Three-dimensional modeling system and method based on laser radar

Abstract

The invention relates to the technical field of three-dimensional modeling, in particular to a three-dimensional modeling system and method based on a laser radar, and the system comprises a point cloud boundary extraction module, a boundary error analysis module, a self-adaptive cutting repair module and a stitching structure optimization module. According to the method, three-dimensional coordinates, normal vectors, reflection intensity, distances and indexes of boundary points are called layer by layer, a Sobel operator and a Canny algorithm are combined to cooperatively recognize a boundary and an error region, then local region repair is completed based on least square plane fitting and spatial residual analysis, a tension vector is analyzed through dynamic planning, and stitching path adjustment is executed. On the cooperative basis of multi-type point cloud features, a spatial topological structure and boundary continuity dynamic modeling, the problems of local damage, connection errors and instable stitching paths of point cloud boundaries in complex scenes are solved, and the integrity of boundary structures, the coherence of stitching transition and the rationality of spatial geometric expression are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional modeling technology, and in particular to a three-dimensional modeling system and method based on laser radar. Background Art

[0002] The field of three-dimensional modeling technology involves digitally representing the real world or design targets as three-dimensional models with spatial attributes. It is widely used in multiple application scenarios such as computer graphics, virtual reality, industrial design, building information modeling, medical image processing and digital twins. Its core issues include three-dimensional data acquisition, geometric shape construction, texture mapping, spatial relationship expression and conversion relationship with two-dimensional images. Currently, this technical field has formed a complete technical system from point cloud acquisition, mesh generation to three-dimensional model optimization. Among them, the traditional three-dimensional modeling system refers to a tool platform for constructing digital three-dimensional geometric shape structures. The technical matter targeted by this patent subject is how to generate a three-dimensional model with a spatial geometric structure based on input data. In traditional methods, it is usually based on image sequences or depth information, with the help of spatial point cloud data obtained by structured light measurement, multi-view stereo reconstruction or laser scanning, and the construction of three-dimensional entities is completed through specific modeling processes such as triangulated mesh reconstruction and surface fitting.

[0003] In the 3D modeling process, existing technologies usually focus on the collection and geometric reconstruction of the entire point cloud. There is a weakening trend in the identification and repair of local anomalies and error areas at the point cloud boundaries. They rely on fixed thresholds or general algorithms to process edge details, resulting in limited continuity of boundary areas in high-noise, complex structures, and dense point cloud environments. The stitching path is prone to jumps and geometric details are missing. This is common in scanning building surface details, road edges, or areas with multiple obstacles in urban environments. Since error points cannot be effectively identified and repaired, the 3D model boundaries are discontinuous, geometric deformation occurs, and the spatial expression ability is reduced in subsequent applications. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a three-dimensional modeling system and method based on laser radar.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a three-dimensional modeling system based on laser radar includes: The point cloud boundary extraction module is used to obtain coordinates, reflection intensity, distance, and index from the lidar point cloud data, call eight neighborhood points, use the Sobel operator to detect edges, and calculate the normal vector of each point based on the neighborhood point set. The module then analyzes and outputs a preliminary boundary point set, which is then passed to the boundary error analysis module. A boundary error analysis module is used to extract distance gradient, intensity change rate, angle difference, and density change from the preliminary boundary point set, call the Canny algorithm for screening, output an error point set, and pass the error point set to the adaptive cutting and repair module; An adaptive cutting and repairing module is used to expand the point set based on the error point set according to the neighborhood expansion rule, call the least squares plane fitting model, construct a spatial residual function, perform plane fitting, and generate an interpolation point set by linear interpolation, and pass the interpolation point set to the stitching structure optimization module; The stitching structure optimization module is used to obtain the tangent direction, angle change, inter-segment distance and strength change through the interpolation point set, call the dynamic programming algorithm to perform tension vector analysis, determine the number of connected edges of the interpolation point, and output a three-dimensional boundary connected structure through linear interpolation adjustment.

[0006] As a further solution of the present invention, the preliminary boundary point set includes the three-dimensional coordinates of the boundary points, the normal vectors of the boundary points, the reflection intensity of the boundary points, and the boundary point index; the error point set specifically includes the error point coordinates, the error point normal vectors, and the error point distribution attributes; the interpolation point set includes the interpolation point coordinates and the interpolation point normal vectors; and the boundary connectivity structure specifically refers to the boundary connectivity path, the boundary segment index, and the connectivity edge set.

[0007] As a further solution of the present invention, the point cloud boundary extraction module includes: The point cloud data acquisition submodule acquires the lidar point cloud data, extracts the coordinates, reflection intensity, distance and index information of each point, and calls the coordinates, reflection intensity, distance and index information, combines and arranges them, and establishes a point cloud attribute matrix; The neighborhood gradient calculation submodule calls the point cloud attribute matrix, indexes the position information of the eight neighborhood points for each point in the matrix, uses the Sobel operator to perform horizontal and vertical convolution operations on the distance values ​​of the multiple neighborhood point sets, and combines the operation values ​​in the two directions to generate the edge response intensity value; The boundary point set generation submodule calls the point cloud attribute matrix and the edge response intensity value, sets the gradient amplitude threshold, compares the edge response intensity value of each point with the gradient amplitude threshold one by one, filters out points whose intensity values ​​exceed the gradient amplitude threshold, and extracts the index of the point in the point cloud attribute matrix to obtain a preliminary boundary point set.

[0008] As a further solution of the present invention, the boundary error analysis module includes: The multidimensional feature extraction submodule extracts the distance and intensity information of each point and its neighboring points based on the preliminary boundary point set, calculates the distance gradient and intensity change rate of multiple points, and detects the angle difference between the point and the line connecting the adjacent points and the density change in the local area. By combining the above four values, a comprehensive feature index of the point set is obtained; The edge point non-maximum suppression submodule calls the point set feature comprehensive index and compares the point set feature comprehensive index value with the two adjacent points along the gradient direction of each point. If the index value of the current point is smaller than that of any adjacent point, it is marked as a non-edge point and removed from the sequence to establish a candidate edge point sequence. The dual-threshold connection screening submodule sets a high gradient threshold and a low gradient threshold according to the candidate edge point sequence, determines points in the sequence with indicators higher than the high threshold as strong edge points, and determines points between the two thresholds as weak edge points. It screens weak edge points that are directly connected to strong edge points or indirectly connected through other weak edge points, combines all strong edge points with the screened weak edge points, and obtains an error point set.

[0009] As a further solution of the present invention, the adaptive cutting and repairing module includes: The neighborhood point set expansion submodule obtains the error point set, takes each error point as the center, defines the range according to the set neighborhood search radius, counts and determines whether the number of point clouds within the radius exceeds the lower limit of the number of points, and merges the point cloud sets that meet the conditions to generate an extended neighborhood point set; The fitting plane construction submodule calls the extended neighborhood point set, constructs the fitting space plane using the least squares criterion, and calculates the orthogonal distance of each point in the point set to the plane and the distance to the geometric center of the error point set using the formula: ; Calculate the deviation measure of the set point cloud and establish the spatial fitting residual value; in, represents the spatial fitting residual value, represents the number of extended neighborhood point sets, Represents the current point number in the point set, Representative Points The orthogonal distance to the fitting plane, Represents the factor used to adjust the influence of center distance, Representative Points The distance to the geometric center of the original error point set, represents the base point cloud density, Representative Points The local point cloud density; The linear interpolation repair submodule sets the residual judgment threshold according to the spatial fitting residual value and identifies the points exceeding the threshold as points to be repaired, obtains the projection coordinates of the points on the least squares fitting plane, and performs linear interpolation operation as a benchmark to generate an interpolation point set.

[0010] As a further solution of the present invention, the suture structure optimization module includes: The point set geometric feature extraction submodule obtains the interpolation point set, calculates the multi-point tangent direction, the distance between adjacent points and the change in association strength, counts the change in the angle between the connecting nodes, defines the tangent direction and the distance between segments as the path basis unit, associates the strength and angle data, and establishes a geometric constraint parameter set; The tension path planning submodule uses a dynamic programming algorithm based on the geometric constraint parameter set, uses strength changes and angle changes as cost factors for tension vector analysis, and performs multi-stage decision-making between interpolation points with tension balance as the optimization goal. It judges and determines the number of connecting edges of multiple interpolation points to obtain the number of node connections. The connection structure generation submodule identifies the interpolation point pairs that need to be connected based on the number of node connections, calls the coordinate information in the geometric constraint parameter set, performs linear interpolation between the point pairs to generate connection edges, and combines all the connection edges to establish a three-dimensional boundary connectivity structure.

[0011] The three-dimensional modeling method based on laser radar is performed based on the above-mentioned three-dimensional modeling system based on laser radar, and includes the following steps: S1: Collect point cloud data through LiDAR to obtain coordinates, reflection intensity, distance, and index parameters, call the eight-neighborhood point set, use the Sobel operator to perform edge detection on the distance parameters, and form a preliminary boundary point set based on the edge data obtained by detection; S2: Based on the preliminary boundary point set, obtain the distance gradient, reflection intensity change rate, angle difference, and point set density change parameters, call the Canny algorithm, perform an edge screening operation on the preliminary boundary point set, and screen out the edge abnormal points to form an error point set; S3: Using the error point set, the point set range is expanded according to the neighborhood expansion rule, the least squares plane fitting model is called, a spatial residual function is established based on the expanded point set, and plane parameter fitting is performed. The fitting result is used to generate an interpolation point set using a linear interpolation method; S4: Through the interpolation point set, the tangent direction, angle change, inter-segment distance and intensity change parameters are detected, the dynamic programming algorithm is called to perform tension vector analysis, the number of edges connected by the interpolation point is determined, the parameters of the boundary connectivity structure are adjusted through linear interpolation, and the three-dimensional boundary connectivity structure is output.

[0012] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, in the overall process of point cloud boundary identification, error area extraction, cutting and repairing, and boundary stitching, the three-dimensional coordinates, normal vectors, reflection intensity, distance and index of the boundary points are called layer by layer, and the Sobel operator and Canny algorithm are combined to collaboratively identify the boundaries and error areas, and then the local area repair is completed based on least squares plane fitting and spatial residual analysis. The tension vector is analyzed by dynamic programming to perform stitching path adjustment. On the basis of the coordination of multi-type point cloud features, spatial topological structure and boundary continuity dynamic modeling, the problems of local damage of point cloud boundaries, connection errors and unstable stitching paths in complex scenes are solved, which significantly improves the integrity of the boundary structure, the coherence of the stitching transition and the rationality of the spatial geometric expression. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a schematic diagram of the process of the three-dimensional modeling system based on laser radar of the present invention; Figure 2 This is a flow chart of the point cloud boundary extraction module of the present invention; Figure 3 This is a flow chart of the boundary error analysis module of the present invention; Figure 4 This is a schematic diagram of the process flow of the adaptive cutting and repair module of the present invention; Figure 5 This is a schematic diagram of the flow of the suture structure optimization module of the present invention. DETAILED DESCRIPTION

[0014] To make the purpose, technical solutions and advantages of the present invention clearer, the following is a detailed description of the technical solutions based on software implementation in conjunction with the system architecture diagram and embodiments. It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present invention and do not constitute a limitation on the scope of protection.

[0015] In the description of this invention, the system architecture relationships or data processing flows indicated by terms such as "layer," "module," "interface," "data flow," "client," and "server" are defined based on the architecture diagrams or flow charts corresponding to the embodiments. This expression is intended solely to clarify the logical relationships between the various elements of the technical solution and does not limit the physical deployment form. The term "plurality" encompasses two or more technical units, including but not limited to scalable elements such as multiple data nodes, processing threads, service instances, or functional components. The specific number will be determined based on the actual business scenario and requires special explanation.

[0016] See also Figure 1 and Figure 2 The present invention provides a technical solution: a three-dimensional modeling system based on laser radar includes: The point cloud boundary extraction module is used to obtain coordinates, reflection intensity, distance, and index from the lidar point cloud data, call eight neighborhood points, use the Sobel operator to detect edges, and calculate the normal vector of each point based on the neighborhood point set. It analyzes and outputs a preliminary boundary point set, which is then passed to the boundary error analysis module. The preliminary boundary point set includes the three-dimensional coordinates of the boundary points, the normal vectors of the boundary points, the reflection intensity of the boundary points, and the index of the boundary points; The point cloud boundary extraction module includes: The point cloud data acquisition submodule acquires the lidar point cloud data, extracts the coordinates, reflection intensity, distance and index information of each point, and calls the coordinates, reflection intensity, distance and index information, combines and arranges them, and establishes a point cloud attribute matrix; This submodule is responsible for acquiring raw data from the LiDAR device. In a specific scan, a handheld, high-precision laser scanner with an accuracy of 0.02mm is used to scan the entire surface of the stone lion, generating a high-density point cloud covering its entire surface. For each laser point in the point cloud, the system automatically extracts its 3D coordinates (X, Y, Z, in millimeters), its reflection intensity (a unitless integer between 0 and 255), its linear distance from the scanner (in millimeters), and its row index in the raw data file.

[0017] In one scan, the data records of five consecutive points collected near the cracks on the stone lion base are as follows: (100.1,205.3,50.7,188,1500.2,101), (100.3,205.4,50.8,190,1500.4,102), (100.5,205.5,50.9,10,1505.1,103), (100.7,205.6,51.0,192,1500.8,104), (100.9,205.7,51.1,193,1501.0,105). In this data sequence, the points The reflection intensity value (10) is much lower than that of its neighboring points (188, 190, 192, 193), and its distance value (1505.1mm) has a jump of more than 4mm compared with the average distance of the neighboring points (about 1500.5mm), which indicates that Located in a physically discontinuous area, such as the edge of a crack. The submodule combines the five attribute information (coordinates, reflection intensity, distance, index) of all the collected points and arranges them in ascending order according to the index number to construct a data set containing 2,500,000 points ( ) for subsequent calls.

[0018] The neighborhood gradient calculation submodule calls the point cloud attribute matrix, indexes the position information of the eight neighborhood points for each point in the matrix, uses the Sobel operator to perform horizontal and vertical convolution operations on the distance values ​​of the multi-neighborhood point set, and combines the operation values ​​in the two directions to generate the edge response intensity value; This submodule calls the point cloud attribute matrix established in the previous step. The system rasterizes the point cloud data on the two-dimensional UV coordinate system based on its projection relationship on the internal sensor of the scanner. For the point with index 103 in the matrix , the system indexes its eight immediate neighbors in the grid, namely the eight neighborhood points. The distance values ​​(unit: mm) of its eight neighboring points are extracted to form a 3×3 distance value matrix.

[0019] by As the center, its neighborhood distance matrix is: ; The system uses a horizontal Sobel operator kernel The operator kernel in the vertical direction , and perform convolution operations on the 3×3 distance value matrix respectively. The convolution operation value in the horizontal direction is: The vertical convolution operation value is: Combining the numerical values ​​of the two directions, we can generate The edge response intensity value of the point, that is, the gradient amplitude This process is repeated for each point in the point cloud attribute matrix, and finally a set of edge response intensity values ​​of all points is generated, which is the same size as the point cloud.

[0020] The boundary point set generation submodule calls the point cloud attribute matrix and edge response intensity value, sets the gradient amplitude threshold, compares the edge response intensity value of each point with the gradient amplitude threshold one by one, filters out points whose intensity value exceeds the gradient amplitude threshold, and extracts the index of the point in the point cloud attribute matrix to obtain the preliminary boundary point set.

[0021] This submodule calls the point cloud attribute matrix and the edge response intensity value set of all points, and sets the gradient amplitude threshold To screen preliminary boundary points. Gradient amplitude threshold The setting is based on an independent experimental verification process. A titanium alloy test block with standard geometric features (such as 90-degree right-angle edges, 10mm diameter circular holes, and bevels) was selected and scanned using the same lidar equipment. The edge response intensity values ​​of all its points were calculated and compared with the known precise CAD model of the test block to obtain the true boundary points. Subsequently, the effect of different threshold settings on the boundary detection performance was tested. The threshold was divided according to the percentile of the intensity value distribution, from the 80th percentile to the 99th percentile, and tested at intervals of 1 percentile.

[0022] Table 1 Experimental data of gradient amplitude threshold selection

[0023] As shown in Table 1, the F1 score is an indicator that comprehensively considers the accuracy and false detection rate. It reaches a peak value of 0.96 when the threshold is set to the 90th percentile of the overall intensity value distribution. This shows that under this setting, the system can identify the true boundary to the greatest extent, while keeping the number of non-boundary points misjudged as boundary points at a low level. Therefore, this setting is applied to the stone lion point cloud data. The edge response intensity values ​​of all 2,500,000 points in the stone lion point cloud are calculated and sorted. The intensity value of the 2,250,000th position (ie, the 90% position) is 2.1, so the gradient amplitude threshold is set. .

[0024] The system compares the edge response intensity value of each point with the threshold Compare them one by one. For example, its intensity value is 2.596, because ,point The points with intensity values ​​lower than 2.1 are considered as non-boundary points and are removed. After finding the points, the indices of these points in the point cloud attribute matrix are extracted and their complete attribute information is obtained based on the indices, including 3D coordinates, normal vectors calculated by performing principal component analysis (PCA) on the neighborhood point set, reflection intensity, and original index. These points together constitute the preliminary boundary point set and are passed to the next module.

[0025] See also Figure 1 and Figure 3 ,Boundary error analysis module is used to extract distance gradient, intensity change rate, angle difference, and density change through the preliminary boundary point set, call the Canny algorithm to filter, output the error point set, and pass the error point set to the adaptive cutting and repairing module; The error point set specifically includes the error point coordinates, error point normal vectors, and error point distribution attributes; The boundary error analysis module includes: The multidimensional feature extraction submodule extracts the distance and intensity information of each point and its neighboring points based on the preliminary boundary point set, calculates the distance gradient and intensity change rate of multiple points, and detects the angle difference between the point and the line connecting the adjacent points and the density change in the local area. By combining the above four values, it obtains the comprehensive feature index of the point set; This submodule quantifies the multi-dimensional features of each point based on the received preliminary boundary point set. (120.5,301.8,88.2), its neighboring points on the boundary sequence are (120.3,301.7,88.1) and (120.8,301.9,88.4). The system is Calculate four features: 1. Distance gradient: This is the point The edge response intensity value calculated in the previous module is 2.8. 2. Intensity change rate: Use the same Sobel operator as the distance gradient calculation, but act on the reflection intensity value matrix of the neighborhood points. The intensity change rate is 35.4. 3. Angle difference: Calculate the vector With vector The angle between The calculated result is 175 degrees, and the angle difference is defined as , which is 5 degrees. 4. Density change: Calculation The number of point clouds within a sphere with a radius of 2 mm around the point The neighboring points are 50 points / mm³. and The local densities were 49 and 47 points / mm³, respectively, with an average of 48 points / mm³. The density variation is defined as The ratio of density to the average density of neighbors, that is, 50 / 48≈1.04.

[0026] To ensure that these eigenvalues ​​with different physical meanings and numerical ranges can participate in subsequent calculations in a unified manner, the system normalizes each eigenvalue before weighted summation and maps it to a unified scale space of [0,100]. The conversion rules here are as follows: the distance gradient and the intensity change rate are linearly scaled according to their actual maximum values ​​in the current preliminary boundary point set; the angle difference (range 0-180 degrees) is converted by dividing by 180 and multiplying by 100; the density change (a ratio close to 1) is converted through a nonlinear mapping function with a center point at 1.0 and a typical change range of [0.5,2.0] mapped to the interval [0,100]. According to this rule, The characteristic values ​​of are converted (the maximum distance gradient in the current data set is 10, the maximum intensity change rate is 100, and the density change is 55 after function mapping): Normalized distance gradient = (2.8 / 10) × 100 = 28.0 Normalized intensity change rate = (35.4 / 100) × 100 = 35.4 Normalized angle difference = (5 / 180) × 100 ≈ 2.78 Normalized Density Change = 55.0 Then, the four normalized values ​​are combined to obtain Comprehensive index of point set features The calculation formula is: . Weight coefficient It is determined by performing logistic regression analysis on a training set containing 10,000 samples (5,000 real crack boundary points annotated by experts and 5,000 edge noise points). The analysis results show that when the weight is set to When , the model has the highest accuracy in distinguishing true boundaries from noise. The comprehensive index of point set features is .

[0027] The edge point non-maximum suppression submodule calls the point set feature comprehensive index and compares the point set feature comprehensive index value with the two adjacent points along the gradient direction of each point. If the index value of the current point is smaller than that of any adjacent point, it is marked as a non-edge point and removed from the sequence to establish a candidate edge point sequence. This submodule calls the point set feature comprehensive index of all preliminary boundary points Sequence. For each point in the sequence (such as ), the system determines its gradient direction, which is determined by the vector sum of the distance gradient and the intensity gradient. In this gradient direction, the system finds Two virtual adjacent points before and after. The previous point The value is 26.5, the latter point The value is 29.2. The system compares these three points Values: 26.5, 27.876, 29.2. Because Index value , does not have the local maximum characteristic, so it is marked as a non-edge point and removed from the sequence. A point is retained only when its value is strictly greater than the values ​​of the two adjacent points before and after it in the gradient direction. This process traverses all preliminary boundary points and eventually establishes a more streamlined and thinner sequence of candidate edge points.

[0028] The dual-threshold connection screening submodule sets the gradient high threshold and gradient low threshold according to the candidate edge point sequence, determines the points in the sequence with indicators higher than the high threshold as strong edge points, and determines the points between the two thresholds as weak edge points. It screens weak edge points that are directly connected to strong edge points or indirectly connected through other weak edge points, combines all strong edge points with the screened weak edge points, and obtains the error point set.

[0029] This submodule processes the candidate edge point sequence generated in the previous step and needs to set the gradient high threshold. With gradient low threshold The setting of these two thresholds is also based on rigorous experiments.

[0030] Threshold setting and experimental verification: We select the cultural relics scanning data containing various complex boundaries (clear, fuzzy, discontinuous) as the test set. Calculate the threshold of all candidate edge points in the test set. value. The candidate range is set to the 60th to 90th percentiles of the distribution of values, The candidate range is set to the 20th to 50th percentile, and Always less than For each ( , ) combination was tested, and the results are shown in Table 2.

[0031] Table 2 Double threshold combination selection experimental data table

[0032] As shown in Table 2, when the high threshold is set at the 80th percentile and the low threshold is set at the 40th percentile, the comprehensive evaluation score is the highest, indicating that the boundary connection effect is most ideal at this time. The 80th percentile of the value distribution is 70.0, and the 40th percentile is 25.0. Therefore, we set , .

[0033] The system traverses the candidate edge point sequence: If a certain point If the value is higher than 70.0, it is judged as a strong edge point and directly added to the final error point set.

[0034] If a certain point The value is between 25.0 and 70.0, that is , it is determined to be a weak edge point.

[0035] If a certain point If the value is lower than 25.0, it is directly eliminated. Finally, the system checks all weak edge points. A weak edge point is retained if it is directly connected to any strong edge point within the eight-neighborhood, or indirectly connected to a strong edge point through a path composed of other weak edge points. All strong edge points are combined with the screened weak edge points to obtain the final error point set. This set accurately outlines the cracks and defects on the surface of the stone lion and records the coordinates, normal vector and distribution properties of each point.

[0036] See also Figure 1 and Figure 4 ,Adaptive cutting and repairing module is used to expand the point set based on the error point set according to the neighborhood expansion rule, call the least squares plane fitting model, construct the spatial residual function, perform plane fitting, linear interpolation to generate the interpolation point set, and pass the interpolation point set to the stitching structure optimization module; The interpolation point set includes the interpolation point coordinates and the interpolation point normal vectors; The adaptive cutting and repair module includes: The neighborhood point set expansion submodule obtains the error point set, takes each error point as the center, defines the range according to the set neighborhood search radius, counts and determines whether the number of point clouds within the radius exceeds the lower limit of the number of points, and merges the point cloud sets that meet the conditions to generate an extended neighborhood point set; This submodule obtains the error point set. The continuous point cloud (a total of 15 points, indexed from E-201 to E-215) that marks the cracks on the front paw of the stone lion in the error point set is used as the processing object. The system sets a neighborhood search radius with each point as the center. The radius The value of is related to the average density of the point cloud and is set to 4 times the average point spacing. After calculation, the average point spacing of the stone lion point cloud is 0.5mm, so it is set mm. At the same time, set the lower limit of the number of points The value of 10 was determined experimentally; plane fitting with fewer than 10 points is statistically unstable. The system searches for all points in the original point cloud data within a sphere with a radius of 2.0 mm, centered at each error point. If the number of points in the neighborhood is at least 10, then that point and its neighbors constitute a valid local point cloud set. All qualified local point cloud sets generated by these 15 error points are merged, and duplicate points are removed, ultimately generating an expanded neighborhood point set of 150 points covering the entire crack area.

[0037] The fitting plane construction submodule calls the extended neighborhood point set, constructs the fitting space plane using the least squares criterion, and calculates the orthogonal distance of each point in the point set to the plane and the distance to the geometric center of the error point set using the formula: ; Calculate the deviation measure of the set point cloud and establish the spatial fitting residual value; in, represents the spatial fitting residual value, represents the number of extended neighborhood point sets, Represents the current point number in the point set, Representative Points The orthogonal distance to the fitting plane, Represents the factor used to adjust the influence of center distance, Representative Points The distance to the geometric center of the original error point set, represents the base point cloud density, Representative Points The local point cloud density; This submodule calls the expanded neighborhood point set and calculates the spatial fitting residual value using the following formula: ; First, explain and assign values ​​to each parameter in the formula: : The spatial fitting residual value is an indicator that quantifies the degree to which the entire extended neighborhood point set deviates from an ideal plane, and the unit is millimeter.

[0038] : The total number of points in the expanded neighborhood point set is determined to be 150 by the previous step.

[0039] : The current point number in the point set, ranging from 1 to 150.

[0040] :point Orthogonal distance to the fitting plane (unit: mm). The best fitting spatial plane equation is constructed by applying the least squares criterion to 150 points. For each point , whose orthogonal distance to the plane is calculated as .

[0041] : Center distance influence factor, a unitless weight coefficient. The setting of this factor is completed through experiments: artificial gaps are created on test samples with different curvatures (flat surface, convex surface with a radius of 500mm, concave surface with a radius of 500mm), and different The repair was performed with different values ​​(0.0, 0.1, 0.2, 0.5, 1.0) and the smoothness of the repaired surface was measured. The results show that when When , the repair results can maintain the best geometric continuity on different surfaces.

[0042] :point The distance to the geometric center of the original error point set (unit: mm). Calculate the arithmetic mean of the coordinates of the original 15 error points to get the geometric center . Calculate the point again arrive The Euclidean distance of .

[0043] : The base point cloud density is obtained by calculating the bounding box volume occupied by the 2,500,000 points of the entire stone lion and dividing the total number of points by the volume, representing the global average density. points / mm³.

[0044] :point The local point cloud density is expressed as points The estimated value was calculated by counting the number of points within a sphere with a radius of 1 mm and centered at φ.

[0045] The operational logic of the formula is to calculate a comprehensive deviation term for each point in the extended neighborhood point set. This term consists of two parts: the orthogonal distance from the point to the fitting plane , and the distance from the point to the center of the error region (Depend on The two distances are both in millimeters and can be added directly. This sum is then multiplied by a density adjustment factor. , which is a unitless coefficient and is used to penalize points located in sparse areas ( ) because the credibility of these points is low. The comprehensive deviation items of all points are summed up and averaged to obtain the final residual value .

[0046] Example: Take three points in the extended neighborhood point set as an example to illustrate: Point 1: Close to the center of the crack. Measurement and calculation: mm, mm, points / mm³.

[0047] Point 2: Far away from the crack center, located on a flat surface. Measurement and calculation: mm, mm, points / mm³.

[0048] Point 3: Located at the edge of the crack, locally sparse. Measurement and calculation yield: mm, mm, Points / mm³. and : The residual term of point 1 =

[0049] The residual term of point 2 =

[0050] The residual term of point 3 = The residual terms of all 150 points are accumulated and then divided by 150 to get the final spatial fitting residual value. The results show that the average integrated distance of the extended neighborhood point set from the fitting plane is 0.55 mm.

[0051] The linear interpolation repair submodule sets the residual judgment threshold according to the spatial fitting residual value and identifies the points exceeding the threshold as points to be repaired. It obtains the projection coordinates of the points on the least squares fitting plane and performs linear interpolation operation as a benchmark to generate an interpolation point set.

[0052] This submodule calculates the spatial fitting residual value Perform subsequent operations. The system sets a residual judgment threshold This threshold is set according to the accuracy requirement of the repair object and is set to twice the average point spacing, that is, The system will and Compare. In this example, This result indicates that the point cloud distribution in this area is relatively flat, making it suitable for interpolation repair based on a single plane. Subsequently, the system identifies the original 15 error points as void areas and obtains the projected coordinates of these points on the constructed least squares fitting plane. Using these projected points as a reference, linear interpolation is performed within the area defined by the healthy boundary points on both sides of the crack to generate a new point set of 50 points that smoothly fills the crack area. These newly generated points constitute the interpolation point set, each of which contains three-dimensional coordinates and a normal vector inherited from the fitting plane.

[0053] See also Figure 1 and Figure 5 ,The suture structure optimization module is used to obtain the tangent direction, angle change, inter-segment distance and strength change through the interpolation point set, call the dynamic programming algorithm to perform tension vector analysis, determine the number of connected edges of the interpolation point, and adjust it through linear interpolation to output the three-dimensional boundary connected structure; The boundary connectivity structure specifically refers to the boundary connectivity path, boundary segment index, and connected edge set; The suture structure optimization module includes: The point set geometric feature extraction submodule obtains the interpolation point set, calculates the multi-point tangent direction, the distance between adjacent points and the change in association strength, counts the change in the angle between the connecting nodes, defines the tangent direction and the distance between segments as the path basis unit, and associates the strength and angle data to establish a geometric constraint parameter set; This submodule obtains the interpolation point set and performs geometric feature analysis on these 50 points. , the system calculates: 1. Tangent direction: through its two nearest neighbors in the ordered sequence and , with vector The direction of the point 2. Inter-segment distance: Calculate the Euclidean distance between adjacent interpolation points, such as 3. Correlation intensity change: The reflection intensity of each interpolation point is obtained by inverse distance weighted interpolation of the intensity values ​​of the original boundary points around it. Then the intensity difference between adjacent interpolation points is calculated. 4. Angle change: Calculate the angle between three consecutive points. The angle between two consecutive line segments. A specific interpolation point , the extracted geometric constraint parameter set is: {tangent direction: (0.707, 0.707, 0), and Distance: 0.48mm, interpolation strength: 155, formed - - Angle: 178 degrees}. This process is performed on all 50 interpolation points to establish a complete set of geometric constraint parameters.

[0054] The tension path planning submodule uses a dynamic programming algorithm based on a set of geometric constraint parameters. It uses strength changes and angle changes as cost factors for tension vector analysis. It uses tension balance as the optimization goal to perform multi-stage decision-making between interpolation points, determine the number of connecting edges between multiple interpolation points, and obtain the number of node connections. This submodule constructs a path planning problem with tension balance as the optimization goal based on a set of geometric constraint parameters. The interpolation points are considered as nodes of the graph, and the potential connections between nodes are considered as edges of the graph. The "cost" or "tension" of an edge is defined as the geometric roughness introduced by connecting two nodes. and Cost is defined by a function that converts geometric discontinuities into numerical values: ; In order to uniformly handle factors of different properties, before calculating the total cost, each factor is passed through the function Convert to unitless penalty points. The conversion rules are: Angle Factor : Quantify the angle change introduced after the connection (deviation from 180 degrees), .

[0055] Distance Factor :The distance The average distance between all interpolation points Normalize the difference, .

[0056] Intensity Factor : The difference in reflection intensity between two points Normalize by dividing by the maximum possible intensity range of 255, .

[0057] Weight coefficient The determination of the weight combination was to recruit 20 experienced 3D modelers to double-blindly score 100 groups of restoration results generated by different weight combinations (1-5 points, 5 points represents the most natural and smooth), and then select the weight combination with the highest average score, which was finally determined as .

[0058] The system uses dynamic programming algorithms to perform multi-stage decision making. Define the state From the starting point to the interpolation point The cumulative cost of the optimal (minimum tension) path. The state transition equation is By solving this problem, the system finds a path with the lowest total cost through all interpolation points. This process not only determines the order in which the points are connected but also determines the number of edges each node should have, at possible branching or intersections of the path (applicable to complex cracks). For the current linear crack repair, the result is that, except for the first and last two points, the remaining 48 points have two connected edges. This process ultimately outputs the number of connections for each interpolation node.

[0059] The connection structure generation submodule identifies the interpolation point pairs that need to be connected based on the number of node connections, calls the coordinate information in the geometric constraint parameter set, performs linear interpolation between the point pairs to generate connection edges, and combines all the connection edges to establish a three-dimensional boundary connectivity structure.

[0060] This submodule generates the final connection structure based on the number of node connections determined in the previous step. The system identifies all interpolation point pairs that need to be connected. Based on the results of dynamic programming, the points Should and Connect. System call The coordinate information of the point pair ( ) between and ( ) to generate two connecting edges. This operation is repeated for all pairs of points that need to be connected, and all generated connecting edges are combined. This collection of edges constitutes the final 3D boundary connectivity structure, which is stored and output as boundary connectivity paths, boundary segment indices, and connected edge sets. This structure completely and smoothly closes the cracks in the stone lion model, providing a high-quality geometric foundation for subsequent meshing and texture mapping.

[0061] The three-dimensional modeling method based on laser radar is performed based on the three-dimensional modeling system based on laser radar, and includes the following steps: S1: Collect point cloud data through LiDAR to obtain coordinates, reflection intensity, distance, and index parameters. Then, call the eight-neighborhood point set and use the Sobel operator to perform edge detection on the distance parameter. The edge data obtained by detection forms a preliminary boundary point set. S2: Based on the preliminary boundary point set, the distance gradient, reflection intensity change rate, angle difference, and point set density change parameters are obtained. The Canny algorithm is called to perform edge screening on the preliminary boundary point set. The edge outliers obtained by screening are used to form an error point set. S3: Based on the error point set, the point set range is expanded according to the neighborhood expansion rule, the least squares plane fitting model is called, the spatial residual function is established based on the expanded point set, and the plane parameters are fitted. The fitting results are used to generate the interpolation point set using linear interpolation; S4: Through the interpolation point set, the tangent direction, angle change, inter-segment distance and intensity change parameters are detected, the dynamic programming algorithm is called to perform tension vector analysis, the number of edges connected by the interpolation point is determined, the parameters of the boundary connectivity structure are adjusted through linear interpolation, and the three-dimensional boundary connectivity structure is output.

[0062] The above examples illustrate preferred implementations of the present invention. Any equivalent adjustments to the technical solutions based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithmic logic using different programming languages, service-oriented reconfiguration of functional modules, adjusting data interaction protocols, optimizing resource scheduling strategies, and other technical improvements. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture level that does not depart from the core technology of the present invention shall be deemed to be within the scope of protection defined by the claims of the present invention.

Claims

1. A three-dimensional modeling system based on laser radar, characterized in that: The system comprises: The point cloud boundary extraction module is used to obtain coordinates, reflection intensity, distance, and index from the lidar point cloud data, call eight neighborhood points, use the Sobel operator to detect edges, and calculate the normal vector of each point based on the neighborhood point set. The module then analyzes and outputs a preliminary boundary point set, which is then passed to the boundary error analysis module. A boundary error analysis module is used to extract distance gradient, intensity change rate, angle difference, and density change from the preliminary boundary point set, call the Canny algorithm for screening, output an error point set, and pass the error point set to the adaptive cutting and repair module; An adaptive cutting and repairing module is used to expand the point set based on the error point set according to the neighborhood expansion rule, call the least squares plane fitting model, construct a spatial residual function, perform plane fitting, and generate an interpolation point set by linear interpolation, and pass the interpolation point set to the stitching structure optimization module; The stitching structure optimization module is used to obtain the tangent direction, angle change, inter-segment distance and strength change through the interpolation point set, call the dynamic programming algorithm to perform tension vector analysis, determine the number of connected edges of the interpolation point, and output a three-dimensional boundary connected structure through linear interpolation adjustment.

2. The laser radar-based three-dimensional modeling system according to claim 1, characterized in that: The preliminary boundary point set includes the three-dimensional coordinates of the boundary points, the normal vectors of the boundary points, the reflection intensity of the boundary points, and the index of the boundary points. The error point set specifically includes the error point coordinates, the error point normal vectors, and the error point distribution attributes. The interpolation point set includes the interpolation point coordinates and the interpolation point normal vectors. The boundary connectivity structure specifically refers to the boundary connectivity path, the boundary segment index, and the connectivity edge set.

3. The laser radar-based three-dimensional modeling system according to claim 2, characterized in that: The point cloud boundary extraction module includes: The point cloud data acquisition submodule acquires the lidar point cloud data, extracts the coordinates, reflection intensity, distance and index information of each point, and calls the coordinates, reflection intensity, distance and index information, combines and arranges them, and establishes a point cloud attribute matrix; The neighborhood gradient calculation submodule calls the point cloud attribute matrix, indexes the position information of the eight neighborhood points for each point in the matrix, uses the Sobel operator to perform horizontal and vertical convolution operations on the distance values ​​of the multiple neighborhood point sets, and combines the operation values ​​in the two directions to generate the edge response intensity value; The boundary point set generation submodule calls the point cloud attribute matrix and the edge response intensity value, sets the gradient amplitude threshold, compares the edge response intensity value of each point with the gradient amplitude threshold one by one, filters out points whose intensity values ​​exceed the gradient amplitude threshold, and extracts the index of the point in the point cloud attribute matrix to obtain a preliminary boundary point set.

4. The laser radar-based three-dimensional modeling system according to claim 3, characterized in that: The boundary error analysis module includes: The multidimensional feature extraction submodule extracts the distance and intensity information of each point and its neighboring points based on the preliminary boundary point set, calculates the distance gradient and intensity change rate of multiple points, and detects the angle difference between the point and the line connecting the adjacent points and the density change in the local area. By combining the above four values, a comprehensive feature index of the point set is obtained; The edge point non-maximum suppression submodule calls the point set feature comprehensive index and compares the point set feature comprehensive index value with the two adjacent points along the gradient direction of each point. If the index value of the current point is smaller than that of any adjacent point, it is marked as a non-edge point and removed from the sequence to establish a candidate edge point sequence. The dual-threshold connection screening submodule sets a high gradient threshold and a low gradient threshold according to the candidate edge point sequence, determines points in the sequence with indicators higher than the high threshold as strong edge points, and determines points between the two thresholds as weak edge points. It screens weak edge points that are directly connected to strong edge points or indirectly connected through other weak edge points, combines all strong edge points with the screened weak edge points, and obtains an error point set.

5. The laser radar-based three-dimensional modeling system according to claim 4, characterized in that: The adaptive cutting and repairing module includes: The neighborhood point set expansion submodule obtains the error point set, takes each error point as the center, defines the range according to the set neighborhood search radius, counts and determines whether the number of point clouds within the radius exceeds the lower limit of the number of points, and merges the point cloud sets that meet the conditions to generate an extended neighborhood point set; The fitting plane construction submodule calls the extended neighborhood point set, constructs the fitting space plane using the least squares criterion, and calculates the orthogonal distance of each point in the point set to the plane and the distance to the geometric center of the error point set using the formula: ； Calculate the deviation measure of the set point cloud and establish the spatial fitting residual value; in, represents the spatial fitting residual value, represents the number of extended neighborhood point sets, Represents the current point number in the point set, Representative Points The orthogonal distance to the fitting plane, Represents the factor used to adjust the influence of center distance, Representative Points The distance to the geometric center of the original error point set, represents the base point cloud density, Representative Points The local point cloud density; The linear interpolation repair submodule sets the residual judgment threshold according to the spatial fitting residual value and identifies the points exceeding the threshold as points to be repaired, obtains the projection coordinates of the points on the least squares fitting plane, and performs linear interpolation operation as a benchmark to generate an interpolation point set.

6. The laser radar-based three-dimensional modeling system according to claim 5, characterized in that: The suture structure optimization module includes: The point set geometric feature extraction submodule obtains the interpolation point set, calculates the multi-point tangent direction, the distance between adjacent points and the change in association strength, counts the change in the angle between the connecting nodes, defines the tangent direction and the distance between segments as the path basis unit, associates the strength and angle data, and establishes a geometric constraint parameter set; The tension path planning submodule uses a dynamic programming algorithm based on the geometric constraint parameter set, uses strength changes and angle changes as cost factors for tension vector analysis, and performs multi-stage decision-making between interpolation points with tension balance as the optimization goal. It judges and determines the number of connecting edges of multiple interpolation points to obtain the number of node connections. The connection structure generation submodule identifies the interpolation point pairs that need to be connected based on the number of node connections, calls the coordinate information in the geometric constraint parameter set, performs linear interpolation between the point pairs to generate connection edges, and combines all the connection edges to establish a three-dimensional boundary connectivity structure.

7. A three-dimensional modeling method based on laser radar, characterized in that: The method is used to implement the laser radar-based three-dimensional modeling system according to any one of claims 1 to 6, comprising the following steps: S1: Collect point cloud data through LiDAR to obtain coordinates, reflection intensity, distance, and index parameters, call the eight-neighborhood point set, use the Sobel operator to perform edge detection on the distance parameters, and form a preliminary boundary point set based on the edge data obtained by detection; S2: Based on the preliminary boundary point set, obtain the distance gradient, reflection intensity change rate, angle difference, and point set density change parameters, call the Canny algorithm, perform an edge screening operation on the preliminary boundary point set, and screen out the edge abnormal points to form an error point set; S3: Using the error point set, the point set range is expanded according to the neighborhood expansion rule, the least squares plane fitting model is called, a spatial residual function is established based on the expanded point set, and plane parameter fitting is performed. The fitting result is used to generate an interpolation point set using a linear interpolation method; S4: Through the interpolation point set, the tangent direction, angle change, inter-segment distance and intensity change parameters are detected, the dynamic programming algorithm is called to perform tension vector analysis, the number of edges connected by the interpolation point is determined, the parameters of the boundary connectivity structure are adjusted through linear interpolation, and the three-dimensional boundary connectivity structure is output.

Images