Gaussian modeling method combined with laser scanning point cloud data processing

Abstract

The invention provides a Gaussian modeling method combined with laser scanning point cloud data processing, and relates to the technical field of three-dimensional modeling, and the method comprises the steps: carrying out the registration of a plurality of point cloud data sets through employing a point cloud fusion algorithm, and obtaining a target point cloud data set; performing data precision, position deviation and noise influence analysis on the plurality of point cloud data sets, determining point cloud data precision, a position deviation coefficient and a noise influence coefficient, and comprehensively evaluating and determining point cloud acquisition precision; setting a data dimension reduction proportion according to the point cloud acquisition precision, and performing covariance matrix dimension reduction on the target point cloud data set; and selecting a predetermined kernel function, performing data fitting and Gaussian process regression modeling on the dimension reduction point cloud data set, and outputting a surface fitting model. According to the method, the technical problems that the requirements for storage and computing resources are too high and efficient three-dimensional modeling cannot be realized due to the fact that a large number of covariance matrixes need to be calculated in a traditional Gaussian modeling method can be solved; the dimensionality of data can be effectively reduced, and the efficiency of three-dimensional modeling is remarkably improved.

Description

Technical Field

[0001] The present application relates to the technical field of three-dimensional modeling, and in particular to a Gaussian modeling method combined with laser scanning point cloud data processing. Background Art

[0002] With the widespread application of laser scanning technology, point cloud data has become an important data source in the field of 3D modeling and spatial analysis. In the fields of computer vision, remote sensing, building information modeling, geographic information system, etc., point cloud data is used to create high-precision 3D models.

[0003] Traditional Gaussian modeling methods based on point cloud data (such as Gaussian process regression) have been proven to be effective in smoothing and surface modeling point cloud data. However, as the scale of point cloud data continues to increase, these traditional methods face enormous computational and storage pressures when processing high-resolution, large-scale point cloud data. Summary of the invention

[0004] The purpose of this application is to provide a Gaussian modeling method combined with laser scanning point cloud data processing to solve the technical problem that the traditional Gaussian modeling method based on point cloud data needs to calculate a large number of covariance matrices, resulting in excessive storage and computing resource requirements and inability to achieve efficient three-dimensional modeling.

[0005] In view of the above problems, the present application provides a Gaussian modeling method combined with laser scanning point cloud data processing, including: performing multiple laser scans on the target object to obtain multiple point cloud data sets, and using a multi-resolution point cloud fusion algorithm to perform data registration on the multiple point cloud data sets to obtain a target point cloud data set, wherein each point cloud data set corresponds to a scanning angle; performing data accuracy, position deviation and noise influence analysis on the multiple point cloud data sets respectively to determine the point cloud data accuracy, position deviation coefficient and noise influence coefficient, and comprehensively evaluate to determine the point cloud acquisition accuracy; setting a data dimensionality reduction ratio according to the point cloud acquisition accuracy, performing covariance matrix dimensionality reduction on the target point cloud data set, and obtaining a reduced dimensionality point cloud data set; selecting a predetermined kernel function, performing data fitting and Gaussian process regression modeling on the reduced dimensionality point cloud data set, and outputting a surface fitting model of the target object.

[0006] Optionally, the Gaussian modeling method combined with laser scanning point cloud data processing also includes: randomly selecting a first point cloud data set at a first scanning angle from the multiple point cloud data sets, and calculating a first average point spacing and a first area coverage rate of the first point cloud data set, and determining a first data accuracy based on the first average point spacing and the first area coverage rate; obtaining a first standard point cloud data set of the target object at the first scanning angle, performing a similarity traversal comparison on the first standard point cloud data set and the first point cloud data set, and determining a first position deviation coefficient based on the comparison result; calculating a first point cloud standard deviation and a first signal-to-noise ratio of the first point cloud data set, and determining a first noise influence coefficient based on the first point cloud standard deviation and the first signal-to-noise ratio; and continuing to perform data accuracy, position deviation and noise influence analysis on other point cloud data sets in the multiple point cloud data sets to obtain multiple point cloud data accuracies, multiple position deviation coefficients and multiple noise influence coefficients.

[0007] Optionally, the Gaussian modeling method combined with laser scanning point cloud data processing also includes: randomly selecting any area in the first point cloud data set according to a predetermined size feature and setting it as a first point cloud area, wherein the predetermined size feature is a 5 by 5 point cloud area; performing random box selection in the first standard point cloud data set according to the predetermined size feature to obtain a first standard point cloud area; performing distribution feature extraction and distribution similarity analysis on the first point cloud area and the first standard point cloud area to obtain a first area similarity; continuing to perform random box selection and similarity analysis in the first standard point cloud data set according to the predetermined size feature until a predetermined number of box selections is reached, outputting multiple standard point cloud areas and multiple area similarities, and selecting a standard point cloud area with the maximum area similarity as a first mapping point cloud area; calculating the coordinate deviation distance between the first point cloud area and the first mapping point cloud area in the target object to obtain a first position deviation distance; continuing to perform position deviation distance calculation on other point cloud areas in the first point cloud data set to obtain multiple position deviation distances, and performing mean calculation on the multiple position deviation distances to obtain the first position deviation coefficient.

[0008] Optionally, the Gaussian modeling method combined with laser scanning point cloud data processing also includes: performing dimensionless processing on the multiple point cloud data accuracies, multiple position deviation coefficients and multiple noise influence coefficients to obtain multiple normalized point cloud data accuracies, multiple normalized position deviation coefficients and multiple normalized noise influence coefficients; configuring parameter influence weights according to the influence of the point cloud data accuracy, position deviation coefficient and noise influence coefficient on the point cloud acquisition accuracy; configuring angle influence weights according to multiple scanning angles, and performing weighted calculation on the multiple normalized point cloud data accuracies, multiple normalized position deviation coefficients and multiple normalized noise influence coefficients according to the angle influence weights and parameter influence weights, and outputting the point cloud acquisition accuracy.

[0009] Optionally, the Gaussian modeling method combined with laser scanning point cloud data processing also includes: pre-training a dimensionality reduction ratio analysis channel, and inputting the point cloud acquisition accuracy into the dimensionality reduction ratio analysis channel, and outputting the data dimensionality reduction ratio; performing covariance matrix dimensionality reduction on the target point cloud data set according to the data dimensionality reduction ratio, and outputting the reduced dimensionality point cloud data set.

[0010] Optionally, the Gaussian modeling method combined with laser scanning point cloud data processing also includes: obtaining a sample point cloud acquisition accuracy set based on the point cloud acquisition records in a historical time zone, and screening the data dimensionality reduction ratios under different sample point cloud acquisition accuracies to obtain multiple data dimensionality reduction ratios that meet the predetermined modeling accuracy, calculating the sample data dimensionality reduction ratio by the mean, and constructing a sample data dimensionality reduction ratio set; using the sample point cloud acquisition accuracy set and the sample data dimensionality reduction ratio set to perform supervised learning on the random forest until convergence, to obtain a trained dimensionality reduction ratio analysis channel.

[0011] Optionally, the Gaussian modeling method combined with laser scanning point cloud data processing also includes: converting the target point cloud data set into a multi-dimensional point cloud matrix, and calculating the target covariance matrix; determining the number of principal components according to the data dimensionality reduction ratio, using the principal component analysis algorithm to perform dimensionality reduction processing on the target covariance matrix, and outputting the reduced dimensionality point cloud data set.

[0012] Optionally, the Gaussian modeling method combined with laser scanning point cloud data processing also includes: selecting an adaptive kernel function based on the multiple normalized noise influence coefficients, and using the adaptive kernel function to perform similarity calculation on the reduced dimensionality point cloud data set to generate a reduced dimensionality covariance matrix; performing data fitting and Gaussian process regression modeling on the reduced dimensionality covariance matrix to generate a surface fitting model of the target object.

[0013] The technical solution provided in this application has at least the following technical effects or advantages: By performing multiple laser scans on the target object, multiple point cloud data sets are obtained, and the multiple point cloud data sets are aligned using a multi-resolution point cloud fusion algorithm to obtain a target point cloud data set, wherein each point cloud data set corresponds to a scanning angle; then, the multiple point cloud data sets are respectively analyzed for data accuracy, position deviation and noise influence, the point cloud data accuracy, position deviation coefficient and noise influence coefficient are determined, and the point cloud acquisition accuracy is determined by comprehensive evaluation; then, the data dimension reduction ratio is set according to the point cloud acquisition accuracy, and the covariance matrix dimension reduction is performed on the target point cloud data set to obtain a reduced dimension point cloud data set; finally, a predetermined kernel function is selected, and data fitting and Gaussian process regression modeling are performed on the reduced dimension point cloud data set to output a surface fitting model of the target object; that is, by setting a corresponding data dimension reduction ratio according to the point cloud data acquisition accuracy, the data dimension can be effectively reduced while retaining the main features and information of the data, and the efficiency of three-dimensional modeling can be significantly improved, so that the target point cloud data can be efficiently and accurately processed and modeled, and a continuous, smooth and high-precision target surface model can be output.

[0014] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0016] Figure 1 A schematic diagram of a process of a Gaussian modeling method combined with laser scanning point cloud data processing in this application; Figure 2 A schematic diagram of a process for determining a first position deviation coefficient in a Gaussian modeling method combined with laser scanning point cloud data processing in the present application. DETAILED DESCRIPTION

[0017] This application provides a Gaussian modeling method combined with laser scanning point cloud data processing to solve the technical problem that the traditional Gaussian modeling method based on point cloud data needs to calculate a large number of covariance matrices, resulting in excessive storage and computing resource requirements and inability to achieve efficient three-dimensional modeling. By setting a suitable data dimensionality reduction ratio according to the point cloud data acquisition accuracy, the data dimension can be effectively reduced while retaining the main features and information of the data, significantly improving the efficiency of three-dimensional modeling, so that the target point cloud data can be efficiently and accurately processed and modeled, and a continuous, smooth and high-precision target surface model can be output.

[0018] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0019] For examples, please see the attached Figure 1 , the present application provides a Gaussian modeling method combined with laser scanning point cloud data processing, which specifically includes the following steps: S100: Perform multiple laser scans on a target object to obtain multiple point cloud data sets, and use a multi-resolution point cloud fusion algorithm to perform data registration on the multiple point cloud data sets to obtain a target point cloud data set, wherein each point cloud data set corresponds to a scanning angle.

[0020] Specifically, first, select laser scanning equipment suitable for the target object, such as laser radar equipment and 3D laser scanners. These equipment can accurately obtain three-dimensional point cloud data of the target object at different angles and distances; then configure the scanning equipment to ensure that the scanning accuracy, scanning range and sampling rate of the laser scanning equipment meet the task requirements. The scanning equipment needs to be able to perform multiple scans in all directions of the target object to ensure the comprehensiveness and detail of the data; then use the selected laser scanning equipment to perform multiple laser scans on the target object (such as target buildings, power grid towers, etc.), where the scanning angle of each laser scan is different, that is, according to the geometric shape and characteristics of the target object, a reasonable scanning path and scanning angle are designed to ensure that different scanning angles can cover all surfaces of the target object to avoid occlusion or blind spots; through multiple scans, point cloud data are obtained from multiple angles, and each scan will generate a point cloud data set, which includes the spatial coordinates and other additional information (such as intensity, color, etc.) obtained from the scanning position and angle, and multiple point cloud data sets are obtained.

[0021] Next, a multi-resolution point cloud fusion algorithm is used to perform data registration on the multiple point cloud data sets. The multi-resolution point cloud fusion algorithm registers the point clouds at multiple resolution levels, first performing a coarse registration, and then refining the registration by gradually increasing the resolution. First, the point cloud data is downsampled, and the number of points in the point cloud is reduced using methods such as the voxel grid method to reduce the complexity of the data, which makes the registration process more efficient, especially when computing resources are limited. Next, a coarse registration is performed on the downsampled point cloud data, usually using a feature matching-based method to find the preliminary transformation (translation, rotation) between different point clouds. The main purpose of this step is to estimate the approximate pose (position and attitude); then the resolution of the point cloud is gradually restored. As the resolution increases, the details of the data set are gradually retained, and the accuracy of the registration is also improved; further use higher resolution point cloud data for registration, and finely align the point cloud based on the initial pose obtained by coarse registration. This can be done through iterative algorithms (such as ICP, NDT, etc.) to accurately calculate the relative transformation between each point cloud, eliminate errors, and obtain more accurate registration results; point cloud data obtained at multiple scanning angles can be fused into a unified point cloud data set after registration. During the fusion process, it may be necessary to remove duplicate points, fill in missing areas, or perform interpolation processing to generate a complete target point cloud data set. Finally, after multiple laser scans and multi-resolution point cloud fusion algorithm processing, a unified target point cloud data set is obtained. This data set is generated by effectively registering and fusing multiple point cloud data sets, contains information from different scanning angles, and has high accuracy and completeness.

[0022] By combining the multi-resolution point cloud fusion algorithm, the problem of point cloud data registration and fusion obtained by multiple laser scans is effectively solved, providing an efficient and accurate point cloud data processing method, and finally generating a unified and high-quality target point cloud data set.

[0023] S200: Analyze the data accuracy, position deviation and noise impact of the multiple point cloud data sets respectively, determine the point cloud data accuracy, position deviation coefficient and noise impact coefficient, and comprehensively evaluate and determine the point cloud acquisition accuracy.

[0024] Furthermore, step S200 of the present application also includes: S210: Randomly select a first point cloud data set at a first scanning angle from the multiple point cloud data sets, calculate a first average point spacing and a first area coverage of the first point cloud data set, and determine a first data accuracy based on the first average point spacing and the first area coverage.

[0025] Specifically, first, a data set is randomly selected from multiple point cloud data sets, and the data set corresponds to a first scanning angle (any one of multiple scanning angles) of the target object; then, a first average point spacing and a first area coverage rate of the first point cloud data set are calculated. The average point spacing refers to calculating the distance between every two points for all points in the first point cloud data set, and taking the average of all distances. The first average point spacing reflects the density and accuracy of the point cloud data. A smaller average point spacing means that the density of the point cloud data is higher and the accuracy is better; the area coverage rate of the point cloud data is used to measure the proportion of the area covered in the scanning data and the uniformity of the points in the area. The higher the area coverage rate, the higher the scanning angle. The more complete the area covered by the data under the degree, the higher the accuracy. The regional coverage rate usually refers to the distribution of points in the point cloud data in a certain area. For example, first, the point cloud data set is projected onto a two-dimensional plane, and the area is divided into grids; then the number of points in each grid cell is counted, and the filling degree of the grid cell (that is, the density of points in each grid) is calculated. The coverage rate is the ratio of the number of grid cells filled with points in the point cloud data to the total number of grid cells, where the covered grid cells refer to the grid cells containing points, and the first regional coverage rate is obtained. The first regional coverage rate reflects the distribution range and uniformity of the point cloud data set in space. A high coverage rate means that the data can fully cover the surface of the target object with fewer missing areas.

[0026] Then, the data accuracy of the point cloud data is comprehensively evaluated by combining the first average point spacing and the first area coverage. For example, a smaller average point spacing and a higher area coverage usually indicate higher data accuracy. The first average point spacing and the first area coverage are combined to define a comprehensive indicator (such as weighted average) to calculate the data accuracy. If the point cloud data density is high and the coverage area is complete, the data accuracy is high; if the cloud data is sparse or the coverage area is incomplete, the data accuracy is low. By calculating the first average point spacing and the first area coverage of the first point cloud data set, the accuracy of the data set can be comprehensively evaluated, providing a basis for subsequent point cloud data processing, model construction and optimization.

[0027] S220: Acquire a first standard point cloud dataset of the target object at the first scanning angle, perform a similarity traversal comparison on the first standard point cloud dataset and the first point cloud dataset, and determine a first position deviation coefficient according to the comparison result.

[0028] Further, if Figure 2 As shown, step S220 of the present application also includes: S221: According to a predetermined size feature, randomly select any area in the first point cloud data set as the first point cloud area, wherein the predetermined size feature is a 5×5 point cloud area; S222: According to the predetermined size feature, randomly select in the first standard point cloud data set to obtain a first standard point cloud area; S223: Perform distribution feature extraction and distribution similarity analysis on the first point cloud area and the first standard point cloud area to obtain a first area similarity; S224: Continue to perform random selection and similarity analysis in the first standard point cloud data set according to the predetermined size feature until a predetermined number of selections is reached, output multiple standard point cloud areas and multiple area similarities, and select the standard point cloud area with the maximum area similarity as the first mapping point cloud area; S225: Calculate the coordinate deviation distance between the first point cloud area and the first mapping point cloud area in the target object to obtain a first position deviation distance; S226: Continue to perform position deviation distance calculation on other point cloud areas in the first point cloud data set to obtain multiple position deviation distances, and calculate the mean of the multiple position deviation distances to obtain the first position deviation coefficient.

[0029] Specifically, first, obtain a predetermined size feature, which is a 5 by 5 point cloud area, and the size of this area is determined by the spatial coordinates of each point in the point cloud data; then, according to the predetermined size feature, randomly select any area that meets the predetermined size feature in the first point cloud data set as the first point cloud area. Then, according to the predetermined size feature, randomly select in the first standard point cloud data set to obtain the first standard point cloud area that meets the predetermined size feature. Then, extract distribution features for the first point cloud area and the first standard point cloud area respectively. The distribution features may include the density of points, the local curvature of the point cloud, the spatial distribution pattern of the point cloud, etc.; and use the Pearson correlation coefficient to evaluate the similarity between the two areas to obtain the first area similarity.

[0030] Next, continue to perform random selection and similarity analysis in the first standard point cloud data set according to the predetermined size feature, that is, continue to perform multiple random selections in the first standard point cloud data set according to the predetermined size feature, and repeat the distribution feature extraction and similarity analysis for the standard point cloud area obtained by each selection until the predetermined number of selections is reached (which can be set according to the actual scenario, such as 1000 times), and output multiple standard point cloud areas and multiple regional similarities; then select the standard point cloud area with the maximum regional similarity as the first mapping point cloud area. Further calculate the coordinate deviation distance between the first point cloud area and the first mapping point cloud area in three-dimensional space, use the Euclidean distance to measure the difference between the two, and obtain the first position deviation distance.

[0031] Further, for other point cloud areas in the first point cloud data set, the deviation distances between them and the corresponding mapping areas are calculated using the same method, and the average of multiple position deviation distances is calculated to obtain a first position deviation coefficient, that is, the average position deviation of the point cloud data set.

[0032] S230: Calculate the first point cloud standard deviation and the first signal-to-noise ratio of the first point cloud data set, and determine the first noise influence coefficient according to the first point cloud standard deviation and the first signal-to-noise ratio; S240: Continue to perform data accuracy, position deviation and noise influence analysis on other point cloud data sets in the multiple point cloud data sets to obtain multiple point cloud data accuracy, multiple position deviation coefficients and multiple noise influence coefficients.

[0033] Specifically, each point in the point cloud data set has a spatial coordinate, and there may be errors or noise. In order to evaluate these errors, the standard deviation of the point cloud data set can be calculated; then the first point cloud standard deviation of the first point cloud data set can be calculated, and the standard deviation comprehensive index of the three-dimensional point cloud can be used to calculate the overall standard deviation. The first point cloud standard deviation represents the overall discreteness or noise level of the point cloud data set. The signal-to-noise ratio (SNR) is used to measure the ratio of signal to noise. The stronger the signal, the smaller the impact relative to the noise. For point cloud data, the signal can be the true position of the point cloud, and the noise is the error caused by the measurement error. The first signal-to-noise ratio of the first point cloud data set is calculated. Finally, the first noise influence coefficient is determined based on the first point cloud standard deviation and the first signal-to-noise ratio. The noise influence coefficient reflects the degree of influence of noise in the point cloud data on the data quality. The larger the noise influence coefficient, the smaller the influence of noise in the data and the better the data quality.

[0034] Then, using the same method as mentioned above to obtain the first data accuracy, the first position deviation coefficient and the first noise influence coefficient, continue to perform data accuracy, position deviation and noise influence analysis on other point cloud data sets in the multiple point cloud data sets to obtain multiple point cloud data accuracy, multiple position deviation coefficients and multiple noise influence coefficients of the multiple point cloud data sets.

[0035] Furthermore, step S200 of the present application also includes: S250: Perform dimensionless processing on the multiple point cloud data accuracies, multiple position deviation coefficients and multiple noise influence coefficients to obtain multiple normalized point cloud data accuracies, multiple normalized position deviation coefficients and multiple normalized noise influence coefficients; S260: Configure parameter influence weights according to the influence of point cloud data accuracy, position deviation coefficient and noise influence coefficient on point cloud acquisition accuracy; S270: Configure angle influence weights according to multiple scanning angles, perform weighted calculation on the multiple normalized point cloud data accuracies, multiple normalized position deviation coefficients and multiple normalized noise influence coefficients according to the angle influence weights and parameter influence weights, and output the point cloud acquisition accuracy.

[0036] Specifically, first, the multiple point cloud data accuracies, multiple position deviation coefficients, and multiple noise influence coefficients are dimensionlessly processed, with the purpose of eliminating the influence between different dimensions and units, so that each indicator has a unified standard, which is convenient for subsequent comprehensive analysis and weighted calculation, such as normalizing the point cloud data accuracy through standard deviation standardization so that its range falls between 0 and 1; and obtaining multiple normalized point cloud data accuracies, multiple normalized position deviation coefficients, and multiple normalized noise influence coefficients. By dimensionless processing, it is ensured that the range of all data is on the same scale, so that the subsequent weighted calculation and influence analysis can be carried out reasonably, and each coefficient is expressed on a unified scale, which is convenient for comprehensive evaluation.

[0037] Next, the influence of different indicators (data accuracy, position deviation, noise impact) on acquisition accuracy is configured and weighted. The influence of each indicator on acquisition accuracy can be assigned according to its importance in practical applications to obtain parameter influence weights. For example, the weight of each indicator is configured through expert experience. Common weight configuration methods include manual configuration based on experience and automatic optimization based on data analysis. By reasonably configuring the weights, the contribution of each indicator to the point cloud acquisition accuracy can be quantified, ensuring that the comprehensive evaluation results accurately reflect the quality of the point cloud data.

[0038] On the other hand, since the point cloud data comes from multiple scanning angles, the point cloud data of each scanning angle may have different description capabilities for the target object. Generally speaking, some angles may have better capture capabilities for some details of the target object, while other angles may have occlusion or large errors. According to the impact of multiple scanning angles on the target object, the angle influence weight of each scanning angle is configured. For example, if the data of a certain angle is more accurate or has a higher coverage, a higher weight can be configured for it. Through experience or data analysis (for example, based on factors such as the size and density of the area covered by the point cloud), a weight is assigned to each scanning angle. The angle weight can be adjusted based on factors such as scanning position, viewing angle, and data coverage. Then, according to the angle influence weight and parameter influence weight, the multiple normalized point cloud data accuracies, multiple normalized position deviation coefficients, and multiple normalized noise influence coefficients are weighted and calculated to output the point cloud acquisition accuracy. By comprehensively considering the influence weights of multiple scanning angles and the influence weights of various data accuracy parameters, the overall point cloud acquisition accuracy can be accurately evaluated, and a reliable point cloud acquisition accuracy result is finally output, which is convenient for subsequent point cloud data fusion, modeling, and optimization.

[0039] S300: Setting a data dimensionality reduction ratio according to the point cloud acquisition accuracy, performing covariance matrix dimensionality reduction on the target point cloud data set, and obtaining a dimensionality-reduced point cloud data set.

[0040] Furthermore, step S300 of the present application also includes: S310: pre-training a dimension reduction ratio analysis channel, inputting the point cloud acquisition accuracy into the dimension reduction ratio analysis channel, and outputting the data dimension reduction ratio.

[0041] Furthermore, step S310 of the present application also includes: S311: According to the point cloud collection records in the historical time zone, a sample point cloud collection accuracy set is obtained, and the data dimensionality reduction ratios under different sample point cloud collection accuracies are screened to obtain multiple data dimensionality reduction ratios that meet the predetermined modeling accuracy. The sample data dimensionality reduction ratio is obtained by average calculation, and a sample data dimensionality reduction ratio set is constructed; S312: The sample point cloud collection accuracy set and the sample data dimensionality reduction ratio set are used to perform supervised learning on the random forest until convergence, and a trained dimensionality reduction ratio analysis channel is obtained.

[0042] Specifically, first, based on the point cloud acquisition records in the historical time zone (such as the last three months), the acquisition accuracy of multiple point cloud data sets is collected. These data may come from different scanning tasks, and each task contains information such as acquisition accuracy, scanning angle, and noise level; then, for each acquisition record, the acquisition accuracy can be quantified by calculating the standard deviation, position deviation, noise influence coefficient, etc. of the data to ensure that the acquisition accuracy set contains the acquisition accuracy information of the point cloud data under different conditions; then, based on the point cloud acquisition accuracy, dimensionality reduction processing is performed for each sample point cloud data set, and an appropriate dimensionality reduction ratio is selected according to the target modeling accuracy requirements, that is, for each sample data set, a data dimensionality reduction ratio suitable for its acquisition accuracy is selected to ensure that the data after dimensionality reduction can meet the accuracy requirements in the subsequent modeling process. By analyzing multiple sample point cloud data sets, a series of data dimensionality reduction ratios that are suitable for different acquisition accuracies are obtained, and the sample data dimensionality reduction ratio is obtained by averaging multiple data dimensionality reduction ratios, and then all dimensionality reduction ratios that meet the predetermined modeling accuracy requirements are integrated into a set to form a sample data dimensionality reduction ratio set.

[0043] Then, the sample point cloud acquisition accuracy set and the sample data dimensionality reduction ratio set are used as sample training data, and the sample data is supervised and learned using a random forest regression or classification model. Random forest models data through multiple decision trees, and can effectively learn patterns from complex data sets. In this step, the input feature is the sample point cloud acquisition accuracy, and the target variable is the corresponding dimensionality reduction ratio; wherein, the random forest model will learn the corresponding dimensionality reduction ratio under different acquisition accuracy according to the accuracy information of the sample point cloud data. During the training process, the model optimizes its prediction ability through multiple iterations until the model converges, that is, reaches the predetermined accuracy or stop condition; the random forest algorithm will gradually adjust the parameters based on the training data set until convergence. Cross-validation and other methods are usually used to evaluate the generalization ability of the model and ensure that the training process can effectively fit the data. When the training error of the model meets the preset threshold and further training cannot significantly improve the performance, the model is considered to be converged, and a training-completed dimensionality reduction ratio analysis channel is obtained. Through the supervised learning of random forest, a dimensionality reduction ratio analysis channel is obtained, which can automatically determine the appropriate dimensionality reduction ratio according to the point cloud acquisition accuracy, thereby improving the efficiency and accuracy of point cloud data modeling.

[0044] Finally, the point cloud acquisition accuracy is input into the dimension reduction ratio analysis channel for analysis, and the data dimension reduction ratio is output.

[0045] S320: Performing covariance matrix dimensionality reduction on the target point cloud dataset according to the data dimensionality reduction ratio, and outputting the dimensionality reduced point cloud dataset.

[0046] Furthermore, step S320 of the present application also includes: S321: Convert the target point cloud data set into a multi-dimensional point cloud matrix, and calculate the target covariance matrix; S322: Determine the number of principal components according to the data dimensionality reduction ratio, use the principal component analysis algorithm to reduce the dimension of the target covariance matrix, and output the reduced-dimensional point cloud data set.

[0047] Specifically, first, the target point cloud data set is converted into a multidimensional point cloud matrix. For example, assuming that the target point cloud data set contains N points, each point has a D-dimensional feature. In order to apply principal component analysis to reduce the dimension of the point cloud data, the point cloud data set must first be converted into a multidimensional matrix, such as converting the target point cloud data set into an N-by-D matrix X, where each row represents a point and contains the D-dimensional coordinates of the point. The covariance matrix describes the correlation between the dimensions in the point cloud data set. When calculating the covariance matrix, the data must first be centered, that is, the mean of each dimension is subtracted from the dimension. The covariance matrix reveals the linear relationship between the dimensions, which helps the subsequent dimensionality reduction step.

[0048] Next, according to the data dimensionality reduction ratio (for example, retaining 60% of the variance), determine the number of principal components k that need to be retained in the principal component analysis. The number of principal components retained determines the dimension of the data after dimensionality reduction. First, perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors. The eigenvalues ​​reflect the variance of each principal component, so the number of principal components to be retained can be determined by the proportion of variance occupied by the cumulative eigenvalues. Then, according to the number of principal components, the principal component analysis algorithm is used to reduce the dimension of the target covariance matrix, that is, after determining the number of principal components k, the first k eigenvectors can be selected to construct a reduced dimension matrix; the original data matrix is ​​projected onto these principal components to obtain a reduced dimension data set; the reduced dimension data set retains the main features in the data and reduces the dimension. By reducing the dimension of point cloud data, the computational complexity is reduced, and sufficient information is retained, ensuring efficient and accurate three-dimensional modeling and data processing.

[0049] S400: Select a predetermined kernel function, perform data fitting and Gaussian process regression modeling on the dimension-reduced point cloud data set, and output a surface fitting model of the target object.

[0050] Furthermore, step S400 of the present application also includes: S410: Select an adaptive kernel function according to the multiple normalized noise influence coefficients, and use the adaptive kernel function to perform similarity calculation on the reduced dimensionality point cloud data set to generate a reduced dimensionality covariance matrix; S420: Perform data fitting and Gaussian process regression modeling on the reduced dimensionality covariance matrix to generate a surface fitting model of the target object.

[0051] Specifically, first, an adaptive kernel function is selected according to the multiple normalized noise influence coefficients, such as selecting a suitable kernel function to adapt the point cloud data set according to the normalized noise influence coefficient. For example, the point cloud data with less noise may select the RBF kernel, while the data with more noise may select the Matern kernel or other kernel functions adapted to the data with stronger noise. Then, the adaptive kernel function is used to calculate the similarity between every two points in the reduced dimensionality point cloud data set, and the covariance matrix after dimensionality reduction is constructed according to the result of the similarity calculation. The covariance matrix describes the similarity between each pair of points in the point cloud data set, which is the basis for subsequent Gaussian process regression modeling.

[0052] Then, the point cloud data is fitted using the covariance matrix after dimensionality reduction and the Gaussian process regression (GPR) model. Gaussian process regression is a non-parametric method that predicts the output value by calculating the similarity (covariance matrix) between input data points. The point cloud data is fitted using the Gaussian process regression method to generate a surface fitting model, which not only provides the predicted value of the point cloud surface, but also gives the prediction uncertainty (i.e., covariance) of each point, which is used to evaluate the reliability of the surface fitting. A three-dimensional surface fitting model of the target object is generated through the Gaussian process regression model. The model reflects the trend and shape of the point cloud data and can provide uncertainty measurement for each predicted point. Through Gaussian process regression modeling, a smooth and continuous surface fitting model of the target object is generated.

[0053] In summary, the Gaussian modeling method combined with laser scanning point cloud data processing provided by this application has the following technical effects: By performing multiple laser scans on the target object, multiple point cloud data sets are obtained, and the multiple point cloud data sets are aligned using a multi-resolution point cloud fusion algorithm to obtain a target point cloud data set, wherein each point cloud data set corresponds to a scanning angle; then, the multiple point cloud data sets are respectively analyzed for data accuracy, position deviation and noise influence, the point cloud data accuracy, position deviation coefficient and noise influence coefficient are determined, and the point cloud acquisition accuracy is determined by comprehensive evaluation; then, the data dimension reduction ratio is set according to the point cloud acquisition accuracy, and the covariance matrix dimension reduction is performed on the target point cloud data set to obtain a reduced dimension point cloud data set; finally, a predetermined kernel function is selected, and data fitting and Gaussian process regression modeling are performed on the reduced dimension point cloud data set to output a surface fitting model of the target object; that is, by setting a corresponding data dimension reduction ratio according to the point cloud data acquisition accuracy, the data dimension can be effectively reduced while retaining the main features and information of the data, and the efficiency of three-dimensional modeling can be significantly improved, so that the target point cloud data can be efficiently and accurately processed and modeled, and a continuous, smooth and high-precision target surface model can be output.

[0054] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0055] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A Gaussian modeling method combined with laser scanning point cloud data processing, characterized in that: Methods include: Perform multiple laser scans on the target object to obtain multiple point cloud data sets, and use a multi-resolution point cloud fusion algorithm to perform data registration on the multiple point cloud data sets to obtain a target point cloud data set, wherein each point cloud data set corresponds to a scanning angle; Performing data accuracy, position deviation and noise impact analysis on the multiple point cloud data sets respectively, determining the point cloud data accuracy, position deviation coefficient and noise impact coefficient, and comprehensively evaluating and determining the point cloud acquisition accuracy; Setting a data dimensionality reduction ratio according to the point cloud acquisition accuracy, performing covariance matrix dimensionality reduction on the target point cloud data set, and obtaining a dimensionality-reduced point cloud data set; A predetermined kernel function is selected to perform data fitting and Gaussian process regression modeling on the dimension-reduced point cloud data set, and a surface fitting model of the target object is output.

2. The Gaussian modeling method combined with laser scanning point cloud data processing according to claim 1, characterized in that: The data accuracy, position deviation and noise impact analysis are performed on the multiple point cloud data sets respectively, including: Randomly selecting a first point cloud data set at a first scanning angle from the multiple point cloud data sets, calculating a first average point spacing and a first area coverage rate of the first point cloud data set, and determining a first data accuracy according to the first average point spacing and the first area coverage rate; Acquire a first standard point cloud data set of the target object at the first scanning angle, perform a similarity traversal comparison between the first standard point cloud data set and the first point cloud data set, and determine a first position deviation coefficient according to the comparison result; Calculating a first point cloud standard deviation and a first signal-to-noise ratio of the first point cloud data set, and determining a first noise influence coefficient according to the first point cloud standard deviation and the first signal-to-noise ratio; Continue to perform data accuracy, position deviation and noise impact analysis on other point cloud data sets in the multiple point cloud data sets to obtain multiple point cloud data accuracy, multiple position deviation coefficients and multiple noise impact coefficients.

3. The Gaussian modeling method combined with laser scanning point cloud data processing according to claim 2 is characterized in that: Performing a similarity traversal comparison on the first standard point cloud data set and the first point cloud data set, and determining a first position deviation coefficient according to the comparison result, includes: According to a predetermined size feature, randomly selecting any area in the first point cloud data set as a first point cloud area, wherein the predetermined size feature is a 5×5 point cloud area; Performing a random selection in the first standard point cloud data set according to the predetermined size feature to obtain a first standard point cloud area; Performing distribution feature extraction and distribution similarity analysis on the first point cloud region and the first standard point cloud region to obtain a first region similarity; Continue to perform random box selection and similarity analysis in the first standard point cloud data set according to the predetermined size feature until a predetermined number of box selections is reached, output a plurality of standard point cloud regions and a plurality of region similarities, and select a standard point cloud region with the maximum region similarity as the first mapping point cloud region; Calculating a coordinate deviation distance between the first point cloud area and the first mapping point cloud area in the target object to obtain a first position deviation distance; Continue to perform position deviation distance calculation on other point cloud areas in the first point cloud data set to obtain multiple position deviation distances, and perform mean calculation on the multiple position deviation distances to obtain the first position deviation coefficient.

4. The Gaussian modeling method combined with laser scanning point cloud data processing according to claim 3 is characterized in that: Comprehensive evaluation to determine the accuracy of point cloud acquisition, including: Performing dimensionless processing on the plurality of point cloud data accuracies, the plurality of position deviation coefficients, and the plurality of noise influence coefficients to obtain a plurality of normalized point cloud data accuracies, a plurality of normalized position deviation coefficients, and a plurality of normalized noise influence coefficients; Configure parameter influence weights according to the influence of point cloud data accuracy, position deviation coefficient, and noise influence coefficient on point cloud acquisition accuracy; Angle influence weights are configured according to multiple scanning angles, and weighted calculations are performed on the multiple normalized point cloud data accuracies, multiple normalized position deviation coefficients, and multiple normalized noise influence coefficients according to the angle influence weights and parameter influence weights to output the point cloud acquisition accuracy.

5. The Gaussian modeling method combined with laser scanning point cloud data processing according to claim 1, characterized in that: The data dimension reduction ratio is set according to the point cloud acquisition accuracy, and the covariance matrix dimension reduction is performed on the target point cloud data set to obtain the dimension-reduced point cloud data set, including: Pre-training a dimension reduction ratio analysis channel, inputting the point cloud acquisition accuracy into the dimension reduction ratio analysis channel, and outputting the data dimension reduction ratio; According to the data dimensionality reduction ratio, covariance matrix dimensionality reduction is performed on the target point cloud data set, and the reduced dimensionality point cloud data set is output.

6. The Gaussian modeling method combined with laser scanning point cloud data processing according to claim 5, characterized in that: Pre-trained dimensionality reduction ratio analysis channel, including: According to the point cloud collection records in the historical time zone, a sample point cloud collection accuracy set is obtained, and the data dimensionality reduction ratios under different sample point cloud collection accuracies are screened to obtain multiple data dimensionality reduction ratios that meet the predetermined modeling accuracy. The sample data dimensionality reduction ratio is obtained by mean calculation, and a sample data dimensionality reduction ratio set is constructed; The sample point cloud acquisition accuracy set and the sample data dimensionality reduction ratio set are used to perform supervised learning on the random forest until convergence, thereby obtaining a trained dimensionality reduction ratio analysis channel.

7. The Gaussian modeling method combined with laser scanning point cloud data processing according to claim 5, characterized in that: According to the data dimensionality reduction ratio, the covariance matrix dimensionality reduction is performed on the target point cloud data set, and the dimensionality reduction point cloud data set is output, including: Convert the target point cloud data set into a multi-dimensional point cloud matrix, and calculate a target covariance matrix; The number of principal components is determined according to the data dimensionality reduction ratio, the target covariance matrix is ​​subjected to dimensionality reduction processing using a principal component analysis algorithm, and the dimensionality-reduced point cloud data set is output.

8. The Gaussian modeling method combined with laser scanning point cloud data processing according to claim 4, characterized in that: Selecting a predetermined kernel function, performing data fitting and Gaussian process regression modeling on the dimension-reduced point cloud data set, and outputting a surface fitting model of the target object, including: Selecting an adaptive kernel function according to the multiple normalized noise influence coefficients, and using the adaptive kernel function to perform similarity calculation on the reduced-dimensional point cloud data set to generate a reduced-dimensional covariance matrix; Data fitting and Gaussian process regression modeling are performed on the reduced dimension covariance matrix to generate a surface fitting model of the target object.

Images