Method and device for establishing garden landscape model based on data fusion technology

Abstract

This application relates to the field of landscape design and digital modeling technology, and discloses a method and apparatus for establishing a landscape model based on data fusion technology. The method includes the following steps: collecting original point cloud data and images of the site; extracting key growth morphology parameters through data registration, semantic segmentation, and individual vegetation segmentation; constructing a multi-dimensional quantitative evaluation model of the landscape's ecology and aesthetics based on these parameters; transforming the design problem into a multi-objective optimization mathematical model and solving it to generate a Pareto optimal solution set; finally, converting the specific design scheme selected from the solution set into a parametric building information model and integrating it with a digital terrain model to generate an interactive three-dimensional integrated landscape model. This invention can objectively and efficiently generate optimized design schemes that take into account both ecological and aesthetic values, automating the design evaluation and optimization adjustment process, and improving the efficiency and optimization level of design scheme generation.

Description

Technical Field

[0001] This invention relates to the field of landscape design and digital modeling technology, specifically to a method and apparatus for establishing landscape models based on data fusion technology. Background Technology

[0002] Landscape design is a comprehensive discipline that integrates ecology, engineering and aesthetics. Its goal is to create a living environment that combines ecological function and aesthetic value. The design process involves understanding the current state of the site, selecting plant materials, organizing the spatial layout, and predicting the overall effect. It is a complex decision-making process.

[0003] In current technological practices, the landscape design process largely relies on the personal experience of designers and two-dimensional computer-aided design software. The evaluation of the merits of design schemes lacks unified and objective quantitative standards and is highly subjective. This makes the design adjustment process a process of repeated trial and error, which is time-consuming and inefficient. At the same time, existing design methods are based on simplified site survey data, which makes it difficult to accurately reflect the real spatial state of complex terrain and existing vegetation, which may lead to the design scheme not matching the actual construction conditions.

[0004] Although 3D modeling software is used for visualizing design schemes, these tools are primarily for result display and do not provide intelligent support for the generation and optimization of design schemes. The design, evaluation, and optimization stages are disconnected, making it difficult for designers to explore numerous potential design possibilities and ensuring that the scheme achieves an optimal or near-optimal balance across multiple objectives, such as ecology and aesthetics. The industry urgently needs a technical method that can deeply integrate high-precision 3D site data with the landscape design process, establish an objective quantitative evaluation system, and achieve automated generation and multi-objective collaborative optimization of design schemes. Therefore, this invention proposes a method and apparatus for establishing garden landscape models based on data fusion technology to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and apparatus for establishing garden landscape models based on data fusion technology, which solves the problems of insufficient utilization of site information, strong subjectivity in design evaluation, and low efficiency in optimization and adjustment in the existing garden landscape design process.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for establishing a garden landscape model based on data fusion technology, the method comprising:

[0007] S1. Acquisition of multi-source spatial information of the site: Obtain the original point cloud dataset and high-resolution texture image set representing the target garden landscape site.

[0008] S2. Registration of multi-site point cloud data: The original point cloud dataset is integrated to generate a complete 3D point cloud model of the site.

[0009] S3. Semantic segmentation of landscape elements: Assign landscape element category labels to the points in the complete site 3D point cloud model, and decompose the model into multiple point cloud subsets including ground point cloud subsets and vegetation point cloud subsets.

[0010] S4. Vegetation Individual Segmentation and Key Parameter Extraction: The vegetation point cloud subset is segmented into multiple individual point cloud clusters, and key growth morphology parameters are calculated.

[0011] S5. Construct a multi-dimensional quantitative evaluation model for garden landscapes: Based on the key growth morphology parameters and preset information, calculate ecological performance indicators and aesthetic quality indicators.

[0012] S6. Generation and adjustment of landscape design schemes based on multi-objective optimization: The landscape design problem is transformed into a multi-objective optimization mathematical model with the ecological performance index and the aesthetic quality index as objective functions, and the multi-objective optimization mathematical model is solved to generate a Pareto optimal solution set.

[0013] S7. Parametric modeling and 3D visualization steps of the design scheme: Generate an interactive 3D integrated landscape model from the specific design scheme selected from the Pareto optimal solution set.

[0014] Preferably, the multi-site point cloud data registration is achieved through a two-stage strategy from coarse to fine. In the coarse registration stage, corresponding feature points between different original point cloud datasets are identified and matched to calculate the initial transformation matrix. In the fine registration stage, the iterative nearest-point algorithm is applied to optimize the pose, with the goal of finding the optimal fine rotation matrix. Translation vector The objective function is to minimize the root mean square error between the transformed source point cloud and the target point cloud. Defined as:

[0015] ;

[0016] in, These are points in the source point cloud after the initial transformation. In the target point cloud, through Tree and other data structures can be used to quickly search for points. The closest corresponding point in space, This represents the number of valid point pairs that were successfully matched in this iteration.

[0017] Preferably, the semantic segmentation step of the landscape elements first calculates the local geometric features, such as curvature, for each point in the point cloud. Planarity and linearity These three eigenvalues (satisfy To calculate:

[0018] Curvature: Planarity: Linearity: ;

[0019] Subsequently, these geometric features are combined into a multidimensional feature vector, and a trained semantic classification model is used to output a predicted class label for each point.

[0020] Preferably, the vegetation individual segmentation and key parameter extraction step first generates a canopy height model using a subset of vegetation point clouds and a subset of ground point clouds, and then normalizes the height of the Z-coordinate of the vegetation points. (in This represents the original elevation of the vegetation point. To achieve the corresponding ground elevation, the canopy height model is then used to generate two-dimensional polygons representing the projection contours of independent tree canopies using a watershed segmentation algorithm. These polygons are then associated with a subset of the three-dimensional vegetation point cloud to form independent individual point cloud clusters.

[0021] Preferably, the calculation of ecological performance indicators and aesthetic quality indicators specifically involves: the calculation of the ecological performance indicators includes: calculating the vertical projection area of ​​the vegetation canopy. Total site area The proportion is used to determine the vegetation coverage rate ; and the application of the Shannon-Wiener index to calculate the species diversity index. Its formula is: ;

[0022] in, This represents the total number of plant species. For species The relative abundance of.

[0023] The calculation of the aesthetic quality indicators includes: determining the conformity to the golden ratio by calculating the deviation between the proportions of key landscape dimensions and the golden ratio values; and calculating color harmony based on the harmony rules of Eaton's color theory. For complementary color harmony, the harmony evaluation function can be defined as:

[0024] ;

[0025] in, and These are the representations of the two colors in the CIELAB color space. and These are the weighting coefficients.

[0026] Preferably, the solution to the multi-objective optimization mathematical model is specifically achieved by applying a non-dominated sorting genetic algorithm, aiming to simultaneously maximize the total score including ecological performance. and aesthetic quality Two-dimensional vector objective function: ;

[0027] in, , , All weighting coefficients are positive.

[0028] Preferably, the parametric modeling and 3D visualization steps of the design scheme specifically include: converting the specific design scheme into a building information model containing parametric information;

[0029] The building information model and digital terrain model containing parametric information are integrated, and scene rendering and interactive presentation are performed to generate the interactive three-dimensional integrated landscape model.

[0030] Preferably, a landscape model building device based on data fusion technology includes:

[0031] The data acquisition module is used to acquire a set of original point cloud datasets and a set of high-resolution texture images representing the target garden landscape site;

[0032] The data processing module is used to integrate the original point cloud dataset to generate a complete three-dimensional point cloud model of the site, assign landscape element category labels to the points in the model, decompose them into multiple point cloud subsets, then segment the point cloud subsets to form multiple individual point cloud clusters, and calculate key growth morphology parameters.

[0033] The evaluation modeling module is used to calculate ecological performance indicators and aesthetic quality indicators based on the key growth morphology parameters, and to construct a multi-objective optimization mathematical model with the ecological performance indicators and aesthetic quality indicators as objective functions.

[0034] The optimization solution module is used to solve the multi-objective optimization mathematical model to generate a Pareto optimal solution set;

[0035] The visualization module is used to generate an interactive three-dimensional integrated landscape model from the specific design schemes selected from the Pareto optimal solution set.

[0036] Preferably, the assessment modeling module specifically includes an ecological assessment unit and an aesthetic assessment unit;

[0037] The optimization solution module is configured to use a non-dominated sorting genetic algorithm to solve the multi-objective optimization mathematical model.

[0038] Preferably, a third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the method of any of the technical solutions in steps S1-S7 of the present invention.

[0039] This invention provides a method and apparatus for establishing a garden landscape model based on data fusion technology. It has the following beneficial effects:

[0040] This invention collects high-precision three-dimensional point cloud data of the site, performs semantic segmentation and individual vegetation segmentation on the point cloud, extracts key growth morphology parameters, and constructs a digital model of the garden landscape site. This ensures that the design is based on the accurate current state of the site, rather than simplified two-dimensional drawings or manual estimation, thereby improving the objectivity and accuracy of landscape analysis and evaluation.

[0041] This invention transforms the landscape design problem into a multi-objective optimization mathematical model with quantified ecological performance indicators and aesthetic quality indicators as objective functions, and applies a non-dominated sorting genetic algorithm to solve it. This can automatically explore a huge design possibility space and generate a set of Pareto optimal solutions that achieve high levels of ecological and aesthetic value. This improves the efficiency and optimization level of design scheme generation and overcomes the limitations of traditional manual trial and error methods, which are inefficient and difficult to achieve global optimum.

[0042] This invention can automatically generate an interactive 3D integrated landscape model that integrates parametric building information model and digital terrain model from specific design schemes selected from the Pareto optimal solution set. The interactive 3D integrated landscape model is not only geometrically accurate, but also provides designers and decision-makers with an intuitive and immersive way to review schemes through scene rendering and interactive presentation. It facilitates the examination of design effects from any angle, thereby promoting communication, accelerating decision-making, and reducing the cost of modifications caused by misunderstandings in the later stages. Attached Figure Description

[0043] Figure 1 This is a flowchart of the method of the present invention;

[0044] Figure 2 This is a schematic diagram of the multi-site cloud data registration process of the present invention;

[0045] Figure 3 This is a flowchart illustrating the point cloud semantic segmentation method based on geometric features of the present invention.

[0046] Figure 4 This is a schematic diagram of the process for vegetation individual segmentation and key parameter extraction in this invention.

[0047] Figure 5 This is a flowchart illustrating the ecological performance index quantification method of the present invention;

[0048] Figure 6 This is a flowchart illustrating the method for quantifying aesthetic quality indicators according to the present invention.

[0049] Figure 7 This is a schematic diagram illustrating the mathematical modeling process of the multi-objective optimization problem in this invention;

[0050] Figure 8 This is a flowchart illustrating the Pareto optimal solution set solution method of the present invention;

[0051] Figure 9 This is a schematic diagram of the parametric modeling process of the present invention;

[0052] Figure 10 This is a flowchart illustrating the integrated landscape model generation and visualization method of the present invention;

[0053] Figure 11 This is the complete process of the semantic segmentation stage of the present invention;

[0054] Figure 12 This is a visualization result of the ground point cloud subset extracted from the site's three-dimensional point cloud model according to the present invention;

[0055] Figure 13 This is an overview of the site semantic segmentation results of the present invention. Detailed Implementation

[0056] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] See attached document Figure 1 This invention provides a method for establishing a garden landscape model based on data fusion technology. In a specific embodiment, the method can be executed by a site information acquisition device, which includes multiple functional units that are interconnected or communicate with each other.

[0058] The method in this embodiment first performs the step of collecting multi-source spatial information of the site, which is jointly executed by the spatial data acquisition unit and the texture data acquisition unit.

[0059] The function of the spatial data acquisition unit is to acquire the three-dimensional spatial geometric information of the target garden landscape site. In a specific implementation, the spatial data acquisition unit works in collaboration with two different types of laser scanning devices. First, multiple measuring stations are set up in the open area of ​​the preset garden landscape site. The surrounding environment is scanned at each measuring station using a terrestrial three-dimensional laser scanner (TLS) to generate a first original point cloud dataset. Each first original point cloud dataset is located in the independent local coordinate system of its corresponding measuring station.

[0060] Next, in the garden landscape site, for blind spots generated by the ground 3D laser scanner, or complex areas with dense vegetation and large terrain undulations, the spatial data acquisition unit uses a handheld laser scanner (HLS) to perform mobile supplementary scanning to generate a second original point cloud dataset, which is also located in its own local coordinate system.

[0061] Through the above operations, the spatial data acquisition unit ultimately outputs a raw point cloud dataset. ,in , The total number of datasets obtained from all station scans and handheld scans, for each raw point cloud dataset. All of them contain the three-dimensional coordinate information of objects within a local area of ​​the site.

[0062] The function of the texture data acquisition unit is to acquire color texture information corresponding to the three-dimensional spatial geometric information. In one specific implementation, the texture data acquisition unit uses a digital camera to collect the scene corresponding to the scanning range of each station of the ground-based 3D laser scanner and the scanning path of the handheld laser scanner, generating a high-resolution texture image set. ,in , This collection of high-resolution texture images represents the total number of images captured. This provides data input for subsequent texture mapping and color analysis of the 3D model.

[0063] See attached document Figure 2 After obtaining a collection containing multiple raw point cloud datasets Subsequently, the method of the present invention performs a multi-site cloud data registration step, which is performed by a data registration unit. Its function is to transform and integrate all the original point cloud datasets that are in their own independent local coordinate systems into a unified global coordinate system to generate a complete three-dimensional point cloud model of the site.

[0064] In one specific implementation, the data registration unit adopts a two-stage registration strategy from coarse to fine. First, an original point cloud dataset is selected, for example: As a reference, its coordinate system is defined as the target global coordinate system. Then, another original point cloud dataset to be registered is selected, for example: , as the source point cloud.

[0065] The first stage of the two-stage registration strategy, from coarse to fine, is coarse registration. The data registration unit identifies and matches corresponding feature points in the source and target point clouds, and calculates the initial transformation matrix. Specifically, the operator performs the initial transformation on the source point cloud... and target point cloud In the visualization interface, manually select at least three non-collinear pairs of corresponding points, such as building corners or fixed landmarks. Based on the coordinates of these pairs of corresponding points, calculate an initial rotation matrix using the least squares method. Translation vector Apply the initial transformation matrix to the source point cloud To roughly align it with the target point cloud The location.

[0066] The second stage of the coarse-to-fine two-stage registration strategy is fine registration. After coarse registration, the data registration unit executes the iterative closest point algorithm to iteratively optimize the pose of the source point cloud to achieve higher registration accuracy. The goal of the iterative closest point algorithm is to find an optimal fine rotation matrix. Translation vector The objective function is to minimize the root mean square error between the transformed source point cloud and the target point cloud. Defined as:

[0067] ;

[0068] in, It is a point in the source point cloud after the initial transformation. In the target point cloud, through Tree and other data structures can be used to quickly search for points. The closest corresponding point in space, This represents the number of valid point pairs that were successfully matched in this iteration.

[0069] The specific iterative steps of the data registration unit performing the iterative nearest point algorithm include:

[0070] Establish the corresponding point set: For a point in the source point cloud after the initial transformation Find the nearest corresponding point in the target point cloud. This forms an initial set of corresponding point pairs.

[0071] Eliminate incorrect matches: Calculate each pair of points The Euclidean distance between them is determined, and a distance threshold is set. If the distance between a pair of points is greater than this threshold If it is a mismatch, it is considered an incorrect match and removed from the corresponding point set, resulting in a set containing... The set of optimized points for each pair of valid points.

[0072] Calculate the transformation matrix: Based on the optimization point set, use the singular value decomposition method to solve for the objective function. Minimize the rotation matrix Translation vector .

[0073] Apply the transformation and check for convergence: calculate the transformation matrix. and Applied to the source point cloud, the convergence condition is then checked. The convergence condition can be set as follows: the change in the transformation matrix between two iterations is less than a preset threshold. or objective function The change is less than the preset threshold Or the preset maximum number of iterations has been reached. If the convergence condition is not met, the spatial data and texture data acquisition will continue to be carried out; if the condition is met, the iteration will terminate.

[0074] The aforementioned two-stage registration process, from coarse to fine, is applied sequentially to each of the remaining original point cloud datasets, for example, in... and After registration, the registered combined point cloud is used as the new target point cloud, and then... As the source point cloud, registration is performed, and so on, until the set is completed. All of them All original point cloud datasets were integrated into the global coordinate system.

[0075] The data registration unit ultimately outputs a single, coordinate-uniform, and complete 3D point cloud model of the site. This provides a data foundation for subsequent landscape element segmentation and parametric extraction steps.

[0076] See attached document Figure 3 and Figure 11 A complete 3D point cloud model of the site was generated in the data registration unit. Subsequently, the method of the present invention performs a semantic segmentation step of landscape elements, which is performed by a semantic segmentation unit whose function is to create a complete three-dimensional point cloud model of the site. Each point in the cloud is assigned a predefined landscape feature category label, thereby decomposing the unordered point cloud data into multiple point cloud subsets with semantic information.

[0077] In one specific implementation, the semantic segmentation unit first processes the complete site 3D point cloud model. Each point in Calculate its local geometric features; for this, for each point... The k-nearest neighbor search algorithm is used to determine its surrounding neighbors. A set of valid point pairs constitutes a local neighborhood point set. Based on local neighborhood point sets Calculate the following set of geometric feature descriptors:

[0078] normal vector By analyzing the local neighborhood point set Principal component analysis is performed to calculate the covariance matrix. The eigenvector corresponding to the smallest eigenvalue of the covariance matrix is ​​defined as a point. Normal vector: .

[0079] curvature Using the three eigenvalues ​​of the covariance matrix (satisfy To calculate curvature Defined as: .

[0080] Planarity and linearity : Also calculated based on three eigenvalues, planarity and linearity They are defined as follows: and .

[0081] Elevation characteristics Calculation point The Z-coordinate value and its local neighborhood point set The difference of the minimum Z coordinate values ​​of all points within the area is used as its normalized elevation feature.

[0082] Subsequently, the semantic segmentation unit combines the calculated geometric feature descriptors into a multidimensional feature vector. For each point Its eigenvectors Represented as: .

[0083] After obtaining the feature vectors of all points, the semantic segmentation unit classifies each point using a trained semantic classification model. In one specific implementation, the semantic classification model is a random forest classifier. The training process of the semantic classification model is as follows: First, from the complete site 3D point cloud model... A representative subset is selected as the training dataset. Then, each point in the training dataset is manually labeled with its true category label, such as ground, low vegetation, tall vegetation, building, water body, etc. Finally, the feature vector of each point in the training dataset and its corresponding true category label are input into the random forest algorithm for training to generate a trained semantic classification model.

[0084] When performing classification, for the complete site 3D point cloud model Each point in The semantic segmentation unit sets its feature vector The feature vector is input into a trained semantic classification model, which then uses its internal set of decision trees to classify the feature vector and outputs a predicted class label. The process iterates through All points in the.

[0085] After classifying all points, the semantic segmentation unit uses the category label obtained for each point. The complete site 3D point cloud model Divided into multiple independent subsets of point clouds, including the ground point cloud subset. Vegetation Cloud Subset Building point cloud subset These semantically labeled subsets of point clouds will serve as data inputs for subsequent parametric extraction and modeling steps.

[0086] See attached document Figure 4 After the semantic segmentation unit generates a subset of point clouds with semantic labels, the method of this invention performs vegetation individualization segmentation and key parameter extraction steps. These steps are performed by a parameter extraction unit, whose function is to first segment the vegetation point cloud as a whole. The data is divided into multiple individual point cloud clusters representing individual trees, and then the key growth morphology parameters of each individual point cloud cluster are calculated.

[0087] In one specific implementation, the parameter extraction unit first processes a subset of the vegetation point cloud. The tree height calculation is performed by normalizing the height to eliminate the influence of terrain undulations. This process utilizes the ground point cloud subset segmented in the previous step. The parameter extraction unit extracts a subset of the ground point cloud. Construct a digital terrain model, and then, for the vegetation point cloud subset... Each point in By querying the digital terrain model using its horizontal (X, Y) coordinates, the corresponding ground elevation can be obtained. The height normalized value of this point Calculated as:

[0088] ;

[0089] in, It is a point The original Z-coordinates, the parameter extraction unit normalizes the height values ​​of all vegetation points. Projected onto a two-dimensional horizontal grid, a canopy height model is generated. The value of each cell in the canopy height model grid represents the maximum normalized height of vegetation points within the grid area.

[0090] Next, the parameter extraction unit performs individual tree segmentation based on the generated canopy height model. In one specific implementation, the watershed segmentation algorithm is used. First, the canopy height model grid is inverted, so that the vertices of the canopy (local maximum points in the original canopy height model) become local minimum points in the new grid, i.e., the bottom of the catchment basin. Then, the watershed algorithm starts from these local minimum points and simulates the flooding process. During the flooding process, at the points where the expanded areas of different catchment basins meet, the watershed segmentation algorithm constructs watershed lines. These watershed lines constitute the boundaries between different canopies. The result of the watershed segmentation algorithm is the generation of a set of two-dimensional polygons, each polygon representing the projected outline of an independent canopy on the horizontal plane.

[0091] After obtaining the two-dimensional contour polygons of the tree canopy, the parameter extraction unit correlates these two-dimensional contours back to a three-dimensional subset of vegetation point clouds. For each two-dimensional contour polygon, the parameter extraction unit filters out all vegetation point clouds whose horizontal coordinates (X, Y) fall inside the polygon, forming an independent single point cloud cluster. This process traverses all two-dimensional contour polygons, ultimately generating a set of individual point cloud clusters. ,in It represents the total number of trees that were successfully divided within the site.

[0092] Finally, the parameter extraction unit performs parameter extraction on each individual point cloud cluster. Calculate its key parameters:

[0093] (1) Tree height ( ): ;

[0094] in, It is a single point cloud cluster The maximum Z-coordinate value of all points, This is obtained by querying a digital terrain model, which shows the ground elevation corresponding to the location of the tree's root base. The location of the root base can be determined by... The horizontal position of the point with the smallest Z-coordinate is determined.

[0095] (2) Crown diameter ( First, divide the individual point cloud clusters into groups. All points are projected onto a two-dimensional XY plane to form a two-dimensional point set. Then, the minimum circumcircle of the two-dimensional point set is calculated. The diameter of the minimum circumcircle is defined as the crown diameter of the tree. .

[0096] The parameter extraction unit ultimately outputs the geometric and positional parameters of each identified individual tree, including its center coordinates and tree height. Crown diameter These parameterized vegetation information will serve as input data for subsequent landscape assessment and optimization steps.

[0097] See attached document Figure 5 , Figure 12 and Figure 13 After obtaining parameterized landscape element information, the method of the present invention performs the step of constructing a multi-dimensional quantitative evaluation model of garden landscape. In a specific embodiment, the step of constructing a multi-dimensional quantitative evaluation model of garden landscape first involves the ecological evaluation unit performing quantitative calculations of ecological performance indicators. The function of the ecological evaluation unit is to receive a garden landscape design scheme to be evaluated and, based on a preset ecological model, calculate and output a set of numerical indicators characterizing the ecological performance of the scheme.

[0098] The landscape design proposals to be evaluated include: the total area of ​​the site. The plan also includes the location, geometry, and species information of all vegetation elements.

[0099] The ecological assessment unit first performs vegetation cover rate calculation. The vegetation coverage rate index is used to quantify the degree to which vegetation covers the site surface. The calculation process includes:

[0100] (1) First, identify the horizontal projection area of ​​all vegetation elements in the design scheme. For arbor vegetation, its horizontal projection area is determined by the diameter of its crown. The calculation is for a circular area; for shrubs or grasslands, the horizontal projection area is determined according to the boundary polygon defined in the design scheme.

[0101] (2) Then, calculate the total area of ​​the union of all these independent horizontal projection areas, which is done through geometric operations to ensure that the overlapping area between vegetation (e.g., the canopy of trees and the shrubs below) is not counted repeatedly. This total area of ​​the union is defined as the effective vegetation cover area. .

[0102] (3) Finally, calculate the effective vegetation cover area. Total site area The ratio of the two values ​​gives the vegetation coverage. : ;

[0103] The ecological assessment unit then performs the species diversity index. The Shannon-Wiener index is used to quantify the richness and evenness of plant species distribution in a design scheme. In one specific implementation, the Shannon-Wiener index is used for calculation, and the calculation process includes:

[0104] First, iterate through all the individual plants in the design scheme and count the total number of plant species included in the scheme. And count the number of each species Number of individuals .

[0105] Then, calculate the total number of individual plants in the scheme. : ;

[0106] Next, calculate each species relative abundance That is, the proportion of individuals of this species to the total number of individuals: ;

[0107] Finally, the relative abundance of all species was calculated. Substituting the values ​​into the Shannon-Wiener index calculation formula, we obtain the species diversity index. : ;

[0108] After completing the above calculations, the ecological assessment unit outputs a numerical vegetation cover rate. and species diversity index These two indicators will serve as objective criteria for evaluating the ecological performance of the landscape design scheme and as one of the input parameters for subsequent optimization steps.

[0109] See attached document Figure 6 After the ecological assessment unit completes the quantification of ecological performance indicators, or in parallel with it, the method of the present invention then performs the quantification calculation of aesthetic quality indicators by the aesthetic assessment unit. The function of the aesthetic assessment unit is to receive the same landscape design scheme to be evaluated, and calculate and output a set of numerical indicators that characterize the aesthetic composition quality of the scheme according to the preset aesthetic model.

[0110] In addition to vegetation information, the landscape design scheme to be evaluated also includes the location and geometric dimensions of non-vegetated landscape elements such as garden paths, water bodies, and landscape structures.

[0111] The aesthetic evaluation unit first performs a test on the conformity of the golden ratio. The calculation of the golden ratio conformity index is used to quantify the size proportions of key visual elements in a design scheme relative to the golden ratio. In one specific implementation, the calculation process for the degree of proximity includes:

[0112] Key element identification: First, identify a set of predefined key aesthetic elements from the design scheme, such as the length and width of a specific garden path, the length and width of a rectangular water feature, and the length, width, and height dimensions of a landscape pavilion.

[0113] Ratio Calculation: For each identified key element, calculate the ratio of its two principal dimensions. For example, for a rectangular body of water, its ratio It is the ratio of its length to its width.

[0114] Single-element compliance score: for each calculated ratio Its relationship with the golden ratio is calculated using an exponential decay function. (its value is) (Compliance score) :

[0115] ;in, It is a positive sensitivity coefficient used to define the rate at which the score decays with proportional deviation. The larger the value, the lower the tolerance for deviations from the golden ratio.

[0116] Overall compliance aggregation: compliance scores for all key elements. The final golden ratio conformity is obtained by performing a weighted average. : ;

[0117] in, It is the total number of all the key elements being evaluated. It is the first The weight of each element can be set according to its visual importance in the overall landscape.

[0118] The aesthetic evaluation unit then performs color harmony assessment. The calculation of the color harmony index is used to quantify whether the color matching of the main ornamental plants in the scheme during their peak flowering or main viewing period conforms to the preset color harmony theory. In a specific implementation, the calculation process includes:

[0119] Dominant color extraction: For each species designated as the main ornamental plant in the scheme, its corresponding reference texture image (derived from the high-resolution texture image set collected in Part 1) is used. The color regions of the flowers or leaves are analyzed using the K-means clustering algorithm to extract the main color tone of the plant and represent it in the form of RGB color values.

[0120] Color space conversion: Convert all extracted RGB primary colors to a representation in the CIELAB color space, which includes luminance. ,as well as (Red-Green Components) and (Yellow-Blue component) Two chromaticity components, this conversion process is completed using standard color space conversion formulas.

[0121] Harmony evaluation: The main color scheme of any two main ornamental plants in the design. and The representations in CIELAB space are as follows: and Through a harmony evaluation function Calculate the color harmony scores between them; for example, for the complementary color harmony model, the evaluation function... It can be defined as:

[0122] ;

[0123] in, and These are weighting coefficients. The evaluation function reaches its maximum value when the chromaticity components of the two colors are opposites of each other and their brightness is close. Similarly, evaluation functions for other harmonious models such as adjacent colors and ternary colors can be defined.

[0124] Overall Harmony Aggregation: The color harmony scores of all the main ornamental plants in the scheme are averaged in pairs to obtain the final color harmony score. .

[0125] After completing the above calculations, the aesthetic evaluation unit outputs a numerical value for the golden ratio conformity. and color harmony These two indicators will serve as objective criteria for evaluating the aesthetic quality of the garden landscape design scheme.

[0126] See attached document Figure 7 After the ecological assessment unit and the aesthetic assessment unit respectively quantify the ecological and aesthetic performance of the design scheme, the method of the present invention performs a landscape design scheme generation and adjustment step based on multi-objective optimization. The landscape design scheme generation and adjustment step is first performed by a mathematical modeling unit, whose function is to abstract and transform the complex decision-making process of garden landscape design into a structured multi-objective optimization problem that can be solved by an algorithm.

[0127] In one specific implementation, the mathematical modeling unit first defines the design variable vector of the design scheme. The design variable vector is a set of all parameters that can be adjusted by the optimization algorithm to fully describe a design scheme. It is constructed as a high-dimensional vector, and its specific components include:

[0128] Vegetation layout variables: for the preset vegetation layout variables in the site One possible planting site, for each planting site ( Allocate an integer variable , used to indicate the plant species planted at that point, This indicates that no plants should be planted at this location, and This indicates that the species being planted is numbered in the species bank. Plants.

[0129] Geometric variables of hardscape elements: For hardscape elements such as water bodies, garden paths, or structures, their geometric shape is defined by a set of coordinate variables. For example, for a garden path defined by a B-spline curve, the variables are the two-dimensional or three-dimensional coordinates of all its control points. ;

[0130] For a rectangular water body, the variables are its center point coordinates, length, and width. All these variables are concatenated to form the final design variable vector. Next, the mathematical modeling unit defines the objective function. The objective function is used to quantify a vector of design variables. The merits of the determined design scheme are determined by the fact that ecological performance and aesthetic quality are two goals that need to be achieved simultaneously. Therefore, the objective function is defined as a two-dimensional vector function, whose goal is to maximize both components at the same time.

[0131] ;

[0132] The first component The total ecological performance score is defined as the weighted sum of the ecological indicators calculated by the aforementioned ecological assessment units: ;

[0133] Second component The overall aesthetic quality score is defined as the weighted sum of the various aesthetic indicators calculated by the aforementioned aesthetic evaluation units: ;

[0134] Here, All weighting coefficients are positive, and their values ​​can be preset according to the design focus of the specific project. All of these indicate that their values ​​depend on the design variable vector. The calculation results.

[0135] Finally, the mathematical modeling unit defines a set of constraints. Any valid design solution must satisfy all constraints, including:

[0136] Cost constraint: The total cost of the design scheme cannot exceed the pre-set project budget. Total cost function The constraint, summed from the procurement costs of all plants and the construction costs of all hardscape elements in the plan, is expressed as: ;

[0137] Ecological suitability constraints: For species numbered in the species bank in the scheme... For plants, the environmental factors of the planting location (such as slope and aspect extracted from the digital terrain model, as well as preset light conditions) must be within the suitable range for that species.

[0138] Spatial layout constraints:

[0139] Any two plant individuals and The distance between them must be greater than the minimum safe spacing determined by the canopy width at their respective maturity stages. .

[0140] All geometric entities of all design elements (plants, water features, garden paths, etc.) must not intrude into predefined restricted areas (such as buildings, existing facilities, etc.).

[0141] The width and slope of garden paths must meet the specifications for pedestrian or barrier-free design.

[0142] Through the above steps, the mathematical modeling unit completely transforms the landscape design problem into a well-defined multi-objective optimization mathematical model that includes design variables, objective functions, and constraints.

[0143] See attached document Figure 8 After the mathematical modeling unit transforms the landscape design problem into a well-defined multi-objective optimization model, the method of the present invention is then executed by an optimization solving unit. The function of the optimization solving unit is to receive the mathematical model and apply a multi-objective evolutionary algorithm to search for and generate a set of landscape design schemes that perform well in both ecological and aesthetic objectives and are mutually independent.

[0144] In one specific implementation, the optimization unit employs a non-dominated sorting genetic algorithm to solve the multi-objective optimization problem. The algorithm's execution process includes the following steps:

[0145] Population initialization: The optimization solving unit first generates an initial population. It contains Each individual is a vector of design variables as defined above. In this example, the generation process of each individual in the population is as follows: its value is randomly generated within the domain of the design variable, and immediately passed through the constraints. During verification, any randomly generated individual that does not satisfy all constraints will be discarded and regenerated until an individual that fully satisfies all constraints is obtained. This process is repeated. Next, the initial population is formed. .

[0146] Iterative evolution process: The optimized solution unit then executes an iterative loop, with an algebraic counter. Starting from 0, in each generation, perform the following operations:

[0147] Offspring population generation: based on the current parent population Generate a size of the same offspring population This process is accomplished by repeatedly performing selection, crossover, and mutation operations:

[0148] Selection: A binary tournament selection method was used from... Two parent individuals are selected based on their non-dominance level and crowding level (these two concepts will be defined later).

[0149] Crossover: Perform a crossover operation on the two selected parent individuals to generate two offspring individuals. This is relevant to the design variable vector. For the integer variable part representing plant species, uniform crossover is used; for the continuous variable part representing geometric coordinates, simulated binary crossover is used.

[0150] Mutation: Perform a mutation operation on the generated offspring individuals with a preset low probability (mutation rate). For integer variables, randomly change to another valid species number; for continuous variables, use polynomial mutation.

[0151] Each generated offspring also needs to be subject to constraints. Verification is required to ensure its validity.

[0152] Merging and Evaluation: Merging parent populations and offspring population Merge into a size of Combined populations ,for For each individual in the equation, calculate its performance in both objective functions. and The rating value on the platform.

[0153] Non-dominated ordering: for combined populations Perform a non-dominated sort, dividing the region into multiple disjoint non-dominated fronts. The sorting rule is: First, find... All individuals not dominated by any other individual constitute the first frontier. Then from Remove from The individuals are identified, and from the remaining individuals, all non-dominated individuals are identified to form the second frontier. And so on, until... All individuals are assigned to a frontier. Each individual thus acquires a non-dominant rank, with lower ranks indicating a more superior individual.

[0154] Crowding calculation: In order to maintain population diversity among individuals of the same non-dominant rank, for each frontier... Individuals within the cluster calculate their crowding distance for the frontier. For an individual in the solution set, its crowding degree is the sum of the side lengths of the cubes formed by its two neighboring individuals in each objective function dimension. Individuals located on the edge of the solution set have their crowding degree set to infinity.

[0155] Elite selection: Based on non-dominance hierarchy and crowding, from composite populations Select Individuals constitute the new paternal population of the next generation. The selection process is as follows: ranked from lowest to highest frontier level (i.e., from...). (Start) Place the individuals from the entire front edge into the system sequentially. until The scale reached If one joins a certain cutting-edge Afterwards, the population size exceeded That would be incorrect. Instead of adding it all at once, add it based on... Sort the individuals by crowding in descending order and select the individuals with higher crowding to fill in the blanks. The remaining empty spaces, until the scale reaches .

[0156] Termination and Output: The above iterative evolution process continues until a preset termination condition is met, such as the number of iterations. Reaching the maximum preset value When the algorithm terminates, the optimization unit will determine the first non-dominated front in the final population. As the result, output it.

[0157] The output is a Pareto optimal solution set, which contains multiple different landscape design schemes. These schemes are all effective and exhibit different trade-offs between ecological and aesthetic objectives. This solution set will be provided to designers as a high-quality and diverse set of alternative design schemes.

[0158] See attached document Figure 9 After the optimization unit generates a Pareto optimal solution set containing multiple alternative design schemes, the method of this invention performs parametric modeling and 3D visualization steps for the design schemes. Parametric modeling and 3D visualization are first performed by a parametric modeling unit, whose function is to receive a vector of design variables selected from the Pareto optimal solution set. The specific design scheme is represented and transformed into a building information model containing parametric information.

[0159] In one specific implementation, the execution of the parametric modeling unit relies on a pre-built library of building information model parametric families stored in a database. The construction process of the parametric family library includes:

[0160] Define plant parameter families: Create a corresponding parameter family file for each plant in the species library. Each plant parameter family is geometrically composed of a component representing the trunk and a component representing the crown. The parameter family file contains a set of predefined type parameters and instance parameters, which correspond to the parameters extracted or calculated in the preceding steps. Specifically, the parameters include:

[0161] A length type instance parameter named tree height, the value of which controls the height of the tree trunk component;

[0162] A length type instance parameter named crown diameter, whose value is used to control the diameter or range of the crown component;

[0163] A length type instance parameter named "height above ground" has a value used to set the height of the bottom of the canopy component above the ground.

[0164] An integer parameter named Species ID, whose value matches the plant's ID in the species database;

[0165] A currency type parameter called Unit Cost, whose value represents the cost of purchasing or cultivating this species of plant.

[0166] Define hardscape parametric families: Create corresponding parametric family files for hardscape elements such as paths and water features. For example, a path-based family contains a length type parameter named path width; a rectangular water feature family contains two length type parameters named length and width.

[0167] Upon receiving a selected design scheme (i.e., a specific design variable vector) After that, the parametric modeling unit performs the following application steps:

[0168] First, the parametric modeling unit parses the design variable vector. Extract the location information and species ID of all plant elements. And the geometric definition information of all hardscape elements.

[0169] Then, for each plant element, the parametric modeling unit performs the following operations:

[0170] According to Species IDs extracted from the data The corresponding plant parameter family file is retrieved and located in the BIM parameter family library.

[0171] At the specified coordinates location in the BIM project environment, load and create an instance of the plant parametric family.

[0172] From the calculation results of the aforementioned parameter extraction unit, obtain the tree height corresponding to this individual plant. and crown diameter The value.

[0173] The obtained tree height value Assign the value to the parameter named tree height in the newly created family instance; then retrieve the obtained crown diameter value. The value is assigned to a parameter called crown diameter. Through this assignment operation, the geometry of the family instance is automatically adjusted to match the actual morphological parameters of the individual plant.

[0174] For each hardscape element, such as a garden path defined by control points, the parametric modeling unit performs the following operations:

[0175] Based on the type of garden path, the corresponding garden path parametric family file is retrieved and located in the BIM parametric family library.

[0176] Create an instance of the garden path family within the Building Information Modeling project environment.

[0177] from The system parses the control point coordinate sequence that defines the center line of the garden path and assigns this sequence to the path definition parameters of the garden path family instance, thereby generating the geometric path of the garden path.

[0178] from The width value of the garden path is parsed out and assigned to the parameter named path width in the garden path family instance.

[0179] By repeatedly performing the above operations on all elements in the design scheme, the parametric modeling unit finally generates a set of landscape element models in the building information modeling environment. These models are precisely placed and their geometric and non-geometric properties are precisely controlled by parameters. These models constitute a digital and parametric expression of the optimized design scheme.

[0180] See attached document Figure 10 After the parametric modeling unit generates a building information model containing a set of parametric landscape elements, the method of the present invention is finally executed by a visualization unit to generate and visualize the integrated landscape model. The function of the visualization unit is to integrate all the separate landscape model elements and generate a three-dimensional digital scene that can be interactively viewed by the user.

[0181] In one specific implementation, the execution process of the visualization unit includes the following steps:

[0182] Terrain surface model generation: The visualization unit first receives a subset of ground point clouds generated in the semantic segmentation step. Based on a subset of ground point clouds, the Delaunay triangulation algorithm is applied to the unit to connect discrete ground points into a continuous, irregular triangular network composed of multiple triangular patches. The triangular network constitutes the basic digital terrain model of the site.

[0183] Model Integration and Localization: The visualization unit then integrates all parametric family instances (including vegetation and hardscape) generated by the parametric modeling unit into a model containing the terrain. In the unified 3D scene, for each family instance, the cell performs a precise vertical positioning operation: it obtains the horizontal base point coordinates (X, Y) of the instance, and then... The triangular facet containing the base point is retrieved from the triangular mesh, and the precise elevation Z value of the base point on the terrain surface is calculated by linear interpolation. The calculated Z value is set as the reference elevation of the family instance to ensure that all landscape elements are accurately placed on the terrain surface.

[0184] Material and texture mapping: In order to generate visually realistic models, the visualization unit performs material assignment and texture mapping operations.

[0185] Terrain texture mapping: The unit receives a set of high-resolution texture images acquired during the information acquisition step. And the associated camera parameters (interior and exterior orientation elements), through the application of a texture projection algorithm, precisely project and fit these two-dimensional texture images onto the three-dimensional terrain model. The above gives the terrain surface realistic colors and details.

[0186] Element Material Assignment: For each parametric family instance, the unit assigns a material from a preset material library based on its species or type information. For example, for a birch plant family instance, the trunk component is assigned a birch bark material containing a normal map and a color map, and the crown component is assigned a birch leaf material containing an opacity channel. For a water family instance, a water material with specific refractive index, reflectivity, and transparency is assigned.

[0187] Scene rendering and interactive presentation: After model integration and material assignment are completed, the visualization unit renders the entire scene in real time within a 3D rendering engine. This process includes:

[0188] Lighting environment settings: Configure at least one light source in the scene, such as a parallel light source to simulate sunlight and define its direction vector, color and intensity; and an ambient light to simulate ambient diffuse light to define the overall brightness of the scene.

[0189] Virtual camera configuration: Define a virtual camera in the scene and provide a user interface to control the parameters of the virtual camera, enabling interactive browsing of the scene. The control functions provided include:

[0190] Translation: Moving the camera position in a plane perpendicular to the camera's line of sight in response to specific user input (e.g., pressing the middle mouse button and moving it).

[0191] Orbit: Responds to specific user input (e.g., pressing the left mouse button and moving) to rotate the camera around a preset focus or scene center point.

[0192] Zooming: Responding to specific user input (e.g., scrolling the mouse wheel) causes the camera to move forward or backward along the user's line of sight, thus zooming in or out of the scene.

[0193] Through the above steps, the visualization unit ultimately outputs a three-dimensional integrated landscape model on a display device, which integrates terrain, vegetation, and hardscape, has complete material and lighting effects, and allows users to freely navigate and examine it from any perspective.

Claims

1. A method for establishing a garden landscape model based on data fusion technology, characterized in that, Includes the following steps: S1. Acquisition of multi-source spatial information of the site, obtaining a set of original point cloud datasets and a set of high-resolution texture images representing the target garden landscape site; S2. Registration of multi-site cloud data: The original point cloud dataset is integrated to generate a complete 3D point cloud model of the site. S3. Semantic segmentation of landscape elements: Assign landscape element category labels to points in the complete site 3D point cloud model, and decompose them into multiple point cloud subsets including ground point cloud subset and vegetation point cloud subset. S4. Vegetation individual segmentation and key parameter extraction: The vegetation point cloud subset is segmented into multiple individual point cloud clusters, and key growth morphology parameters are calculated. S5. Construct a multi-dimensional quantitative evaluation model for garden landscapes, and calculate ecological performance indicators and aesthetic quality indicators based on key growth morphology parameters and preset information. S6. Generation and adjustment of landscape design schemes based on multi-objective optimization: The landscape design problem is transformed into a multi-objective optimization mathematical model with ecological performance indicators and aesthetic quality indicators as objective functions, and the multi-objective optimization mathematical model is solved to generate a Pareto optimal solution set. S7. Parametric modeling and 3D visualization of the design scheme will generate an interactive 3D integrated landscape model from the specific design scheme selected from the Pareto optimal solution set.

2. The method for establishing a garden landscape model based on data fusion technology according to claim 1, characterized in that, The S2 step specifically includes: Coarse registration is performed between different original point cloud datasets within the set of original point cloud datasets by identifying and matching feature points with the same name and by calculating the initial transformation matrix. After coarse registration, the iterative nearest point algorithm is applied to perform pose optimization on the original point cloud dataset that has been coarsely registered and aligned, and then fine registration is performed.

3. The method for establishing a garden landscape model based on data fusion technology according to claim 1, characterized in that, The S3 step specifically includes: Local geometric features of points in a complete 3D point cloud model of the site are calculated and combined into multi-dimensional feature vectors. Using a trained semantic classification model, the multidimensional feature vector of each point is classified to output the predicted class label. Based on the predicted category labels in the output, points with the same category labels are grouped together, and the site's 3D point cloud model is decomposed into multiple point cloud subsets, including ground point cloud subsets and vegetation point cloud subsets.

4. The method for establishing a garden landscape model based on data fusion technology according to claim 1, characterized in that, The S4 step specifically includes: A canopy height model is generated using a subset of vegetation point clouds and a subset of ground point clouds; Based on the generated canopy height model, a watershed segmentation algorithm is used to generate two-dimensional polygons representing the projection contours of individual tree canopies; The generated two-dimensional polygons are associated back with the three-dimensional vegetation point cloud subsets to form independent individual point cloud clusters, and geometric analysis is performed on the formed independent individual point cloud clusters to calculate the key growth morphology parameters.

5. The method for establishing a garden landscape model based on data fusion technology according to claim 1, characterized in that, In step S5, the calculation of ecological performance indicators and aesthetic quality indicators specifically includes: The ecological performance indicators include: determining the vegetation coverage rate by calculating the ratio of the vertical projection area of ​​the vegetation canopy to the total area of ​​the site, and calculating the species diversity index by applying the Shannon-Wiener index; The aesthetic quality indicators include: determining the conformity of the golden ratio by calculating the deviation between the key dimensions of the landscape and the golden ratio value, and calculating the color harmony based on the harmony rules of Eaton's color theory.

6. The method for establishing a garden landscape model based on data fusion technology according to claim 1, characterized in that, In step S6, solving the multi-objective optimization mathematical model specifically involves applying a non-dominated sorting genetic algorithm to transform ecological performance indicators and aesthetic quality indicators into a Pareto optimal solution set.

7. The method for establishing a garden landscape model based on data fusion technology according to claim 1, characterized in that, The S7 step specifically includes: The specific design schemes in the Pareto optimal solution set are generated and then transformed into building information models containing parametric information. By integrating the building information model and digital terrain model containing parametric information, scene rendering and interactive presentation are performed to generate the interactive three-dimensional integrated landscape model.

8. A garden landscape model building device based on data fusion technology, characterized in that, The data acquisition module is used to acquire a set of original point cloud datasets and a set of high-resolution texture images representing the target garden landscape site; The data processing module is used to integrate the original point cloud dataset to generate a complete 3D point cloud model of the site, assign landscape element category labels to the points in the model, decompose the model into multiple point cloud subsets, then segment the point cloud subsets to form multiple individual point cloud clusters, and calculate key growth morphology parameters. The evaluation modeling module is used to calculate ecological performance indicators and aesthetic quality indicators based on key growth morphology parameters, and to construct a multi-objective optimization mathematical model with ecological performance indicators and aesthetic quality indicators as objective functions. The optimization solution module is used to solve multi-objective optimization mathematical models and generate Pareto optimal solution sets. The visualization module is used to generate an interactive 3D integrated landscape model from specific design schemes selected from the Pareto optimal solution set.

9. The garden landscape model building device based on data fusion technology according to claim 8, characterized in that, The assessment modeling module specifically includes an ecological assessment unit and an aesthetic assessment unit; The optimization solution module uses a non-dominated sorting genetic algorithm to solve the multi-objective optimization mathematical model and generate a Pareto optimal solution set.

10. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a computer program, wherein when executed by a processor, the computer program implements the method as described in any one of claims 1 to 7.

Images