices cannot be merged. Since the indices generated in step S02 are all octree spatial indices, they all have the same 3D spatial index. Furthermore, the essence of multi-feature layer fusion is merging the LOD trees corresponding to multiple layers. According to tree merging theory, the merged trees must have parallel child nodes, and then the merging process starts from the first parallel child node and proceeds layer by layer until a merged tree is formed. Therefore, multiple layers must have at least one identical or similar LOD level to ensure LOD level overlap.

[0037] Specifically, first, select LOD models in the same group whose base layers have the same 3D spatial index and whose LOD levels overlap to obtain the LOD models to be aligned; then, align the root nodes of the LOD models to be aligned; finally, adjust the position of the LOD models to align the nodes of the same level.

[0038] For example: As shown in Figure 2, A1 and C1 are independent models. Models A1 and C1 can be merged into B1 to simplify the overall model structure. Align the root nodes of the LOD trees of the same level, find the overlapping parts of the LOD level, and adjust the relative positions of the three trees according to the principle of the same or similar LOD level so that the nodes of the same level are aligned. The alignment result is shown in Figure 3.

[0039] Step S05: Determine the primary and secondary models in each group of LOD models, and merge the models based on the depth difference between the primary and secondary models.

[0040] The primary and secondary models are determined based on the root node level of the tree structure of the LOD models to be merged. The model with the highest root node level is designated as the primary model, and the others are designated as secondary models. Next, the depth difference between the subtree with the candidate leaf node as its vertex and the tree to be merged is calculated. The leaf node with the depth difference closest to 0 is selected. If two leaf nodes have the same absolute depth difference, the leaf node with the negative depth difference is selected. Finally, the root node of the secondary model's LOD tree is merged into the corresponding leaf node of the primary model's LOD tree according to the fusion rules.

[0041] The fusion rule is that conditions 1, 2, and 3 must be met simultaneously, where, Condition 1 states that if at least one location in a LOD tree has a leaf node, then the merged leaf node includes the values ​​of all LOD tree nodes at that location, where the value represents the number of geographic features contained in the node. Specifically, if tree B has no node and tree A has node 'a', the merged result will have node 'a'; if tree B has node 'b' and tree A has no node, the merged result will have node 'b'; if tree B has node 'b' and tree A has node 'a', the merged result will have node 'b+a'. Tree A and tree B represent the LOD trees of the two models to be merged.

[0042] Condition 2: If, after merging, the value of a parent node is not equal to the sum of the values ​​of all its secondary child nodes, then delete all child nodes of that parent node; Condition 3: All parent nodes of the top nodes of the subtrees participating in the merging of the main model LOD tree need to be increased by the value of the top node of the secondary model LOD tree.

[0043] For example, consider the LOD tree of the primary model A2 and the LOD tree of the secondary model B2 in Figure 4.

[0044] The root node of tree A2 is at a higher level than that of tree B2, so tree A2 is positioned as the main model. Furthermore, when the root node a11 of tree A2 is used as a merged leaf node, the depth difference between the leaves of tree A2 and tree B2 is closest to 0, i.e., 3-4=-1.

[0045] For the first layer of parallel nodes, there are only nodes a11 and b11. When merging, the value under node a11 (i.e., the number of geographic elements contained under node a11) and the value of node b11 need to be added together and then assigned to node a11.

[0046] For the second-level parallel nodes, since node b11 has no left leaf node, the value of node a21 remains unchanged during merging. Therefore, the value of node a22 and the value of node b21 need to be added together and then assigned to node a22.

[0047] For the third-level parallel nodes, since nodes a22 and b21 each have their own left and right nodes, when merging, the values ​​of nodes a31 and b31 need to be added together and then assigned to node a31, and the values ​​of nodes a32 and b32 need to be added together and then assigned to node a32.

[0048] For nodes b41 and b42, which only exist in the C-tree in the fourth-level parallel nodes, it is inappropriate to use them as the left and right nodes of node a31, respectively. The value of the new node a31 is not equal to the sum of the values ​​of nodes b41 and b42. According to the model fusion rules, this does not satisfy the requirement that the value of any parent node in a LOD tree equals the sum of the values ​​of its child nodes. Therefore, nodes b41 and b42 must be removed. Similarly, since the new node a11 adds the value of node b11, all parent nodes of node a11 must also add the value of node b11. The model LOD tree after fusion using the above method for the primary and secondary models shown in Figure 3 is shown in Figure 5.

[0049] Step S06: Generate a 3D Tiles format model file and tile set metadata of the 3D scene based on all hash files and the merged model of the 3D scene.

[0050] Specifically, by constructing 3D tile data, traversing all hash files of the 3D scene, reading the complete 3D scene, constructing a glTF model using the triangulation of the 3D scene, and saving information such as vertices, normals, their topological relationships, and textures, a 3D Tiles format model file and tile set metadata are generated. In this embodiment, a CMPT format file header is generated, a CMPT field is added to the magic attribute to specify the tile type, and the text length and binary data length are added to featureTable and batchTable respectively. The number of tiles merged in CMPT is added to tilesLength. The overall scene is named scene, and the metadata file of the entire tile set is named tilesset.json. The overall scene contains m local scenes, and the metadata file of each local scene is named tilessetX.json (X∈[1,M]), representing scene 1 to scene m; a local scene contains multiple 3D models, and the metadata file of each model is named tilessetX_Y.json (Y∈[1,N]), representing that the model belongs to the Yth model of local scene X. The model folder contains sub-model folders for all LOD levels of the model, and each sub-model folder stores the corresponding 3D tiles. Each model tile is named X_Y_Z (Z∈[1,L]), which means that the tile belongs to the Lth level tile of the Yth model of the local scene X. The tile suffix and tile type are consistent.

[0051] Ultimately, all 3D Tiles format model files and tile set metadata are deployed on a server on the Internet or local area network.

[0052] This approach combines LOD trees with octree indexes to merge multiple independent model tiles into a single model tile, reducing model complexity and thus model data, thereby lowering the time cost of model parsing.

[0053] A second aspect of this invention provides a 3D scene tile model optimization device, comprising a memory and a controller connected in sequence. The memory stores a computer program, and the controller reads the computer program and executes the 3D scene tile model optimization method described in the first aspect. Specifically, the memory may include, but is not limited to, random-access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; the controller may not be limited to using a microcontroller of the STM32F105 series. Furthermore, the computer device may also include, but is not limited to, a power supply unit, a display screen, and other necessary components.

[0054] A third aspect of the present invention provides a computer-readable storage medium storing instructions, wherein when the instructions are executed on a computer, the controller is configured to read the computer program and execute the three-dimensional scene tile model optimization method described in the first aspect.

[0055] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing a 3D scene tile model, characterized in that, Includes the following steps: Acquire the 3D scene and construct a global triangulation of the 3D scene; The triangular mesh is divided into nodes to generate an octree index and a hash file of the corresponding sub-scenes of the triangular mesh. The octree index includes a parent node and child nodes corresponding to the parent node. Each parent node contains a set of LOD models of tree space index. The LOD models are grouped according to their features and attributes to obtain at least one group of LOD models; Perform the same level of node alignment operation on each group of LOD models; Identify the primary and secondary models in each group of LOD models, and merge the models based on the depth difference between the primary and secondary models; Generate a 3DTiles format model file and tile set metadata of the 3D scene based on all hash files and the merged model of the 3D scene; The step of grouping the LOD models according to their features and attributes includes: Select feature layers whose spatial distribution is less than the second threshold and which have the same or related feature types to obtain at least one set of LOD models to be aligned; The step of performing the same-level node alignment operation on each group of LOD models includes: Select LOD models from the same group whose base layers have the same 3D spatial index and whose LOD levels overlap to obtain the LOD models to be aligned; then align the root nodes of the LOD models to be aligned. Adjust the position of the LOD model to align nodes of the same level; The model merging based on the depth difference between the primary and secondary models includes: Calculate the depth difference between the secondary model and the primary model; Select the leaf node whose depth difference is closest to 0 from the main model, and merge the root node of the LOD tree of the secondary model into the leaf node of the LOD tree of the main model according to the fusion rule; The fusion rule is that conditions 1, 2, and 3 must be satisfied simultaneously. in, Condition 1 is: if at least one location in a LOD tree has a leaf node, then the merged leaf node includes the values ​​of all LOD tree nodes at that location, where the value is the number of geographic features contained in the node; Condition 2: If, after merging, the value of a parent node is not equal to the sum of the values ​​of all its secondary child nodes, then delete all child nodes of that parent node; Condition 3: All parent nodes of the top nodes of the subtrees participating in the merging of the main model LOD tree need to be increased by the value of the top node of the secondary model LOD tree.

2. The method for optimizing a three-dimensional scene tile model according to claim 1, characterized in that, The step of partitioning the triangular network into nodes to generate an octree index includes: Traverse each triangulation in the triangulation network and determine whether the triangulation network meets the partitioning condition. The partitioning condition is that the node where the triangulation network is located has reached the maximum recursion depth or the number of triangles in the node where the triangulation network is located is greater than a first threshold. In response to the triangulation satisfying the partitioning condition, the current node is partitioned using the octree indexing method to obtain the octree index. The octree index records the parent node and eight child nodes corresponding to the parent node and not intersecting with each other. Each node is a LOD tree. Recursively traverse all child nodes of the parent node, and determine whether the child node satisfies the partitioning condition, until all nodes no longer satisfy the partitioning condition.

3. The method for optimizing a three-dimensional scene tile model according to claim 1, characterized in that: The determination of the primary and secondary models in each group of LOD models includes: In each LOD model group, the one with the highest root node level is determined as the primary model, and the rest are secondary models.

4. The method for optimizing a three-dimensional scene tile model according to claim 1, characterized in that, The step of selecting the leaf node whose depth difference is closest to 0 from the main model includes: If the absolute values ​​of the depth differences between two leaf nodes are the same, then the leaf node with the negative depth difference is selected.

5. A three-dimensional scene tile model optimization device, comprising a memory and a controller connected in sequence, wherein the memory stores a computer program, characterized in that: The controller is used to read the computer program and execute the three-dimensional scene tile model optimization method according to any one of claims 1-4.

6. A computer-readable storage medium storing instructions thereon, characterized in that: When the instructions are executed on a computer, the method for optimizing a three-dimensional scene tile model as described in any one of claims 1-4 is performed.

Images