Scene multi-level detail generation method in CIM platform
By using a Gaussian sphere-based method to simplify the scene surface mesh, the problem of balancing high-precision rendering and low load in digital city visualization is solved, achieving efficient multi-level detail generation while preserving important geometric features.
Patent Information
- Application Number
- CN202411121103.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-11-26
AI Technical Summary
Existing technologies struggle to reduce the computer rendering load while rendering buildings with high precision in digital city visualization, and are prone to losing important geometric features.
A Gaussian sphere-based method for simplifying scene surface meshes is adopted. This method generates a multi-level detail model by dividing the facet family into normals, merging approximately coplanar faces, searching for hyperface boundary information, and deleting details with low visual importance.
Finding a balance between computational efficiency and visual quality, effectively simplifying mesh models, and generating multi-level detail models suitable for various application needs.
Smart Images

Figure CN119027613A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to a method for generating multi-level details of a scene in a CIM platform. Background Art
[0002] With the emergence and widespread application of three-dimensional digital cities, digital city visualization has become a research hotspot. Digital cities are virtual simulations of real cities based on computer technology. The scene range corresponding to digital cities is large, which contains many buildings. For digital city visualization, on the one hand, we want to render as many high-precision buildings as possible, and on the other hand, we want to reduce the computer rendering load. In order to find a balance between high precision and low load, it is necessary to explore an efficient scheduling strategy, among which the level of detail (LOD) model is a widely used balance strategy. The previous level of detail model methods all judge whether to retain geometric elements based on the spatial distance between the nodes of the scene mesh model, or between the point and the patch, or between the point and the edge. During the drawing process, by controlling the error threshold, the projection error between the simplified mesh model and the original model in the screen space is small enough to keep the simplified model visually equivalent to the original model. In the scene model, some geometric structures have small geometric dimensions in the scene space, but they play an important role in defining the geometric features of the scene. If the distance in the scene space is used as the criterion alone, when a larger threshold is taken, these scene geometric features will inevitably be lost.
[0003] Since there is a strict correspondence between the geometric features of the scene model and the normal changes of the model surface, simplifying the coplanar features of the facets in the scene mesh model can simplify the scene model to the greatest extent while effectively maintaining the geometric features of the scene model. For this reason, this patent proposes a scene surface mesh simplification method based on a Gaussian sphere. Summary of the invention
[0004] The present invention aims at the deficiencies in the prior art and provides the following technical solutions:
[0005] A method for generating multi-level details of a scene in a CIM platform includes the following steps:
[0006] A1, divide the model surface into some facet families according to the normal direction of the facets in the network model;
[0007] A2, merging adjacent facets within a family of approximately coplanar facets and searching for boundary information of the resulting hyperface;
[0008] A3, retriangulate the hypersurface. For the adjacent fixed points on the hypersurface boundary, merge the model surface according to a certain importance criterion, and delete the surface details with small area and low visual importance;
[0009] A4, uses the strategy of merging eight-connected domain patches to generate a multi-level detail model.
[0010] As an improvement of the above technical solution, the detailed steps of dividing the model surface into facet families are as follows:
[0011] B1, divide the Gaussian sphere;
[0012] B2, processing the divided facets to determine whether any pair of normal angles among several connected facets is less than the coplanar error threshold set by the user;
[0013] Merge the patches that meet the conditions into a hyperplane;
[0014] B3, shrinking hyperplane boundary;
[0015] B4, process the hyperplane boundary: linearize the hyperplane boundary; merge adjacent boundary points;
[0016] B5, retriangulate the processed hyperplane boundary;
[0017] B6, generate M by merging the hypersurfaces in each Gaussian sphere grid in the eight-connected neighborhood of the Gaussian sphere. i Layer of detail model.
[0018] As an improvement of the above technical solution, in step B2, a simplified scene grid model M is generated. MaxLevel-1 , it is necessary to classify the scenery patches in the scene according to the bottom-level Gaussian sphere mesh, and classify them into the corresponding Gaussian sphere mesh according to the normal of each patch in the scenery mesh model to form a patch family.
[0019] As an improvement of the above technical solution, in step B3, when searching for the hyperplane boundary, a closed loop is searched in the boundary point table, and points belonging to the loop are defined as boundary points of the loop.
[0020] As an improvement of the above technical solution, in step B4, when the hyperplane boundary is linearized, within the allowable error range, the hyperplane is defined as a plane, that is, until two hyperplanes intersect on a straight line, the common boundary of two adjacent hyperplanes is approximately a straight line.
[0021] As an improvement to the above technical solution, when merging adjacent boundary points, the merging process is performed within the boundary ring of the hypersurface.
[0022] As an improvement of the above technical solution, in step B6, the known M i Layer detail model, generate M i-1The process of layer merging is achieved by merging the hypersurfaces in each Gaussian sphere grid in the eight-connected neighborhood of the Gaussian sphere, and the hypersurface merging is achieved through the following steps:
[0023] C1: Take each hyperface in the central child node as a seed and expand to the adjacent hyperface. If the adjacent face of the current hyperface is a hyperface in its node, then expand until the current hyperface has no more adjacent faces that can be merged.
[0024] C2: Taking each hyper-surface in the surrounding child nodes as a seed, expand to the adjacent hyper-surfaces;
[0025] C3: Put M i The hyperfaces in the layer nodes are transferred to their respective parent nodes, i.e., M i-1 Then, each hypersurface is straightened and triangulated to generate a hypersurface that satisfies M. i-1 Layer threshold M i-1 Layer simplifies the mesh;
[0026] C4: Recursively generate simplified meshes at other levels until the M0 level.
[0027] Beneficial effects of the present invention:
[0028] The algorithm proposed in the present invention performs coplanar merging and mesh simplification based on normals, thereby generating a multi-level detail model. The algorithm consists of hypersurface boundary search, adjacent boundary point merging, re-triangulation and other parts, and the algorithm uses the size of the angle between the normals of adjacent facets as a criterion for judging coplanarity. First, the model surface is divided into a number of facet families according to the normals of the facets in the mesh model, and then the adjacent facets are merged in the approximately coplanar facet families, and the boundary information of the hypersurface formed after the merging is searched, and then the hypersurface is re-triangulated. For the adjacent vertices located on the hypersurface boundary, the algorithm merges the model surface according to a certain importance criterion, and can delete surface details with small area and low visual importance. Finally, the strategy of merging eight-connected neighborhood facets is adopted to generate a multi-level detail model, so that the algorithm finds a balance between computational efficiency and model visual quality, and can effectively optimize and simplify the mesh model to generate a multi-level detail model suitable for various application requirements. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 is a flow chart of the present invention;
[0030] Figure 2 It is a hierarchical grid and a planar nine-tree structure diagram of the present invention;
[0031] Figure 3 A mapping structure diagram of the cube surface grid to the spherical surface of the present invention;
[0032] Figure 4It is a structural diagram of the hypersurface boundary linearization of the present invention;
[0033] Figure 5 A retriangulated structure diagram of the hypersurface of the present invention;
[0034] Figure 6 This is the eight-connected network node merging graph of the present invention. DETAILED DESCRIPTION
[0035] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0036] See also Figure 1-6 , the present invention provides a technical solution:
[0037] A method for generating multi-level details of a scene in a CIM platform includes the following steps:
[0038] A1, divide the model surface into some facet families according to the normal direction of the facets in the network model;
[0039] A2, merging adjacent facets within a family of approximately coplanar facets and searching for boundary information of the resulting hyperface;
[0040] A3, retriangulate the hypersurface. For the adjacent fixed points on the hypersurface boundary, merge the model surface according to a certain importance criterion, and delete the surface details with small area and low visual importance;
[0041] A4, uses the strategy of merging eight-connected domain patches to generate a multi-level detail model.
[0042] In this implementation scheme, the proposed algorithm performs coplanar merging and mesh simplification based on normals to generate a multi-level detail model. The algorithm consists of hypersurface boundary search, adjacent boundary point merging, and re-triangulation. The algorithm uses the size of the angle between the normals of adjacent facets as the criterion for judging coplanarity. First, the facets in the mesh model are divided into some facet families according to their normals. Then, the adjacent facets are merged in the approximately coplanar facet families, and the boundary information of the hypersurface formed after the merger is searched. Then, the hypersurface is re-triangulated. For the adjacent vertices located on the hypersurface boundary, the algorithm merges them according to certain importance criteria, and can delete surface details with small area and low visual importance. Finally, the strategy of merging eight-connected neighborhood facets is adopted to generate a multi-level detail model. Therefore, the algorithm finds a balance between computational efficiency and model visual quality, and can effectively optimize and simplify the mesh model to generate a multi-level detail model suitable for various application requirements.
[0043] Specifically, the detailed steps for dividing the model surface into patch families are as follows:
[0044] B1, divide the Gaussian sphere;
[0045] B2, processing the divided facets to determine whether any pair of normal angles among several connected facets is less than the coplanar error threshold set by the user;
[0046] Merge the patches that meet the conditions into a hyperplane;
[0047] B3, shrinking hyperplane boundary;
[0048] B4, process the hyperplane boundary: linearize the hyperplane boundary; merge adjacent boundary points;
[0049] B5, retriangulate the processed hyperplane boundary;
[0050] B6, generate M by merging the hypersurfaces in each Gaussian sphere grid in the eight-connected neighborhood of the Gaussian sphere. i Layer of detail model.
[0051] Specifically, in step B1, the plane nine-branch tree represents a hierarchical data structure, such as Figure 2 As shown, the square is the root node of the tree, and its side length is 3n. The quadrilateral is divided into 9 small squares, and its side length is 1 / 3 of the original square side length, that is, 3 n -1 , and then, recursively decompose these 9 small squares to obtain a planar nine-branch tree.
[0052] In this embodiment, first, a cube is mapped to a spherical surface, and then each face of the cube is divided into a planar nine-branch tree, such as Figure 3 As shown, the segmentation result is mapped onto the sphere to obtain a hierarchically segmented sphere. Figure 3 It is represented as a surface of a cube to be mapped onto the Gaussian sphere. The six spherical meshes formed by the six surfaces of the cube are the hierarchical spherical mesh S0 layer.
[0053] The grid marked with mode 1 is obtained by subdividing the S0 layer and belongs to the S1 layer; the grid marked with mode 2 is obtained by subdividing the S1 layer and belongs to the S2 layer; the grid marked with mode 3 is obtained by subdividing the S2 layer and belongs to the S3 layer, ... Thus, the entire spherical grid forms a forest structure with the 6 nodes of the S0 layer as the root.
[0054] Specifically, in step B2, the most refined simplified scene grid model M is generated. MaxLevel-1, it is necessary to classify the scenery patches in the scene according to the bottom-level Gaussian sphere mesh, and classify them into the corresponding Gaussian sphere mesh according to the normal of each patch in the scenery mesh model to form a patch family.
[0055] In this embodiment, after being classified into the corresponding Gaussian sphere mesh to form a facet family, the faces in the facet family may be connected or disconnected, wherein the connected faces constitute a hypersurface. Therefore, a facet family corresponds to one or more hypersurfaces.
[0056] Specifically, in step B3, when searching for the hyperplane boundary, a closed loop is searched in the boundary point table, and points belonging to the loop are defined as boundary points of the loop.
[0057] In this embodiment, the ring here is the outer boundary of a hypersurface, or the boundary of the inner hole of a hypersurface. There are many methods to determine the inner and outer boundaries. The algorithm of the present invention adopts a method of first projecting the ring onto the maximum projection plane, and then determining whether it is clockwise or counterclockwise.
[0058] Specifically, in step B4, when the hyperplane boundary is linearized, within the allowable error range, the hyperplane is defined as a plane, that is, until two hyperplanes intersect on a straight line, the common boundary of two adjacent hyperplanes is approximately a straight line.
[0059] In this embodiment, within the allowable error range, the hypersurface can be considered as a plane. Note that two planes intersect on a straight line, so the common boundary of two adjacent hypersurfaces should be approximately a straight line. Based on this principle, we further simplify the boundary of each hypersurface, such as Figure 4 As shown, the common boundaries of adjacent hypersurfaces are replaced by straight lines.
[0060] Specifically, when adjacent boundary points are merged, the merging process is performed within the boundary ring of the hypersurface.
[0061] In this embodiment, in the process of merging adjacent boundary points, points with low importance are moved to points with high importance. Corner points can only be used as target points, that is, edge deletion can only be achieved by moving other points to the corner points. If both points are corner points, the merging operation is not performed because the algorithm only merges very close points. Therefore, the geometric and topological features of the original scene model can be maintained while achieving simplification.
[0062] Specifically, in step B5, if Figure 5As shown, the hyperplane is retriangulated, that is, given a boundary ring of a region, there are many ways to triangulate the region. The algorithm proposed in the present invention is to first project the boundary ring to the maximum projection surface, and then perform Delaunay triangulation in 2D space. Delaunay triangulation in 2D space is a common geometric calculation method for generating a triangular mesh that does not have any vertices within its circumscribed circle. A sufficiently large super triangle is selected to completely contain all vertices in the area to be triangulated. Usually, the vertices of this super triangle are a large triangle outside the area, such as an equilateral triangle with a sufficiently large side length, to ensure that all points are outside it, and the vertices in the area to be triangulated are inserted into the current Delaunay triangular mesh one by one in order. For each inserted vertex, its position in the current triangular mesh is found to ensure that the newly formed triangle meets the Delaunay condition.
[0063] Specifically, in step B6, the known M i Layer detail model, generate M i-1 The process of layer merging is achieved by merging the hypersurfaces in each Gaussian sphere grid in the eight-connected neighborhood of the Gaussian sphere, and the hypersurface merging is achieved through the following steps:
[0064] C1: Take each hyperface in the central child node as a seed and expand to the adjacent hyperface. If the adjacent face of the current hyperface is a hyperface in its node, then expand until the current hyperface has no more adjacent faces that can be merged.
[0065] C2: Taking each hyper-surface in the surrounding child nodes as a seed, expand to the adjacent hyper-surfaces;
[0066] C3: Put M i The hyperfaces in the layer nodes are transferred to their respective parent nodes, i.e., M i-1 Then, each hypersurface is straightened and triangulated to generate a hypersurface that satisfies M. i-1 Layer threshold M i-1 Layer simplifies the mesh;
[0067] C4: Recursively generate simplified meshes at other levels until the M0 level.
[0068] In this embodiment, a multi-level detail model is generated by merging hypersurfaces in each Gaussian sphere mesh in the eight-connected neighborhood of the Gaussian sphere. In C1, the current hypersurface is merged with its adjacent hypersurfaces to form a new hypersurface. This expansion is performed recursively until the current hypersurface has no more neighbors that can be merged.
[0069] In C2, Figure 6As shown, the grids marked by mode 6 are the surrounding child nodes of the parent node. In addition to the faces within the grid itself, they can merge the grid nodes marked by mode 7 in the surrounding grids, that is, the hypersurfaces within the grid nodes in the eight-connected neighborhood except the central child node of its parent node. This expansion is performed recursively until the current hypersurface no longer has adjacent hypersurfaces that can be merged.
[0070] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them.
Claims
1. A method for generating multi-level details of a scene in a CIM platform, characterized in that: The following steps are involved: A1, divide the model surface into some facet families according to the normal direction of the facets in the network model; A2, merging adjacent facets within a family of approximately coplanar facets and searching for boundary information of the resulting hyperface; A3, retriangulate the hypersurface. For the adjacent fixed points on the hypersurface boundary, merge the model surface according to a certain importance criterion, and delete the surface details with small area and low visual importance; A4, uses the strategy of merging eight-connected domain patches to generate a multi-level detail model.
2. A method for generating multi-level details of a scene in a CIM platform according to claim 1, characterized in that: The detailed steps for dividing the model surface into patch families are as follows: B1, divide the Gaussian sphere; B2, processing the divided facets to determine whether any pair of normal angles among several connected facets is less than the coplanar error threshold set by the user; Merge the patches that meet the conditions into a hyperplane; B3, shrinking hyperplane boundary; B4, process the hyperplane boundary: linearize the hyperplane boundary; Merge adjacent boundary points; B5, retriangulate the processed hyperplane boundary; B6, generate M by merging the hypersurfaces in each Gaussian sphere grid in the eight-connected neighborhood of the Gaussian sphere. i Layer of detail model.
3. A method for generating multi-level details of a scene in a CIM platform according to claim 2, characterized in that: In step B2, a simplified scene mesh model M is generated. MaxLevel-1 , it is necessary to classify the scenery patches in the scene according to the bottom-level Gaussian sphere mesh, and classify them into the corresponding Gaussian sphere mesh according to the normal of each patch in the scenery mesh model to form a patch family.
4. A method for generating multi-level details of a scene in a CIM platform according to claim 2, characterized in that: In step B3, when searching for the hyperplane boundary, a closed loop is searched in the boundary point table, and points belonging to the loop are defined as boundary points of the loop.
5. A method for generating multi-level details of a scene in a CIM platform according to claim 2, characterized in that: In step B4, when the hyperplane boundary is linearized, within the allowable error range, the hyperplane is defined as a plane, that is, until two hyperplanes intersect on a straight line, the common boundary of two adjacent hyperplanes is approximately a straight line.
6. A method for generating multi-level details of a scene in a CIM platform according to claim 5, characterized in that: When merging adjacent boundary points, the merging process is performed within the boundary ring of the hypersurface.
7. A method for generating multi-level details of a scene in a CIM platform according to claim 2, characterized in that: exist In step B6, the known M i Layer detail model, generate M i-1 The process of layer merging is achieved by merging the hypersurfaces in each Gaussian sphere grid in the eight-connected neighborhood of the Gaussian sphere, and the hypersurface merging is achieved through the following steps: C1: Take each hyperface in the central child node as a seed and expand to the adjacent hyperface. If the adjacent face of the current hyperface is a hyperface in its node, then expand until the current hyperface has no more adjacent faces that can be merged. C2: Taking each hyper-surface in the surrounding child nodes as a seed, expand to the adjacent hyper-surfaces; C3: Put M i The hyperfaces in the layer nodes are transferred to their respective parent nodes, i.e., M i-1 Then, each hypersurface is straightened and triangulated to generate a hypersurface that satisfies M. i-1 Layer threshold M i-1 Layer simplifies the mesh; C4: Recursively generate simplified meshes at other levels until the M0 level.
