A fast three-dimensional modeling method, medium and system based on underwater point cloud data

Abstract

The present invention provides a fast three-dimensional modeling method, medium and system based on underwater point cloud data, belonging to the technical field of underwater depth measurement, including: First, obtain low-resolution point cloud data through spatial downsampling, and perform multivariate mixed decomposition on it to obtain stable components and variable components. Construct an initial three-dimensional mesh model based on the stable components, and then use an adaptive block strategy to divide it into multiple sub-regions. Process the original point cloud data by sub-region to obtain multiple high-resolution point cloud data blocks. Next, perform anomaly recognition and noise reduction processing on each high-resolution point cloud data block, reconstruct the corrected high-resolution point cloud data block, and perform mesh reconstruction and boundary registration to construct a complete high-resolution three-dimensional model of the underwater scene. After processing, a final high-quality three-dimensional model of the underwater scene is obtained, solving the technical problem that existing three-dimensional underwater models often cannot fully reflect the detailed features of the underwater terrain.

Description

Technical Field

[0001] The present invention belongs to the technical field of seabed depth measurement, and in particular, relates to a fast three-dimensional modeling method, medium and system based on seabed point cloud data. Background Art

[0002] Seabed topography mapping is one of the basic tasks in the fields of marine scientific research and underwater engineering construction. With the continuous deepening of marine development activities, human demand for seabed topography information is also growing. Traditional seabed topography surveys usually rely on acoustic measurement technology, such as multi-beam echo sounders, single-beam echo sounders, etc. These technologies can obtain three-dimensional point cloud data of seabed topography, providing a basis for subsequent three-dimensional modeling and analysis. However, due to the complexity of the marine environment and the limitations of measurement equipment, these traditional methods still have certain limitations in data acquisition efficiency, resolution and accuracy. For example, the scanning range and resolution of acoustic measurement equipment are limited by many factors, such as water depth, ocean currents, turbulence, etc. In deep-sea environments, it is difficult for measurement equipment to obtain high-resolution seabed topography data. Therefore, the existing three-dimensional seabed models often cannot fully reflect the detailed characteristics of the seabed topography, and it is difficult to meet the needs of underwater engineering construction, marine resource exploration, etc. for high-precision seabed topography information. Summary of the invention

[0003] In view of this, the present invention provides a rapid three-dimensional modeling method, medium and system based on seabed point cloud data, which can solve the technical problem that existing three-dimensional seabed models often cannot fully reflect the detailed features of the seabed topography.

[0004] The present invention is achieved in that:

[0005] The present invention provides a rapid three-dimensional modeling method based on seabed point cloud data, which includes the following steps:

[0006] S01, collecting seabed point cloud data, and performing spatial downsampling processing on the seabed point cloud data to obtain low-resolution point cloud data;

[0007] S02, performing multivariate mixed decomposition calculation on the low-resolution point cloud data to obtain low-resolution point cloud stable component data and low-resolution point cloud variable component data;

[0008] S03, constructing an initial three-dimensional grid model of the seabed scene based on the low-resolution point cloud stable component data, and calculating the spatial curvature distribution of the initial three-dimensional grid model;

[0009] S04, adaptively dividing the initial three-dimensional grid model into blocks according to the spatial curvature distribution, and dividing the initial three-dimensional grid model into a plurality of sub-regions;

[0010] S05. Spatially partition the original seabed point cloud data according to the multiple sub-regions to obtain multiple high-resolution point cloud data blocks;

[0011] S06. Perform multivariate mixture decomposition calculation on each of the high-resolution point cloud data blocks to respectively obtain the corresponding point cloud stable component data and point cloud variable component data;

[0012] S07. Perform Fourier transform on the point cloud variable component data of each of the high-resolution point cloud data blocks, establish a point cloud spatial distribution feature matrix, extract the main eigenvector, and construct a point cloud anomaly recognition model;

[0013] S08. Calculate the anomaly coefficient matrix of each of the high-resolution point cloud data blocks according to the point cloud anomaly recognition model, perform noise reduction processing on the point cloud variable component data, and obtain the point cloud corrected component data;

[0014] S09. Combine and reconstruct the point cloud stable component data and the point cloud corrected component data of each of the high-resolution point cloud data blocks to obtain the corrected high-resolution point cloud data blocks;

[0015] S10. Perform grid reconstruction on each of the corrected high-resolution point cloud data blocks to generate a high-resolution sub-region three-dimensional model;

[0016] S11. Perform boundary registration and stitching on the high-resolution sub-region three-dimensional models to construct a complete high-resolution seabed scene three-dimensional model;

[0017] S12. Perform global geometric optimization and texture mapping on the high-resolution seabed scene three-dimensional model to obtain the final seabed scene three-dimensional model.

[0018] Based on the above technical solutions, a fast three-dimensional modeling method based on seabed point cloud data of the present invention can also be improved as follows:

[0019] Among them, the step S01 specifically includes: collecting the original point cloud data of the seabed area and performing voxel downsampling processing on the original point cloud data; wherein, the voxel downsampling processing is to divide the three-dimensional space into several voxel units, perform statistical calculation on the point cloud data inside each voxel unit to obtain the average value or median value of the points inside each voxel unit; the size of the voxel unit is adjustable within the range of 0.01 meters to 1 meter, and the balance between the calculation efficiency and the retention degree of the seabed terrain details is achieved by adjusting the size of the voxel unit.

[0020] Further, the step S02 specifically includes: performing multivariate mixture decomposition calculation on the low-resolution point cloud data, and decomposing the low-resolution point cloud data into low-resolution point cloud stable component data and low-resolution point cloud variable component data; wherein, the low-resolution point cloud stable component data represents the basic shape of the seabed terrain and is linearly represented by using 5 to 10 basis functions; the low-resolution point cloud variable component data represents the minute changes in the seabed terrain and is obtained by subtracting the low-resolution point cloud stable component data from the low-resolution point cloud data.

[0021] The step S03 specifically includes: constructing an initial three-dimensional grid model of the seabed scene based on the low-resolution point cloud stable component data; calculating the spatial curvature distribution on the surface of the initial three-dimensional grid model, and the spatial curvature distribution is obtained by calculating the gradient operator and Hessian matrix of each point of the initial three-dimensional grid model; the spatial curvature distribution is used to characterize the complexity and change characteristics of the seabed terrain.

[0022] The step S04 specifically includes: calculating the curvature mean and curvature standard deviation according to the spatial curvature distribution; taking the weighted sum of the curvature mean and the curvature standard deviation as the block threshold, and the value range of the weight coefficient in the weighted sum is 1 to 3; dividing the grid cells with curvature values higher than the block threshold into separate sub-regions; improving the geometric accuracy of the three-dimensional model in complex regions through the adaptive block strategy.

[0023] The step S05 specifically includes: performing spatial partitioning on the original seabed point cloud data according to the sub-region partitioning scheme of the initial three-dimensional grid model; dividing the partitioned original seabed point cloud data into multiple high-resolution point cloud data blocks; the high-resolution point cloud data blocks retain the complete spatial distribution information of the original seabed point cloud data.

[0024] The step S06 specifically includes: performing multivariate mixture decomposition calculation on each high-resolution point cloud data block; linearly decomposing each high-resolution point cloud data block by using 10 to 20 detail basis functions to obtain the corresponding point cloud stable component data; obtaining the point cloud variable component data by subtracting the point cloud stable component data from the high-resolution point cloud data block.

[0025] The step S07 specifically includes: performing three-dimensional discrete Fourier transform on the point cloud variable component data of each high-resolution point cloud data block; establishing a point cloud spatial distribution feature matrix according to the result of the three-dimensional discrete Fourier transform; extracting the main eigenvector by performing eigenanalysis on the point cloud spatial distribution feature matrix; constructing a point cloud anomaly recognition model based on the main eigenvector.

[0026] The specific steps of S08 include: calculating the anomaly coefficient matrix of each high-resolution point cloud data block according to the point cloud anomaly recognition model; selecting 3 to 5 main eigenvectors to construct the anomaly coefficient matrix; using the anomaly coefficient matrix to perform noise reduction processing on the point cloud change component data to obtain the point cloud correction component data.

[0027] The specific steps of S09 include: linearly superimposing the point cloud stable component data and the point cloud correction component data of each high-resolution point cloud data block; obtaining the corrected high-resolution point cloud data block through the linear superposition; the corrected high-resolution point cloud data block has higher spatial accuracy and lower noise level.

[0028] The specific steps of S10 include: processing each corrected high-resolution point cloud data block using a point cloud meshing algorithm; the point cloud meshing algorithm includes a Poisson surface reconstruction algorithm or a spherical projection algorithm; converting the discrete point cloud data into a continuous three-dimensional grid surface model through the point cloud meshing algorithm.

[0029] The specific steps of S11 include: extracting the boundary point pairs of the high-resolution sub-region three-dimensional model; establishing a boundary registration error function, the boundary registration error function includes a rigid body transformation term and a regularization term, and the coefficient value range of the regularization term is from 0.1 to 1; realizing the precise splicing of each high-resolution sub-region three-dimensional model by minimizing the boundary registration error function.

[0030] The specific steps of S12 include: establishing a smoothing error function, the smoothing error function includes a vertex displacement term and a normal vector term; the weight coefficient value range of the smoothing error function is from 0.01 to 0.1; geometrically optimizing the high-resolution seabed scene three-dimensional model by minimizing the smoothing error function; performing texture mapping processing on the optimized three-dimensional model.

[0031] Furthermore, the calculation process of the spatial downsampling processing in step S01 is specifically represented as follows:

[0032] ;

[0033] In the formula, is the original point cloud data set; is the downsampled point cloud data set; is the voxel downsampling function; is the voxel size, and the value range is [0.01m, 1m]; is the downsampling error.

[0034] Furthermore, the multivariate mixed decomposition calculation in step S02 is specifically represented as follows:

[0035] ;

[0036] ;

[0037] ;

[0038] In the formula, is the stable component of the low-resolution point cloud; is the variable component of the low-resolution point cloud; is the th basis function; is the basis function coefficient; is the number of basis functions, generally taking 5 - 10.

[0039] Furthermore, the calculation of the spatial curvature distribution in step S03 is specifically expressed as follows:

[0040] ;

[0041] In the formula, is the curvature value at point ; is the implicit function of the point cloud fitting; is the gradient operator; is the Hessian matrix; is the curvature calculation error.

[0042] Furthermore, the adaptive block calculation in step S04 is specifically expressed as follows:

[0043] ;

[0044] In the formula, is the block threshold; is the mean curvature; is the standard deviation of the curvature; is the weight coefficient, with a value range of [1, 3].

[0045] Furthermore, the high-resolution point cloud decomposition calculation in step S06 is specifically expressed as follows:

[0046] ;

[0047] ;

[0048] ;

[0049] In the formula, is the sub-region index; is the th high-resolution point cloud data block; is the stable component; is the variable component; is the th detail basis function; is the coefficient of the detail basis function; is the number of detail basis functions, generally taking 10 - 20.

[0050] Furthermore, the Fourier transform calculation in step S07 is specifically expressed as follows:

[0051] ;

[0052] In the formula, is the result of the three-dimensional Fourier transform; is the number of spatial sampling points; is the imaginary unit; is the frequency domain coordinate.

[0053] Furthermore, the calculation of the anomaly coefficient matrix in step S08 is specifically expressed as follows:

[0054] ;

[0055] ;

[0056] In the formula, is the anomaly coefficient matrix; is the eigenvalue; is the eigenvector; is the number of main features, generally taking 3 - 5; is the corrected variable component.

[0057] Furthermore, the point cloud reconstruction calculation in step S09 is specifically expressed as follows:

[0058] ;

[0059] In the formula, is the reconstructed high-resolution point cloud; is the reconstruction error.

[0060] Furthermore, the boundary registration calculation in step S11 is specifically expressed as follows:

[0061] ;

[0062] In the formula, is the registration error; is the number of sub-regions; is the th number of boundary points of the sub-region; is the rigid body transformation matrix; is the corresponding point pair; is the regularization coefficient, and its value range is [0.1, 1]; is the identity matrix.

[0063] Furthermore, the geometric optimization calculation in step S12 is specifically expressed as follows:

[0064] ;

[0065] In the formula, is the smoothing error; is the vertex set; is the patch set; is the Laplace operator; is the vertex displacement; is the patch normal vector; is the target normal vector; is the weight coefficient, and its value range is [0.01, 0.1].

[0066] The second aspect of the present invention provides a computer-readable storage medium, wherein program instructions are stored in the computer-readable storage medium, and when the program instructions run on a computer, they are used to execute the above-mentioned fast three-dimensional modeling method based on underwater lidar point cloud data.

[0067] The third aspect of the present invention provides a fast three-dimensional modeling system based on underwater lidar point cloud data, which includes the above-mentioned computer-readable storage medium.

[0068] Compared with the prior art, the beneficial effects of the fast three-dimensional modeling method, medium and system based on underwater lidar point cloud data provided by the present invention are as follows:

[0069] 1. The geometric accuracy and resolution of the three-dimensional underwater model are improved. By adaptively refining the grid of the original underwater lidar point cloud data, the detailed features of complex underwater terrain can be effectively captured. Compared with the existing methods based on acoustic measurement, the three-dimensional model constructed by the method of the present invention can describe the underwater terrain more realistically and in more detail. This is of great significance for application scenarios such as underwater engineering design and resource exploration.

[0070] 2. The ability of the three-dimensional underwater model to express dynamic changes is enhanced. By decomposing the point cloud data into stable components and variable components, the present invention method can depict both the basic shape and the tiny changes of the underwater terrain. This ability is of great value for applications such as monitoring the evolution of underwater terrain and evaluating the impact of underwater engineering.

[0071] 3. The automation degree and efficiency of 3D modeling are improved. Compared with the traditional data processing process that involves more manual participation, the method of the present invention can extract a high-quality 3D seabed model from the original point cloud data more automatically. This not only improves work efficiency but also reduces the subjective errors introduced by humans.

[0072] 4. The application scenarios of the 3D seabed model are expanded. With its characteristics of high resolution and high precision, the 3D seabed model constructed by the method of the present invention can provide more reliable basic data support for fields such as ocean exploration, underwater engineering, and resource development. This will promote the further development of related application technologies.

[0073] Generally speaking, the fast 3D modeling method based on seabed point cloud data proposed by the present invention has made remarkable technological progress in improving modeling efficiency, enhancing model expression ability, expanding application scenarios, etc., and solves the technical problem that the existing 3D seabed models often cannot fully reflect the detailed features of the seabed terrain. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 is a flowchart of the method provided by the present invention;

[0075] Figure 2 is a heat map of the curvature distribution of a partial area of the seabed terrain in the embodiment;

[0076] Figure 3 is a result diagram of the spectral feature analysis of the point cloud variation component in the embodiment;

[0077] Figure 4 is a statistical distribution histogram of the registration error of the sub-region boundary in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0078] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0079] As Figure 1 shown, it is a flowchart of a fast 3D modeling method based on seabed point cloud data provided by the first aspect of the present invention. This method includes the following steps:

[0080] S01. Collect seabed point cloud data, perform spatial downsampling processing on the seabed point cloud data to obtain low-resolution point cloud data;

[0081] S02. Perform multivariate mixed decomposition calculation on the low-resolution point cloud data to obtain low-resolution point cloud stable component data and low-resolution point cloud variation component data;

[0082] S03. Construct an initial three-dimensional grid model of the seabed scene based on the stable component data of the low-resolution point cloud, and calculate the spatial curvature distribution of the initial three-dimensional grid model;

[0083] S04. Perform adaptive partitioning on the initial three-dimensional grid model according to the spatial curvature distribution, and divide the initial three-dimensional grid model into multiple sub-regions;

[0084] S05. Perform spatial partitioning on the original seabed point cloud data according to multiple sub-regions to obtain multiple high-resolution point cloud data blocks;

[0085] S06. Perform multivariate mixture decomposition calculation on each high-resolution point cloud data block to obtain the corresponding point cloud stable component data and point cloud variable component data respectively;

[0086] S07. Perform Fourier transform on the point cloud variable component data of each high-resolution point cloud data block, establish a point cloud spatial distribution feature matrix, extract the main eigenvector, and construct a point cloud anomaly recognition model;

[0087] S08. Calculate the anomaly coefficient matrix of each high-resolution point cloud data block according to the point cloud anomaly recognition model, perform noise reduction processing on the point cloud variable component data, and obtain the point cloud corrected component data;

[0088] S09. Combine and reconstruct the point cloud stable component data and the point cloud corrected component data of each high-resolution point cloud data block to obtain the corrected high-resolution point cloud data block;

[0089] S10. Perform grid reconstruction on each corrected high-resolution point cloud data block to generate a high-resolution sub-region three-dimensional model;

[0090] S11. Perform boundary registration and splicing on the high-resolution sub-region three-dimensional models to construct a complete high-resolution seabed scene three-dimensional model;

[0091] S12. Perform global geometric optimization and texture mapping on the high-resolution seabed scene three-dimensional model to obtain the final seabed scene three-dimensional model.

[0092] The following describes the specific implementation manners of the above steps in detail:

[0093] The specific implementation manner of step S01 is to first collect the original point cloud data of the seabed area . These original point cloud data usually have a high sampling density and resolution. In order to reduce the computational overhead of subsequent processing, it is necessary to process the original point cloud data Perform spatial downsampling. Specifically, the voxel downsampling method can be used. This method divides the three-dimensional space into a number of voxel units, and statistically processes the point cloud data inside each voxel, such as calculating the average or median value of the points inside the voxel, etc., so as to obtain the downsampled low-resolution point cloud data. The voxel size can be adjusted within the range of to achieve the goal of balancing the calculation efficiency and retaining the details of the seabed terrain. The downsampling process can be expressed as:

[0094] ;

[0095] where is the voxel downsampling function, and is the error introduced during the downsampling process.

[0096] The specific implementation of step S02 is to perform multivariate mixture decomposition calculation on the low-resolution point cloud data obtained in step S01. This calculation process can decompose the low-resolution point cloud data into two parts: the stable component and the variable component . Among them, the stable component represents the basic shape of the seabed terrain and can be linearly represented using a set of basis functions ; the variable component describes the minute changes in the seabed terrain. The mathematical description is as follows:

[0097] ;

[0098] ;

[0099] ;

[0100] where are the basis function coefficients, is the number of basis functions, usually taking ones. The purpose of this step is to decompose the low-resolution point cloud data into stable and variable components that can be processed independently.

[0101] The specific implementation of step S03 is, first, based on the low-resolution point cloud stable component obtained in step S02, construct an initial three-dimensional grid model of the seabed scene. In order to evaluate the geometric characteristics of this initial model, it is necessary to calculate the spatial curvature distribution on the model surface Curvature is an important indicator reflecting the complexity of geometric shapes and can be used to guide subsequent adaptive mesh refinement. The calculation formula for curvature is as follows:

[0102] ;

[0103] where, is the implicit function fitted by the point cloud, is the gradient operator, is the Hessian matrix, is the error introduced in the curvature calculation process. By calculating the curvature values of each point in the initial three-dimensional mesh model, the complexity and variation characteristics of the seabed terrain can be better understood.

[0104] The specific implementation of step S04 is to perform adaptive partitioning on the initial three-dimensional mesh model according to the spatial curvature distribution calculated in step S03. The purpose is to subdivide the areas with higher complexity while maintaining a lower mesh density in the flat areas. The specific partitioning strategy is as follows:

[0105] ;

[0106] where, is the partitioning threshold, is the mean value of curvature, is the standard deviation of curvature, is the weight coefficient, and its value range is . When the curvature value of a certain mesh cell is higher than this threshold , it is divided into a sub-region for subdivision processing. Through this adaptive partitioning strategy, the geometric accuracy of the three-dimensional model in complex regions can be effectively improved while maintaining a relatively low computational cost.

[0107] The specific implementation of step S05 is to perform spatial partitioning on the original seabed point cloud data according to the sub-region partitioning scheme obtained in step S04, thereby obtaining multiple high-resolution point cloud data blocks . These high-resolution point cloud data blocks will be used for subsequent fine-grained modeling.

[0108] The specific implementation of step S06 is to perform multivariate mixture decomposition calculation on each high-resolution point cloud data block to obtain the corresponding stable component and variable component . This step is similar to step S02, except that here the high-resolution point cloud data is processed instead of the previous low-resolution point cloud data. The mathematical description is as follows:

[0109] ;

[0110] ;

[0111] ;

[0112] Among them, is the th detail basis function, is the corresponding coefficient, is the number of detail basis functions, usually taking ones. The purpose of this step is to further separate the stable terrain information and changing information in the high-resolution point cloud data, laying a foundation for subsequent point cloud anomaly detection and noise suppression.

[0113] The specific implementation of step S07 is to perform a three-dimensional discrete Fourier transform (3D-DFT) on the changing component of each high-resolution point cloud data block, thereby establishing the point cloud spatial distribution feature matrix . By analyzing this feature matrix , the main spectral feature vectors can be extracted, and a point cloud anomaly recognition model can be constructed. The mathematical description is as follows:

[0114] ;

[0115] Among them, is the number of spatial sampling points, is the imaginary unit, is the frequency domain coordinate. By performing feature analysis on , the main spectral feature vectors can be extracted, and these vectors will form the basis of the point cloud anomaly recognition model.

[0116] The specific implementation of step S08 is to calculate the anomaly coefficient matrix of each high-resolution point cloud data block according to the point cloud anomaly recognition model established in step S07. Then use this anomaly coefficient matrix to perform noise reduction processing on the point cloud changing component , and obtain the corrected point cloud changing component . The mathematical description is as follows:

[0117] ;

[0118] ;

[0119] Among them, is the eigenvalue, is the eigenvector, is the number of main eigenvectors, usually taking pieces. Through this outlier point cloud suppression method based on principal component analysis, the noise and interference components in the high-resolution point cloud data can be effectively reduced.

[0120] The specific implementation of step S09 is to combine the stable components and the corrected variable components of each high-resolution point cloud data block to reconstruct and obtain the final corrected high-resolution point cloud data block . The mathematical description is as follows:

[0121] ;

[0122] where is the error introduced during the reconstruction process. The purpose of this step is to fuse the stable components and the corrected variable components obtained in the previous steps to obtain the final high-quality point cloud data block.

[0123] The specific implementation of step S10 is to perform mesh reconstruction on each corrected high-resolution point cloud data block to generate the corresponding three-dimensional model of the high-resolution sub-region. Various common point cloud meshing algorithms can be used in this step, such as Poisson surface reconstruction, spherical projection, etc. Through this mesh reconstruction process, the discrete point cloud data can be converted into a continuous three-dimensional mesh surface model.

[0124] The specific implementation of step S11 is to perform boundary registration and stitching on the three-dimensional models of each high-resolution sub-region generated in step S10 to construct the complete three-dimensional model of the high-resolution seabed scene. The mathematical description of the boundary registration is as follows:

[0125] ;

[0126] where is the registration error, is the number of sub-regions, is the number of boundary points of the th sub-region, is the rigid body transformation matrix, is the corresponding pair of boundary points, is the regularization coefficient, and its value range is , is the identity matrix. By minimizing this registration error function , the precise stitching between the three-dimensional models of each sub-region can be achieved.

[0127] The specific implementation of step S12 is to perform global geometric optimization and texture mapping on the three-dimensional model of the high-resolution seabed scene constructed in step S11 to further improve the visual quality of the model. The mathematical description of geometric optimization is as follows:

[0128] ;

[0129] where is the smoothing error, is the vertex set, is the face set, is the Laplace operator, is the vertex displacement, is the face normal vector, is the target normal vector, is the weight coefficient, and its value range is . By minimizing this smoothing error function , the geometric shape of the three-dimensional model can be optimized to make it smoother and more natural. At the same time, realistic surface details can also be added to the three-dimensional model through techniques such as texture mapping to further enhance the visual effect.

[0130] In summary, the fast three-dimensional modeling method based on seabed point cloud data proposed by the present invention involves multiple mathematical formulas and algorithm implementation details. This method can effectively extract the stable features and dynamic change information of the seabed terrain from the original high-density point cloud data, and finally construct a high-quality three-dimensional seabed scene model through technical means such as adaptive mesh refinement, point cloud anomaly detection, and geometric optimization.

[0131] The second aspect of the present invention provides a computer-readable storage medium, wherein program instructions are stored in the computer-readable storage medium, and when the program instructions run on a computer, they are used to execute the above-mentioned fast three-dimensional modeling method based on seabed point cloud data.

[0132] The third aspect of the present invention provides a fast three-dimensional modeling system based on seabed point cloud data, which includes the above-mentioned computer-readable storage medium.

[0133] Specifically, the principle of the present invention is: by effectively processing and analyzing the original high-resolution seabed point cloud data, stable terrain features and dynamic change information are extracted from it, and finally a high-quality three-dimensional seabed scene model is constructed. This method mainly includes the following key technical links:

[0134] 1. Spatial downsampling. Since the original seabed point cloud data is usually large in quantity and dense, directly performing 3D modeling will face huge computational overhead. Therefore, the present invention first performs spatial downsampling on the original point cloud data to obtain low-resolution point cloud data. This step can significantly reduce the computational complexity of subsequent processing while retaining key terrain features.

[0135] 2. Point cloud decomposition. The downsampled low-resolution point cloud data is separated into stable terrain features (stable components) and dynamically changing information (changing components) through the method of multivariate mixture decomposition. This decomposition can provide a favorable data basis for subsequent adaptive mesh refinement and abnormal point cloud detection.

[0136] 3. Adaptive mesh refinement. Based on the stable components of the low-resolution point cloud, an initial 3D mesh model is first constructed. Then, according to the spatial curvature distribution of the mesh cells, an adaptive block strategy is adopted to locally refine the mesh. This can provide a more refined geometric expression for complex terrain areas while ensuring the overall computational efficiency.

[0137] 4. Frequency domain analysis and anomaly detection. For each high-resolution point cloud data block, its spectral features are extracted using the 3D discrete Fourier transform. By analyzing these spectral features, a point cloud anomaly recognition model can be constructed to effectively suppress the noise and interference components in the original point cloud data.

[0138] 5. Point cloud reconstruction and mesh generation. The stable components and the changing components that have passed through anomaly detection are combined and reconstructed to obtain the final high-quality point cloud data. Based on these point cloud data, a corresponding 3D model is generated using the mesh reconstruction method. At the same time, by performing boundary registration and stitching on the sub-region 3D models, a complete high-resolution seabed scene model can be constructed.

[0139] 6. Global optimization and texture mapping. To further enhance the visual effect of the 3D seabed model, global geometric optimization and texture mapping processing can be performed finally. This includes optimizing the geometric shape of the model using methods such as Laplacian smoothing, and enhancing the surface details of the model through texture mapping.

[0140] Through the organic combination of the above technical links, the method of the present invention can efficiently extract stable terrain features and dynamically changing information from the original seabed point cloud data, and construct a high-resolution and high-precision 3D seabed scene model. This method makes full use of the inherent characteristics of the point cloud data, and overcomes the deficiencies of existing 3D modeling methods based on acoustic measurements in complex terrain modeling, dynamic change expression, etc. through means such as adaptive mesh refinement and frequency domain analysis.

[0141] It should be noted that the explanations of relevant variables in this specific embodiment are shown in Table 1 below:

[0142] Table 1 Variable Explanation Table

[0143]

[0144] The following provides an embodiment of a specific application scenario of the present invention: Taking the seabed terrain mapping task in a certain sea area as an example, the specific implementation process of the method of the present invention is introduced. This sea area is located in the southeast coast of China, with a water depth range between 50 - 1000 meters. The terrain is complex and changeable, including steep seabed mountains and flat continental shelf areas. In order to obtain a high-precision three-dimensional seabed model of this sea area, the ocean department decides to use the fast three-dimensional modeling method based on seabed point cloud data proposed by the present invention for mapping.

[0145] First, the mapping team used a multibeam echosounder to comprehensively collect seabed terrain data in this sea area. Through several days of continuous operation, approximately 250 million seabed point cloud data were obtained in total. Due to the high acquisition density of the original point cloud data, huge computational overhead will be generated in the subsequent three-dimensional modeling process. For this reason, the mapping team first performed spatial downsampling on the original point cloud data.

[0146] Specifically, the mapping team divided the entire sea area into 1 m x 1 m x 1 m voxel grids, and statistically processed the point cloud data inside each voxel, extracting the average coordinate value at the center position of the voxel. The purpose of doing this is to significantly reduce the computational complexity of subsequent processing while retaining the key features of the seabed terrain. After this step, the original 250 million point cloud data were compressed into approximately 30 million low-resolution point cloud data . Table 2 shows the comparison between the original point cloud data and the downsampled point cloud data.

[0147] Table 2 Point Cloud Data Downsampling Results

[0148] Index Original point cloud data Downsampled point cloud data Number of points in the point cloud 250 million 30 million Average point spacing 0.2 m 1 m Computational complexity Extremely high Relatively low

[0149] Next, the mapping team performed multivariate mixed decomposition calculation on the downsampled low-resolution point cloud data. Through this step, the point cloud data can be decomposed into a point cloud stable component reflecting the stable features of the seabed terrain , and a point cloud variable component describing the dynamic changes of the seabed terrain .

[0150] Specifically, the mapping team chose to use 5 basis functions to fit the point cloud stable component . By solving the coefficients of the basis functions through the least squares method , the following mathematical expression can be obtained:

[0151] ;

[0152] ;

[0153] Among them, the basis function is selected based on the analysis of the low-resolution point cloud data and can better represent the basic shape of the seabed terrain. The point cloud variation component reflects the subtle undulations relative to the stable terrain. Through this decomposition method, the surveying and mapping team obtained the stable characteristics and dynamic information of the seabed terrain.

[0154] Based on the stable component of the point cloud , the surveying and mapping team constructed an initial three-dimensional grid model. To evaluate the geometric properties of the grid model, the team calculated the spatial curvature of each grid cell. The larger the curvature value, the more complex the terrain in that area. By analyzing the global curvature distribution, the team found that there are many steep terrain features in this sea area, such as underwater mountains and deep sea trenches.

[0155] As Figure 2 shown, it is the heat map of the curvature distribution of a partial area of the seabed terrain, which shows the curvature value distribution of different areas through the depth of color, helping to identify complex terrain areas.

[0156] To better depict these complex terrain areas, the surveying and mapping team adopted a strategy of adaptive grid refinement. The specific approach is to first calculate the average curvature and the standard deviation of curvature of the initial grid model, and then set an adaptive block threshold . For those grid cells with curvature values higher than this threshold, the team divides them into more detailed sub-regions for further processing. Through this adaptive refinement method, a high-resolution grid model consisting of 30,000 sub-regions is finally obtained.

[0157] Next, the surveying and mapping team further processed the point cloud data block of each high-resolution sub-region. First, the surveying and mapping team used 10 detail basis functions to perform multivariate mixed decomposition on the point cloud data block , so as to obtain the stable component and the variation component :

[0158] ;

[0159] ;

[0160] ;

[0161] Among them, the detailed basis function is selected considering the topographic characteristics of the sub-region, and can more precisely describe the local topographic changes.

[0162] For the point cloud variation component of each sub-region, the surveying and mapping team further uses the three-dimensional discrete Fourier transform to extract its spectral characteristics. By analyzing these spectral characteristics, the team found that most of the high-frequency components are related to the noise and interference of the seabed topography. Therefore, the surveying and mapping team constructed an outlier point cloud recognition model based on principal component analysis for effective noise reduction processing.

[0163] Specifically, the team extracted the first 3 main spectral feature vectors and their corresponding eigenvalues , and constructed the following outlier coefficient matrix :

[0164] ;

[0165] ;

[0166] Among them, represents the point cloud variation component after noise suppression.

[0167] As Figure 3 shown, it shows the spectral feature analysis results of the point cloud variation component, and shows the spectral distributions of the three main feature vectors.

[0168] With the processed stable component and variation component of each sub-region, the surveying and mapping team reconstructed and combined them to obtain the final high-resolution point cloud data block :

[0169] ;

[0170] Based on these high-resolution point cloud data, the surveying and mapping team used the Poisson surface reconstruction algorithm to grid each sub-region and generated the corresponding high-resolution three-dimensional grid model. In order to stitch these sub-region models into a complete seabed scene model, the team also implemented the step of boundary registration. Specifically, the surveying and mapping team optimized the geometric consistency between the sub-region models by minimizing the following registration error function :

[0171] ;

[0172] Among them, is the rigid body transformation matrix of the th sub-region model, and

[0173] is the corresponding pair of boundary points. Through this registration optimization, the surveying and mapping team finally accurately pieced together 30,000 sub-region models into a complete three-dimensional model of the high-resolution seabed scene. Figure 4 As shown in

[0174] is the statistical distribution histogram of the sub-region boundary registration error, showing the error distribution during the registration process. :

[0175] ;

[0176] Among them, is the Laplace operator, is the vertex displacement, are the actual patch normal vector and the target normal vector respectively. By minimizing this error function, the geometric shape of the overall model can be made smoother and more natural.

[0177] Finally, the surveying and mapping team pasted realistic textures on the three-dimensional model using high-resolution aerial images. This not only improves the visual effect of the model but also further enhances its ability to restore the actual seabed environment. Table 3 lists the main parameters of the construction process of the three-dimensional model of this sea area.

[0178] Table 3 Main parameters of the construction process of the three-dimensional seabed model

[0179]

[0180] Through the above steps, the surveying and mapping team finally constructed a high-precision three-dimensional seabed scene model covering the entire surveyed sea area with a resolution of 1 m. This model not only accurately depicts complex seabed terrain features, such as steep underwater mountains and deep and wide trenches, etc., but also can effectively reflect the dynamic change information of the seabed terrain.

[0181] This three-dimensional seabed model not only provides reliable basic data support for subsequent marine scientific research, but also brings important reference value to relevant underwater engineering construction projects. For example, the marine department can use this model to simulate and optimize the underwater pipeline laying scheme, and the water resources exploration department can also carry out precise exploration activities based on this model. In short, the rapid three-dimensional modeling method based on seabed point cloud data proposed by the present invention brings new technical support for various applications in the marine field.

[0182] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should be covered within the protection scope of the present invention.

Claims

1. A rapid three-dimensional modeling method based on seabed point cloud data, characterized in that: The following steps are involved: S01, collecting seabed point cloud data, and performing spatial downsampling processing on the seabed point cloud data to obtain low-resolution point cloud data; S02, performing multivariate mixed decomposition calculation on the low-resolution point cloud data to obtain low-resolution point cloud stable component data and low-resolution point cloud variable component data; wherein the low-resolution point cloud stable component data represents the basic form of the seabed topography, and is linearly represented using 5 to 10 basis functions; the low-resolution point cloud variable component data represents a slight change in the seabed topography, and is obtained by subtracting the low-resolution point cloud stable component data from the low-resolution point cloud data; S03, constructing an initial three-dimensional grid model of the seabed scene based on the low-resolution point cloud stable component data, and calculating the spatial curvature distribution of the initial three-dimensional grid model; S04, adaptively dividing the initial three-dimensional grid model into blocks according to the spatial curvature distribution, and dividing the initial three-dimensional grid model into a plurality of sub-regions; S05, spatially partitioning the original seabed point cloud data according to the multiple sub-regions to obtain multiple high-resolution point cloud data blocks; S06, performing multivariate mixed decomposition calculation on each of the high-resolution point cloud data blocks to obtain corresponding point cloud stable component data and point cloud variable component data respectively; specifically, using 10 to 20 detail basis functions to perform linear decomposition on each of the high-resolution point cloud data blocks to obtain corresponding point cloud stable component data; and obtaining point cloud variable component data by subtracting the point cloud stable component data from the high-resolution point cloud data block; S07, performing Fourier transform on the point cloud change component data of each high-resolution point cloud data block, establishing a point cloud spatial distribution feature matrix, extracting the main feature vector, and constructing a point cloud anomaly recognition model; S08: Calculate the anomaly coefficient matrix of each high-resolution point cloud data block according to the point cloud anomaly recognition model , perform noise reduction on the point cloud change component data to obtain the point cloud correction component data; where, ;in, is the main eigenvector, is the characteristic value, is the number of main eigenvectors, which is 3-5; S09, combining and reconstructing the point cloud stabilization component data and the point cloud correction component data of each of the high-resolution point cloud data blocks to obtain a corrected high-resolution point cloud data block; S10, reconstructing a grid for each of the corrected high-resolution point cloud data blocks to generate a high-resolution sub-region three-dimensional model; S11, performing boundary registration and splicing on the high-resolution sub-region three-dimensional models to construct a complete high-resolution three-dimensional model of the seabed scene; S12, performing global geometry optimization and texture mapping on the high-resolution three-dimensional model of the seabed scene to obtain a final three-dimensional model of the seabed scene.

2. A rapid three-dimensional modeling method based on seabed point cloud data according to claim 1, characterized in that: The step S01 specifically includes: collecting original point cloud data of the seabed area, and performing voxel downsampling processing on the original point cloud data; wherein the voxel downsampling processing is to divide the three-dimensional space into a number of voxel units, perform statistical calculations on the point cloud data inside each voxel unit, and obtain the average value or median value of the points inside each voxel unit; the size of the voxel unit is adjustable within the range of 0.01 meters to 1 meter, and a balance between computational efficiency and the degree of retention of seabed terrain details is achieved by adjusting the size of the voxel unit.

3. The rapid three-dimensional modeling method based on seabed point cloud data according to claim 2 is characterized in that: The step S07 specifically includes: performing a three-dimensional discrete Fourier transform on the point cloud change component data of each of the high-resolution point cloud data blocks; establishing a point cloud spatial distribution feature matrix based on the results of the three-dimensional discrete Fourier transform; extracting a main eigenvector by performing feature analysis on the point cloud spatial distribution feature matrix; and constructing a point cloud anomaly recognition model based on the main eigenvector.

4. The rapid three-dimensional modeling method based on seabed point cloud data according to claim 3 is characterized in that: The step S09 specifically includes: linearly superimposing the point cloud stabilization component data and the point cloud correction component data of each of the high-resolution point cloud data blocks; obtaining a corrected high-resolution point cloud data block through the linear superposition; the corrected high-resolution point cloud data block has higher spatial accuracy and lower noise level.

5. The rapid three-dimensional modeling method based on seabed point cloud data according to claim 4 is characterized in that: The step S11 specifically includes: extracting boundary point pairs of the high-resolution sub-region three-dimensional model; establishing a boundary registration error function, wherein the boundary registration error function includes a rigid body transformation term and a regularization term, and the coefficient of the regularization term ranges from 0.1 to 1; and achieving accurate splicing of each of the high-resolution sub-region three-dimensional models by minimizing the boundary registration error function.

6. The rapid three-dimensional modeling method based on seabed point cloud data according to claim 5, characterized in that: The step S12 specifically includes: establishing a smoothing error function, wherein the smoothing error function includes a vertex displacement term and a normal vector term; a weight coefficient of the smoothing error function ranges from 0.01 to 0.1; geometrically optimizing the high-resolution seabed scene three-dimensional model by minimizing the smoothing error function; and performing texture mapping processing on the optimized three-dimensional model.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed in a computer, they are used to execute the rapid three-dimensional modeling method based on seabed point cloud data as described in any one of claims 1 to 6.

8. A rapid three-dimensional modeling system based on seabed point cloud data, characterized in that: A computer-readable storage medium comprising the method of claim 7.

Images