Mountainous area complex terrain three-dimensional modeling method based on digital twinning

Abstract

The invention relates to the technical field of digital twinning, in particular to a mountainous area complex terrain three-dimensional modeling method based on digital twinning, which comprises the following steps: acquiring first point cloud data and second point cloud data, and preprocessing the point cloud data to obtain first point cloud data and second point cloud data; the first point cloud data and the second point cloud data are point cloud data acquired by scanning the mountain terrain based on different visual angles; feature point extraction is carried out on the preprocessed first point cloud data and the preprocessed second point cloud data, and a first feature point set and a second feature point set which are used for representing mountain terrain features are obtained; determining a target transformation matrix by iteratively calculating the coincidence rate of the first feature point set and the second feature point set; registering the first point cloud data and the second point cloud data according to the target transformation matrix; and according to the registered first point cloud data and second point cloud data, constructing a three-dimensional model of the mountain terrain on a digital twin platform. According to the method, the precision and efficiency of three-dimensional modeling of the mountainous complex terrain can be improved.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, specifically to a method for three-dimensional modeling of complex mountainous terrain based on digital twins. Background Technology

[0002] With the gradual penetration of digital twin technology into the field of engineering construction, 3D modeling of complex mountainous terrain has become an important foundation for supporting the development of water conservancy projects in mountainous areas. Whether it is the site selection of water conservancy hubs, the design of reservoir dams, or the planning of water diversion channels, it is necessary to restore the topography, hydrogeological conditions, and surrounding environment of mountainous areas through high-precision terrain modeling. This provides a visual reference for engineering decisions, helps water conservancy projects in mountainous areas achieve digital transformation, and improves the scientific nature of engineering planning and construction.

[0003] Currently, the key to achieving multi-source point cloud data fusion lies in point cloud registration technology. This technology aims to find the spatial transformation relationship between point clouds from different perspectives, enabling them to be accurately aligned in a unified coordinate system. However, mountainous terrain is characterized by dramatic undulations, severe occlusion by ground features, massive amounts of point cloud data, and significant noise interference. Traditional Iterative Closest Point (ICP) registration methods rely on the initial overlapping area. When dealing with small overlapping areas from multiple perspectives, such as modeling a river in a dense forest, the river point cloud data is abundant when data is collected directly above, but less or even completely lost when measured from the side due to occlusion by the dense forest. This makes accurate registration difficult, leading to discontinuities in the model. Consequently, existing registration methods generally suffer from low computational efficiency and insufficient model reconstruction accuracy when handling such complex scenarios, failing to meet the dual requirements of modeling speed and accuracy in practical engineering applications. Summary of the Invention

[0004] To address the technical problems of low efficiency and unstable accuracy in multi-view point cloud data registration in complex mountainous terrain scenarios, the present invention aims to provide a 3D modeling method for complex mountainous terrain based on digital twins. The specific technical solution adopted is as follows: Firstly, a method for 3D modeling of complex mountainous terrain based on digital twins is provided. This method includes: acquiring first and second point cloud data, and preprocessing the point cloud data, wherein the first and second point cloud data are point cloud data collected from scanning the mountainous terrain from different perspectives; extracting feature points from the preprocessed first and second point cloud data respectively to obtain a first feature point set and a second feature point set used to characterize the mountainous terrain features; determining the target transformation matrix by iteratively calculating the overlap rate between the first and second feature point sets; registering the first and second point cloud data according to the target transformation matrix; and constructing a 3D model of the mountainous terrain on a digital twin platform based on the registered first and second point cloud data.

[0005] In one possible design, the point cloud data is preprocessed, including: normalizing the coordinate and color information of each sampling point in the point cloud data; determining the average distance between each sampling point and multiple neighboring sampling points for each sampling point in the point cloud data; determining a noise threshold based on the average distances of all sampling points; and removing sampling points whose average distance is greater than the noise threshold.

[0006] In one possible design, feature points are extracted from the preprocessed point cloud data to obtain a feature point set for characterizing the terrain features of the mountainous area. This includes: for each sampling point in the preprocessed point cloud data, determining the normal curvature between the sampling point and multiple neighboring sampling points, whereby the normal curvature is used to characterize the degree of bending change of the local terrain; and filtering out sampling points among multiple neighboring sampling points whose absolute value of normal curvature is less than a preset curvature threshold to obtain the feature point set.

[0007] In one possible design, the target transformation matrix is ​​determined by iteratively calculating the overlap rate between the first and second feature point sets. This includes: in each iteration, performing a bidirectional neighbor search on both the first and second feature point sets to obtain initial corresponding point pairs; determining valid overlapping point pairs from the initial corresponding point pairs based on a preset deviation threshold; determining the overlap rate for the current iteration based on the number of valid overlapping point pairs, the number of sampling points in the first feature point set, and the number of sampling points in the second feature point set; if the overlap rate is less than the preset overlap rate threshold, determining spatial transformation parameters based on the valid overlapping point pairs, updating the first feature point set based on the spatial transformation parameters, and performing the next iteration calculation based on the updated first and second feature point sets, where the spatial transformation parameters include rotation and translation matrices; and if the overlap rate is greater than or equal to the preset overlap rate threshold, determining the current spatial transformation parameters as the target transformation matrix.

[0008] In one possible design, a bidirectional nearest neighbor search is performed on the first feature point set and the second feature point set to obtain initial corresponding point pairs between the first feature point set and the second feature point set. This includes: constructing a first spatial index structure for the first feature point set based on a first dimension; searching for the nearest neighbor of each sample point in the first feature point set in the second feature point set using the first spatial index structure to form a first set of corresponding point pairs; constructing a second spatial index structure for the second feature point set based on a second dimension; searching for the nearest neighbor of each sample point in the second feature point set in the first feature point set using the second spatial index structure to form a second set of corresponding point pairs; and determining the intersection of the first set of corresponding point pairs and the second set of corresponding point pairs as the initial corresponding point pairs.

[0009] In one possible design, the above-mentioned digital twin-based 3D modeling method for complex mountainous terrain further includes: determining the information entropy of multiple candidate dimensions in the first feature point set, and determining the candidate dimension with the largest information entropy as the first dimension, wherein the multiple candidate dimensions include spatial coordinate dimension, gray value dimension and reflection intensity dimension; determining the information entropy of multiple candidate dimensions in the second feature point set, and determining the candidate dimension with the largest information entropy as the second dimension.

[0010] In one possible design, based on a preset deviation threshold, valid overlapping point pairs are determined from the initial corresponding point pairs, including: for any corresponding point pair in the initial corresponding point pairs, if the Euclidean distance between the sampling points in the first feature point set and the sampling points in the second feature point set is less than the preset deviation threshold, the corresponding point pair is determined to be a valid overlapping point pair.

[0011] In one possible design, spatial transformation parameters are determined based on effective coincident point pairs, including: determining the centroids of the first feature point set and the second feature point set; decentrifuging the sampling points in the effective coincident point pairs; calculating the covariance matrix of the decentrifuged sampling points; and performing matrix decomposition on the covariance matrix to obtain the rotation matrix and translation matrix.

[0012] In one possible design, a 3D model of the mountainous terrain is constructed on a digital twin platform based on the registered first and second point cloud data. This includes: constructing a triangular mesh using a preset mesh reconstruction algorithm on the registered first and second point cloud data; constructing a 3D model of the mountainous terrain based on the triangular mesh and exporting the 3D model as a file in a preset general model format; and importing the 3D model into the digital twin platform for visualization.

[0013] In one possible design, the above-mentioned digital twin-based 3D modeling method for complex mountainous terrain further includes: when the overlap rate is less than a preset overlap rate threshold and the number of iterations exceeds a preset maximum number, the current spatial transformation parameters are determined as the target transformation matrix.

[0014] The present invention has the following beneficial effects: In the digital twin-based 3D modeling method for complex mountainous terrain provided by this invention, the point cloud data is effectively purified through a preprocessing process, laying a high-quality data foundation for subsequent processing. Furthermore, a targeted feature point extraction strategy significantly reduces the amount of data computation while preserving key terrain features, improving processing efficiency. By introducing an iterative registration mechanism guided by overlap rate, the spatial transformation parameters of the point cloud can be adaptively optimized, effectively overcoming the registration difficulties caused by occlusion and limited overlapping areas in mountainous terrain, ensuring seamless fusion of point cloud data from different perspectives. Finally, through deep integration with the digital twin platform, a high-precision 3D model that can be used in practical engineering applications is directly generated, significantly improving modeling accuracy and efficiency. This effectively solves the problems of incomplete terrain representation, large registration deviations, and insufficient model practicality in traditional technologies, providing accurate and reliable visualization support for decision-making in site selection, design, and construction of water conservancy projects in mountainous areas, assisting in the digital transformation of water conservancy projects, and reducing engineering planning and construction risks. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating a method for three-dimensional modeling of complex mountainous terrain based on digital twins, provided in one embodiment of the present invention. Figure 2 This is a schematic diagram of the sampling point position relationship provided in one embodiment of the present invention; Figure 3 This is a flowchart illustrating a method for three-dimensional modeling of complex mountainous terrain based on digital twins, provided as an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a digital twin-based 3D modeling method for complex mountainous terrain proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] In embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0019] In the description of this invention, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" and "more than one" refer to two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] The following description, in conjunction with the accompanying drawings, details a specific scheme for a digital twin-based 3D modeling method for complex mountainous terrain provided by this invention.

[0022] Please see Figure 1 The diagram shows a flowchart of a three-dimensional modeling method for complex mountainous terrain based on digital twins, provided by an embodiment of the present invention, including steps S101-S105.

[0023] S101. Obtain the first point cloud data and the second point cloud data, and preprocess the point cloud data.

[0024] Among them, the first point cloud data and the second point cloud data are point cloud data collected from mountain terrain scanning based on different perspectives.

[0025] As one possible approach, a drone equipped with a lidar device is used to perform point cloud data acquisition. During the acquisition process, by adjusting the drone's flight path and altitude, two sets of point cloud data based on different perspectives are obtained, which are defined as the first point cloud data and the second point cloud data.

[0026] In some embodiments, the lidar device scans the mountainous terrain surface by emitting a laser beam. Each sampling point in the point cloud data contains its three-dimensional coordinate information in a real-world coordinate system (such as WGS-84 (World Geodetic System-1984 Coordinate System) or an engineering-independent coordinate system), and synchronously records the color information (such as RGB channel data) and attribute information (such as reflection intensity) of the sampling point.

[0027] Furthermore, the collected first and second point cloud data are preprocessed respectively.

[0028] In some embodiments, the preprocessing operations performed on the first point cloud data and the second point cloud data are the same, and the preprocessing operations for the first point cloud data or the second point cloud data are as follows.

[0029] First, the point cloud data is backed up by creating geographic coordinate copies of the first and second point cloud data. These copies will always retain the original real geographic coordinates and will be used for the final model's geographic positioning and visualization.

[0030] Furthermore, the coordinate and color information of each sampling point in the point cloud data is normalized (for the computational copy). This processing is temporary and does not change the geographic coordinate copy.

[0031] Normalizing the coordinate information of the sampling points includes: calculating the coordinates of all sampling points in each point cloud dataset at the specified coordinates. Minimum value in three dimensions ( , , ) and maximum value ( , , For each sample point in the dataset, its normalized coordinates ( , , The calculation is as follows: =( - ) / ( - ()( (Similarly for dimensions), this operation linearly maps point cloud coordinates to an approximate range of [0,1]. Simultaneously, the normalization scaling parameters for each dataset are recorded. (Right now( - , - , - Translation parameters (Right now( , , This is used to subsequently transform the calculated translation vector back to the true geographic scale.

[0032] The normalization process for the color information of the sampling points involves dividing the channel data of the three RGB color channels of the sampling points by 255. It should be noted that the above normalization process is a well-known technique to those skilled in the art and will not be elaborated upon here.

[0033] Furthermore, for each sampling point in the point cloud data, the average distance between the sampling point and multiple neighboring sampling points is determined.

[0034] The average distance between a sampling point and multiple neighboring sampling points can be calculated using the following formula: In the formula, For the point cloud data The average distance between a sampling point and multiple neighboring sampling points The number of neighboring sampling points. Indicates the first The coordinates of each sampling point For the first The coordinates of the neighboring sampling points Indicates the calculation of the first The sampling point and the first Euclidean distance between neighboring sampling points Indicates the first Each sampling point is respectively with The sum of the distances between each neighboring sampling point in the neighborhood sampling points.

[0035] It should be noted that the number of neighborhood sampling points The number of neighboring sampling points can be determined based on the density of the point cloud data. Higher density results in a larger number of neighboring sampling points, while lower density results in a smaller number. Alternatively, a fixed value can be set based on experience, such as 15. The neighboring sampling points can be determined using the k-nearest neighbor (k-NN) method based on the k-th nearest neighbor (k-NN) algorithm. The location of each sampling point was determined. One sampling point.

[0036] Furthermore, the noise threshold is determined based on the average distance corresponding to all sampling points.

[0037] Wherein, the noise determination threshold is expressed as , This represents the mean of the average distances between all sampling points. The standard deviation of the average distance of all sampling points. This represents the weighting coefficient, which is set according to the noise distribution characteristics of the point cloud data. It typically ranges from 1.0 to 2.0, with smaller values ​​used when the noise distribution is uniform and larger values ​​used when the distribution is discrete.

[0038] In some embodiments, based on typical mountainous terrain point cloud data, The possible value is 1.5, when When the coefficient is 1.5, it can effectively remove significant noise points such as flying birds and suspended particles while avoiding the accidental removal of terrain edge feature points, achieving a balance between denoising effect and feature preservation. In practical applications, this coefficient can be adjusted within the range of 1.0-2.0 according to the specific terrain complexity. Smaller values ​​can be used for flat terrain, such as 1 or 1.1, while larger values ​​are suitable for densely vegetated areas, such as 1.9 or 2.

[0039] Furthermore, sampling points with an average distance greater than the noise threshold are removed.

[0040] Among them, In the case of, The corresponding sampling points are removed.

[0041] It should be noted that, due to the complex terrain of mountainous areas, there are often irrelevant targets such as birds, suspended particles, or man-made structures that are not part of the mountainous terrain. Therefore, the average distance between the sampling point and multiple neighboring sampling points is [not specified]. The larger the value, the greater the difference between the sampling point and its surrounding neighboring sampling points, and the less likely it is to belong to the point cloud data of mountainous terrain. If it is greater than the noise judgment threshold, it is determined to be a noise point and is removed.

[0042] In some embodiments, it can also be determined whether a sampling point is a noise point that needs to be removed by determining whether the number of sampling points within a preset range centered on any sampling point is less than a preset number. For example, if the number of sampling points within a 1-meter range centered on any sampling point is 3, which is less than the preset number of 5, then it is determined as a noise point and removed.

[0043] After the above preprocessing, we obtain a geographic coordinate copy (preserving the original real coordinates for the final output) and a cleaned computation copy (normalized and denoised for subsequent feature extraction, registration and all other computation steps).

[0044] S102. Extract feature points from the preprocessed first point cloud data and second point cloud data respectively to obtain the first feature point set and the second feature point set used to characterize the terrain features of the mountainous area.

[0045] It should be noted that the preprocessed first and second point cloud data still contain a large number of "feature-similar redundant points" caused by the complex terrain structure of mountainous areas (such as gentle slopes and near-planar terrain areas). These redundant points are located close to each other and have the same terrain representation ability, which will increase the amount of subsequent calculations and may also interfere with the accurate matching of corresponding point pairs.

[0046] As one possible implementation, for the preprocessed first point cloud data and second point cloud data, the ability of the sampling points to represent the terrain features of the mountainous area is quantified, redundant points are eliminated, and sampling points that can reflect the key features of the terrain are retained, forming the first feature point set corresponding to the first point cloud data and the second feature point set corresponding to the second point cloud data, respectively.

[0047] In some embodiments, for each sampling point in the preprocessed point cloud data, the normal curvature between the sampling point and multiple neighboring sampling points is determined, wherein the normal curvature is used to characterize the degree of curvature variation of the local terrain. Further, sampling points among the multiple neighboring sampling points whose absolute value of normal curvature is less than a preset curvature threshold are filtered to obtain a feature point set.

[0048] It should be noted that when determining multiple neighboring sampling points of a sampling point, sampling points existing within a preset range centered on the sampling point can be defined as neighboring sampling points. The preset range can be determined based on the density of sampling points in the point cloud data. The higher the density, the smaller the preset range, and the lower the density, the larger the preset range.

[0049] Among them, a sample point can be constructed based on any one of its multiple neighboring sample points, such as... Figure 2 The normal section circle shown is as follows: Figure 2 As shown, A point is any sampling point in the point cloud data. Point is the sampling point The first of multiple neighboring sampling points The vector of neighboring sampling points Sampling points normal vector, vector Sampling points The normal vector, For vectors With a straight line The angle between them For vectors with vector The angle between them.

[0050] based on Figure 2 As shown, the formula for calculating the normal curvature between a sampling point and any of its neighboring sampling points is as follows: In the formula, The normal vector is represented as sampling points Compared to the first Normal curvature of neighboring sampling points, Indicates the calculation of sampling points and neighboring sampling points The Euclidean distance between them, the negative sign indicates the direction of curvature bending, not the magnitude, and refers to the sampling point. To neighboring sampling points The curvature is directed downwards. This indicates the calculation of the absolute value of the normal curvature. This is a preset curvature threshold.

[0051] In some embodiments, before calculating the normal curvature between a sampling point and its neighboring sampling points using the above formula, it is first determined that... and Is it 0? If the value is 0, it indicates that the two points coincide, and the normal curvature of the sampling point and its neighboring sampling points is directly set to 0; if If the value is 0, it means that the neighboring sampling point is directly above or below the sampling point, and the calculation of the normal curvature of the neighboring sampling point is skipped.

[0052] It should be noted that if redundant points are considered to exist among the sampling points and their neighbors, then two adjacent sampling points can be approximated as being located on the same plane. The curvature of two points on the same plane is very small. Based on general practical engineering experience, the curvature of most smooth terrains is concentrated between 0.001 and 0.005. This invention is based on the fact that very smooth planar terrain is extremely rare in mountainous areas; therefore, the preset curvature threshold is set to... .

[0053] Based on the above embodiments, redundant points in the first point cloud data and the second point cloud data are filtered to obtain the first feature point set and the second feature point set.

[0054] S103. Determine the target transformation matrix by iteratively calculating the overlap rate between the first feature point set and the second feature point set.

[0055] As one possible implementation, a bidirectional nearest neighbor search is first performed in each iteration to obtain initial corresponding point pairs. Then, based on a preset deviation threshold, the initial corresponding point pairs are geometrically verified to select valid overlapping point pairs that meet the spatial consistency requirements. Subsequently, the overlap rate of the current iteration is calculated based on the number of valid overlapping point pairs and the size of the feature point set. Finally, the iteration is terminated by judging whether the overlap rate reaches a preset overlap rate threshold: if the preset overlap rate threshold is not reached and the maximum number of iterations is not exceeded, the spatial transformation parameters including rotation and translation matrices are calculated based on the valid overlapping point pairs, the spatial position of the first feature point set is updated, and the next iteration continues; if the overlap rate reaches the preset overlap rate threshold or exceeds the maximum number of iterations, the current spatial transformation parameters are determined as the target transformation matrix.

[0056] In some embodiments, when obtaining the initial corresponding point pairs, a sphere query method can be used to search for neighboring points. Here, each sampling point in the first feature point set is taken as the center of the sphere, and a preset search radius is set based on the point cloud density. All sampling points falling into the sphere are found in the second feature point set, and the nearest sampling point is taken as the corresponding point to obtain the first corresponding point pair set. Similarly, a sphere query is performed based on the second feature point set to obtain the second corresponding point pair set. The intersection can be further taken to obtain the initial corresponding point pairs.

[0057] It should be noted that the specific implementation method of determining the target transformation matrix by iteratively calculating the overlap rate between the first feature point set and the second feature point set can be referred to the description of steps S201-S205 in the subsequent embodiments of this invention, and will not be repeated here.

[0058] S104. Register the first point cloud data and the second point cloud data according to the target transformation matrix.

[0059] As one possible implementation, during the registration process, the second point cloud data is used as the reference standard, that is, the coordinate system of the second point cloud data is used as the target coordinate system. The spatial position remains unchanged, and spatial transformation operations are performed only on the first point cloud data. Assume the three-dimensional coordinates of any sampling point in the first point cloud data are... The rotation matrix in the target transformation matrix is (A 3×3 matrix, used to represent the combination of rotation angles about the x, y, and z axes), the translation matrix is... (A 3×1 vector, used to represent the translation along the x, y, and z axes). Further, based on the normalization scaling parameters used during coordinate normalization in step S101 above, the translation matrix... Transformed to a true geographical scale. ,in, It is a translation matrix. This is the normalized scaling parameter.

[0060] Then the coordinates of the transformed sampling point .in, For rotation operations, the sampling points are adjusted via matrix multiplication. The spatial angle is adjusted to ensure that the overall orientation of the first point cloud data is consistent with that of the second point cloud data; a translation matrix is ​​then superimposed. Operations (i.e., adding vectors) The component then adjusts the overall position of the point cloud after rotation to align the two in space.

[0061] Furthermore, after performing the aforementioned spatial transformation operation on all sampling points in the first point cloud data, the transformed first point cloud data is obtained. The transformed first point cloud data and the second point cloud data are in the same three-dimensional coordinate system, and the sampling points in areas with significant terrain features spatially overlap. After the registration process is completed, the transformed first point cloud data and the second point cloud data are fused to form a complete point cloud set covering the target mountainous terrain.

[0062] Understandably, in this set, the sampling points in the overlapping areas complement each other due to spatial alignment, while the sampling points in the non-overlapping areas fill in the gaps in the terrain data, ultimately resulting in registered point cloud data that can completely and consistently represent the complex terrain of the mountainous area, providing high-quality input for the subsequent construction of the 3D model.

[0063] S105. Based on the registered first point cloud data and second point cloud data, construct a three-dimensional model of the mountainous terrain on the digital twin platform.

[0064] As one possible implementation, a triangular mesh is constructed using a preset mesh reconstruction algorithm on the registered first and second point cloud data, and a three-dimensional model of the mountainous terrain is constructed based on the triangular mesh. The three-dimensional model is then exported as a file in a preset general model format.

[0065] The registered first and second point cloud data are merged into a unified point cloud dataset. A triangular mesh model is constructed using a preset mesh reconstruction algorithm. This process connects adjacent sampling points to form a set of triangular facets, thereby constructing a complete terrain surface representation. A 3D terrain model with geometric structure and texture information is then generated based on the triangular mesh model and exported as a file in a preset universal model format, including but not limited to 3D Tiles, OSGB, or glTF formats.

[0066] Furthermore, the 3D model is imported into the digital twin platform for visualization.

[0067] The digital twin platform is a professional software platform (such as the CesiumJS digital twin platform) that supports the loading, rendering, and interactive operation of 3D models. It features adaptability to multiple 3D model formats, support for spatial querying, and dynamic annotation. During the import process, the digital twin platform automatically parses the geometric data and attribute information of the model file, reconstructs the 3D scene of the mountainous terrain through a 3D rendering engine, and maps the 3D coordinates of the model to the real geographic coordinates, ensuring that the spatial position of the model in the platform is consistent with the actual mountainous terrain. The digital twin platform supports interactive operations such as scaling, rotating, and translating the 3D model. Engineers can adjust the perspective to view the overall terrain or local details (such as the depth of deep valleys, the slope of steep cliffs, and the direction of river valleys). Simultaneously, the digital twin platform can associate the model's attribute data (such as the reflection intensity of sampling points and terrain elevation values), and supports obtaining terrain parameters for specific locations by clicking on the model surface. This provides intuitive data support for engineering decisions such as the site selection of water conservancy projects, the planning of reservoir dam axes, and the design of water diversion channel routes, realizing the transformation of mountainous terrain data from point cloud data into a visual decision-making tool.

[0068] Understandably, in the digital twin-based 3D modeling method for complex mountainous terrain provided by this invention, the preprocessing process effectively purifies the point cloud data, laying a high-quality data foundation for subsequent processing. Furthermore, through targeted feature point extraction strategies, the computational load is significantly reduced while retaining key terrain features, improving processing efficiency. By introducing an iterative registration mechanism guided by overlap rate, the spatial transformation parameters of the point cloud can be adaptively optimized, effectively overcoming the registration difficulties caused by occlusion and limited overlapping areas in mountainous terrain, ensuring seamless fusion of point cloud data from different perspectives. Finally, through deep integration with the digital twin platform, a high-precision 3D model that can be used in practical engineering applications is directly generated, significantly improving modeling accuracy and efficiency. This effectively solves the problems of incomplete terrain representation, large registration deviations, and insufficient model practicality in traditional technologies, providing accurate and reliable visualization support for decision-making in site selection, design, and construction of water conservancy projects in mountainous areas, assisting in the digital transformation of water conservancy projects, and reducing engineering planning and construction risks.

[0069] In one design, the present invention provides a specific implementation method for determining the target transformation matrix by iteratively calculating the overlap rate of the first feature point set and the second feature point set, such as... Figure 3 As shown, it includes steps S201-S205.

[0070] S201. In each iteration, a bidirectional neighbor search is performed on the first feature point set and the second feature point set to obtain the initial corresponding point pairs between the first feature point set and the second feature point set.

[0071] As one possible implementation, in each iteration, for the first feature point set, a first spatial index structure is constructed for the first feature point set according to the first dimension, and then the nearest neighbor of each sampling point in the first feature point set in the second feature point set is searched through the first spatial index structure to form a first set of corresponding point pairs.

[0072] First, the information entropy of multiple candidate dimensions in the first feature point set is determined, and the candidate dimension with the largest information entropy is determined as the first dimension. The multiple candidate dimensions include spatial coordinate dimension, gray value dimension and reflection intensity dimension.

[0073] In some embodiments, for a first feature point set, the information entropy of each candidate dimension of the first feature point set is first calculated, and the candidate dimension with the largest information entropy is determined as the first dimension used to construct the first spatial index structure. For example, the bidirectional nearest neighbor search method used is the bidirectional multidimensional binary tree (K-Dimensional Tree, KD-tree) search method. If the candidate dimension with the largest information entropy is the x-coordinate dimension, then the first feature point set is divided according to the x-coordinate dimension to construct the first spatial index structure KD-tree. The coordinates of each point in the second feature point set are taken as the target, and the search is recursively performed from the root node in the KD-tree. According to the nearest neighbor search method of the KD-tree, a coordinate that is "closest" to the target point is finally found in the KD-tree. The coordinates of the target point and the coordinates of the matched point are combined as a set of matching point pairs. The above process is repeated. For each sampling point in the first feature point set, a matching sampling point is found in the second feature point set, and all matching point pairs are determined as the first set of corresponding point pairs, denoted as . .

[0074] For the second feature point set, a second spatial index structure is constructed based on the second dimension. Then, the nearest neighbor of each sampling point in the second feature point set is searched in the first feature point set through the second spatial index structure to form a second set of corresponding point pairs.

[0075] In some embodiments, the second set of corresponding point pairs is determined using the method described above for determining the first set of corresponding point pairs for the first set of feature points, denoted as . This will not be elaborated upon here.

[0076] Furthermore, based on the determined first set of corresponding points... and the second set of corresponding point pairs Determine their intersection These are the initial corresponding point pairs.

[0077] S202. Based on the preset deviation threshold, determine the effective overlapping point pairs from the initial corresponding point pairs.

[0078] As one possible implementation, for any pair of corresponding points in the initial pair of corresponding points, the Euclidean distance between the sampling points in the first feature point set and the sampling points in the second feature point set is determined, and if the Euclidean distance is less than a preset deviation threshold, the pair of corresponding points is determined as a valid pair of overlapping points for subsequent calculation of the overlap rate.

[0079] In some embodiments, for any corresponding point pair, a spherical region can be determined with the sampling point in the first feature point set or the sampling point in the second feature point set as the center and a preset deviation threshold as the radius. If the sampling point in the second feature point set or the sampling point in the first feature point set falls into the spherical region, then the corresponding point pair is determined to be a valid overlapping point pair.

[0080] It should be noted that the preset deviation threshold can be a fixed value set based on experience, or it can be dynamically adjusted based on the distribution characteristics of the Euclidean distance of all initial corresponding point pairs in each iteration, such as in the... During the next iteration, the preset deviation threshold can be set to ,in, For the first The mean of the Euclidean distances between all initial corresponding point pairs during the next iteration. For the first The standard deviation of the Euclidean distance of all initial corresponding point pairs during the next iteration. This is an adjustable scaling factor, which can be set to a value between 1.5 and 3.0 based on experience.

[0081] Understandably, due to the limited overlap of point cloud sets collected from different perspectives in complex mountainous areas, there is often a certain deviation in the overlap between the first and second feature point sets. Therefore, by introducing a geometric consistency verification mechanism based on Euclidean distance and a preset deviation threshold, the accuracy and reliability of corresponding point pair selection during point cloud registration are effectively improved. Secondary selection of initial corresponding point pairs through distance constraints ensures that only point pairs that truly meet the 3D spatial proximity condition can be determined as valid overlapping points. This significantly reduces the risk of mismatches caused by uneven point cloud density and similar local features, and enhances robustness against noise points and outliers in complex mountainous terrain. This selection mechanism provides a high-quality, high-reliability input data foundation for subsequent transformation matrix calculations, fundamentally guaranteeing the accuracy and stability of the registration results. This allows the final constructed 3D terrain model to more realistically and completely reproduce the detailed features and spatial structure of the mountainous terrain.

[0082] S203. Determine the overlap rate of the current iteration based on the number of valid overlapping point pairs, the number of sampling points in the first feature point set, and the number of sampling points in the second feature point set.

[0083] In some embodiments, the overlap rate is calculated using the following formula: In the formula, Indicates the first The overlap rate is calculated in the next iteration. The number of valid overlapping point pairs, For the first feature point set The number of sampling points in the middle For the second feature point set The number of sampling points in the middle Indicates taking the first feature point set Number of sampling points With the second feature point set Number of sampling points The minimum value in.

[0084] S204. When the overlap rate is less than the preset overlap rate threshold, determine the spatial transformation parameters based on the effective overlapping point pairs, update the first feature point set based on the spatial transformation parameters, and perform the next iteration calculation based on the updated first feature point set and the second feature point set.

[0085] The spatial transformation parameters include rotation and translation matrices.

[0086] As one possible implementation, based on the overlap rate determined in step S203 above, it is compared with a preset overlap rate threshold. If the overlap rate is less than the preset threshold, the centroids of the first and second feature point sets are calculated and determined, and the sampling points in the effective overlapping point pairs are decentroided. Furthermore, the covariance matrix is ​​calculated for the decentroided sampling points, and the Singular Value Decomposition (SVD) method is used to decompose the covariance matrix to obtain rotation and translation matrices as spatial transformation parameters.

[0087] Wherein, assume the first feature point set Second feature point set Based on this, the centroid of the first feature point set is calculated as follows: , For the first feature point set, the first One sampling point, The number of sampling points in the first feature point set is ; the centroid of the second feature point set is . , For the second feature point set, the first One sampling point, This represents the number of sampling points in the second feature point set. Further, the sampling points in the effective overlapping point pairs are decentroided to obtain the decentroided point set. and ,in, , The sampling points that are valid coincident point pairs in the first feature point set. This is the sampling point after centroid removal; , The sampling points that are valid coincident point pairs in the second feature point set. This is the sampling point after centroid removal.

[0088] Furthermore, the centroid-free point set is calculated. and The covariance matrix between them: In the formula, The number of valid overlapping point pairs, For point set and The covariance matrix between them is a 3×3 matrix, used to represent... and The distribution relationship between them.

[0089] Furthermore, regarding the covariance matrix... Perform singular value decomposition: In the formula, and They are 3×3 orthogonal matrices, and their column vectors are respectively and eigenvectors, It is a diagonal matrix containing non-negative singular values.

[0090] Furthermore, the rotation matrix is ​​calculated. : The rotation matrix is ​​calculated. Then, the translation matrix can be calculated based on the centroids of the first and second feature point sets. : In summary, the spatial transformation parameters are determined based on the effective coincident point pairs of the current iteration, which are then used to update the first feature point set.

[0091] Furthermore, the first feature point set is updated based on the obtained spatial transformation parameters, and then the next iteration calculation is performed based on the updated first feature point set and the second feature point set.

[0092] In some embodiments, if the overlap rate is less than a preset overlap rate threshold, it is determined whether the current iteration number exceeds a preset maximum number. If it has exceeded the preset maximum number, the current spatial transformation parameter is determined as the target transformation matrix.

[0093] It should be noted that, based on experience in regional monitoring of water conservancy projects, the recommended minimum overlap rate should be greater than or equal to 50%. Therefore, this application sets the preset overlap rate threshold to 50%, and the preset maximum number of iterations can be set to 10 based on experience. This invention does not impose specific limitations on this.

[0094] S205. If the overlap rate is greater than or equal to the preset overlap rate threshold, the current spatial transformation parameters are determined as the target transformation matrix.

[0095] Understandably, if the overlap rate is greater than or equal to the preset overlap rate threshold, it indicates that the overlap rate of the modeling region has reached the requirement for accurate modeling. The iteration is terminated, and the current spatial transformation parameters are used to determine the target transformation matrix for the registration of the first point cloud data and the second point cloud data.

[0096] Understandably, this invention, through a combination of bidirectional nearest neighbor search and spatial consistency verification, can adaptively identify and utilize the effective spatial correspondence between two sets of feature points. In each iteration, it filters out highly reliable overlapping point pairs based on geometric constraints and calculates accurate spatial transformation parameters accordingly. This iterative strategy guided by the overlap rate not only effectively overcomes the sensitivity of traditional methods to initial position, but more importantly, it can continuously expand the effective matching range through gradual optimization even with a limited overlapping area in the point cloud, ultimately achieving stable convergence. This mechanism significantly improves the adaptability and robustness of the registration process under complex terrain conditions, ensuring high-precision registration results even in mountainous scenes with vegetation occlusion and significant terrain undulations, providing a reliable spatial data foundation for subsequent 3D modeling.

[0097] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0098] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for 3D modeling of complex mountainous terrain based on digital twins, characterized in that, The method includes: Acquire first point cloud data and second point cloud data, and preprocess the point cloud data. The first point cloud data and the second point cloud data are point cloud data collected from mountain terrain scanning based on different perspectives. Feature points are extracted from the preprocessed first point cloud data and second point cloud data respectively to obtain a first feature point set and a second feature point set used to characterize the terrain features of the mountainous area. The target transformation matrix is ​​determined by iteratively calculating the overlap rate between the first feature point set and the second feature point set. The first point cloud data and the second point cloud data are registered according to the target transformation matrix; Based on the registered first point cloud data and second point cloud data, a three-dimensional model of the mountainous terrain is constructed on the digital twin platform.

2. The method for 3D modeling of complex mountainous terrain based on digital twins according to claim 1, characterized in that, Preprocessing of point cloud data includes: The coordinate and color information of each sampling point in the point cloud data are normalized. For each sampling point in the point cloud data, determine the average distance between the sampling point and multiple neighboring sampling points; The noise threshold is determined based on the average distance corresponding to all sampling points. Sampling points whose average distance is greater than the noise threshold are removed.

3. The method for 3D modeling of complex mountainous terrain based on digital twins according to claim 1, characterized in that, Feature points are extracted from the preprocessed point cloud data to obtain a feature point set used to characterize the terrain features of the mountainous area, including: For each sampling point in the preprocessed point cloud data, the normal curvature between the sampling point and multiple neighboring sampling points is determined. The normal curvature is used to characterize the degree of curvature change of the local terrain. The feature point set is obtained by filtering out the sampling points whose absolute value of the normal curvature is less than a preset curvature threshold from the plurality of neighboring sampling points.

4. The method for 3D modeling of complex mountainous terrain based on digital twins according to claim 1, characterized in that, The target transformation matrix is ​​determined by iteratively calculating the overlap rate between the first feature point set and the second feature point set, including: In each iteration, a bidirectional neighbor search is performed on the first feature point set and the second feature point set to obtain the initial corresponding point pairs between the first feature point set and the second feature point set. Based on a preset deviation threshold, valid overlapping point pairs are determined from the initial corresponding point pairs; The overlap rate of the current iteration is determined based on the number of valid overlapping point pairs, the number of sampling points in the first feature point set, and the number of sampling points in the second feature point set. If the overlap rate is less than a preset overlap rate threshold, spatial transformation parameters are determined based on the effective overlapping point pairs, and the first feature point set is updated based on the spatial transformation parameters. The next iteration calculation is performed based on the updated first feature point set and the second feature point set. The spatial transformation parameters include rotation matrix and translation matrix. If the overlap rate is greater than or equal to the preset overlap rate threshold, the current spatial transformation parameters are determined as the target transformation matrix.

5. The method for three-dimensional modeling of complex mountainous terrain based on digital twins according to claim 4, characterized in that, A bidirectional neighbor search is performed on the first feature point set and the second feature point set to obtain initial corresponding point pairs between the first feature point set and the second feature point set, including: Construct a first spatial index structure for the first feature point set based on the first dimension; The first spatial index structure is used to search for the nearest neighbor of each sampling point in the first feature point set in the second feature point set, forming a first set of corresponding point pairs; Construct a second spatial index structure for the second feature point set based on the second dimension; The second spatial index structure is used to search for the nearest neighbor of each sampling point in the second feature point set in the first feature point set, forming a second set of corresponding point pairs; The intersection of the first set of corresponding point pairs and the second set of corresponding point pairs is determined as the initial set of corresponding point pairs.

6. The method for three-dimensional modeling of complex mountainous terrain based on digital twins according to claim 5, characterized in that, The method further includes: Determine the information entropy of multiple candidate dimensions in the first feature point set, and determine the candidate dimension with the largest information entropy as the first dimension. The multiple candidate dimensions include spatial coordinate dimension, gray value dimension and reflection intensity dimension. Determine the information entropy of multiple candidate dimensions in the second feature point set, and determine the candidate dimension with the largest information entropy as the second dimension.

7. The method for three-dimensional modeling of complex mountainous terrain based on digital twins according to claim 4, characterized in that, Based on a preset deviation threshold, valid overlapping point pairs are determined from the initial corresponding point pairs, including: For any of the initial corresponding point pairs, if the Euclidean distance between the sampling points in the first feature point set and the sampling points in the second feature point set is less than the preset deviation threshold, the corresponding point pair is determined to be the effective coincident point pair.

8. The method for three-dimensional modeling of complex mountainous terrain based on digital twins according to claim 4, characterized in that, Determining spatial transformation parameters based on the effective coincidence point pairs includes: Determine the centroids of the first feature point set and the second feature point set; The sampling points in the effective overlapping point pairs are decentrifuged; The covariance matrix is ​​calculated for the centroid-decentered sampling points, and the covariance matrix is ​​decomposed to obtain the rotation matrix and translation matrix.

9. The method for three-dimensional modeling of complex mountainous terrain based on digital twins according to claim 1, characterized in that, Based on the registered first point cloud data and second point cloud data, a three-dimensional model of the mountainous terrain is constructed on the digital twin platform, including: A triangular mesh is constructed using a preset mesh reconstruction algorithm on the registered first point cloud data and second point cloud data; A three-dimensional model of the mountainous terrain is constructed based on the triangular mesh, and the three-dimensional model is exported as a file in a preset general model format; The 3D model is imported into the digital twin platform for visualization.

10. The method for three-dimensional modeling of complex mountainous terrain based on digital twins according to claim 4, characterized in that, The method further includes: If the overlap rate is less than the preset overlap rate threshold and the number of iterations exceeds the preset maximum number of iterations, the current spatial transformation parameters are determined as the target transformation matrix.

Images