Plant three-dimensional model lightweight method, system, device and storage medium

Abstract

The application provides a plant three-dimensional model lightweight method, system, device and storage medium, and relates to the technical field of model processing. The method comprises the following steps: dividing a plant three-dimensional model into multiple components based on structural characteristics; performing a surface reduction operation on a first component based on a first surface reduction mode to obtain a first surface reduction part; wherein the first component is any one of the components, and the first surface reduction mode is determined based on the structural characteristics of the first component; and generating a lightweight model based on multiple first surface reduction parts. The method can reduce the model file size by adopting an adaptive lightweight strategy for different components of the plant three-dimensional model, and can improve the efficiency of storing the plant three-dimensional model and transmitting the plant three-dimensional model data, thereby improving the smoothness of model use.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of model processing, in particular to a plant three-dimensional model lightweight method, system, device and storage medium. BACKGROUND

[0002] Plant design is an important part of landscape design. After landscape design turns from two-dimensional design to three-dimensional design, plant design also turns from representing plants through two-dimensional blocks to representing plants through plant three-dimensional models. Compared with the two-dimensional design, plant three-dimensional models can more realistically and comprehensively present the visual effect and ecological function of plants in the landscape, significantly improve the expressiveness and communication efficiency of the design, and also provide accurate guidance in the construction phase.

[0003] A plant three-dimensional model is usually created by a three-dimensional plant modeling software such as Speedtree or Blender, and is composed of a trunk, branches, leaves, flowers, roots, textures and materials, etc. A simple plant three-dimensional model is composed of about 10,000 to 50,000 triangular facets and low-resolution textures, and the file size is between 5MB and 20MB. A high-precision plant model is composed of 100,000 to 1,000,000 triangular facets and high-resolution textures, and the file size is usually between 20MB and 100MB, or even larger. Therefore, the plant three-dimensional model in the prior art has the problem of large file size and low use fluency. SUMMARY

[0004] The present application provides a plant three-dimensional model lightweight method, device, equipment and storage medium, which can use adaptive lightweight strategies for different components of the plant three-dimensional model to reduce the model surface, and perform multi-level optimization processing on the model, thereby improving the ability to reduce the model file size, improving the efficiency of storing the plant three-dimensional model and transmitting the plant three-dimensional model data, and improving the fluency of using the model.

[0005] In a first aspect, the present application provides a plant three-dimensional model lightweight method, which can include:

[0006] dividing the plant three-dimensional model into a plurality of components based on structural features;

[0007] performing a surface reduction operation on the first component based on a first surface reduction manner to obtain a first surface reduction part; wherein the first component is any of the components, and the first surface reduction manner is determined based on the structural features of the first component;

[0008] generating a lightweight model based on a plurality of the first surface reduction parts.

[0009] In the above implementation manner, by adopting an adaptive lightweight strategy for different components of the plant three-dimensional model to perform model surface reduction, the model is optimized in multiple levels, which can improve the ability to reduce the model file size, improve the efficiency of storing the plant three-dimensional model and transmitting the plant three-dimensional model data, and thus improve the smoothness of model use.

[0010] In some embodiments, the plurality of components includes trunk and branch parts, and the surface reduction operation on the first component based on the first surface reduction manner can include:

[0011] In the case where the first component is the trunk and branch part, a model skeleton curve is extracted from the trunk and branch part;

[0012] The model skeleton curve is classified in stages to determine a main trunk part and a multi-level branch part in the trunk and branch part, and a topological relationship of the model skeleton curve is established based on the main trunk part and the multi-level branch part;

[0013] The mesh surface of the trunk and branch part is classified according to the topological relationship to determine a classification result of each mesh surface;

[0014] The surface reduction ratio of each part in the main trunk part and the multi-level branch part is determined according to the classification result, and a surface reduction operation is performed according to the surface reduction ratio.

[0015] In the above implementation process, the mesh optimization method based on region importance recognition can identify parts that are less important or do not affect the overall visual effect in the trunk and branch, and perform surface reduction on these parts. By reducing the number of triangular facets in these parts, the complexity of the trunk and branch area in the model can be reduced, thereby improving the rendering efficiency of the model and improving the smoothness of model use.

[0016] In some embodiments, the plurality of components includes a root system part, and the surface reduction operation on the first component based on the first surface reduction manner can include:

[0017] In the case where the first component is the root system part, a model skeleton curve is extracted from the root system part;

[0018] determining a maximum value of the first direction coordinate and a vertex corresponding to the maximum value by traversing the grid vertices of the root system part, determining all first end points from the model skeleton curve, determining a main root as a skeleton curve on which all the first end points are located, and determining a side root as a remaining skeleton curve, and establishing a topological relationship of the model skeleton curve based on the main root and the side root; wherein the first end point is an end point whose absolute difference value between the end point first direction coordinate and the maximum first direction coordinate is less than a preset tolerance value;

[0019] classifying the grid faces of the root system part according to the topological relationship, and determining a classification result of each grid face;

[0020] fitting the root system based on a plurality of preset geometric bodies for each classified grid face of the root system part, and calculating a union of the plurality of preset geometric bodies based on a Boolean operation to reduce an overlapping part between the preset geometric bodies.

[0021] In the above implementation process, the geometric replacement method can reduce the number of triangular faces in the model by replacing the complex root system part with a simpler geometric shape, thereby reducing the size of the processed model file, improving the efficiency of storage and data transmission, and improving the smoothness of model use. Although the geometric replacement method simplifies the geometric shape of the root system, reasonable replacement and layout in the embodiments of the present application can still maintain the coherence and authenticity of the model in vision, which helps to maintain the overall visual effect of the model and reduces the visual distortion of the plant three-dimensional model caused by simplification processing.

[0022] In some embodiments, the plurality of component parts includes leaf and flower parts, and the decimation operation on the first component part based on the first decimation method can include:

[0023] In the case where the first component part is the leaf and flower part, the model grid faces of the leaf and flower part are divided into a first important region and a second important region;

[0024] determining a first error matrix based on the first weight and determining a second error matrix based on the second weight;

[0025] performing edge collapse operation on the first important region based on the first error matrix and performing edge collapse operation on the second important region based on the second error matrix.

[0026] In the implementation process, the errors of different parts of the leaves and flowers are calculated by using the region weighted error simplification algorithm, and simplification is performed based on the errors, so that the number of triangular facets in the plant three-dimensional model can be reduced. The region weighted error simplification algorithm can better maintain the detailed features of the leaves and flowers, and can improve the visual effect of the model and maintain the realism of the model on the basis of improving the storage of the plant three-dimensional model and improving the efficiency of data transmission of the plant three-dimensional model.

[0027] In some embodiments, the plant three-dimensional model can be divided into a plurality of component parts based on structural features, which can include:

[0028] The structure of the plant three-dimensional model is identified and labeled based on a target semantic segmentation model, and the plant three-dimensional model is divided into a plurality of component parts according to the identification and labeling results of the plant three-dimensional model.

[0029] In the above implementation mode, the plant three-dimensional model is stored according to the structure hierarchy based on the deep learning semantic segmentation technology, and the most suitable lightweight strategy is used for different parts, so that the model size can be reduced while the visual quality is maintained.

[0030] In some embodiments, the lightweight model can be generated based on a plurality of first reduced surface parts, which can include:

[0031] For one of the first reduced surface parts, the corresponding part in the plant three-dimensional model is replaced according to the first reduced surface part, so that the lightweight model is obtained after all the first reduced surface parts are replaced.

[0032] In some embodiments, the method can further include:

[0033] Determining the distance from the model reference point of the lightweight model to the camera position of the current view;

[0034] In the case that the lightweight model is in the focal point area of the current view, the mapping resolution of the lightweight model is determined according to the distance from the model reference point to the camera position of the current view and the state parameters of the focal point area, and the lightweight model is loaded based on the mapping resolution.

[0035] In a second aspect, the embodiments of the present application provide a plant three-dimensional model lightweight system, which can include:

[0036] The division module is configured to divide the plant three-dimensional model into a plurality of component parts based on structural features;

[0037] The reducing module is configured to perform a reducing operation on the first component part based on a first reducing mode to obtain a first reduced part, wherein the first component part is any of the component parts, and the first reducing mode is determined based on a structural feature of the first component part.

[0038] The generating module is configured to generate a lightweight model based on the first reduced parts.

[0039] In a third aspect, an embodiment of the present application provides a computer device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method described in the above description when executing the computer program.

[0040] In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the muscle function evaluation method described in the above description.

[0041] Compared with the prior art, the present application has the beneficial effects that: by adopting individualized lightweight strategies for different component parts of a plant three-dimensional model, the volume of the model file is reduced by 50%-70%. The storage and transmission costs are reduced, and the model is suitable for resource-limited devices and complex landscape project scenarios. The loading speed of the lightweight model in Rhinoceros is significantly accelerated, and the interface lag and delay problems are reduced. The user experience is improved, especially in large-scale landscape projects containing a large number of plant models, the design operation is more smooth, and the adaptability of the generated lightweight model is stronger, which can better support visualization, construction guidance, virtual display and other purposes in landscape design. In addition, different first reducing modes such as topology analysis and geometric substitution method can avoid the common geometric distortion and structural damage problems in the grid simplification process. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 A step schematic diagram of the plant three-dimensional model lightweight method provided by the embodiment of the present application.

[0043] Figure 2 A schematic diagram of the plant three-dimensional model lightweight system provided by the embodiment of the present application. DETAILED DESCRIPTION

[0044] The present application will be further described in detail below in combination with test examples and specific embodiments. However, it should not be understood that the above-mentioned subject matter of the present application is limited to the following embodiments, and any technology realized based on the content of the present application belongs to the protection scope of the present application.

[0045] In the description of specific embodiments of the present application, the terms of "upper", "lower", "left", "right", "center", "inner", "outer", "side" and the like indicating the orientation or positional relationship are expressed based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship when the product / device / apparatus is usually used. These terms of orientation or positional relationship are only for the convenience of describing the present application or simplifying the description in specific embodiments, for the purpose of facilitating the quick understanding of the scheme by the skilled person, and are not intended to indicate or imply that a specific device / component / element must have a specific orientation, or be constructed and operated in a specific positional relationship, and therefore cannot be understood as a limitation of the present application.

[0046] In the description of specific embodiments of the present application, the technical terms "first", "second" and the like only distinguish one entity or operation from another entity or operation, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of specific embodiments of the present application, the meaning of "multiple" is two and more than two, unless otherwise specifically limited.

[0047] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearance of the phrase in various places in the specification are not necessarily all referring to the same embodiment, or are necessarily mutually exclusive or alternative embodiments to other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0048] In the process of research, the applicant found that under the prior art scheme, Rhinoceros mainly uses a mesh simplification algorithm to simplify the entire model to simplify the plant three-dimensional model. Mesh simplification combines, deletes, optimizes, and processes the vertices, edges, and faces of the quadrilateral mesh and triangular facets of the model, thereby reducing the number of redundant meshes to reduce the model file size. This mesh simplification algorithm is to simplify the entire model, and cannot intelligently identify the structural hierarchy of the plant three-dimensional model, resulting in some visually important details being over-simplified or lost, causing the simplified plant three-dimensional model to look unnatural or lack realism. In this mesh simplification algorithm, the simplification priority is determined based on error measurement, and the simplification process does not fully consider the visual importance of each part of the model, resulting in uneven simplification. Important parts such as tree trunks and branches may be over-simplified, resulting in a lack of necessary details in these parts, while less important parts such as roots and some areas of leaves may be simplified too little, resulting in a lack of obvious simplification efficiency improvement, and even affecting the visual effect of the plant three-dimensional model. The mesh simplification algorithm does not fully consider the hierarchical structure and detailed connection relationship of the plant three-dimensional model when simplifying the mesh, such as the connection between the tree trunk and the branch, the connection between the branch and the leaf, etc. When simplifying the mesh, especially using vertex merging or edge folding methods, the topological structure of the model may be damaged, causing geometric distortion, and even affecting subsequent analysis and calculation, such as plant growth simulation, light calculation, etc. Finally, since the mesh simplification algorithm simplifies the entire model, a large amount of calculation is required to determine which facets can be simplified, especially for high-precision models or very complex plant three-dimensional models, which will bring a large computational overhead, resulting in a high computational cost of the simplification process itself. In some resource-limited devices or environments, this overhead may affect performance, resulting in an increase in processing time and affecting the smoothness of large-scale scenes.

[0049] Therefore, based on this, the embodiment of the present application provides a plant three-dimensional model lightweight method, which selects the most suitable lightweight strategy for each component of the model, thereby performing multi-level optimization processing on the model to minimize the model file size and improve the smoothness of the model use.

[0050] Embodiment 1

[0051] Please refer to Figure 1 , Figure 1A schematic diagram of the steps of the plant three-dimensional model lightweight method provided in the embodiments of the present application. The plant three-dimensional model lightweight method provided in the embodiments of the present application can be applied to Rhinoceros software, which is a 3D computer graphics and computer-aided design application software. The main geometric model of Rhinoceros is based on the NURBS mathematical model, and the mathematical precise representation of generating curves and free-form surfaces in computer graphics can be realized.

[0052] The following embodiments are described with the use of Rhinoceros software. The plant three-dimensional model lightweight method can include:

[0053] S1, dividing the plant three-dimensional model into multiple components based on structural characteristics.

[0054] The structural characteristics refer to the morphological characteristics of plants. The three-dimensional structure of plants usually includes trunk, branch, root system, leaf and flower, etc. These parts have different characteristics in morphology, size and position, and therefore can be used as the basis for division.

[0055] S2, performing a surface reduction operation on the first component based on a first surface reduction method to obtain a first surface reduction part.

[0056] The first component is any component, and the first surface reduction method is determined based on the structural characteristics of the first component. The first surface reduction method is determined based on the structural characteristics of the first component, which means that according to the multiple components divided from the plant three-dimensional model, a corresponding first surface reduction method is selected for each component. In the embodiments of the present application, the first surface reduction method can be one of multiple surface reduction methods, which will be described later, and will not be expanded here.

[0057] S3, generating a lightweight model based on the multiple first surface reduction parts.

[0058] In the embodiments of the present application, by using personalized lightweight strategies for different components of the plant three-dimensional model, the volume of the model file is reduced by 50%-70%. The storage and transmission costs are reduced, making the model suitable for resource-limited devices and complex landscape project scenarios. The loading speed of the lightweight model in Rhinoceros is significantly accelerated, reducing the interface lag and delay problems. The user experience is improved, especially in large-scale landscape projects containing a large number of plant models, enhancing the smoothness of design operations. The generated lightweight model has better adaptability and can better support visualization, construction guidance, virtual display and other purposes in landscape design. In addition, different first surface reduction methods such as topology analysis and geometric substitution method can avoid the common geometric distortion and structural damage problems in the grid simplification process.

[0059] In the above implementation process, by adopting an adaptive lightweight strategy for different components of the plant three-dimensional model to reduce the surface of the model, the model is optimized at multiple levels, which can improve the ability to reduce the size of the model file, improve the storage of plant three-dimensional models, and improve the efficiency of plant three-dimensional model data transmission, thereby improving the smoothness of model use.

[0060] Embodiment 2

[0061] This embodiment is a specific implementation of the above embodiment 1, which is a first surface reduction method for reducing the surface of the first component.

[0062] In the case where the first component is the trunk and branch part, the first surface reduction method can be a mesh optimization method based on region importance recognition; the first surface reduction method for reducing the surface of the first component can include:

[0063] S201, in the case where the first component is the trunk and branch part, extracting the model skeleton curve from the trunk and branch part.

[0064] Wherein, the mesh object of the trunk and branch part can be determined from the division result of the plant three-dimensional model, and the mesh object named trunk and branch is loaded in Rhinoceros. The implementation of extracting the model skeleton curve from the trunk and branch part can include: obtaining the naked edge of the mesh through the “Mesh.GetNakedEdges” method provided by RhinoCommon, the naked edge refers to the edge that is only shared by one face of the mesh. The naked edge is converted into a Nurbs curve through the “ToNurbsCurve” method, which involves obtaining the two endpoints of the edge, and using interpolation or fitting method to create a smooth curve. The redundant information in the curve is simplified through the “Curve.Simplify” method, while retaining its main features and shape to obtain the simplified curve, which is the model skeleton curve of the trunk and branch part mesh, representing the main shape and structure of the mesh.

[0065] S202, classifying the model skeleton curve step by step, respectively determining the main stem part and multi-level branch part in the trunk and branch part, and establishing the topological relationship of the model skeleton curve based on the main stem part and the multi-level branch part.

[0066] In the skeleton curve, the longest curve can be considered as the main stem, the curve directly connected with the main stem is the first level branch, the curve directly connected with the first level branch is the second level branch, and the curve directly connected with the second level branch is the third level branch. Determining the main stem part and the multi-level branch part in the stem and branch part, and establishing the topological relationship of the model skeleton curve based on the main stem part and the multi-level branch part can include: obtaining the length of each line segment in the skeleton curve by using the "GetLength" function, and finding the longest curve by using the "Max" function, and marking the longest curve as the main stem. The distance from the end point of other curves to the main stem is obtained by using the "Curve.ClosestPoint" method, and when the distance is less than a preset tolerance value, it is considered that the point is on the main stem or close enough to the main stem, and the curve where the point is located is marked as the first level branch; the preset tolerance value can be set to 0.01 m, 0.02 m, 0.03 m, etc. Finally, the curves are classified level by level by using the recursive method. For the curves that have been marked as the first level branch, the "Curve.ClosestPoint" method and the tolerance value are used again to judge the relationship between other unmarked curves and the first level branches, so as to identify the second level branches. In this way, the third level branches, the fourth level branches, etc. can be continuously identified, so as to establish the topological relationship of the model skeleton curve.

[0067] S203, classifying the mesh surfaces of the stem and branch part according to the topological relationship, and determining the classification result of each mesh surface.

[0068] The step of classifying the model mesh surfaces of the trunk and branches according to the skeleton curves with topological relations can include: establishing a minimum axis-aligned bounding box for each curve based on a "curve.GetBoundingBox" method provided by RhinoCommon, and setting the accurate parameter to True, so that the bounding box is calculated according to the position of the curve in the current transformation state (such as rotation, translation, etc.), to ensure more accurate results; wherein RhinoCommon is a core development library of Rhinoceros software, which provides a comprehensive tool set for developers, allowing developers to create Rhinoceros-based applications, plug-ins and scripts to extend the functions of Rhino. The bounding box range is extended by the "BoundingBox.Inflate(tolerance)" method to tolerate a certain tolerance value (tolerance), wherein the tolerance value of the main stem is 1m, the tolerance value of the first-level branch is 0.5m, the tolerance value of the second-level branch is 0.3m, and the tolerance value of the third-level branch is 0.1m; all mesh surface vertices are traversed, and the "BoundingBox.Contains" method is used to determine whether the vertex falls within the bounding box of the curve, at the same time, the "BoundingBox" method is used to establish a bounding box for all mesh surfaces, and the "BoundingBox.Intersects" method is used to detect whether the mesh surface bounding box intersects with the skeleton curve bounding box, if either condition is met, it is considered that the mesh surface is associated with the current curve, and it is classified into the corresponding topological relation classification.

[0069] S204, determining the surface reduction ratio of each part of the main stem part and the multi-level branch part according to the classification result, and performing surface reduction operation according to the surface reduction ratio.

[0070] The skeleton curve based on the existing topological relationship classifies the model grid surface of the trunk and branches. The "curve.GetBoundingBox" method provided by RhinoCommon is used to establish the minimum axis-aligned bounding box for each curve, and the accurate parameter is set to True, so that the bounding box is calculated according to the position of the curve in the current transformation state (such as rotation, translation, etc.), ensuring more accurate results. The "BoundingBox.Inflate(tolerance)" method is used to expand the bounding box range to tolerate a certain tolerance value (tolerance), wherein the tolerance value of the main stem can be 1 m, the tolerance value of the first-order branch can be 0.5 m, the tolerance value of the second-order branch can be 0.3 m, and the tolerance value of the third-order branch can be 0.1 m. All mesh surface vertices are traversed, and the "BoundingBox.Contains" method is used to determine whether the vertex falls within the bounding box of the curve. At the same time, the "BoundingBox" method is used to establish the bounding box for all mesh surfaces, and the "BoundingBox.Intersects" method is used to detect whether the mesh surface bounding box intersects with the skeleton curve bounding box. If either condition is met, it is considered that the mesh surface is associated with the current curve, and it is classified into the corresponding topological relationship classification.

[0071] The surface reduction ratio represents the visual importance of the three-dimensional model of the plant. The visual importance of the trunk, the first-order branch, the second-order branch, and the third-order branch decreases in turn, so the trunk surface reduction ratio can be set to 20%, the first-order branch surface reduction ratio can be set to 40%, the second-order branch surface reduction ratio can be set to 60%, and the third-order branch surface reduction ratio can be set to 80%.

[0072] The "Mesh.Vertices.AddVertices" method and the "Mesh.Faces.AddFace" method provided in RhinoCommon are used to extract and reconstruct the sub-meshes of each classification, and the "Normals.ComputeNormals" method and the "Mesh.Compact" method are used to optimize the normal calculation and vertex structure of the mesh. Based on the surface reduction ratio set by the classification, the target mesh face number is calculated, and the "Mesh.Reduce" method is called to simplify the sub-mesh. Finally, the simplified mesh is stored and output according to the classification.

[0073] In the above implementation process, the mesh optimization method based on region importance recognition can identify the less important or non-affecting overall visual effect part in the trunk and branches, and perform surface reduction processing. By reducing the number of triangular faces in these parts, the complexity of the trunk and branch regions in the model can be reduced, thereby improving the rendering efficiency of the model and enhancing the smoothness of the model usage.

[0074] Example 3

[0075] The embodiment is a specific implementation of the first component reduction operation based on the first reduction method in Embodiment 1.

[0076] In the case where the first component is a root system, the first reduction method can be a geometric replacement method. The first component reduction operation based on the first reduction method to obtain the first reduced component can include:

[0077] S205, in the case where the first component is a root system, extracting a model skeleton curve from the root system.

[0078] Extracting the model skeleton curve from the root system can include: loading a mesh object named root system, obtaining the naked edges of the mesh through the “Mesh.GetNakedEdges” method provided by RhinoCommon, and converting the naked edges into Nurbs curves through the “ToNurbsCurve” method; simplifying the redundant information in the curve through the “Curve.Simplify” method while retaining its main features and shape to obtain the simplified curve, which is the model skeleton curve of the root system.

[0079] S206, traversing the grid vertices of the root system to determine the maximum value of the first direction coordinate and the vertex corresponding to the maximum value, determining all first endpoints from the model skeleton curve, determining the skeleton curve where all first endpoints are located as the main root, and the remaining skeleton curves as the side roots, and establishing the topological relationship of the model skeleton curve based on the main root and the side roots; wherein the first endpoint is an endpoint whose absolute difference between the first direction coordinate and the maximum first direction coordinate is less than a preset tolerance value.

[0080] Since the main root is necessarily connected to the trunk, its starting point is the point with the maximum Z coordinate among the root system grid vertices. Traverse the root system grid vertices to find the maximum Z coordinate and its vertex. Get all endpoint coordinates of the skeleton curve, and use the “np.abs” method provided by the NumPy library to calculate the absolute difference between the Z coordinates of all endpoints in the skeleton curve and the maximum Z coordinate. Filter out the point set whose absolute difference is less than the tolerance value (tolerance=1cm) through the Boolean index, and mark the skeleton curve where these endpoints are located as the main root, and the remaining skeleton curves as the side roots, thereby establishing the topological relationship of the root system skeleton curve.

[0081] S207, classifying the grid faces of the root system according to the topological relationship to determine the classification result of each grid face.

[0082] The skeleton curve based on the existing topological relationship is used to classify the grid surface of the root system model. The "curve.GetBoundingBox" method provided by RhinoCommon is used to establish the minimum axis-aligned bounding box for each curve, and the accurate parameter is set to True, so that the bounding box is calculated based on the position of the curve in the current transformation state (such as rotation, translation, etc.), ensuring more accurate results. The "BoundingBox.Inflate(tolerance)" method is used to expand the bounding box range to tolerate a certain tolerance value (tolerance), wherein the tolerance value of the main root is 30 cm, and the tolerance value of the lateral root is 15 cm. All grid surface vertices are traversed, and the "BoundingBox.Contains" method is used to determine whether the vertex falls within the bounding box of a certain curve. At the same time, the "BoundingBox" method is used to establish a bounding box for all grid surfaces, and the "BoundingBox.Intersects" method is used to detect whether the grid surface bounding box intersects with the skeleton curve bounding box. If either condition is met, it is considered that the grid surface is associated with the current curve, and it is classified into the corresponding topological relationship category.

[0083] In S208, for each classified root system part, a plurality of preset geometric bodies are fitted based on the root system part, and a union of the plurality of preset geometric bodies is calculated based on a Boolean operation to reduce overlapping parts between the preset geometric bodies.

[0084] The preset geometric body can be an ellipsoid. For projects with relatively simple root systems or low spatial analysis requirements, a single geometric body can be used to fit the root system. The "GetBoundingBox" method provided by RhinoCommon is used to establish a minimum axis-aligned bounding box for the root system grid surface, and the accurate parameter is set to True, so that the bounding box is calculated based on the position of the grid surface in the current transformation state (such as rotation, translation, etc.), ensuring more accurate results. The "BoundingBox.Center" method is used to calculate the center point coordinates of the bounding box to determine the center point position of the fitted ellipsoid. The dimensions of the bounding box are used to determine the radii of the long axis, medium axis and short axis of the ellipsoid. The "Ellipsoid" class is used to specify the center point and axis vector to construct an ellipsoid model, and the "Ellipsoid.ToBrep" method is used to convert it to a three-dimensional geometric entity.

[0085] For projects with relatively complex root systems or high spatial analysis requirements, the root system can be divided into main roots and lateral roots by analyzing the topological relationship, and then a plurality of ellipsoids are fitted to replace the segments.

[0086] The minimum axis-aligned bounding box for different types of root system mesh faces is established by the "GetBoundingBox" method provided by RhinoCommon, and the accurate parameter is set to True, so that the bounding box is calculated according to the position of the mesh face in the current transformation state (such as rotation, translation, etc.), ensuring more accurate results; the center point coordinates of the bounding box are calculated using "BoundingBox.Center" to determine the center point position of the fitted ellipsoid; the radii of the long axis, medium axis and short axis of the ellipsoid are determined according to the size of the bounding box; the ellipsoid model is constructed by specifying the center point and axis vector through the "Ellipsoid" class, and is converted into a three-dimensional geometric entity by using the "Ellipsoid.ToBrep" method; the union of multiple ellipsoids is calculated by Boolean operation to reduce the overlapping parts of the ellipsoids.

[0087] In the above implementation process, the method of geometric replacement can reduce the number of triangular facets in the model by replacing complex root system parts with simpler geometric shapes, thereby reducing the size of the processed model file, improving the efficiency of storage and data transmission, and improving the smoothness of model use. Although the geometric replacement method simplifies the geometry of the root system, reasonable replacement and layout in the embodiments of the present application can still maintain the coherence and authenticity of the model in vision, which helps to maintain the overall visual effect of the model and reduces the visual distortion of the plant three-dimensional model caused by simplification.

[0088] Embodiment 4

[0089] This embodiment is a specific implementation of the above-mentioned embodiment 1, which is a first reduction operation on the first component based on a first reduction method.

[0090] In the case where the first component is a leaf and flower part, the first reduction method can be a region weighted error simplification algorithm. The first reduction operation on the first component based on the first reduction method to obtain the first reduced part can include:

[0091] S209、In the case where the first component is a leaf and flower part, the model mesh of the leaf and flower part is divided into a first important region and a second important region.

[0092] The edges of the leaves and flowers have a significant influence on the overall model profile in vision and are preferentially retained in the simplification process, therefore, the model mesh of the leaves and flowers needs to be first divided into a first important region and a second important region. The first important region represents a high importance region, and the second important region represents a low importance region.

[0093] The vertices of the naked edges of the model mesh surface (i.e. edges not shared with other mesh surfaces) are extracted as key points for identifying high importance areas by the "GetNakedEdges" method provided by RhinoCommon; all vertices of the model mesh surface are obtained by the "Mesh.Vertices" method; the Euclidean distance d of each vertex in the mesh from all boundary points is calculated by the "np.linalg.norm" method in NumPy, and when the distance d is less than or equal to 0.02 m, the vertex is marked as a first importance area, and when the distance d is greater than 0.02 m, the vertex is marked as a low importance area.

[0094] S210, determining a first error matrix based on the set first weight and a second error matrix based on the set second weight.

[0095] The high importance area is the most visually critical part, and a higher weight needs to be set to ensure that the details of these areas are better preserved during the simplification process. The low importance area is a relatively flat and simple part of the mesh, which does not greatly affect the final visual effect, and a lower weight can be set to allow more simplification operations. In the embodiment of the application, the weight of the high importance area can be set to 4, and the weight of the low importance area can be set to 2.

[0096] S211, performing edge collapse operations on the first importance area based on the first error matrix and performing edge collapse operations on the second importance area based on the second error matrix.

[0097] Wherein, the first error matrix based on the area weight for each vertex in the first importance area and the second error matrix based on the area weight for each vertex in the second importance area can be calculated by the "compute_qem_matrix" method, and the current error minimum edge is selected in combination with the "find_min_error_edge" method; the edge collapse operation is performed by the "collapse_edge" method to gradually reduce the number of surfaces of the model; the original model in the current document is replaced by the simplified model using the "sc.doc.Objects.Replace" method provided by RhinoScriptSyntax.

[0098] In the above implementation process, the errors of different parts of the leaves and flowers are calculated by the area weighted error simplification algorithm, and the simplification processing is performed based on these errors, which can reduce the number of triangular facets in the plant three-dimensional model. The area weighted error simplification algorithm can better maintain the detailed features of the leaves and flowers, and can maintain the realism of the model and improve the visual effect of the model on the basis of improving the storage of the plant three-dimensional model and improving the efficiency of the plant three-dimensional model data transmission.

[0099] Embodiment 5

[0100] The embodiment is a specific implementation of the above-mentioned embodiment 1 based on structural features to divide the plant three-dimensional model into multiple components.

[0101] Dividing the plant three-dimensional model into multiple components based on structural features can include:

[0102] Identifying and labeling the structure of the plant three-dimensional model based on the target semantic segmentation model, and dividing the plant three-dimensional model into multiple components according to the identification and labeling results of the plant three-dimensional model.

[0103] Wherein, identifying and labeling the structure of the plant three-dimensional model based on the target semantic segmentation model, and dividing the plant three-dimensional model into multiple components according to the identification and labeling results of the plant three-dimensional model can include:

[0104] Data preparation: Prepare a plant three-dimensional model with clear trunk and branch, leaf, flower, root structure, use 3DSlicer three-dimensional labeling tool, create Trunk and Branches, Leaves, Flower, Roots four categories in "Segment Editor" panel; label each part of the plant three-dimensional model by manual labeling and threshold labeling; normalize the model by adjusting the model size, coordinates, attributes to ensure the consistency of the model data; voxelize the model to divide it into fixed size three-dimensional grid; generate diversified data samples through data enhancement such as rotation, scaling, gray adjustment, elastic deformation, etc. to provide high quality samples for model training.

[0105] Model construction: adopt 3D U-Net based convolutional neural network, this deep learning model structure contains encoding path and decoding path. In the encoding path, each layer contains two 3x3x3 convolution kernels to extract local region features, and a ReLu activation function is connected after each convolution to filter out unimportant feature information, while batch normalization (BN) is introduced before ReLu activation function to standardize the input data and keep the stability of data distribution, improve the stability of model and reduce overfitting. After convolution, the feature map resolution is halved by maximum pooling operation to capture global information, and a total of 3 pooling operations are performed in the encoding path. In the decoding path, the feature map resolution is gradually increased by upsampling operation, and each layer contains two 3x3x3 convolution kernels to halve the feature number. The high-resolution features in the encoding path are directly transmitted to the decoding path through skip connection to supplement the possible missing detail information in the decoding path. Finally, the last layer in the decoding path is a 1x1 convolution to output a single feature map representing the model prediction value.

[0106] Model training: Adjust the class weights by weighted Softmax loss function, ignore unlabeled voxels, and make the model pay more attention to the four important classes of trunk and branch, leaf, flower, and root system. Then input the voxelized three-dimensional data block into the model for training batch by batch. At the same time, through data augmentation such as rotation, scaling, grayscale adjustment, elastic deformation, etc. to improve the generalization ability of the model. In the initial stage of model training, use Adam optimizer to quickly converge, and when the model has approached convergence, use SGD optimizer for fine-tuning to further improve the segmentation accuracy of the model. Finally, evaluate the performance on the validation set, adjust the hyperparameters according to the results, and optimize the model performance.

[0107] The model can be represented as:

[0108]

[0109] where the loss L is calculated by the weight w i The prediction error of each class is weighted and summed; N is the total number of samples; C is the total number of classes; such as background, trunk, leaf, etc. w i is the weight of the i-th class, used to adjust the attention of the model to different classes. is the true class label of sample n (One-Hot encoding), and yi is 1 for the corresponding class of the teammate, and the rest is 0. is the predicted probability of class i for sample n, calculated by the Softmax function. The Softmax function is represented as:

[0110]

[0111] where, is the predicted score of class i by the model (logits).

[0112] Model deployment: In Rhinoceros, load the pre-trained semantic segmentation model through the "init" method, and use the "model.eval" method to switch the model to evaluation mode to ensure stable behavior during inference. Receive input data through the "predict" method, and use "torch.no_grad" to disable gradient calculation to improve inference efficiency, and output the predicted structure.

[0113] Model application: Get all objects in the document through the "doc.Objects" method provided in RhinoCommon; filter out the mesh objects of geometry type Mesh using the "isinstance" method and extract the objects into a list and return; use the "mesh.Vertices" method to get the geometric coordinates of all vertices of the mesh object and store the coordinates in the form of a list; use the "np.array" function to convert the list into a multi-dimensional array object, and use the "torch.tensor" function to convert the array into a tensor format acceptable by the semantic segmentation model. Call the semantic segmentation model to classify and predict the vertices, and label each class, such as tree trunk and branch as 0, leaf as 1, etc.; use the "squeeze" and "numpy" methods provided by Pytorch to optimize the prediction results and return the classification label array. Call the ".NET Framework library "Color.FromArgb(R, G, B)" to create an RGB color label for each class, such as tree trunk and branch RGB value is 139, 69, 19, leaf RGB value is 34, 139, 34, flower RGB value is 255, 105, 180, root system RGB value is 101, 67, 33; traverse the vertex classification list to get the color value corresponding to each vertex; use the "mesh.VertexColors.Add" method in RhinoCommon to assign the color value to the mesh vertex according to the category, and update the color information of the vertex; return the mesh object containing the color information, and the user can intuitively view the model structure segmentation and recognition results through the color information.

[0114] Classification storage: Build a dictionary structure, where the key is the classification label of the vertex, and the value is an empty mesh object; traverse the vertex classification list, use the "enumerate" function to get the vertex index and its corresponding classification label, and add the vertex to the mesh object corresponding to the classification through the "mesh.Vertices.Add" method provided by RhinoCommon; traverse the mesh face, use the "mesh.Faces" method to get the vertex index and its corresponding classification label of each face; use the "set" function to form a set of vertex classification labels of each face and remove duplicate elements, and use the "len" function to check the length of the set to determine whether the vertices of the face belong to the same category. If they belong to the same category, add the face to the corresponding classification mesh through the "Mesh.Faces.AddFace" method, realize the separation and storage of the four parts of the plant three-dimensional model, namely the trunk, branch, leaf, and root system, and provide an operation basis for selecting the most suitable lightweight strategy for multi-level optimization processing according to the characteristics of different parts of the model.

[0115] In the implementation manner of identifying and marking the structure of the plant three-dimensional model based on the target semantic segmentation model, and dividing the plant three-dimensional model into multiple components according to the identification and marking results, the plant three-dimensional model is stored separately according to the structure level based on the deep learning semantic segmentation technology, and the most suitable lightweight strategy is adopted for different parts, so that the model size can be reduced while the visual quality is maintained.

[0116] Embodiment 6

[0117] This embodiment is a specific implementation of generating a lightweight model in Embodiment 1.

[0118] The way of generating a lightweight model can include:

[0119] For a first surface reduction part, the corresponding part in the plant three-dimensional model is replaced according to the first surface reduction part, so that the lightweight model is obtained after all the first surface reduction parts are replaced.

[0120] Embodiment 7

[0121] This embodiment is a specific implementation of optimizing the map and material by the map resolution adaptive algorithm in Embodiment 1.

[0122] The step of optimizing the map and material by the map resolution adaptive algorithm can include:

[0123] S41, determine the distance from the model reference point of the lightweight model to the camera position of the current view.

[0124] Wherein, the "GetBoundingBox" method provided by RhinoCommon is used to establish the minimum axis-aligned bounding box for the model, and the accurate parameter is set to True, so that the bounding box is calculated according to the position of the mesh surface in the current transformation state (such as rotation, translation, etc.), ensuring more accurate results; the center point coordinates of the axis-aligned bounding box of the model are obtained by the "BoundingBox.Center" method, which are used as the reference point of the model; the camera position of the current view (the observation point coordinates of the user in the three-dimensional view) is obtained, and the distance from the model reference point to the user observation point is calculated by "DistanceTo", which is the distance d from the lightweight model to the camera position of the current view.

[0125] Then it can be determined whether the model is in the focal point area of the line of sight, and the determination method can include: subtracting the reference point coordinates from the observation point coordinates to obtain the target direction vector (target vector) from the camera to the model center, that is, the direction vector = model center point coordinates - camera position coordinates; obtaining the line of sight direction vector (line of sight vector) of the camera through "viewport.CameraDirection"; calculating the dot product of the target vector and the line of sight vector after standardizing the two vectors, and determining the included angle θ between the two vectors through the inverse cosine function; determining whether the model is in the focal point area through the preset horizontal and vertical angle thresholds (horizontal angle 45°, vertical angle 30°), if θ is within the range, it is considered that the model is in the focal point area.

[0126] wherein the included angle formula is:

[0127]

[0128] u and v represent the target vector and the line of sight vector respectively; since the two vectors have been standardized, the formula can be simplified as:

[0129]

[0130] S42, in the case that the lightweight model is in the focal point area of the current view, determining the texture resolution of the lightweight model according to the distance from the model reference point to the camera position of the current view and the state parameter of the focal point area, and loading the lightweight model based on the texture resolution.

[0131] A "get_texture_resolution" method is defined, which dynamically adjusts the model texture resolution by inputting the distance parameter d between the camera and the model and the Boolean parameter (True or False) whether it is in the focal point area, combining the preset rule logic, and the rule logic is as shown in Table 1; finally, the model file is loaded and applied according to the resolution.

[0132] Table 1

[0133]

[0134] Please refer to Figure 2 , Figure 2 A schematic diagram of a plant three-dimensional model lightweight system provided by an embodiment of the present application. The plant three-dimensional model lightweight system 20 can include:

[0135] The division module 21 is used to divide the plant three-dimensional model into multiple components based on structural features;

[0136] The reducing module 22 is configured to perform a reducing operation on the first component part based on a first reducing mode to obtain a first reduced part, wherein the first component part is any component part, and the first reducing mode is determined based on a structural feature of the first component part.

[0137] The generating module 23 is configured to generate the lightweight model based on the plurality of first reduced parts.

[0138] Optionally, the reducing module 22 can be specifically configured to:

[0139] In a case where the first component part is a trunk and branch part, a model skeleton curve is extracted from the trunk and branch part;

[0140] The model skeleton curve is classified in stages to determine a main stem part and a multi-stage branch part in the trunk and branch part, and a topological relationship of the model skeleton curve is established based on the main stem part and the multi-stage branch part;

[0141] The mesh surface of the trunk and branch part is classified according to the topological relationship to determine a classification result of each mesh surface;

[0142] A reducing ratio of each part in the main stem part and the multi-stage branch part is determined according to the classification result, and a reducing operation is performed according to the reducing ratio.

[0143] Optionally, the reducing module 22 can be specifically configured to:

[0144] In a case where the first component part is a root part, a model skeleton curve is extracted from the root part;

[0145] The mesh vertices of the root part are traversed to determine a maximum value of a first direction coordinate and a vertex corresponding to the maximum value, all first end points are determined from the model skeleton curve, a main root is determined as a skeleton curve on which all the first end points are located, and a side root is determined as the remaining skeleton curves, and a topological relationship of the model skeleton curve is established based on the main root and the side root, wherein the first end point is an end point whose absolute difference between a first direction coordinate of the end point and a maximum first direction coordinate is less than a preset tolerance value.

[0146] The mesh surface of the root part is classified according to the topological relationship to determine a classification result of each mesh surface;

[0147] For each classified mesh surface of the root part, the root part is fitted based on a plurality of preset geometric bodies, and a union set of the plurality of preset geometric bodies is calculated based on a Boolean operation to reduce an overlapping part between the preset geometric bodies.

[0148] Optionally, the reducing module 22 can be specifically configured to:

[0149] In the case that the first component is a leaf and flower part, the model mesh surface of the leaf and flower part is divided into a first important region and a second important region;

[0150] The first error matrix is determined based on the set first weight, and the second error matrix is determined based on the set second weight;

[0151] The edge collapse operation is performed on the first important region based on the first error matrix, and the edge collapse operation is performed on the second important region based on the second error matrix.

[0152] Optionally, the division module 21 can be specifically configured to:

[0153] The structure of the plant three-dimensional model is identified and labeled based on the target semantic segmentation model, and the plant three-dimensional model is divided into a plurality of components according to the identification and labeling results of the plant three-dimensional model.

[0154] Optionally, the generation module 23 can be specifically configured to:

[0155] The distance from the model reference point of the lightweight model to the camera position of the current view is determined.

[0156] In the case that the lightweight model is in the focus area of the current view, the map resolution of the lightweight model is determined according to the distance from the model reference point of the lightweight model to the camera position of the current view and the state parameter of the focus area, and the lightweight model is loaded based on the map resolution.

[0157] It should be understood that the various modules of the device provided in the above embodiments are only used as an example for the division of the functional modules in the above description, and in actual application, the above functions can be completed by different functional modules according to the needs, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0158] Each functional module in the above embodiments can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit, and the integrated unit can be realized in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only used for mutual distinction, and do not limit the protection scope of the embodiments of the application.

[0159] Based on the same application concept, the embodiments of the present application also provide a computer device, which can include a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to realize the method as described in the above description.

[0160] Based on the same application concept, the embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the method described in the foregoing description.

[0161] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the foregoing application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A lightweight method for plant 3D models, characterized in that, include: The three-dimensional plant model is divided into multiple components based on structural features; A first surface reduction operation is performed on the first component based on a first surface reduction method to obtain a first surface-reduced portion; wherein, the first component is any of the aforementioned components, and the first surface reduction method is determined based on the structural features of the first component; A lightweight model is generated based on multiple first subtractive surfaces; The plurality of components include a trunk and branches. The process of performing a surface reduction operation on the first component based on a first surface reduction method to obtain a first surface-reduced portion includes: When the first component is the trunk and branches, the model skeleton curve is extracted from the trunk and branches. The model skeleton curves are classified step by step to determine the main trunk and multi-level branch parts in the trunk and branch parts respectively, and the topological relationship of the model skeleton curves is established based on the main trunk and multi-level branch parts. The grid surfaces of the trunk and branches are classified according to the topological relationship, and the classification result of each grid surface is determined. Based on the classification results, determine the reduction ratio of each part in the main trunk and the multi-level branches, and perform the reduction operation according to the reduction ratio.

2. The method according to claim 1, characterized in that, The plurality of components includes a root system portion, and the first reduction operation on the first component based on the first reduction method to obtain the first reduced-surface portion includes: When the first component is the root system portion, the model skeleton curve is extracted from the root system portion; Traverse the mesh vertices of the root system, determine the maximum value of the first direction coordinate and the vertex corresponding to the maximum value, determine all first endpoints from the model skeleton curves, determine the skeleton curves where all first endpoints are located as principal roots, and the remaining skeleton curves as lateral roots, and establish the topological relationship of the model skeleton curves based on the principal roots and the lateral roots; wherein, the first endpoint is the endpoint where the absolute difference between the first direction coordinate of the endpoint and the maximum first direction coordinate is less than a preset tolerance value; The grid surfaces of the root system are classified according to the topological relationship, and the classification result of each grid surface is determined. For each classified root system mesh surface, the root system is fitted based on multiple preset geometries, and the union of the multiple preset geometries is calculated based on Boolean operations to reduce the overlap between the preset geometries.

3. The method according to claim 1, characterized in that, The plurality of components include leaf and flower parts, and the process of performing a surface reduction operation on the first component based on a first surface reduction method to obtain a first surface-reduced part includes: When the first component is the leaf and flower part, the model mesh surface of the leaf and flower part is divided into a first important region and a second important region; A first error matrix is ​​determined based on a first set weight, and a second error matrix is ​​determined based on a second set weight. The first important region is subjected to edge collapse operation based on the first error matrix, and the second important region is subjected to edge collapse operation based on the second error matrix.

4. The method according to claim 1, characterized in that, The plant 3D model is divided into multiple components based on structural features, including: The structure of the plant 3D model is identified and labeled based on the target semantic segmentation model, and the plant 3D model is divided into multiple components according to the identification and labeling results.

5. The method according to claim 1, characterized in that, The generation of a lightweight model based on multiple first subtracted surface portions includes: For each of the first facet reduction portions, the corresponding portion in the three-dimensional plant model is replaced according to the first facet reduction portion, so that the lightweight model is obtained after all the first facet reduction portions have been replaced.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: Determine the distance from the model reference point of the lightweight model to the camera position in the current view; When the lightweight model is in the focus area of ​​the current view, the texture resolution of the lightweight model is determined based on the distance from the model reference point to the camera position of the current view and the state parameters of the focus area, and the lightweight model is loaded based on the texture resolution.

7. A lightweight system for three-dimensional plant models, characterized in that, include: The partitioning module is used to divide the 3D plant model into multiple components based on structural features; A surface reduction module is used to perform a surface reduction operation on a first component based on a first surface reduction method to obtain a first surface-reduced portion; wherein, the first component is any of the aforementioned components, and the first surface reduction method is determined based on the structural features of the first component; The generation module is used to generate a lightweight model based on multiple first subtracted surface portions; The plurality of components include a trunk and branches, and the surface reduction module is specifically used for: When the first component is the trunk and branches, the model skeleton curve is extracted from the trunk and branches. The model skeleton curves are classified step by step to determine the main trunk and multi-level branch parts in the trunk and branch parts respectively, and the topological relationship of the model skeleton curves is established based on the main trunk and multi-level branch parts. The grid surfaces of the trunk and branches are classified according to the topological relationship, and the classification result of each grid surface is determined. Based on the classification results, determine the reduction ratio of each part in the main trunk and the multi-level branches, and perform the reduction operation according to the reduction ratio.

8. A computer device, characterized in that, The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as claimed in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 6.

Images