Three-dimensional shape modeling method for spoil field based on time sequence LiDAR data

Abstract

The invention relates to a spoil ground three-dimensional form modeling method based on time sequence LiDAR data. The method comprises the steps that spoil ground terrain collaborative data collection is carried out by using unmanned aerial vehicle oblique photogrammetry and laser radar technologies; importing the image data into three-dimensional modeling software to generate an orthoimage, calculating an exterior orientation element and an interior orientation element of the image through aerial triangulation, and establishing a geometrical relationship between the images; dense matching is carried out by adopting a multi-view stereo matching algorithm, and point cloud optimization is realized by fusing statistical filtering and radius filtering; adopting a progressive triangulation network encryption algorithm to construct and optimize the density and the structure of the initial irregular triangulation network; and carrying out texture mapping, image registration and seam optimization on the optimized triangulated irregular network to realize the construction of a live-action three-dimensional model. According to the scheme provided by the invention, the details of the spoil site can be effectively reserved, the features are accurately restored, the visual authenticity of the three-dimensional model is improved, and the actual condition of the spoil site is accurately reflected.

Description

Technical Field

[0001] This application relates to the field of three-dimensional morphological modeling technology, and in particular to a method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data. Background Technology

[0002] For major engineering earth and rock infrastructure such as spoil heaps, dynamic monitoring and accurate modeling of their three-dimensional morphology are directly related to the scientific and effective nature of landslide disaster early warning, sustainable maintenance of regional geological environment, and engineering safety management. It is an important technical support for ensuring ecological security and engineering stability.

[0003] With the development of surveying and mapping technology and computer vision technology, 3D morphological modeling technology has undergone iterative upgrades from traditional manual surveying to digital modeling. Technologies such as UAV remote sensing, LiDAR, and oblique photogrammetry are widely used in 3D morphological modeling.

[0004] However, in special scenarios like spoil heaps, single technical methods are easily constrained by environmental factors. Spoil heap point cloud data contains non-ground noise points such as vegetation and buildings, as well as outliers caused by measurement errors, making it difficult to retain terrain details while removing noise, directly affecting modeling accuracy. When dealing with typical abrupt terrain changes in spoil heaps such as steep slopes and gullies, problems such as model breakage, over-smoothing, or excessive redundant points easily occur, making it impossible to accurately reproduce the true structure of the terrain. Furthermore, for factors such as changes in lighting and differences in image acquisition perspectives, there are significant color differences between different images, resulting in texture seams in the 3D model's texture map, poor visual realism, and an inability to accurately reflect the actual surface condition of the spoil heap.

[0005] Therefore, there is an urgent need for a three-dimensional morphological modeling method for spoil heaps that combines high efficiency, high accuracy, and realism, in order to break through the existing technological bottlenecks. Summary of the Invention

[0006] To address or partially address the problems existing in related technologies, this application provides a method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data. This method aims to solve the problems that the detailed processing, accurate feature restoration, and visual realism in the three-dimensional morphological modeling of spoil heaps cannot accurately reflect the actual condition of the spoil heaps.

[0007] This application provides a method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data, including: UAV oblique photogrammetry and lidar technology were used to collect collaborative topographic data of spoil disposal sites. The acquired image data is imported into 3D modeling software for preliminary denoising, correction and stitching to generate orthophotos. The exterior and interior orientation elements of the images are calculated by aerial triangulation to establish the geometric relationship between the images. Dense point cloud generation includes dense matching and point cloud optimization. Dense matching is performed using a multi-view stereo matching algorithm, and point cloud optimization is achieved by fusing statistical filtering and radius filtering. A progressive triangulation encryption algorithm is used to construct and optimize the density and structure of the initial irregular triangulation network, thereby improving the accuracy and detail representation capability of the digital elevation model under complex terrain conditions. Texture mapping, image registration, and seam optimization are performed on the optimized irregular triangular mesh to construct a realistic 3D model. Texture mapping uses color information from UAV images to fit the surface of the irregular triangular mesh, giving the model a realistic visual effect. Image registration calculates the correspondence between each image and the irregular triangular mesh based on camera pose parameters obtained from aerial triangulation. Seam optimization uses a weighted mixing algorithm to process overlapping areas, eliminating color differences and seams between images.

[0008] Optionally, in some implementations, a progressive triangulation encryption algorithm is used to dynamically optimize the density and structure of irregular triangulation networks, including: The progressive triangulation encryption algorithm is an efficient terrain modeling method based on LiDAR point cloud data. Its core objective is to improve the accuracy and detail representation of digital elevation models under complex terrain conditions by dynamically optimizing the density and structure of the triangulation network. The progressive triangulation encryption algorithm addresses the model breakage or oversmoothing problems that easily occur in triangulation construction methods in abrupt terrain areas such as steep slopes and gullies. It adaptively enhances the local point cloud density through an iterative encryption strategy, while effectively eliminating redundant points to reduce computational complexity. The progressive triangulation encryption algorithm consists of four stages: initial seed point selection, iterative triangulation encryption, terrain feature constraint optimization, and final model verification. Based on the point cloud spatial distribution uniformity criterion, a sparse and representative seed point set is selected to construct an initial irregular triangular network. The terrain feature lines are identified through curvature analysis to ensure that the initial irregular triangular network retains the macroscopic terrain outline. Based on the initial irregular triangulation network, point clouds with elevation residuals exceeding the threshold are inserted layer by layer. The local triangular patches are updated using the Delaunay triangulation algorithm criterion. At the same time, the terrain roughness index is introduced to dynamically adjust the densification threshold to avoid excessive subdivision of flat areas. After each iteration, the difference between the interpolation error and the normal vector of the adjacent triangular facet is calculated to determine whether further refinement or termination of local optimization is needed. By combining the point cloud classification module, non-ground points such as vegetation and buildings are removed, the density of triangular mesh nodes in areas of abrupt terrain changes is forcibly increased, and the continuity of terrain structure is maintained by constraining edge connection rules.

[0009] Optionally, in some implementations, point cloud optimization is achieved by fusing statistical filtering and radius filtering, including: To address the issue that noisy points and outliers in point cloud data can significantly reduce the accuracy of subsequent modeling and analysis, statistical filtering and radius filtering are two efficient denoising methods that improve the quality of point cloud data. By fusing statistical filtering and radius filtering for multi-level denoising, robustness in complex scenarios can be improved. Statistical filtering is a point cloud denoising method based on local neighborhood statistical analysis. Its core principle is that effective point clouds exhibit statistical consistency in spatial distribution, while noise points, deviating from the neighborhood distribution pattern, can be identified and removed. The neighborhood is defined as the distribution of each point in the point cloud. Select the k nearest neighbors to form a local neighborhood set N(k,k). ), k is a preset parameter, usually taken as 20~50; the distance statistic calculation is the calculation of... Average distance to all neighboring points and standard deviation The formula is: (3) (4) In the formula, For point With neighboring points The Euclidean distance; Anomaly detection is achieved by setting a threshold. , This is an empirical coefficient. The value is 2, if the point average neighborhood distance If the value exceeds T, it is identified as a noise point and removed. The performance of statistical filtering is highly dependent on the parameters k and With proper setting, a smaller k value is suitable for dense point clouds and can preserve detailed features; a larger k value is suitable for sparse regions and enhances statistical stability, but may produce excessive smoothing. Radius filtering is a point cloud denoising method based on spatial density. It identifies and removes isolated noise points by analyzing the distribution density of points within a local neighborhood. The process includes neighborhood definition, density determination, and parameter setting. Neighborhood definition involves defining the neighborhood for each point in the point cloud. A spherical neighborhood space is constructed with the neighborhood as the center and a preset search radius r. Density is determined by counting the number of points within the neighborhood. ,like Less than the preset threshold Then determine For noise points, the mathematical expression for the removal criteria is: (5) In the formula, 1 is an indicator function.

[0010] Optionally, in some implementations, a multi-view stereo matching algorithm is used for dense matching, including: The formula for calculating dense matching is: (2) In the formula, The matching cost is given by p, where p is the number of image points and d is the disparity. For neighborhood windows, For the weight function, This is a similarity measurement function.

[0011] Optionally, in some implementations, the optimized irregular triangular mesh undergoes texture mapping, image registration, and seam optimization, including: Data preparation is aligned with input to ensure that all original UAV images with irregular triangular meshes are obtained, and high-precision camera parameters for each image are calculated through aerial triangulation. Color consistency preprocessing is performed. Due to changes in lighting and automatic white balance factors, there are color differences between different images. Color equalization is performed on the entire image set. Mesh projection and visibility calculation utilize camera intrinsic and extrinsic parameters to project each vertex of the triangulation onto every image that can see it; image selection uses frontal detection, considering only images where the angle between the camera's line of sight and the triangle's normal direction is small; occlusion detection uses Z-Buffer or ray tracing techniques to determine whether a triangle is occluded by other parts of the model in a given image, and only fully visible triangles are included as candidates. Based on the camera pose parameters obtained from aerial triangulation, the correspondence between each image and the irregular triangular mesh is calculated. First, the three-dimensional triangular mesh is unfolded onto a two-dimensional plane, and a two-dimensional (u, v) coordinate is assigned to each three-dimensional vertex to minimize the distortion and stretching of the unfolded triangle. For complex, large models, a single texture map cannot contain all the details. Therefore, the UV space is divided into multiple texture tiles, and each texture tile generates an independent texture image. Optimal viewpoint selection: For each triangle in the texture tile, select the best one or several images from all images that can see the triangle as the color source and arrange them compactly in the triangle; the image selection criteria are: the larger the pixel area occupied by the triangle in the image, the clearer the details, and the higher the priority; the smaller the angle between the camera's line of sight and the triangle's normal, the smaller the texture deformation and the higher the quality; prioritize images that are not occluded and are far from the image edge. Seam optimization: The main causes of seams are that when adjacent texture tiles are rendered separately, discontinuities appear at the boundaries due to differences in color sources or processing; within the same texture tile, adjacent triangles may come from different source images, and if these images have differences in color or brightness, visible seams will be formed. For a given texture image, instead of using only the color from the optimal viewpoint, the color values ​​from multiple candidate viewpoints are fused. The contribution weight of each candidate viewpoint is determined based on the viewpoint angle, image resolution, and distance from the image center. Among these factors, the viewpoint angle is higher when it is directly facing the triangle; the image resolution is higher when it is higher; and the distance from the image center is higher when it is closer to the image center and the less lens distortion there is. Final color These are the candidate colors. Its weight The weighted average and weighted mixed calculation formulas are as follows: (6) At the boundary of the tiles, a transition zone is set to allow the colors of adjacent tiles to blend smoothly within this area, thereby eliminating hard boundaries; After generating all texture maps, perform an overall color balance to ensure that the colors of the entire model are consistent. Export the triangular mesh with UV coordinates and the generated texture image together into a common 3D format to obtain a complete 3D model with realistic visual effects.

[0012] Optionally, in some implementations, the exterior and interior orientation elements of the image are calculated using aerial triangulation, including: The exterior and interior orientation elements of the images are calculated using an aerial triangulation algorithm to establish the geometric relationships between the images. The calculation formula is as follows: (1) In the formula, (X, Y, Z) are the coordinates of the ground point. , , ( ) represents the coordinates of the photography center. R is the scaling factor, R is the rotation matrix, and (x, y, z) are the image point coordinates.

[0013] Optionally, in some implementations, the collaborative collection of spoil heap topography data includes: The topographic data of the spoil disposal site was collected by using a drone equipped with a lidar system, and high-precision three-dimensional geospatial data was obtained by using active laser scanning and multi-sensor fusion technology. Through non-repetitive scanning technology, it achieves full coverage of a horizontal field of view of 70.4° and a vertical field of view of 4.5°, with a point frequency of 240,000 pts / s in single-echo mode, and supports three-echo penetration of the vegetation canopy. The UAV RTK positioning system synchronizes in real time with the built-in IMU of the lidar. By fusing GNSS, IMU and laser ranging data through a tightly coupled Kalman filter algorithm, the three-dimensional spatial coordinates (X, Y, Z) of each laser point are generated. The RGB camera and LiDAR are triggered in a spatiotemporal synchronization to achieve true color coloring of point clouds.

[0014] The technical solution provided in this application may include the following beneficial effects: A multi-view stereo matching algorithm and point cloud optimization combining statistical filtering and radius filtering are employed to effectively remove noise points while preserving terrain details, thus improving point cloud quality. A progressive triangulation algorithm is used to construct and optimize irregular triangulation networks, dynamically adjusting the network density and structure to avoid model breakage or excessive smoothing under complex terrain conditions, significantly improving the accuracy and detail representation of the digital elevation model. Texture mapping, image registration, and seam optimization are performed on the optimized triangulation networks, and a weighted mixing algorithm is used to eliminate color differences and seams between images, giving the 3D model a realistic visual effect and accurately reflecting the actual surface conditions of the spoil heap.

[0015] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0016] The above and other objects, features and advantages of this application will become more apparent from the more detailed description of exemplary embodiments thereof in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments thereof.

[0017] Figure 1 This is a flowchart illustrating a method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data, as shown in an embodiment of this application. Figure 2 This is a schematic diagram comparing the point cloud denoising effect of the three-dimensional morphological modeling method for spoil heaps based on temporal LiDAR data, as shown in the embodiments of this application. Figure 3 This is a schematic diagram comparing point cloud data before and after the progressive triangulation encryption algorithm of the three-dimensional morphology modeling method for spoil heaps based on time-series LiDAR data, as shown in the embodiments of this application. Figure 4 This is a schematic diagram of point cloud data RGB and texture image of the three-dimensional morphological modeling method for spoil heaps based on temporal LiDAR data, as shown in the embodiments of this application. Detailed Implementation

[0018] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to make this application more thorough and complete, and to fully convey the scope of this application to those skilled in the art.

[0019] In special scenarios like spoil heaps, single technical methods are easily constrained by environmental factors. Spoil heap point cloud data contains non-ground noise points such as vegetation and buildings, as well as outliers caused by measurement errors, making it difficult to retain terrain details while removing noise, directly affecting modeling accuracy. When dealing with typical abrupt terrain changes in spoil heaps such as steep slopes and gullies, problems such as model breakage, over-smoothing, or excessive redundant points easily occur, making it impossible to accurately reproduce the true structure of the terrain. Furthermore, factors such as changes in lighting and differences in image acquisition perspectives result in significant color differences between different images, causing texture seams in the 3D model's texture map, poor visual realism, and an inability to accurately reflect the actual surface condition of the spoil heap.

[0020] To address the aforementioned issues, this application provides a method for 3D morphological modeling of spoil heaps based on time-series LiDAR data. It employs a multi-view stereo matching algorithm and point cloud optimization that integrates statistical filtering and radius filtering to effectively remove noise points while preserving terrain details, thus improving point cloud quality. A progressive triangulation algorithm is used to construct and optimize irregular triangulation networks, dynamically adjusting the network density and structure to avoid model breakage or excessive smoothing under complex terrain conditions, significantly improving the accuracy and detail representation of the digital elevation model. By performing texture mapping, image registration, and seam optimization on the optimized triangulation network, a weighted mixing algorithm is used to eliminate color differences and seams between images, giving the 3D model a realistic visual effect and accurately reflecting the actual surface conditions of the spoil heap.

[0021] The technical solutions of the embodiments of this application are described in detail below with reference to the accompanying drawings.

[0022] Figure 1 This is a flowchart illustrating a method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data, as shown in an embodiment of this application.

[0023] See Figure 1 A method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data, comprising: S101. Use UAV oblique photogrammetry and lidar technology to collect collaborative topographic data of spoil disposal sites; Specifically, the collaborative collection of topographic data for spoil disposal sites includes: The topographic data of the spoil disposal site was collected using a DJI M300 drone equipped with an L1 lidar system, which utilizes active laser scanning and multi-sensor fusion technology to obtain high-precision three-dimensional geospatial data. Through non-repetitive scanning technology, it achieves full coverage of a horizontal field of view of 70.4° and a vertical field of view of 4.5°, with a point frequency of 240,000 pts / s in single-echo mode, and supports three-echo penetration of the vegetation canopy. The UAV RTK positioning system synchronizes in real time with the built-in IMU of the lidar. By fusing GNSS, IMU and laser ranging data through a tightly coupled Kalman filter algorithm, the three-dimensional spatial coordinates (X, Y, Z) of each laser point are generated. The RGB camera and LiDAR are triggered in a spatiotemporal synchronization to achieve true color coloring of point clouds.

[0024] High-precision geospatial data acquisition using the DJI M300 drone requires the deployment of ground-based RTK control points to achieve centimeter-level positioning accuracy. The drone flight process integrates air-to-ground cooperative positioning technology, as detailed below: Thirteen ground-based RTK control points were evenly distributed within the study area, with a spacing of ≤30 m, covering the boundary and central area of ​​the survey area. Control points were marked with fixed PVC boards and L-shaped markers sprayed onto the ground, and real-world photographs and RTK coordinate attribute tables were recorded simultaneously at each point.

[0025] By loading control point coordinates using DJI Pilot 2, a terrain-following flight path was designed, with a flight altitude of 120 m, a flight speed of 9 m / s, a heading overlap of 85%, and a lateral overlap of 80%. The Qianxun Location FindCM service was enabled to achieve real-time differential positioning via network RTK, and the L1 camera of the H20T gimbal was simultaneously activated.

[0026] After the drone is manually taken off and reaches a safe altitude of 50 m, it switches to an automatic flight path and transmits visible light and LiDAR data back in real time through the OcuSync 3.0 image transmission system.

[0027] After exporting the original imagery, aerial triangulation was performed using DJI Terra. Ten of the 13 control points were used as adjustment benchmarks, and the remaining three were used as checkpoints to evaluate absolute accuracy.

[0028] S102. Import the acquired image data into 3D modeling software for preliminary denoising, correction, and stitching to generate orthophotos. Calculate the exterior and interior orientation elements of the images using aerial triangulation to establish geometric relationships between them. During this process, stable feature points are automatically extracted based on overlapping areas, and feature matching is performed between multiple views. Subsequently, a topological connection structure between images is constructed based on the feature point matching relationships, and abnormal connections with large matching errors are removed. On this basis, combined with ground control points and high-precision UAV positioning data, unified geometric constraints and overall optimization are applied to all images to ensure consistency in spatial position and attitude parameters, thus forming a complete and stable image geometric relationship network.

[0029] Specifically, the processing of the acquired image data includes image import and preprocessing and aerial triangulation calculation. In the import and preprocessing stage, the acquired image data is imported into 3D modeling software for preliminary denoising, correction and stitching to generate high-precision orthophotos. The exterior and interior orientation elements of the imagery are calculated using aerial triangulation, including: The exterior and interior orientation elements of the images are calculated using an aerial triangulation algorithm to establish the geometric relationships between the images. The calculation formula is as follows: (1) In the formula, (X, Y, Z) are the coordinates of the ground point. , , ( ) represents the coordinates of the photography center. R is the scaling factor, R is the rotation matrix, and (x, y, z) are the image point coordinates.

[0030] S103. Dense point cloud generation includes dense matching and point cloud optimization. Dense matching is performed using a multi-view stereo matching algorithm, and point cloud optimization is achieved by fusing statistical filtering and radius filtering. Specifically, dense matching employs a multi-view stereo matching algorithm, including: Dense matching uses a multi-view stereo matching algorithm to generate dense point clouds. The specific steps are as follows: Based on the aerial triangulation results, stereo correction is performed on multiple images with overlapping areas to ensure that the image points of the same name are located within the corresponding epipolar constraint range. Within the overlapping region of multi-view images, a matching window is constructed using pixels as the basic unit, and the similarity cost between images from different viewpoints is calculated to obtain candidate matching relationships. The matching results are filtered by multi-view consistency constraints to remove mismatched pixels; Based on the disparity information obtained from the final matching, the three-dimensional coordinates of the corresponding ground points are calculated using the camera geometric model to generate a dense three-dimensional point cloud.

[0031] The formula for calculating dense matching is: (2) In the formula, The matching cost is given by p, where p is the number of image points and d is the disparity. For neighborhood windows, For the weight function, This is a similarity measurement function.

[0032] Specifically, point cloud optimization is achieved by fusing statistical filtering and radius filtering, including: To address the issue that noisy points and outliers in point cloud data can significantly reduce the accuracy of subsequent modeling and analysis, statistical filtering and radius filtering are two efficient denoising methods that improve the quality of point cloud data. By fusing statistical filtering and radius filtering for multi-level denoising, robustness in complex scenarios can be enhanced. The multi-level denoising effect is significant, for example... Figure 2 As shown; Statistical filtering is a point cloud denoising method based on local neighborhood statistical analysis. Its core principle is that effective point clouds exhibit statistical consistency in spatial distribution, while noise points, deviating from the neighborhood distribution pattern, can be identified and removed. The neighborhood is defined as the distribution of each point in the point cloud. Select the k nearest neighbors to form a local neighborhood set N(k,k). ), k is a preset parameter, usually taken as 20~50; the distance statistic calculation is the calculation of... Average distance to all neighboring points and standard deviation The formula is: (3) (4) In the formula, For point With neighboring points The Euclidean distance; Anomaly detection is achieved by setting a threshold. , This is an empirical coefficient. The value is 2, if the point average neighborhood distance If the value exceeds T, it is identified as a noise point and removed. The performance of statistical filtering is highly dependent on the parameters k and With proper setting, a smaller k value is suitable for dense point clouds and can preserve detailed features; a larger k value is suitable for sparse regions and enhances statistical stability, but may produce excessive smoothing. Radius filtering is a point cloud denoising method based on spatial density. It identifies and removes isolated noise points by analyzing the distribution density of points within a local neighborhood. The process includes neighborhood definition, density determination, and parameter setting. Neighborhood definition involves defining the neighborhood for each point in the point cloud. A spherical neighborhood space is constructed with the neighborhood as the center and a preset search radius r. Density is determined by counting the number of points within the neighborhood. ,like Less than the preset threshold Then determine For noise points, the mathematical expression for the removal criteria is: (5) In the formula, 1 is an indicator function.

[0033] S104. The density and structure of the initial irregular triangular mesh are constructed and optimized using a progressive triangular mesh encryption algorithm to improve the accuracy and detail representation capability of the digital elevation model under complex terrain conditions. Specifically, the progressive triangular mesh encryption algorithm is used to dynamically optimize the density and structure of irregular triangular meshes, including: The progressive triangulation encryption algorithm is an efficient terrain modeling method based on LiDAR point cloud data. Its core objective is to improve the accuracy and detail representation of digital elevation models under complex terrain conditions by dynamically optimizing the density and structure of the triangulation network. The progressive triangulation encryption algorithm addresses the model breakage or oversmoothing problems that easily occur in triangulation construction methods in abrupt terrain areas such as steep slopes and gullies. It adaptively enhances the local point cloud density through an iterative encryption strategy, while effectively eliminating redundant points to reduce computational complexity. The progressive triangulation encryption algorithm consists of four stages: initial seed point selection, iterative triangulation encryption, terrain feature constraint optimization, and final model verification. Based on the point cloud spatial distribution uniformity criterion, a sparse and representative seed point set is selected to construct an initial irregular triangular network. The terrain feature lines are identified through curvature analysis to ensure that the initial irregular triangular network retains the macroscopic terrain outline. Based on the initial irregular triangulation network, point clouds with elevation residuals exceeding the threshold are inserted layer by layer. The local triangular patches are updated using the Delaunay triangulation algorithm criterion. At the same time, the terrain roughness index is introduced to dynamically adjust the densification threshold to avoid excessive subdivision of flat areas. After each iteration, the difference between the interpolation error and the normal vector of the adjacent triangular facet is calculated to determine whether further refinement or termination of local optimization is needed. By combining the point cloud classification module of 3D modeling software, non-ground points such as vegetation and buildings are removed, and the density of triangulated mesh nodes in areas of abrupt terrain changes is forcibly increased. Terrain structure continuity is maintained through constraint edge connection rules, achieving the following effect: Figure 3 As shown.

[0034] S105. Perform texture mapping, image registration, and seam optimization on the optimized irregular triangular mesh to realize the construction of a real-scene 3D model; Specifically, the optimized irregular triangular mesh undergoes texture mapping, image registration, and seam optimization, including: Data preparation is aligned with input to ensure that all original UAV images with irregular triangular meshes are obtained, and high-precision camera parameters for each image are calculated through aerial triangulation. Color consistency preprocessing is performed. Due to changes in lighting and automatic white balance factors, there are color differences between different images. Color equalization is performed on the entire image set. Mesh projection and visibility calculation utilize camera intrinsic and extrinsic parameters to project each vertex of the triangulation onto every image that can see it; image selection uses frontal detection, considering only images where the angle between the camera's line of sight and the triangle's normal direction is small; occlusion detection uses Z-Buffer or ray tracing techniques to determine whether a triangle is occluded by other parts of the model in a given image, and only fully visible triangles are included as candidates. Based on the camera pose parameters obtained from aerial triangulation, the correspondence between each image and the irregular triangular mesh is calculated. First, the three-dimensional triangular mesh is unfolded onto a two-dimensional plane, and a two-dimensional (u, v) coordinate is assigned to each three-dimensional vertex to minimize the distortion and stretching of the unfolded triangle. For complex, large models, a single texture map cannot contain all the details. Therefore, the UV space is divided into multiple texture tiles, and each texture tile generates an independent texture image. Optimal viewpoint selection: For each triangle in the texture tile, select the best one or several images from all images that can see the triangle as the color source and arrange them compactly in the triangle; the image selection criteria are: the larger the pixel area occupied by the triangle in the image, the clearer the details, and the higher the priority; the smaller the angle between the camera's line of sight and the triangle's normal, the smaller the texture deformation and the higher the quality; prioritize images that are not occluded and are far from the image edge. Seam optimization: The main causes of seams are that when adjacent texture tiles are rendered separately, discontinuities appear at the boundaries due to differences in color sources or processing; within the same texture tile, adjacent triangles may come from different source images, and if these images have differences in color or brightness, visible seams will be formed. For a given texture image, instead of using only the color from the optimal viewpoint, the color values ​​from multiple candidate viewpoints are fused. The contribution weight of each candidate viewpoint is determined based on the viewpoint angle, image resolution, and distance from the image center. Among these factors, the viewpoint angle is higher when it is directly facing the triangle; the image resolution is higher when it is higher; and the distance from the image center is higher when it is closer to the image center and the less lens distortion there is. Final color These are the candidate colors. Its weight The weighted average and weighted mixed calculation formulas are as follows: (6) At the boundary of the tiles, a transition zone is set to allow the colors of adjacent tiles to blend smoothly within this area, thereby eliminating hard boundaries; After generating all texture maps, perform an overall color balance to ensure that the colors of the entire model are consistent. Exporting the triangular mesh with UV coordinates and the generated texture image together into a common 3D format yields a complete 3D model with realistic visual effects, such as... Figure 4 As shown.

[0035] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data, characterized in that: UAV oblique photogrammetry and lidar technology were used to collect collaborative topographic data of spoil disposal sites. The acquired image data is imported into 3D modeling software for preliminary denoising, correction and stitching to generate orthophotos. The exterior and interior orientation elements of the images are calculated by aerial triangulation to establish the geometric relationship between the images. Dense point cloud generation includes dense matching and point cloud optimization. Dense matching is performed using a multi-view stereo matching algorithm, and point cloud optimization is achieved by fusing statistical filtering and radius filtering. A progressive triangulation encryption algorithm is used to construct and optimize the density and structure of the initial irregular triangulation network, thereby improving the accuracy and detail representation capability of the digital elevation model under complex terrain conditions. Texture mapping, image registration, and seam optimization are performed on the optimized irregular triangular mesh to construct a realistic 3D model. Texture mapping uses color information from UAV images to fit the surface of the irregular triangular mesh, giving the model a realistic visual effect. Image registration calculates the correspondence between each image and the irregular triangular mesh based on camera pose parameters obtained from aerial triangulation. Seam optimization uses a weighted mixing algorithm to process overlapping areas, eliminating color differences and seams between images.

2. The method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data according to claim 1, characterized in that, The method of dynamically optimizing the density and structure of irregular triangular networks using a progressive triangular network encryption algorithm includes: The progressive triangulation encryption algorithm consists of four stages: initial seed point selection, iterative triangulation encryption, terrain feature constraint optimization, and final model verification. Based on the point cloud spatial distribution uniformity criterion, a sparse and representative seed point set is selected to construct an initial irregular triangular network. The terrain feature lines are identified through curvature analysis to ensure that the initial irregular triangular network retains the macroscopic terrain outline. Based on the initial irregular triangulation network, point clouds with elevation residuals exceeding the threshold are inserted layer by layer. The local triangular patches are updated using the Delaunay triangulation algorithm criterion. At the same time, the terrain roughness index is introduced to dynamically adjust the densification threshold to avoid excessive subdivision of flat areas. After each iteration, the difference between the interpolation error and the normal vector of the adjacent triangular facet is calculated to determine whether further refinement or termination of local optimization is needed. By combining the point cloud classification module, vegetation and building non-ground points are removed, the density of triangular mesh nodes in areas of abrupt terrain changes is forcibly increased, and the continuity of terrain structure is maintained by constraining edge connection rules.

3. The method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data according to claim 1, characterized in that, The point cloud optimization achieved by fusing statistical filtering and radius filtering includes: Multi-stage denoising is performed by fusing statistical filtering and radius filtering to improve robustness in complex scenarios; Statistical filtering works by identifying and removing noise points that deviate from their neighborhood distribution, as valid point clouds exhibit statistical consistency in spatial distribution. The neighborhood is defined as the distribution of each point in the point cloud. Select the k nearest neighbors to form a local neighborhood set N(k,k). ), k is a preset parameter, usually taken as 20~50; the distance statistic calculation is the calculation of... Average distance to all neighboring points and standard deviation The formula is: （3） （4） In the formula, For point With neighboring points The Euclidean distance; Anomaly detection is achieved by setting a threshold. , This is an empirical coefficient. The value is 2, if the point average neighborhood distance If the value exceeds T, it is identified as a noise point and removed. The performance of statistical filtering is highly dependent on the parameters k and With proper setting, a smaller k value is suitable for dense point clouds and can preserve detailed features; a larger k value is suitable for sparse regions and enhances statistical stability, but may produce excessive smoothing. Radius filtering identifies and removes isolated noise points by analyzing the distribution density of points within a local neighborhood. The process includes neighborhood definition, density determination, and parameter setting. Neighborhood definition involves defining each point in the point cloud. A spherical neighborhood space is constructed with the neighborhood as the center and a preset search radius r. Density is determined by counting the number of points within the neighborhood. ,like Less than the preset threshold Then determine For noise points, the mathematical expression for the removal criteria is: （5） In the formula, 1 is an indicator function.

4. The method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data according to claim 1, characterized in that, The method of using a multi-view stereo matching algorithm for dense matching includes: The formula for calculating the dense matching is: （2） In the formula, The matching cost is given by p, where p is the number of image points and d is the disparity. For neighborhood windows, For the weight function, This is a similarity measurement function.

5. The method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data according to claim 1, characterized in that, The process of performing texture mapping, image registration, and seam optimization on the optimized irregular triangular mesh includes: Data preparation is aligned with input to ensure that all original UAV images with irregular triangular meshes are obtained, and high-precision camera parameters for each image are calculated through aerial triangulation. Color consistency preprocessing is performed. Due to changes in lighting and automatic white balance factors, there are color differences between different images. Color equalization is performed on the entire image set. Mesh projection and visibility calculation utilize camera intrinsic and extrinsic parameters to project each vertex of the triangulation onto every image that can see it; image selection uses frontal detection, considering only images where the angle between the camera's line of sight and the triangle's normal direction is small; occlusion detection uses Z-Buffer or ray tracing techniques to determine whether a triangle is occluded by other parts of the model in a given image, and only fully visible triangles are included as candidates. Based on the camera pose parameters obtained from aerial triangulation, the correspondence between each image and the irregular triangular mesh is calculated. First, the three-dimensional triangular mesh is unfolded onto a two-dimensional plane, and a two-dimensional (u, v) coordinate is assigned to each three-dimensional vertex to minimize the distortion and stretching of the unfolded triangle. For complex, large models, a single texture map cannot contain all the details. Therefore, the UV space is divided into multiple texture tiles, and each texture tile generates an independent texture image. Optimal viewpoint selection: For each triangle in the texture tile, select the best one or several images from all images that can see the triangle as the color source and arrange them compactly in the triangle; the image selection criteria are: the larger the pixel area occupied by the triangle in the image, the clearer the details, and the higher the priority; the smaller the angle between the camera's line of sight and the triangle's normal, the smaller the texture deformation and the higher the quality; prioritize images that are not occluded and are far from the image edge. Seam optimization: The main causes of seams are that when adjacent texture tiles are rendered separately, discontinuities appear at the boundaries due to differences in color sources or processing; within the same texture tile, adjacent triangles may come from different source images, and if these images have differences in color or brightness, visible seams will be formed. For a given texture image, instead of using only the color from the optimal viewpoint, the color values ​​from multiple candidate viewpoints are fused. The contribution weight of each candidate viewpoint is determined based on the viewpoint angle, image resolution, and distance from the image center. Among these factors, the viewpoint angle is higher when it is directly facing the triangle; the image resolution is higher when it is higher; and the distance from the image center is higher when it is closer to the image center and the less lens distortion there is. Final color These are the candidate colors. Its weight The weighted average and weighted mixed calculation formulas are as follows: （6） At the boundary of the tiles, a transition zone is set to allow the colors of adjacent tiles to blend smoothly within this area, thereby eliminating hard boundaries; After generating all texture maps, perform an overall color balance to ensure that the colors of the entire model are consistent. Export the triangular mesh with UV coordinates and the generated texture image together into a common 3D format to obtain a complete 3D model with realistic visual effects.

6. The method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data according to claim 1, characterized in that, The calculation of the exterior and interior orientation elements of the image through aerial triangulation includes: The exterior and interior orientation elements of the images are calculated using an aerial triangulation algorithm to establish the geometric relationships between the images. The calculation formula is as follows: （1） In the formula, (X, Y, Z) are the coordinates of the ground point. , , ( ) represents the coordinates of the photography center. R is the scaling factor, R is the rotation matrix, and (x, y, z) are the image point coordinates.

7. The method for three-dimensional morphological modeling of spoil heaps based on time-series LiDAR data according to claim 1, characterized in that, The collaborative data collection of the spoil disposal site topography includes: The topographic data of the spoil disposal site was collected by using a drone equipped with a lidar system, and high-precision three-dimensional geospatial data was obtained by using active laser scanning and multi-sensor fusion technology. Through non-repetitive scanning technology, it achieves full coverage of a horizontal field of view of 70.4° and a vertical field of view of 4.5°, with a point frequency of 240,000 pts / s in single-echo mode, and supports three-echo penetration of the vegetation canopy. The UAV RTK positioning system synchronizes in real time with the built-in IMU of the lidar. By fusing GNSS, IMU and laser ranging data through a tightly coupled Kalman filter algorithm, the three-dimensional spatial coordinates (X, Y, Z) of each laser point are generated. The RGB camera and LiDAR are triggered in a spatiotemporal synchronization to achieve true color coloring of point clouds.

Images