CN117971837A - Method for constructing spatial index of live-action three-dimensional model - Google Patents

Abstract

The invention belongs to the technical field of digital twinning, and particularly relates to a method for constructing a spatial index of a live-action three-dimensional model, which comprises the following steps: step 1: resolving tile data of the live-action three-dimensional model, and resolving an LOD hierarchical structure; step 2: establishing a unified LOD hierarchical structure, establishing a unified root tile and LOD parameters, wherein the number of layers is a multiple of 3, and finishing the reformation of the LOD hierarchical structure; step 3: calculating a space bounding box of the live-action three-dimensional model, dividing the whole space into a uniform quadtree structure from top to bottom, numbering grids, filling tile data of each layer into a new grid, and generating a new tile file. According to the invention, the index structure is optimized, the index tree of the large model is split into the local small indexes with basically consistent sizes, so that the method is convenient and quick to load, is favorable for optimizing memory allocation, can reduce the pressure on a network when index files are transmitted, and does not need to inquire from the root node of the whole model, thereby greatly improving the random reading performance of the model.

Description

Method for constructing spatial index of live-action three-dimensional model

Technical Field

The invention belongs to the technical field of digital twinning, and particularly relates to a method for constructing a spatial index of a live-action three-dimensional model.

Background

In recent years, the digital twin technology has been widely applied to various fields such as travel, education, real estate, emergency rescue, smart city, security and the like, and the technology is mostly based on a B/S architecture, so that a three-dimensional model in the real world and various big data information are organically integrated together, and rich immersive experience is brought to users;

In order to meet the requirements of immersive experience and ensure the efficiency of content production, a realistic large-scale scene three-dimensional real model with a fine structure needs to be provided, and large-scale three-dimensional real modeling based on oblique photogrammetry is one of the technical means commonly used at present;

In order to solve the problem, the common method in the industry is to divide model data according to a three-dimensional space, wherein each unit space is called a tile, and the space corresponding to one tile can be subdivided again, so as to construct a hierarchical tree structure; the leaf nodes of the tree correspond to the most primitive three-dimensional model, and the three-dimensional model structure contained in the intermediate nodes of the tree is a simplification of the model of the next layer, so that a Level-of-detail (LOD) structure is obtained. When a user accesses, the application can firstly load the coarsest three-dimensional model, and only a part of the region of interest of the user is loaded with finer data, so that the LOD structure can greatly lighten the pressure on network transmission and model rendering.

However, as the fineness of the model increases, the hierarchy of the spatial index structure also becomes deeper, and this index method also has a performance problem:

1. the existing indexing technology, for example 3DTiles, also performs certain model optimization on the basis of osgb files, and separates the index data from the model data, but does not optimize the index structure itself, the system needs to initially load and maintain a huge index structure in a memory, and when processing a large-scale live-action three-dimensional model, great stress is caused on the network for transmitting the index files themselves.

2. Because of the tree structure, the nodes on different branches need to be traversed downwards from the root node, and when a user switches and enlarges and reduces the model frequently, the application needs to find and load proper model data from the root node again, so that serious performance problems are caused.

3. The tree structure (generally osgb format) output by the current oblique photogrammetry technology is uneven in space division, some space divisions are too dense, some space divisions are too sparse, and the inconsistency causes the inability to realize a uniform data structure, which is very unfavorable for optimizing model loading.

Disclosure of Invention

The invention aims to provide a method for constructing a spatial index of a live-action three-dimensional model, which is characterized in that an index structure is optimized, an index tree of a large model is split into local small indexes with basically consistent sizes, so that the method is convenient and quick to load, is favorable for optimizing memory allocation, can reduce the pressure on a network when an index file is transmitted, and does not need to inquire from a root node of the whole model, thereby greatly improving the random reading performance of the model.

The technical scheme adopted by the invention is as follows:

A method for constructing a spatial index of a live-action three-dimensional model comprises the following steps:

step 1: resolving tile data of the live-action three-dimensional model, and resolving an LOD hierarchical structure;

Step 2: establishing a unified LOD hierarchical structure, establishing a unified root tile and LOD parameters, wherein the number of layers is a multiple of 3, and finishing the reformation of the LOD hierarchical structure;

Step 3: calculating a space bounding box of the live-action three-dimensional model, dividing the whole space into a uniform quadtree structure from top to bottom, numbering grids, filling tile data of each layer into a new grid, generating a new tile file, and finishing tile reforming;

Step 4: firstly, building a root index tree in the first three layers of an LOD hierarchical structure, wherein each grid is a node of the tree, the number of grids of each layer is 1, 4 and 16 respectively, the 16 grids of the third layer of the root index tree are further divided to generate 64 grids, the 64 grids are taken as root nodes, three layers of secondary index trees are continuously built respectively, and the number of grids of each layer is 1, 4 and 16; and so on until the last layer of the LOD hierarchical structure, each index tree is stored into a JSON file;

each index tree contains three aspects of accessibility information, including node content accessibility, node accessibility, secondary index accessibility, accessibility data is saved using a binary array, and the coordinates of the spatial grid are encoded into a morton code as a position number in the binary array using a Z-space filling curve.

Further, the step 1 includes the following steps:

Step 101: using an OSG open source library to read all osgb files in a root directory and subdirectories of the large-scale live-action three-dimensional model;

Step 102: and analyzing the LOD hierarchical structure according to all osgb files of the large-scale live-action three-dimensional model.

Further, the osgb file internally includes LOD detail level metadata, bounding box information, three-dimensional model vertices, and texture information.

Further, the LOD data includes a mode of calculation of whether to load a next level of detail osgb file and an associated next level osgb file.

Further, the step 2 includes the steps of:

Step 201: combining vertex data contained in each subdirectory root tile, then upwards simplifying and constructing a new LOD layer, wherein the uppermost layer is used as a root node of the whole LOD hierarchical structure and is stored into a osgb file which is used as a starting point of initial loading of a model;

step 202: the geometric error of the LOD of the 0 th layer is calculated and used as a threshold value for loading the LOD hierarchical structure of the next layer.

Further, the method for simplifying the LOD layer in step 201 may use any algorithm of vertex clustering, region merging, and geometric element deletion.

Further, the step 3 includes the following steps:

step 301: when the space range of the space bounding box of the live-action three-dimensional model is calculated, the altitude is not considered, only the longitude and latitude range of the model is considered, and finally the range of the live-action three-dimensional model is expressed as:

[long_min,lat_min,Width,Length]

Wherein, front long _min、lat_min represents the left lower corner coordinate of the bounding box, and Width and Length represent the size of the bounding box;

Step 302: dividing the space bounding box into a uniform quadtree structure from top to bottom, wherein the number of layers is N _new, the grids are numbered according to a coordinate system, the numbered formats are (l, x, y), l represents the layer serial number of LOD, x and y represent the numbers of the grids along the XY coordinate axes respectively, and the space range of the space bounding box can be directly calculated to be

Thus, for a grid of vertices (v _x,v_y,v_z) and numbers (l, x, y) contained by a layer of tiles, if:

Then vertex (v _x,v_y,v_z) belongs to the mesh;

Step 303: and (3) generating new tile files by using all vertexes meeting the formula condition in the step (2) together with other associated information.

Further, the layer sequence number of the LOD is counted from 0 and ranges from [0, L-1], x, y E [0,2 ^l -1].

The invention has the technical effects that:

(1) According to the method for constructing the spatial index of the live-action three-dimensional model, the index structure is optimized, the index tree of the large model is split into the local small indexes with basically consistent sizes, so that the method is convenient and quick to load, is favorable for optimizing memory allocation, can reduce the pressure on a network when index files are transmitted, does not need to inquire from the root node of the whole model, and therefore the random reading performance of the model is greatly improved.

Drawings

FIG. 1 is a partial block diagram of a typical oblique photography data output;

FIG. 2 is a graph of uniform spatial division of a quadtree according to the present invention;

FIG. 3 is a schematic diagram of an index structure of the present invention;

fig. 4 is a schematic representation of the content of the index tree of the present invention.

Detailed Description

The present invention will be specifically described with reference to examples below in order to make the objects and advantages of the present invention more apparent. It should be understood that the following text is intended to describe only one or more specific embodiments of the invention and does not limit the scope of the invention strictly as claimed.

As shown in fig. 1-4, a method for constructing a spatial index of a live-action three-dimensional model includes the following steps:

step 1: resolving tile data of the live-action three-dimensional model, and resolving an LOD hierarchical structure;

A large-scale live-action three-dimensional model reconstructed by using the oblique photography technology is saved into a plurality of osgb files, a part of typical oblique photography data is shown in fig. 1, a root directory and contents of two subdirectories are listed in the figure, the root directory comprises 12 subdirectories, the whole model is divided into 12 blocks, each osgb file in the subdirectories represents a Tile, all osgb files in each subdirectory form an independent LOD hierarchical structure, wherein a first osgb file named by a subdirectory name, such as tile_ +000_ +001_ +000.Osgb, represents a root node of the LOD hierarchical structure.

The osgb file internally contains LOD detail level metadata, bounding box (BoundingVolume) information, three-dimensional model vertex and texture information and the like; the LOD data includes a calculation mode (distance mode or pixel mode) of whether to load the next level of detail osgb file and an associated next level osgb file; in this way, a plurality osgb of files are interrelated to represent a large-scale live-action three-dimensional model.

As shown in fig. 1, step 1 is described in detail herein, specifically, step 1 includes the following steps:

Step 101: using an OSG open source library to read all osgb files in a root directory and subdirectories of the large-scale live-action three-dimensional model;

Step 102: and analyzing the LOD hierarchical structure according to all osgb files of the large-scale live-action three-dimensional model.

Step 2: a unified LOD hierarchical structure is established, unified root tiles and LOD parameters are established, the number of layers is a multiple of 3, and the LOD hierarchical structure is reformed.

Specifically, the number of layers of the LOD hierarchy after reforming is closest to the number of layers of the LOD hierarchy before reforming and is a multiple of 3, for example, the number of layers of the LOD hierarchy before reforming is 8, and then the number of layers of the LOD hierarchy after reforming is 9 and is just 3 times 3.

It can be seen from the data structure of fig. 1 that the data of oblique photography lacks a uniform LOD root tile, and the initial model loading needs to load at least the root osgb files in all subdirectories and is transmitted once, once the subdirectories are many, great performance pressure is likely to be caused on the system, and the objective of step2 is to build a uniform LOD hierarchical structure, which specifically includes the following steps:

Step 201: combining the vertex data contained in each subdirectory root tile, then simplifying and constructing a new LOD layer upwards, assuming that the LOD hierarchical structure before reforming has an L-layer structure, then the number of layers of the LOD hierarchical structure after reforming is L _new, then sequentially simplifying and constructing (L _new -L) layers upwards, adopting algorithms such as vertex clustering, region combining, geometric element deleting and the like by the simplifying method, and finally, saving the uppermost layer as a root node of the LOD hierarchical structure to be a osgb file which is equivalent to the initial loading starting point of the model, thereby greatly improving the performance.

Step 202: calculating the geometrical error (geometricError ₀) of the LOD of the 0 th layer, which can be directly obtained by calculation for the LOD of the l layer as a threshold value for loading the LOD hierarchical structure of the next layer:

step 3: calculating a space bounding box of the live-action three-dimensional model, dividing the whole space into a uniform quadtree structure from top to bottom, numbering grids, filling tile data of each layer into a new grid, generating new tile files, and finishing the reformation of tiles, wherein the new tile files are the numbers of the grids, for example, the file name of the grid corresponding to the tile files with the numbers of (l, x, y) is l_x_y.b3dm.

Specifically, the step 3 includes the following steps:

Step 301: considering that the live-action three-dimensional models of oblique photogrammetry are all ground-attached models, the altitude can be omitted, so that when the space scope of the space bounding box of the live-action three-dimensional model is calculated, only the longitude and latitude scope of the model is considered, and finally the scope of the live-action three-dimensional model can be expressed as:

[long_min,lat_min,Width,Length]

wherein, front long _min、lat_min represents the left lower corner coordinates of the bounding box, and Width, length represent the dimensions of the bounding box.

Step 302: according to the principle, the method divides the space bounding box into a uniform quadtree structure from top to bottom, the number of layers is N _new, as shown in FIG. 2, an example of 3-layer quadtree division is shown, grids are numbered according to a coordinate system, the numbered formats are (l, x, y), l represents the layer sequence number of LOD (counted from 0 and in the range of [0, L-1 ]), x and y represent the numbers of the grids along XY coordinate axes (wherein x, y E [0,2 ^l -1 ]), and the space range of the space bounding box can be directly calculated to be

Thus, for a grid of vertices (v _x,v_y,v_z) and numbers (l, x, y) contained by a layer of tiles, if:

then the vertex (v _x,v_y,v_z) belongs to the mesh (note that altitude need not be considered).

Step 303: all vertices meeting the formula conditions in the step 2 are generated into new tile files together with other associated information, such as textures, normals and the like through the formula in the step 2, and the method adopts a b3dm format and is also a mainstream format of the current tile model. To ensure the integrity of certain building models, the spatial extent of the grid may be suitably extended so that there is overlap at the edges. The last grid numbered (l, x, y) corresponds to the file name of the tilefile l_x_y.b3dm. B3dm are stored in tiles subdirectories under the final output directory.

Considering that there may be grids that do not contain any vertices, this step also creates a temporary look-up table to record which grids have corresponding tilefiles, if present, save 1, and if not present save 0.

Step 4: firstly, building a root index tree in the first three layers of an LOD hierarchical structure, wherein each grid is a node of the tree, the number of grids of each layer is 1, 4 and 16 respectively, the 16 grids of the third layer of the root index tree are further divided to generate 64 grids, the 64 grids are taken as root nodes, three layers of secondary index trees are continuously built respectively, and the number of grids of each layer is 1, 4 and 16; and so on until the last layer of the LOD hierarchical structure, each index tree is stored into a JSON file;

each index tree contains three aspects of accessibility information, including node content accessibility, node accessibility, secondary index accessibility, accessibility data is saved using a binary array, and the coordinates of the spatial grid are encoded into a morton code as a position number in the binary array using a Z-space filling curve.

Specifically, for a large-scale scene, the temporary lookup table established in the step 3 also becomes extremely huge, one-time loading is wasteful and inconvenient to find, and the method solves the problem by establishing a novel index system, and specifically comprises the following steps:

step 401: the first three layers of the LOD hierarchy build a root index tree, each grid corresponds to one node of the tree (because of the one-to-one correspondence, grids and nodes are no longer distinguished later herein), so the number of tree nodes per layer is 1, 4, 16, respectively, as shown in fig. 3 b. The root index tree is saved as a separate JSON file with file name 0_0_0.json.

Step 402: according to the principle of the quadtree, the 16 grids of the third layer of the root index tree are continuously divided into 64 grids, then the 64 grids are used as root nodes to continuously establish new three-layer index trees respectively, and the tree structure is the same as that of the 1) step, and the tree is called a secondary index tree. Each secondary index tree is stored as a separate JSON file, and assuming that the number of the corresponding grid of the root node of a certain secondary index tree is (l, x, y), the corresponding index file name is l_x_y.

Step 403: and so on until the last layer of LOD is reached, eventually an index structure as shown in fig. 3a can be built. All index files are stored in the subdirectory indics under the final output directory.

Step 404: and generating an entry index file tileset. Json of the whole model under the final output catalog, wherein the fields comprise geometricError ₀, a model bounding box and the total layer number of the LOD hierarchical structure.

The node content accessibility indicates whether model files can be accessed or not in space grids (total 21) corresponding to all nodes of the current index tree, and the model files comprise two fields: the total number of accessible nodes (count) represents the binary number set (data) accessible to each node (0 represents accessible, 1 represents inaccessible, and length 21) and is stored in a string format of 16-ary code. In addition, if the total number of accessible nodes is 0, data is null.

The node accessibility indicates whether all nodes of the current index tree need to be accessed, and if the accessibility of one node is 1, two possibilities exist:

The first is that the accessibility of the current node content is 1;

The second type is that the accessibility of the lower node of the current node is 1 or the lower node is the root node of the secondary index tree.

Node accessibility includes two fields, the total number of accessible nodes (count) and binary number set (data) representing accessibility of each node (0 represents existence, 1 represents absence and length is 21), and a character string format of 16-system coding is adopted; in addition, if the total number of accessible nodes is 0, data is null.

The secondary index accessibility represents whether the next 64 grids at the bottom layer of the index tree are the root nodes of the new secondary index tree, contains two fields, the total number of secondary indexes (count) represents the binary number group (data) of the secondary index accessibility (0 represents accessibility, 1 represents non-accessibility, and the length is 64), is saved in a string format coded by 16 system, and in addition, if the total number of secondary indexes is 0, the data is null.

FIG. 4 shows an example of an index tree for indicating node accessibility, node content accessibility, and secondary index accessibility. The upper half of fig. 4 is arranged in quadtree space and the lower half is arranged in one-dimensional binary array. Node accessibility and node content accessibility hold information for three levels of nodes of the index tree, thus having a length of 1+4+16=21, while secondary index presence holds information for the next level (level 3) of the index tree, thus having a length of 64.

The method uses a one-dimensional binary array to store the accessibility information, and in order to store two-dimensional grid information in the one-dimensional array, the method adopts a Z filling curve to arrange the sequence of grids (nodes), namely Morton codes, which are formed by staggered arrangement of binary formats of two-dimensional numbers of grids, for example, for grids with the numbers of (1, 3), the binary formats are (01, 11), so that the Morton codes are 1011, the values are 11, namely, the grids with the coordinates of (1, 3) are 11 in the index number of the array (the condition of multiple layers is not considered).

The top half of fig. 4 shows the order in which the Z-fill curve traverses in the two-dimensional grid and the values of the morton codes, node accessibility and node content accessibility since the information of three layers of nodes is preserved, the morton codes of each layer are end-to-end.

The spatial index constructed by the method has the following advantages:

1. the evenly divided space facilitates rapid localization to the desired LOD layer without requiring layer-by-layer access. And the desired tile and index files can be quickly located using the coordinates and hierarchy of the grid.

2. The index tree of the whole large model is split into local small indexes with basically consistent sizes, so that the quick loading is convenient, and the optimization of memory allocation is facilitated.

3. The three accessibility in the index tree can help the application to reject grids which do not need to be accessed, reduce unnecessary network requests, and improve the random reading performance of tiles, thereby greatly improving the speed of traversing and inquiring the whole model.

4. The accessibility of binary array preservation can reduce space occupation, and the adjacent grids of the Morton codes can be found to be adjacent in space from FIG. 4, which is beneficial to the application to quickly inquire the grid accessibility information of an area.

5. If the grid to be accessed is not the root node of the index tree, the number Morton code of the grid is utilized to reversely push out the coordinates of the root node of the index tree to which the grid belongs so as to find the index file, and the query does not need to be started from the root node of the whole model.

In summary, according to the technical scheme, the index structure is optimized, the index tree of the large model is split into the local small indexes with basically consistent sizes, so that the method is convenient and quick to load, is favorable for optimizing memory allocation, can reduce the pressure on a network when index files are transmitted, and does not need to inquire from the root node of the whole model, thereby greatly improving the random reading performance of the model.

The foregoing is merely a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention, which are intended to be comprehended within the scope of the present invention. Structures, devices and methods of operation not specifically described and illustrated herein, unless otherwise indicated and limited, are implemented according to conventional means in the art.

Claims (8)

1. A method for constructing a spatial index of a live-action three-dimensional model is characterized by comprising the following steps: the method comprises the following steps:

step 1: resolving tile data of the live-action three-dimensional model, and resolving an LOD hierarchical structure;

Step 2: establishing a unified LOD hierarchical structure, establishing a unified root tile and LOD parameters, wherein the number of layers is a multiple of 3, and finishing the reformation of the LOD hierarchical structure;

Step 3: calculating a space bounding box of the live-action three-dimensional model, dividing the whole space into a uniform quadtree structure from top to bottom, numbering grids, filling tile data of each layer into a new grid, generating a new tile file, and finishing tile reforming;

Step 4: firstly, building a root index tree in the first three layers of an LOD hierarchical structure, wherein each grid is a node of the tree, the number of grids of each layer is 1, 4 and 16 respectively, the 16 grids of the third layer of the root index tree are further divided to generate 64 grids, the 64 grids are taken as root nodes, three layers of secondary index trees are continuously built respectively, and the number of grids of each layer is 1, 4 and 16; and so on until the last layer of the LOD hierarchical structure, each index tree is stored into a JSON file;

each index tree contains three aspects of accessibility information, including node content accessibility, node accessibility, secondary index accessibility, accessibility data is saved using a binary array, and the coordinates of the spatial grid are encoded into a morton code as a position number in the binary array using a Z-space filling curve.

2. The method for constructing the spatial index of the three-dimensional model of the live-action according to claim 1, which is characterized in that: the step 1 comprises the following steps:

Step 101: using an OSG open source library to read all osgb files in a root directory and subdirectories of the large-scale live-action three-dimensional model;

Step 102: and analyzing the LOD hierarchical structure according to all osgb files of the large-scale live-action three-dimensional model.

3. The method for constructing the spatial index of the three-dimensional model of the live-action according to claim 2, which is characterized in that: the osgb file internally includes LOD detail level metadata, bounding box information, three-dimensional model vertices and texture information.

4. A method for constructing a spatial index of a three-dimensional model of a real scene according to claim 3, wherein: the LOD data includes a calculation mode of whether to load the next level of detail osgb file and the associated next level osgb file.

5. The method for constructing the spatial index of the three-dimensional model of the live-action according to claim 2, which is characterized in that: the step 2 comprises the following steps:

Step 201: combining vertex data contained in each subdirectory root tile, then upwards simplifying and constructing a new LOD layer, wherein the uppermost layer is used as a root node of the whole LOD hierarchical structure and is stored into a osgb file which is used as a starting point of initial loading of a model;

step 202: the geometric error of the LOD of the 0 th layer is calculated and used as a threshold value for loading the LOD hierarchical structure of the next layer.

6. The method for constructing the spatial index of the three-dimensional model of the live-action according to claim 5, wherein the method comprises the following steps: the method for simplifying the LOD layer in step 201 may use any algorithm of vertex clustering, region merging and geometric element deletion.

7. The method for constructing the spatial index of the three-dimensional model of the live-action according to claim 1, which is characterized in that: the step 3 comprises the following steps:

step 301: when the space range of the space bounding box of the live-action three-dimensional model is calculated, the altitude is not considered, only the longitude and latitude range of the model is considered, and finally the range of the live-action three-dimensional model is expressed as:

[long_min,lat_min，Width，Length]

Wherein, front long _min、lat_min represents the left lower corner coordinate of the bounding box, and Width and Length represent the size of the bounding box;

Step 302: dividing the space bounding box into a uniform quadtree structure from top to bottom, wherein the number of layers is N _new, the grids are numbered according to a coordinate system, the numbered formats are (l, x, y), l represents the layer serial number of LOD, x and y represent the numbers of the grids along the XY coordinate axes respectively, and the space range of the space bounding box can be directly calculated to be

Thus, for a grid of vertices (v _x,v_y,v_z) and numbers (l, x, y) contained by a layer of tiles, if:

Then vertex (v _x,v_y,v_z) belongs to the mesh;

Step 303: and (3) generating new tile files by using all vertexes meeting the formula condition in the step (2) together with other associated information.

8. The method for constructing the spatial index of the three-dimensional model of the live-action according to claim 7, wherein the method comprises the following steps: the LOD layer sequence number is counted from 0 and ranges from [0, L-1], x, y E [0,2 ^l -1].