A building stereo extraction method based on multispectral laser radar point cloud data

Abstract

This invention relates to a method for extracting 3D building data based on multispectral LiDAR point cloud data, belonging to the field of multispectral LiDAR point cloud technology. First, the dimensionality features of the multispectral LiDAR point cloud data are extracted. The optimal neighborhood is selected using Shannon entropy theory, and geometric features are extracted from this neighborhood. A support vector machine is used to classify the multispectral LiDAR point cloud data into three categories: line, surface, and discrete. The 2D planar type multispectral LiDAR point cloud data is then extracted. Finally, building data from the multispectral LiDAR point cloud is extracted using height and neighborhood variance filtering. This invention selects the optimal neighborhood for feature extraction, increasing the accuracy of classification and improving the precision of 3D building extraction from multispectral LiDAR point cloud data.

Description

Technical Field

[0001] This invention relates to a method for extracting three-dimensional structures from multispectral lidar point cloud data, belonging to the field of multispectral lidar point cloud technology. Background Technology

[0002] Automatic extraction of building information from remote sensing scenes is a hot topic in remote sensing application research. However, due to various factors such as different imaging conditions, complex background features, and diverse building types, automatic building extraction is extremely difficult. Traditional building extraction based on spectral images is limited by the imaging methods themselves, only able to extract two-dimensional planar information of remote sensing scene features, thus limiting its application in three-dimensional building extraction. Traditional LiDAR, on the other hand, can only extract three-dimensional spatial information of remote sensing scenes, lacking spectral information, and still has limitations. While combining spectral images and LiDAR data can simultaneously acquire two-dimensional spectral information and three-dimensional spatial information, this combined use inevitably brings difficulties in image alignment, as well as problems such as spectral mismatch and spectral drift, posing even greater challenges to practical applications.

[0003] With the development of supercontinuum laser technology, multispectral LiDAR has moved from research and development to application. Using multispectral LiDAR systems with multi-band information detection capabilities, the three-dimensional spatial distribution and spectral information of targets can be acquired simultaneously, providing a data foundation and information guarantee for the fine three-dimensional spatial classification of ground features in remote sensing scenes. The National Medium- and Long-Term Development Plan for Civil Space Infrastructure (2015-2025) lists multispectral LiDAR as a key area requiring forward deployment and development. In multispectral LiDAR remote sensing ground feature classification, the integrated acquisition of three-dimensional spatial and spectral information of the observed target has been achieved, fundamentally solving the problems of mismatch and drift caused by the joint use of spectral images and LiDAR data. It also takes into account the complementary information between the two, providing a data foundation and information guarantee for the extraction of buildings in remote sensing scenes, and will significantly improve the accuracy of three-dimensional building extraction. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for building stereo extraction based on multispectral lidar point cloud data, so as to solve the problem of classification inaccuracy caused by different types of land features being classified into the same neighborhood when selecting the neighborhood of point cloud in traditional methods.

[0005] The technical solution of this invention is: a method for extracting three-dimensional building data based on multispectral lidar point cloud data. First, the dimensionality features of the multispectral lidar point cloud data are extracted. The optimal neighborhood is selected using Shannon entropy theory. Geometric features are then extracted based on this neighborhood. The multispectral lidar point cloud data is divided into three categories: line, surface, and discrete using a support vector machine. The 2D planar type of multispectral lidar point cloud data is extracted. Finally, the multispectral lidar point cloud building data is extracted by height and neighborhood variance filtering.

[0006] The specific steps are as follows:

[0007] Step 1: Filter multispectral LiDAR point cloud data according to spectral angle. For each point in the multispectral LiDAR point cloud data, by setting different spectral angle thresholds, points with spectral angles less than the threshold are included in the neighborhood. By selecting different spectral angle thresholds, six neighborhoods of sizes of 10, 20, 50, 100, 150, and 200 points are respectively selected as the preliminary neighborhoods for each point in the multispectral LiDAR point cloud, so that the optimal neighborhood can be selected from the preliminary neighborhoods in subsequent steps.

[0008] Step 2: For each point in the multispectral lidar point cloud, construct a covariance matrix using the 3D coordinates of the point and its neighbors within the six pre-selected neighborhoods from Step 1. Perform principal component analysis on the covariance matrix to obtain its three eigenvalues ​​in the current neighborhood, which are then sorted in descending order to become λ1, λ2, and λ3. Calculate the dimensionality of the point in its current neighborhood using the following formula:

[0009]

[0010]

[0011]

[0012] Where a 1D It is a one-dimensional linear feature, a 2D It is a two-dimensional planar feature, a 3D For three-dimensional discrete features, a 1D a 2D a 3D The sum of is 1, and the dimensional features represent the probabilities that the point belongs to three different dimensional types, i.e., when the point cloud is linearly distributed, a 1D When a is relatively large and distributed in a planar manner, 2D When a is large and has a discrete distribution 3D Larger;

[0013] Step 3: Extract the optimal neighborhood from the preliminary neighborhood based on Shannon entropy theory;

[0014] Step 4: Re-extract the geometric features of the multispectral lidar point cloud data in the optimal neighborhood. The geometric features include dimensional features, surface roughness, plane slope, and intensity roughness.

[0015]

[0016] Where SR is surface roughness, PS is plane slope, and IR is strength roughness;

[0017] Step 5: Input the geometric features of the multispectral lidar point cloud data into a support vector machine for classification. Classify the multispectral lidar point cloud into three different categories: one-dimensional linear, two-dimensional planar, and three-dimensional discrete. Extract the multispectral lidar point cloud data belonging to the two-dimensional planar type. Through height and neighborhood variance filtering, remove the ground and non-building planes to obtain the multispectral lidar point cloud building data.

[0018] Furthermore, Step 3 specifically includes:

[0019] In Step 1, for each point in the multispectral lidar point cloud data, different sized preliminary neighborhoods are selected, and dimensional features are calculated based on the eigenvalues ​​of its singular decomposition. The entropy value is then calculated using the dimensional features of the lidar point cloud data.

[0020]

[0021] Among them, E f This indicates that the entropy value was obtained using Shannon's entropy theory. Let N represent the neighborhood of the P-th point.

[0022] For each point in the multispectral lidar point cloud data, an optimizer must be applied separately, and the dimensional features of that point must be recalculated by substituting the formula from Step 1 into the optimal neighborhood. Specifically:

[0023]

[0024] in, This indicates that neighborhood optimization is performed on the P-th point, selecting the optimal neighborhood with the minimum entropy value from the pre-selected neighborhoods.

[0025] Furthermore, in Step 5, the extracted 2D planar type multispectral lidar point cloud data includes multispectral lidar point cloud data that are planarly distributed on the ground, building rooftops, and other surfaces.

[0026] To accurately obtain building data, height and variance filtering were used to remove interference information such as ground level, thus accurately extracting the building data.

[0027] Traditional methods sometimes group multispectral lidar point cloud data belonging to different types of land features into the same neighborhood. This invention optimizes neighborhood selection before extracting multispectral lidar point cloud data, ensuring that multispectral lidar data within the same neighborhood originates from the same type of land feature. This improves the classification accuracy of multispectral lidar point cloud data and enables precise extraction of building data.

[0028] The beneficial effects of this invention are: Compared with the prior art, this invention mainly solves the problem of classification inaccuracy caused by classifying different types of land features into the same neighborhood when selecting point cloud neighborhoods in traditional methods. By selecting the optimal neighborhood for feature extraction, the accuracy of classification is increased, and the precision of building stereo extraction from multispectral lidar point cloud data is improved. Attached Figure Description

[0029] Figure 1 This is a flowchart of the steps of the present invention;

[0030] Figure 2 This is a view of multispectral lidar point cloud data of ground features in this invention;

[0031] Figure 3 This invention is the result of classifying multispectral lidar point cloud data using a support vector machine.

[0032] Figure 4 This is the classification data obtained by classifying data in a support vector machine according to the present invention;

[0033] Figure 5 These are buildings in the multispectral lidar point cloud data finally extracted by this invention. Detailed Implementation

[0034] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0035] like Figure 1 As shown, a method for 3D building extraction based on multispectral lidar point cloud data is described. The specific steps of the method are as follows:

[0036] Step 1: Extract multispectral lidar point cloud data from the remote sensing scene, such as Figure 2 The scenario is shown. Multispectral LiDAR point cloud data is filtered according to spectral angle. For each point in the multispectral LiDAR point cloud data, by setting different spectral angle thresholds, points with spectral angles smaller than the threshold are included in the neighborhood. By selecting different spectral angle thresholds, six neighborhoods of sizes of 10, 20, 50, 100, 150, and 200 points are respectively selected as the preliminary neighborhoods for each point in the multispectral LiDAR point cloud, so that the optimal neighborhood can be selected from the preliminary neighborhoods in subsequent steps.

[0037] Step 2: For each point in the multispectral lidar point cloud, construct a covariance matrix using the 3D coordinates of the point and its neighbors within the six pre-selected neighborhoods from Step 1. Perform principal component analysis on the covariance matrix to obtain its three eigenvalues ​​in the current neighborhood, which are then sorted in descending order to become λ1, λ2, and λ3. Calculate the dimensionality of the point in its current neighborhood using the following formula:

[0038]

[0039]

[0040]

[0041] Where a 1D It is a one-dimensional linear feature, a 2D It is a two-dimensional planar feature, a 3D For three-dimensional discrete features, a 1D a 2D a 3D The sum of is 1, and the dimensional features represent the probabilities that the point belongs to three different dimensional types, i.e., when the point cloud is linearly distributed, a 1D When a is relatively large and distributed in a planar manner, 2D When a is large and has a discrete distribution 3D Larger;

[0042] Step 3: Based on Shannon entropy theory, extract the optimal neighborhood from the preliminary neighborhood. For each point in the multispectral LiDAR point cloud data selected in Step 1, calculate the dimensional features based on the eigenvalues ​​of its singular decomposition within the different sized preliminary neighborhoods. Use the dimensional features of the LiDAR point cloud data to calculate the entropy value.

[0043]

[0044] Among them, E f This indicates that the entropy value was obtained using Shannon's entropy theory. Let N represent the neighborhood of the P-th point.

[0045] Each point in the multispectral lidar point cloud data needs to be optimized separately using an optimizer, and the dimensional features of that point are recalculated by substituting the formula from Step 1 into the optimal neighborhood.

[0046] Specifically:

[0047]

[0048] in, This indicates that neighborhood optimization is performed on the P-th point, selecting the optimal neighborhood with the minimum entropy value from the pre-selected neighborhoods.

[0049] Step 4: Re-extract the geometric features of the multispectral lidar point cloud data in the optimal neighborhood. The geometric features include dimensional features, surface roughness, plane slope, and intensity roughness.

[0050]

[0051] Where SR is surface roughness, PS is plane slope, and IR is strength roughness;

[0052] Step 5: Input the geometric features of the multispectral LiDAR point cloud data into a support vector machine for classification, classifying it into three different categories: one-dimensional linear, two-dimensional planar, and three-dimensional discrete multispectral LiDAR point clouds. The classification results are as follows: Figure 3 As shown, the categorized data is as follows: Figure 4 The OA is 97.04%; Kappa is 92.69%. Multispectral LiDAR point cloud data belonging to the two-dimensional planar type is extracted. After height and neighborhood variance filtering, ground and non-building planes are removed to obtain multispectral LiDAR point cloud building data, such as... Figure 5 .

[0053] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for extracting three-dimensional structures from multispectral lidar point cloud data, characterized in that: Step 1: Filter the multispectral lidar point cloud data according to the spectral angle. For each point in the multispectral lidar point cloud data, by setting different spectral angle thresholds, points with spectral angles less than the threshold are included in the neighborhood. By selecting different spectral angle thresholds, six neighborhoods of sizes of 10, 20, 50, 100, 150, and 200 points are selected as the preliminary neighborhoods for each point in the multispectral lidar point cloud, so that the optimal neighborhood can be selected from the preliminary neighborhoods in subsequent steps. Step 2: For each point in the multispectral lidar point cloud, using the three-dimensional coordinates of the point and its neighbors in the six pre-selected neighborhoods from Step 1, construct a covariance matrix. Perform principal component analysis on the covariance matrix to obtain its three eigenvalues ​​in the current neighborhood. Arrange these eigenvalues ​​in descending order, which are λ1, λ2, and λ3. Calculate the dimensionality of the point in its current neighborhood using the formula: Where a 1D It is a one-dimensional linear feature, a 2D It is a two-dimensional planar feature, a 3D For three-dimensional discrete features, a 1D a 2D a 3D The sum of is 1, and the dimensional features represent the probability that the point belongs to three different dimensional types. Step 3: Extract the optimal neighborhood from the preliminary neighborhood based on Shannon entropy theory; Step 4: Re-extract the geometric features of the multispectral lidar point cloud data in the optimal neighborhood. The geometric features include dimensional features, surface roughness, plane slope, and intensity roughness. Where SR is surface roughness, PS is plane slope, and IR is strength roughness; Step 5: Input the geometric features of the multispectral lidar point cloud data into a support vector machine for classification. Classify the multispectral lidar point cloud into three different categories: one-dimensional linear, two-dimensional planar, and three-dimensional discrete. Extract the multispectral lidar point cloud data belonging to the two-dimensional planar type. Through height and neighborhood variance filtering, remove the ground and non-building planes to obtain the multispectral lidar point cloud building data.

2. The method for extracting three-dimensional building data from multispectral lidar point cloud data according to claim 1, characterized in that, Step 3 specifically refers to: In Step 1, for each point in the multispectral lidar point cloud data, different sized preliminary neighborhoods are selected, and dimensional features are calculated based on the eigenvalues ​​of its singular decomposition. The entropy value is then calculated using the dimensional features of the lidar point cloud data. Among them, E f This indicates that the entropy value was obtained using Shannon's entropy theory. Let N represent the neighborhood of the Pth point; For each point in the multispectral lidar point cloud data, it is necessary to optimize it separately through the optimizer, and then re-substitute the formula in Step 1 into the optimal neighborhood to calculate the dimensional features of the point. Specifically: in This indicates that neighborhood optimization is performed on the P-th point, selecting the optimal neighborhood with the minimum entropy value from the pre-selected neighborhoods.

3. The method for extracting three-dimensional building data from multispectral lidar point cloud data according to claim 1, characterized in that: In Step 5, the extracted two-dimensional planar multispectral lidar point cloud data includes multispectral lidar point cloud data that are planarly distributed on the ground, building rooftops, and other surfaces.

Images