Cross-source fusion modeling method and system for forest three-dimensional point cloud model

Abstract

The invention provides a cross-source fusion modeling method and system for a forest three-dimensional point cloud model. The method comprises the following steps: acquiring first information based on camera equipment carried by an unmanned aerial vehicle, wherein the first information comprises tree canopy layer image data above a preset height of a forest and forest underlayer image data below the preset height of the forest; performing image processing based on the first information, and preprocessing point cloud data obtained by processing to obtain preprocessed point cloud data; determining canopy layer forest data and understory forest data based on the preprocessed point cloud data, and obtaining position data of each forest; and carrying out primary registration on the position data of each forest based on an improved Delou triangulation network algorithm, and carrying out secondary registration on a primary registration result through an optimization function to obtain a forest three-dimensional point cloud model. According to the method, the efficiency and precision of forest three-dimensional modeling are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud data processing, and in particular to a cross-source fusion modeling method and system for a three-dimensional point cloud model of a forest. Background Art

[0002] Against the backdrop of global climate change and ecological degradation, forests, as important carbon sinks, require precise measurement of their structural parameters for ecological protection and sustainable management. Traditional vegetation monitoring methods rely primarily on ground-based measurements and remote sensing technologies. However, ground-based measurements are time-consuming, labor-intensive, and costly, making them difficult to apply on a large scale. While remote sensing technology can cover large areas, its resolution and accuracy are insufficient, making it difficult to meet the demands of precise measurement. In recent years, the development of three-dimensional modeling technology, particularly the application of lidar (LiDAR), has provided new avenues for vegetation structure monitoring. However, its high cost and limitations in complex terrain still hinder its widespread application.

[0003] The emergence of drone-based oblique photography and panoramic imaging technologies has provided an efficient and cost-effective solution for 3D forest modeling. Drones can quickly acquire high-resolution images of the tree canopy, while panoramic cameras can capture detailed information about the understory. However, effectively fusing these two perspectives and constructing a high-precision 3D forest model through precise point cloud processing and registration techniques remains a pressing technical challenge. Summary of the Invention

[0004] The purpose of the present invention is to provide a cross-source fusion modeling method and system for a forest 3D point cloud model to improve the above-mentioned problem. To achieve the above-mentioned purpose, the technical solution adopted by the present invention is as follows:

[0005] In a first aspect, the present application provides a cross-source fusion modeling method for a forest 3D point cloud model, comprising:

[0006] Acquiring first information based on a camera device carried by a drone, the first information including canopy image data above a preset height of trees and understory image data below the preset height of trees;

[0007] performing image processing based on the first information, and preprocessing the obtained point cloud data to obtain preprocessed point cloud data;

[0008] Determine canopy tree data and understory tree data based on the pre-processed point cloud data to obtain location data of each tree;

[0009] The position data of each tree is preliminarily registered based on the improved Delaunay triangulation algorithm, and the preliminarily registered result is secondarily registered through the optimization function to obtain a three-dimensional point cloud model of the tree.

[0010] In a second aspect, the present application further provides a cross-source fusion modeling system for a forest 3D point cloud model, characterized by including:

[0011] an acquisition unit, configured to acquire first information based on a camera device carried by a drone, wherein the first information includes image data of a canopy layer above a preset height of the trees and image data of an understory layer below the preset height of the trees;

[0012] a processing unit, configured to perform image processing based on the first information, and preprocess the point cloud data obtained by the processing to obtain preprocessed point cloud data;

[0013] an analysis unit, configured to determine canopy tree data and understory tree data based on the preprocessed point cloud data, and obtain position data of each tree;

[0014] The registration unit is used to perform a preliminary registration on the position data of each tree based on the improved Delaunay triangulation algorithm, and perform a secondary registration on the preliminary registration result through an optimization function to obtain a three-dimensional point cloud model of the tree.

[0015] The beneficial effects of the present invention are:

[0016] The present invention uses drones and panoramic cameras to acquire image data of the canopy and understory layers, respectively, and combines this with an improved Delaunay triangulation algorithm and further point-to-surface registration to construct a high-precision three-dimensional point cloud model of forest trees. This method effectively addresses the problems of low data acquisition efficiency, high cost, and insufficient accuracy in traditional technologies, significantly improving the integrity and accuracy of three-dimensional forest modeling. Furthermore, by optimizing the point cloud preprocessing and registration processes, the present invention further improves the efficiency and reliability of model construction, providing a more scientific and efficient technical approach for forest resource management and carbon sequestration capacity assessment, with broad application prospects and important ecological significance.

[0017] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the embodiments of the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1Schematic diagram of the process of the cross-source fusion modeling method of the forest 3D point cloud model according to an embodiment of the present invention;

[0020] Figure 2 Schematic diagram of the cross-source fusion modeling system structure of the forest three-dimensional point cloud model described in an embodiment of the present invention.

[0021] Markings in the figure: 701, acquisition unit; 702, processing unit; 703, analysis unit; 704, registration unit. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0023] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are used only to distinguish the description and should not be understood as indicating or implying relative importance.

[0024] Example 1:

[0025] like Figure 1 As shown, this embodiment provides a cross-source fusion modeling method for a forest three-dimensional point cloud model.

[0026] See also Figure 1 , the figure shows that the method includes step S1, step S2, step S3 and step S4.

[0027] Step S1: acquiring first information based on a camera device mounted on a drone, wherein the first information includes image data of a canopy layer above a preset height of trees and image data of an understory layer below the preset height of trees;

[0028] It can be understood that in this step, a complete data collection is completed within a specific time range. The canopy image data mainly contains the upper structure information of the forest, such as the tree crowns, branches and leaves, and vegetation coverage of the trees, while the understory image data mainly reflects the structure near the forest floor, including tree trunks, shrubs, and herbaceous plants. In order to obtain these data, the canopy images are mainly collected by professional drones equipped with photography equipment. The drones use continuous photography mode for aerial photography, set the flight altitude and route, and ensure that the overlap rate of each area reaches more than 70%, thereby improving the accuracy of image matching. The understory image data is obtained by a panoramic camera. The panoramic camera collects data along an S-shaped trajectory, records a 360-degree panoramic video containing GPS information, extracts static images from the video, and writes the GPS information into the corresponding image. The combination of these data provides a basis for comprehensive monitoring of forest ecosystems, and provides rich information support for subsequent three-dimensional modeling and analysis.

[0029] Step S2: performing image processing based on the first information, and preprocessing the obtained point cloud data to obtain preprocessed point cloud data;

[0030] It is understandable that the pre-processed point cloud data includes a dense point cloud of the canopy layer and a dense point cloud of the understory layer, ensuring the integrity and high resolution of the pre-processed point cloud data. For example, the canopy layer image data can generate a point cloud reflecting the height and canopy structure of the trees, while the understory layer image data can provide three-dimensional information of the ground vegetation and terrain. This conversion from two-dimensional to three-dimensional can not only more intuitively display the spatial structure of the forest, but also provide accurate three-dimensional coordinate information for subsequent point cloud processing and analysis. The generated pre-processed point cloud data retains the original characteristics of the forest environment, provides rich details for further pre-processing and feature extraction, and enables the system to more comprehensively analyze the various elements and their interrelationships in the forest ecosystem. It is understandable that in this step, the pre-processed point cloud data is subjected to a series of optimization processes through the point cloud pre-processing model, and the specific steps include performing cropping, filtering, ground point separation, normal vector calculation and correction operations on the pre-processed point cloud data in sequence. For example, the cropping operation can remove irrelevant areas and focus on the research target; the filtering operation can remove noise points and outliers, making the point cloud smoother and more accurate; and the ground point separation can clearly distinguish between vegetation and the ground, facilitating subsequent vegetation structure analysis. Normal vector calculation and correction can further optimize the geometric features of the point cloud and ensure the consistency of point clouds from different data sources in spatial coordinates. These preprocessing steps not only improve the accuracy of the point cloud data, but also enhance the operability of the data, making subsequent tree location extraction and three-dimensional modeling more efficient and accurate. In this step, step S2 includes step S21, step S22, step S23, and step S24.

[0031] Step S21: calibrating the canopy layer image data and the understory layer image data to generate point cloud data;

[0032] It can be understood that in this step, by processing the canopy image data and the understory image data and generating pre-processed point cloud data, the system can convert two-dimensional image information into three-dimensional spatial data, thus laying the foundation for three-dimensional forest modeling. Specifically, after completing a complete data collection within a specific time range, the canopy image captured by the drone's aerial camera and the understory image captured by the panoramic camera are imported into the Ag i soft Metashape software to generate a sparse point cloud. On this basis, the pre-processed point cloud data is further generated.

[0033] Step S22: cropping and denoising the point cloud data to obtain denoised point cloud data;

[0034] It can be understood that in this step, by performing a cropping operation on the pre-processed point cloud data, the point cloud data in irrelevant areas can be effectively removed, and the focus can be placed on the target area for research, thereby reducing the amount of computation required for subsequent processing and improving data processing efficiency. Specifically, the cropping area is first defined in the point cloud processing software. After loading the point cloud data, the point cloud data in the target area can be accurately extracted by setting cropping parameters (such as cropping boundaries, cropping directions, etc.) and performing the cropping operation to obtain the cropping result. This process not only helps to remove redundant background information, but also ensures the pertinence and accuracy of subsequent analysis, providing an optimized data basis for subsequent operations such as point cloud filtering and feature extraction.

[0035] Step S23: Separating the denoised point cloud data based on a cloth simulation filtering algorithm to obtain ground points containing only vegetation and / or non-vegetation features;

[0036] As can be understood, the cloth simulation filtering algorithm separates ground points from non-ground points in the denoised point cloud data, thereby distinguishing vegetation point clouds from ground and other non-vegetation feature point clouds. This process provides an accurate data foundation for subsequent 3D modeling and analysis. Specifically, the denoised result is first de-noised and flipped. Next, the cloth grid is initialized and its size is set. The cloth is placed above the highest point in the denoised point cloud. All point cloud data in the denoised result is then projected onto the same horizontal plane, and the nearest point and its elevation are recorded for each cloth point. The displacement of movable grid points due to gravity is calculated and compared with the elevation of the nearest point. Points that meet the criteria are designated as immovable. This process is iterated until the height change of all points is less than the set cloth algorithm threshold. Finally, the ground points are distinguished and extracted by comparing the height difference between the point cloud and the cloth points, resulting in pure vegetation and non-vegetation feature point cloud data.

[0037] Step S24: Calculate the normal vector of each ground point using a neighborhood plane fitting algorithm, and correct the Z axis of the corresponding ground point based on the calculated normal vector to obtain preprocessed point cloud data.

[0038] It can be understood that calculating the normal vectors of ground points using a neighborhood plane fitting algorithm can provide precise directional information for subsequent point cloud correction and analysis, thereby enhancing the geometric characteristics and spatial consistency of the point cloud data. Specifically, a neighborhood is first defined for each ground point in the point cloud. This neighborhood is determined based on a fixed radius or a fixed number of neighboring points. Next, all points within the neighborhood of each ground point are collected and used to perform a plane fit, using the least squares method to find the plane that best fits these points. The normal vector of this plane is the normal vector of the ground point. Correcting the ground points based on the normal vectors in this step ensures the spatial consistency of the point cloud data and provides a more accurate foundation for subsequent modeling and analysis. Specifically, the spatial orientation of each ground point is first determined using the normal vectors. Next, a Z-axis correction is applied to each ground point to ensure that the normal vectors of all points are consistent with the set reference direction or gravity direction. This correction process involves fine-tuning the position of each point in the point cloud data to eliminate deviations caused by measurement errors or sensor bias, ultimately obtaining the preprocessed results.

[0039] Step S3: determining canopy tree data and understory tree data based on the pre-processed point cloud data to obtain location data of each tree;

[0040] It is understandable that this step extracts the canopy point cloud and the understory point cloud from the preprocessing results respectively. The canopy point cloud contains point cloud information of the crown, branches and leaves of the upper forest layer, and the understory point cloud contains point cloud information of the trunks, shrubs and other parts of the lower forest layer, which can provide key information for the multi-dimensional analysis of the forest ecosystem. For example, the canopy point cloud data can reflect the crown width, height and growth status of the trees, while the understory point cloud data can reveal the position of the trunk, the size of the diameter at breast height and the distribution of the vegetation under the forest. This layered extraction method not only helps to accurately locate the position of the trees, but also provides important data support for forest resource management, biomass estimation and ecological protection. In this step, step S3 includes step S31, step S32, step S33 and step S34.

[0041] Step S31, converting the pre-processed point cloud data into a digital surface model and a digital elevation model;

[0042] It can be understood that in this step, the rasterization tool is first selected and the resolution parameters of the raster are set, and then the rasterization operation is performed to convert the point cloud data into a regular grid system to obtain multiple grid cells. By extracting the elevation value and surface data of each grid cell respectively and performing interpolation processing, a digital elevation model and a digital surface model can be generated.

[0043] Step S32: inputting the digital surface model and the digital elevation model into GIS software for subtraction, and extracting the vegetation height information obtained by the subtraction to obtain the crown height information;

[0044] It can be understood that in this step, by subtracting the digital elevation model from the digital surface model, the crown height information of the vegetation can be obtained. The specific operation is that, for each pixel of the digital surface model and each pixel in the digital elevation model, the difference between the elevation value of the digital surface model and the elevation value of the digital elevation model is calculated to obtain the crown height information. This process can not only reveal the vertical structure of the vegetation, but also provide important data support for subsequent vegetation biomass estimation, forest resource management and ecological protection. The crown height model is a key component in forest three-dimensional modeling. Combined with the trunk position information, it can provide a basis for a comprehensive analysis of the forest structure.

[0045] Step S33: identifying and segmenting the crown height information based on a watershed segmentation algorithm to obtain tree positions in the crown layer;

[0046] It can be understood that in this step, the crown height model is processed through the watershed segmentation algorithm, which can realize the identification and segmentation of the tree canopy layer, thereby accurately locating the position of a single tree, and providing important basic data support for subsequent forest structure analysis, tree growth monitoring and forest resource management. In this step, step S33 includes step S331, step S332 and step S333.

[0047] Step S331: Inverting the crown height model to obtain a three-dimensional model containing multiple basins, and presetting a seed point, wherein a point is pre-set as the smallest perforation, and the perforation is used as the seed point;

[0048] It is understood that this step, by inverting the crown height model, can convert it into a three-dimensional regional model containing multiple "basins." This conversion is a key preprocessing step for the watershed segmentation algorithm. In the inverted model, the original crown tops (high points) become low-lying "basins," while the low-lying areas between the crowns become "ridges." This conversion enables the watershed algorithm to effectively segment the crowns by simulating the process of water converging, forming watersheds between the "basins." The inverted three-dimensional regional model provides a suitable terrain foundation for the subsequent segmentation algorithm, enabling the algorithm to more accurately identify and locate the positions of individual trees. By setting the minimum point in each inverted "basin" as a seed point, a starting position is provided for the watershed segmentation algorithm. The seed point is the starting reference point for the algorithm and is located at the lowest point of each crown (i.e., the lowest point of the inverted basin). These seed points, serving as the starting points for segmentation, guide the algorithm from the lowest point upwards to simulate the process of water converging, thereby forming segmentation boundaries between the crowns.

[0049] Step S332: Starting from all the seed points, gradually perform water filling simulation on the three-dimensional model. When the water level rises to the edge of the basin, stop filling the basin, and use each basin area that is in a flooded state as a segmented area.

[0050] It is understood that in this step, starting from all set seed points, the inverted three-dimensional regional model is gradually "watered down" and simulated. This process is the core of the watershed segmentation algorithm. By simulating the upward diffusion of water from the seed point (preset point), the algorithm can dynamically identify the boundaries between tree crowns. Specifically, the water flow gradually rises along the height gradient until it encounters a higher area (i.e., the boundary between tree crowns) and stops. When the simulated water level rises to the edge of each "basin," the water injection operation stops. At this point, each flooded "basin area" is defined as an independent segmentation region. This process marks the completion of the watershed segmentation algorithm, and each segmentation region corresponds to a separate tree crown. This simulation process effectively distinguishes different tree crown areas, ensuring that each crown is accurately segmented. Through the gradual water filling simulation, the algorithm can adaptively handle complex terrain changes, thereby providing accurate segmentation results for subsequent tree location extraction and forest structure analysis.

[0051] Step S333: Assign each segmented area to a tree position, and determine the position of each tree in the canopy layer according to the center point of the segmented area.

[0052] It can be understood that in this step, by counting all the segmented areas and determining the center point of each area, the position of each tree in the canopy layer can be accurately located. Each segmented area corresponds to an independent tree crown, and its boundary is determined by the watershed segmentation algorithm. By calculating the center point of each segmented area (such as the geometric center or mass center), the location information of the trees can be obtained. This process not only realizes the precise positioning of individual trees in the forest, but also provides important basic data support for subsequent forest resource management, forest growth monitoring and ecological research.

[0053] Step S34: Fit the understory trees into a circle based on the least squares method, and use the center of the fitted circular image as the position of the understory trees.

[0054] It is understandable that this step intercepts a point cloud slice containing the tree's diameter at breast height, fits a circle using the least squares method, and uses the coordinates of the center of the fitted circle as the tree position. The canopy tree position point set reflects the top structure of the tree, while the understory tree position point set provides information on the trunk and understory vegetation. By separating these two data sets, forest structures at different levels can be processed and analyzed more accurately, providing more comprehensive data support for subsequent three-dimensional modeling and forest resource management. In this step, step S34 includes step S341 and step S342.

[0055] Step S341: obtaining a point cloud slice containing the DBH of understory trees from the pre-processed point cloud data according to a preset range threshold;

[0056] As you can understand, this step uses a preset range threshold to extract point cloud slices containing the DBH of understory trees from the preprocessed point cloud data, significantly reducing the data volume and facilitating subsequent processing. This step specifically targets the DBH of understory trees, ensuring that subsequent analysis focuses on these specific targets and avoids interference from irrelevant data.

[0057] Step S342: Fit the point cloud in the point cloud slice into a circle based on the least squares method, and use the diameter of the circle as the breast height diameter of the tree, and the center coordinates of the circle as the position coordinates of the understory trees to obtain the position of the understory trees.

[0058] It can be understood that this step uses the coordinates of the center of the fitted circle to determine the position coordinates of the understory trees, achieving accurate positioning of the trees. This step automates the extraction of DBH and position information from point cloud data, reducing manual intervention and improving efficiency.

[0059] Step S4: performing a preliminary registration on the position data of each tree based on the improved Delaunay triangulation algorithm, and performing a second registration on the preliminary registration result through an optimization function to obtain a three-dimensional point cloud model of the tree.

[0060] It can be understood that in this step, the improved Delaunay triangulation algorithm is used to perform preliminary registration of the tree position point cloud data, which can achieve preliminary alignment of the canopy layer and the understory point cloud data. The final preliminary registration result provides a basis for subsequent fine registration, ensuring that the point clouds from different data sources can achieve good alignment at the tree position, providing reliable support for further forest 3D modeling and analysis. In this step, step S4 includes

[0061] Step S41: constructing at least one combination of canopy tree location points and at least one combination of understory tree location points based on the location data of each tree, wherein each combination includes three non-collinear tree location points;

[0062] It can be understood that in this step, by selecting at least two groups of tree location points in the canopy layer and understory tree location point sets (each group contains three non-collinear tree location points), and calculating the distance between each pair of tree location points in each group, distance constraints can be provided for the subsequent Delaunay triangulation construction.

[0063] Step S42: Calculate the distance between each pair of tree locations in each combination, and use the minimum value of the distance calculated in all combinations as the constraint threshold;

[0064] It's understandable that this step effectively filters out non-tree points (such as shrubs or noise) by calculating the minimum distance between any two tree locations across all combinations and selecting a global minimum as the constraint threshold. This improves the accuracy and reliability of point cloud registration. This process provides an important reference for subsequent triangulation and point cloud alignment, ensuring accurate matching of tree locations across different levels of point cloud data.

[0065] Step S43: taking a combination of three possible non-collinear points of all canopy tree positions and understory tree positions as a candidate point group;

[0066] As you can understand, in this step, by exhaustively enumerating all possible combinations of three non-collinear points in both the canopy and understory tree location points, candidate point groups are generated, providing a comprehensive point set foundation for subsequent Delaunay triangulation and point cloud registration. This process ensures that all possible triangle combinations are considered, providing sufficient candidates for subsequent distance constraints and similarity assessments. By exhaustively enumerating all possible combinations, the geometric relationships between tree location points can be more comprehensively captured, ensuring high-precision point cloud registration and 3D modeling.

[0067] Step S44: determining whether the distance between any two non-collinear points in each candidate point group is greater than the constraint threshold; if so, obtaining a triangle corresponding to each candidate point group by connecting any two points in the candidate point group;

[0068] It can be understood that in this step, by judging whether the distance between the two non-collinear points in each candidate point group is greater than the set constraint threshold, the point groups that do not meet the tree spacing characteristics can be effectively filtered out, thereby improving the accuracy and reliability of point cloud registration. Specifically, for each candidate point group, if the distance between the two non-collinear points inside it is greater than the constraint threshold, it is considered that these point groups meet the geometric characteristics of the tree positions, and the points inside them can be connected in pairs to form corresponding triangles. This process provides an accurate triangle combination for the subsequent Delaunay triangulation construction and point cloud alignment, ensuring the accurate matching of point cloud data at different levels in the tree positions. Through this distance constraint, non-tree points (such as shrubs or noise points) can be more effectively identified and filtered out, such as shrubs or noise points, thereby improving the accuracy and efficiency of point cloud registration.

[0069] Step S45: Connect all candidate point groups whose points are greater than the constraint threshold to form a triangular network, and obtain a Delaunay triangulation network of the canopy layer and a Delaunay triangulation network of the understory layer respectively;

[0070] It can be understood that in this step, for all candidate point groups that meet the constraints, the points inside them are connected in pairs to form triangles, thereby constructing the Delaunay triangulation networks of the canopy layer and the understory layer respectively. This process is based on the principle of Delaunay triangulation, and generates a triangulation network with optimal geometric characteristics by connecting points that meet the distance constraints. The Delaunay triangulation network of the canopy layer can reflect the spatial distribution of the top structure of the trees, while the Delaunay triangulation network of the understory layer reveals the layout of the trunks and understory vegetation. By constructing these two levels of triangulation networks separately, the geometric characteristics of the forest structure can be more intuitively represented, while providing an accurate geometric framework for subsequent point cloud registration and three-dimensional modeling.

[0071] Step S46: By comparing the side lengths and angle similarities of triangles in the canopy layer Delaunay triangulation network and the understory layer Delaunay triangulation network, matching triangles are identified, and based on all matching triangles, the position data of each tree is aligned, and non-tree points are filtered through side length and preset angle constraints to obtain an initial registration result.

[0072] It can be understood that in this step, by comparing the side lengths and angle similarities of triangles in the canopy layer Delaunay triangulation network and the understory layer Delaunay triangulation network, matching triangles are identified. This process, based on the similarity of geometric features, can achieve preliminary alignment of the canopy layer point cloud data and the understory layer point cloud data at the tree positions. Specifically, the side length and angle features of all possible three non-collinear point combinations in the canopy layer and understory layer point sets are first calculated; then, constraint thresholds for the side lengths and angles are set, and only point groups that meet these constraints are retained; a Delaunay triangulation network is constructed based on these point groups, and matching triangles are identified by comparing the side lengths and angle similarities of the triangles. Based on these matching triangles, the canopy layer point cloud data and the understory layer point cloud data are aligned at the tree positions, thereby achieving preliminary fusion of point cloud data at different levels. At the same time, by setting constraint conditions for side lengths and angles, non-tree points (such as shrubs or noise points) that do not meet the geometric characteristics of the trees are filtered out, further improving the purity of the point cloud data and the registration accuracy. The final preliminary registration result is a set of point cloud data of the canopy and understory layers that have been aligned at the tree positions.

[0073] It can be understood that in this step, step S46 is followed by step S47.

[0074] Step S47: further register the initial registration result to obtain a three-dimensional point cloud model of the forest;

[0075] It can be understood that in this step, the preliminary registration result is further optimized based on the further registration from point to surface, with the aim of obtaining a high-precision three-dimensional point cloud model of the forest. Specifically, the further registration is performed by projecting the point cloud data after the preliminary registration onto the preliminary registration result (as a reference surface), and ensuring the precise alignment of the point cloud data in space by minimizing the optimization function of the projection error. This process includes: calculating the vertical distance from each point in the point cloud data after the preliminary registration to the preliminary registration result (reference surface), and constructing a minimum error equation to minimize the sum of the squares of these distances. By iteratively adjusting the rotation and translation parameters of the point cloud data after the preliminary registration, the alignment accuracy of the point cloud is gradually optimized until the error reaches the set threshold or converges. After optimization, the point cloud data after the preliminary registration is precisely aligned with the preliminary registration result in space, and finally a three-dimensional point cloud model of the forest is obtained. This process can effectively eliminate the slight deviations that may exist in the preliminary registration stage, and further improve the accuracy and consistency of the point cloud data. Among them, the minimum error equation is specifically expressed as:

[0076]

[0077] Among them, dmin represents the minimum projection error obtained through the optimization process, that is, in the process of point cloud precise registration, the projection error is minimized by adjusting the point cloud data; argmin M , which means finding the transformation parameter M that minimizes the error function; M contains the rotation matrix R and the translation vector t, which are used to describe the rigid body transformation from the point cloud data after the initial registration to the initial registration result; m is the total number of points in the point cloud; i is an index variable for iteration and enumeration; set the point qi in the point cloud data after the initial registration and map it to the initial registration result; the d i represents the point closest to point qi in the initial registration result; the normal vector of the tangent plane where the point in the initial registration result is located is n i ; said (Mq i -d i )n i Represents the point-to-surface error, that is, the point cloud Mq after the initial registration after the transformation of the point cloud data i The distance to the tangent plane of the initial registration result, the physical meaning of the tangent plane represents the point cloud Mq after the initial registration i Projection to normal vector n i The physical significance of the minimum error equation lies in the distance between the transformed points in the initially registered point cloud data and the tangent plane of the initial registration result, that is, the distance between the initially registered point cloud data projected onto the normal vector. By minimizing this distance, the point cloud data can accurately represent the position, shape, and structure of trees, providing high-quality data support for 3D forest modeling, vegetation analysis, and resource management.

[0078] Example 2:

[0079] like Figure 2 As shown, this embodiment provides a cross-source fusion modeling system for a forest 3D point cloud model. Figure 2 The system includes an acquisition unit 701 , a processing unit 702 , an evaluation unit 703 and a construction unit 704 .

[0080] An acquisition unit 701 is configured to acquire first information based on a camera device mounted on a drone, wherein the first information includes image data of a canopy layer above a preset height of trees and image data of an understory layer below the preset height of trees;

[0081] A processing unit 702 is configured to perform image processing based on the first information, and preprocess the point cloud data obtained by the processing to obtain preprocessed point cloud data;

[0082] An analysis unit 703 is configured to determine canopy tree data and understory tree data based on the pre-processed point cloud data to obtain location data of each tree.

[0083] The registration unit 704 is used to perform a preliminary registration on the position data of each tree based on the improved Delaunay triangulation algorithm, and perform a secondary registration on the preliminary registration result through an optimization function to obtain a three-dimensional point cloud model of the tree.

[0084] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

[0085] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A cross-source fusion modeling method for a forest 3D point cloud model, characterized in that: include: Acquiring first information based on a camera device carried by a drone, the first information including canopy image data above a preset height of trees and understory image data below the preset height of trees; performing image processing based on the first information, and preprocessing the obtained point cloud data to obtain preprocessed point cloud data; Determine canopy tree data and understory tree data based on the pre-processed point cloud data to obtain location data of each tree; The position data of each tree is preliminarily registered based on the improved Delaunay triangulation algorithm, and the preliminarily registered result is secondarily registered through the optimization function to obtain a three-dimensional point cloud model of the tree.

2. The cross-source fusion modeling method of a forest 3D point cloud model according to claim 1 is characterized in that: Performing image processing based on the first information and preprocessing the obtained point cloud data includes: Calibrate the canopy layer image data and the understory layer image data to generate point cloud data; Performing cropping and denoising on the point cloud data to obtain denoised point cloud data; Separating the denoised point cloud data based on a cloth simulation filtering algorithm to obtain ground points containing only vegetation and / or non-vegetation features; The normal vector of each ground point is calculated by the plane fitting algorithm of the neighborhood, and the Z axis of the corresponding ground point is corrected based on the calculated normal vector to obtain the preprocessed point cloud data.

3. The cross-source fusion modeling method of a forest 3D point cloud model according to claim 1 is characterized in that: Determining canopy tree data and understory tree data based on the preprocessed point cloud data includes: Converting the pre-processed point cloud data into a digital surface model and a digital elevation model; Inputting the digital surface model and the digital elevation model into GIS software for subtraction, and extracting the vegetation height information obtained by the subtraction to obtain the crown height information; Identifying and segmenting the crown height information based on a watershed segmentation algorithm to obtain tree positions in the crown layer; The understory trees are fitted into a circle based on the least squares method, and the center of the fitted circular image is used as the position of the understory trees.

4. The cross-source fusion modeling method of a forest 3D point cloud model according to claim 3 is characterized in that: The crown height information is identified and segmented based on a watershed segmentation algorithm, including: By inverting the crown height model, a three-dimensional model including multiple basins is obtained, and a seed point is preset, wherein a point is preset as a minimum perforation, and the perforation is used as the seed point; Starting from all the seed points, gradually perform water filling simulation on the three-dimensional model, stop water filling when the water level rises to the edge line of the basin, and use each basin area that is in a flooded state at this time as a segmentation area; Each of the segmented areas corresponds to the position of a tree, and the position of each tree in the canopy layer is determined according to the center point of the segmented area.

5. The cross-source fusion modeling method of a forest 3D point cloud model according to claim 3 is characterized in that: The understory trees are fitted into a circle using the least squares method, and the center of the fitted circular image is used as the location of the understory trees, including: According to the preset range threshold, a point cloud slice containing the DBH of understory trees is obtained from the pre-processed point cloud data; Based on the least squares method, the point cloud in the point cloud slice is fitted into a circle, and the diameter of the circle is used as the breast diameter of the tree, and the center coordinates of the circle are used as the position coordinates of the understory trees to obtain the position of the understory trees.

6. A cross-source fusion modeling system for a forest 3D point cloud model, characterized in that: include: an acquisition unit, configured to acquire first information based on a camera device carried by a drone, wherein the first information includes image data of a canopy layer above a preset height of the trees and image data of an understory layer below the preset height of the trees; a processing unit, configured to perform image processing based on the first information, and preprocess the point cloud data obtained by the processing to obtain preprocessed point cloud data; an analysis unit, configured to determine canopy tree data and understory tree data based on the preprocessed point cloud data, and obtain position data of each tree; The registration unit is used to perform a preliminary registration on the position data of each tree based on the improved Delaunay triangulation algorithm, and perform a secondary registration on the preliminary registration result through an optimization function to obtain a three-dimensional point cloud model of the tree.

7. The cross-source fusion modeling system for forest 3D point cloud models according to claim 6 is characterized in that: The processing unit includes: A first processing subunit is configured to calibrate the canopy layer image data and the understory layer image data to generate point cloud data; A second processing subunit is configured to perform cropping and denoising on the point cloud data to obtain denoised point cloud data; a third processing subunit, configured to separate the denoised point cloud data based on a cloth simulation filtering algorithm to obtain ground points containing only vegetation and / or non-vegetation features; The fourth processing subunit is used to calculate the normal vector of each of the ground points through a plane fitting algorithm of the neighborhood, and correct the Z axis of the corresponding ground point based on the calculated normal vector to obtain preprocessed point cloud data.

8. The cross-source fusion modeling system for forest 3D point cloud models according to claim 6 is characterized in that: The analysis unit comprises: A first analysis subunit, configured to convert the pre-processed point cloud data into a digital surface model and a digital elevation model; The second analysis subunit is used to input the digital surface model and the digital elevation model into the GIS software for subtraction, and extract the vegetation height information obtained by the subtraction to obtain the crown height information; The third analysis subunit is used to identify and segment the crown height information based on a watershed segmentation algorithm to obtain the tree positions in the crown layer; The fourth analysis subunit is used to fit the understory trees into a circle based on the least square method, and use the center of the fitted circular image as the position of the understory trees.

9. The cross-source fusion modeling system for forest 3D point cloud models according to claim 8, characterized in that: The third analysis subunit comprises: a fifth analysis subunit, configured to obtain a three-dimensional model including multiple basins by inverting the crown height model, and presetting a seed point, wherein a point is pre-set as a minimum perforation, and the perforation is used as the seed point; a sixth analysis subunit, configured to perform a water filling simulation on the three-dimensional model step by step starting from all the seed points, stop water filling when the water level rises to the edge of the basin, and use each of the basin areas that is flooded at this time as a segmented area; The seventh analysis subunit is configured to correspond each segmented area to a position of a tree, and determine the position of each tree in the canopy layer according to the center point of the segmented area.

10. The cross-source fusion modeling system for forest 3D point cloud models according to claim 8, characterized in that: The fourth analysis subunit comprises: an eighth analysis subunit, configured to extract, from the pre-processed point cloud data, point cloud slices containing the breast height diameters of understory trees according to a preset range threshold; The ninth analysis subunit is used to fit the point cloud in the point cloud slice into a circle based on the least squares method, and use the diameter of the circle as the breast diameter of the tree, and the center coordinates of the circle as the position coordinates of the understory trees to obtain the position of the understory trees.

Images