Lidar point cloud occlusion repairing method based on neighborhood correlation

Abstract

The application discloses a kind of based on neighborhood correlation laser radar point cloud occlusion repair method, it is related to laser radar remote sensing technical field.Based on neighborhood correlation laser radar point cloud occlusion repair method, steps include data preparation, point cloud data voxelization, voxel attribute labeling and voxel occlusion repair.The application is analyzed by neighborhood correlation to all possible occluded voxels, and whether the voxel is occluded according to dynamic threshold value judgment, based on the judgment result to all occluded voxels are filled and repaired, not only can the range of neighborhood and weight be adaptively adjusted according to the complexity of occlusion area and data distribution, so as to more accurately capture contextual information, improve repair accuracy, so as to enhance the adaptability of algorithm to complex occlusion environment, but also can the specific situation of occlusion area, such as height and density, be adjusted in real time according to the judgment standard, so as to more accurately identify the occlusion missing area, improve the repair effect of the laser radar point cloud occlusion repair method.

Description

Technical Field

[0001] This invention relates to the field of lidar remote sensing technology, specifically to a lidar point cloud occlusion repair method based on neighborhood correlation. Background Technology

[0002] Forest ecosystems play a central role in global carbon cycling, climate regulation, and biodiversity conservation. Among them, coniferous forests, as the most widespread forest type in the Northern Hemisphere, present unique challenges to remote sensing technology due to their structural complexity. Tree canopy index (LAI), a key parameter characterizing vegetation canopy structure, is directly related to photosynthetic efficiency, transpiration rate, and carbon sequestration capacity, and is a core indicator for assessing forest productivity and ecological function. While traditional LAI measurement methods (such as destructive sampling and leaf litter collection) offer high accuracy, they suffer from low efficiency, ecological damage, and difficulty in scaling up, making them unsuitable for modern forestry monitoring needs.

[0003] Passive optical remote sensing (such as multispectral and hyperspectral imagery) uses vegetation indices (such as Normalized Differential Vegetation Index (NDVI) and Modified Soil-Adjusted Vegetation Index (MSAVI)) to retrieve LAI (Local Area Identification). While widely used, this method is limited by signal saturation (especially in dense canopies) and insufficient vertical structure resolution. LiDAR, through active laser pulse emission, can accurately capture the three-dimensional structure of vegetation, compensating for the lack of vertical information in passive remote sensing. Ground-based LiDAR (such as Terrestrial Laser Scanning (TLS) systems) can provide centimeter-level resolution data, but its coverage is limited. Airborne LiDAR is suitable for regional scales, but its flight altitude and conifer aggregation characteristics cause severe data occlusion problems. The main problems are: (1) Canopy penetration limitation. Dense upper needles block the laser beam, lower voxel data is missing, and the LAI inversion value is low. (2) Voxel model defects. Existing algorithms (such as VoxLAD and AMAPVox) rely on the Beer-Lambert law or contact ratio method, assuming that the leaves are randomly distributed. However, the spatial clustering of needles causes the clustering index (Ω) to change dynamically with the zenith angle (especially when the error is significant at >60°), so complex corrections (such as path length distribution models) need to be introduced. (3) Insufficient data integrity. About 30%-50% of the lower voxels in the airborne point cloud are marked as "empty" due to occlusion. Traditional filling methods (such as interpolation) ignore the correlation of the needle neighborhood, further amplifying the error.

[0004] To address the aforementioned problems, this invention proposes a LiDAR point cloud occlusion repair method based on neighborhood correlation, which offers excellent repair results. Summary of the Invention

[0005] To overcome the shortcomings of existing lidar point cloud occlusion repair methods, which require complex corrections and ignore neighborhood correlation when repairing missing point cloud data using traditional filling methods, thus further amplifying errors and resulting in poor repair performance, this invention proposes a lidar point cloud occlusion repair method based on neighborhood correlation with good repair performance.

[0006] The LiDAR point cloud occlusion repair method based on neighborhood correlation includes the following steps: Data preparation involves acquiring LiDAR point cloud data of the site under study using an airborne lidar, and then preprocessing the LiDAR point cloud data to obtain preprocessed LiDAR point cloud data. Point cloud data voxels are generated by vertically layering the preprocessed LiDAR point cloud data according to its height to obtain layered LiDAR point cloud data. Adaptive size voxels are then generated for each height layer of the layered LiDAR point cloud data to obtain voxel data of the plot under study. Voxel attribute annotation maps each point in the layered LiDAR point cloud data to the corresponding voxel in the voxel data of the plot under study, and marks whether the voxel contains points, thus obtaining the voxel attribute data of the plot under study. Voxel occlusion repair involves pre-judging voxels in the voxel attribute data of the plot under study to obtain all potentially occluded voxels. Neighborhood correlation analysis is then performed on all potentially occluded voxels, and a dynamic threshold is used to determine whether the voxel is occluded. Based on the judgment results, all occluded voxels are filled and repaired to obtain repaired voxel attribute data, which is then used for LAI inversion.

[0007] As a preferred aspect of the invention, the specific steps for preprocessing the LiDAR point cloud data to obtain preprocessed LiDAR point cloud data are as follows: Data format conversion: Parses non-standard format point cloud data in LiDAR point cloud data and converts it into a common set of 3D coordinate points; and handles and skips rows with non-standard format in LiDAR point cloud data. For data denoising, a noise threshold is dynamically set based on the local density of points in LiDAR point cloud data. A distance-based filtering method is used to calculate the distance from each point to its nearest neighbor. If the distance exceeds the dynamically set noise threshold, the point is removed. Uniform sampling divides the point cloud space formed by LiDAR point cloud data into a uniform cubic grid, with each grid retaining only one center point.

[0008] As a preferred aspect of the invention, the specific steps for vertically layering the preprocessed LiDAR point cloud data according to height to obtain layered LiDAR point cloud data are as follows: A three-dimensional Cartesian coordinate system is constructed based on the preprocessed LiDAR point cloud data. The x, y, and y axes of this system correspond to the length, width, and height of the preprocessed LiDAR point cloud data, respectively. The height range of the preprocessed LiDAR point cloud data, including its minimum height, is calculated based on this constructed coordinate system. and maximum height ; The height interval of the point cloud data layer is preset as needed, and the height range of the preprocessed LiDAR point cloud data is divided into several layers based on the preset height interval. The preprocessed LiDAR point cloud data is divided into different layers according to height, resulting in layered LiDAR point cloud data, where each layer contains points within a corresponding height range.

[0009] As a preferred aspect of the invention, the specific steps for adaptively scaling the LiDAR point cloud data of each height layer in the layered LiDAR point cloud data to obtain the voxel data of the plot under study are as follows: For each height layer in the layered LiDAR point cloud data, calculate the distance from each point in that height layer to its nearest neighbor, obtain the nearest neighbor distance of each point in that height layer, and take the median of the nearest neighbor distances of all points in that height layer as the voxel size of that height layer. The length, width, and height of the voxel are all set as the voxel dimensions of the height layer, and the horizontal range of the height layer is divided into... The grid is used to obtain voxel data of the plot under study, where , , and among them and These represent the maximum and minimum values ​​of a point within that height layer along the x-axis, respectively. and These represent the maximum and minimum values ​​of a point within that height layer along the y-axis, respectively. This indicates the voxel size of the height layer.

[0010] As a preferred aspect of the invention, the specific steps of mapping each point in the layered LiDAR point cloud data to the corresponding voxel in the voxel data of the plot under study, and marking whether the voxel contains points, to obtain the voxel attribute data of the plot under study are as follows: Determine the voxel partitioning rules, including the length, width, and height of the voxels, find the minimum coordinate value of the layered LiDAR point cloud data, and use it as the starting point for voxel partitioning; For each point in the layered LiDAR point cloud data, the voxel index to which the point belongs is calculated based on the point's coordinates in the three-dimensional rectangular coordinate system and the voxel partitioning rules. The attributes of each voxel are recorded using a hash table or array. If a voxel contains at least one point, it is marked as 1; otherwise, it is marked as 0, thus obtaining the voxel attribute data of the plot under study.

[0011] As a preferred aspect of the invention, the specific steps for performing occlusion pre-judgment on the voxels in the voxel attribute data of the plot to be studied to obtain all potentially occluded voxels are as follows: The data is processed layer by layer from the top layer of the voxel attribute data of the plot under study to the bottom layer. During this process, for each empty voxel marked as 0 in the voxel attribute data of the plot under study, it is first checked whether the upper voxel of the empty voxel is marked as 0. If the upper voxel is marked as 0, the current empty voxel is skipped. If the upper voxel is marked as 1, it is determined that the current empty voxel may be occluded, and the lower voxel of the empty voxel is checked. Check if the lower voxel of the empty voxel is marked as 0. If the lower voxel is marked as 1, skip the current empty voxel. If the lower voxel is marked as 0, determine that the current empty voxel may be occluded and continue to check other empty voxels to obtain all voxels that may be occluded.

[0012] As a preferred aspect of the invention, the specific steps of performing neighborhood correlation analysis on all potentially occluded voxels and determining whether a voxel is occluded based on a dynamic threshold, filling and repairing all occluded voxels based on the determination results, obtaining repaired voxel attribute data, and using it for LAI inversion are as follows: For each voxel that may be occluded, the neighborhood range is dynamically determined based on the point density and fill condition of the current voxel's height layer. When the fill density of this height layer is less than 70%, the neighborhood range is 26 neighborhoods; when the fill density of this height layer is greater than or equal to 70%, the neighborhood range is 8 neighborhoods. Based on the dynamically determined neighborhood range, a neighborhood system is determined for each potentially occluded voxel. The sum of states of the neighborhood system is calculated using a weighted summation method. When the sum of states of the neighborhood system is greater than or equal to the dynamic threshold corresponding to the current voxel, the voxel is determined to be occluded, and its label is filled with 1. When the sum of states of the neighborhood system is less than the dynamic threshold corresponding to the current voxel, the voxel's label is kept at 0. This yields the repaired voxel attribute data. The specific formula for calculating the sum of states of the neighborhood system is as follows: , in Represents the sum of states of the neighborhood system. Represents the first in the neighborhood system The state tag value of each neighboring voxel. Represents the first in the neighborhood system The state weights of each neighboring voxel are assigned hierarchically based on the distance between the current voxel and its neighboring voxels. Indicates the size of the neighborhood.

[0013] As a preferred aspect of the invention, the specific value of the dynamic threshold needs to be based on the degree of occlusion of the current voxel's height layer, and the specific value of the dynamic threshold also decreases with the height of the current voxel's height layer.

[0014] The present invention has the following advantages: 1. This invention performs neighborhood correlation analysis on all potentially occluded voxels and determines whether a voxel is occluded based on a dynamic threshold. Based on the determination results, it fills and repairs all occluded voxels. This not only adaptively adjusts the neighborhood range and weights according to the complexity of the occluded area and the data distribution, thus more accurately capturing contextual information and improving repair accuracy, thereby enhancing the algorithm's adaptability to complex occlusion environments and avoiding the information insufficiency or redundancy problems that may arise from fixed neighborhood analysis, but also adjusts the judgment criteria in real time according to the specific conditions of the occluded area, such as height and density, thereby more accurately identifying occluded and missing areas and reducing misjudgments and omissions. This effectively improves the adaptability and accuracy of the occlusion repair algorithm, enhancing the repair effect of this LiDAR point cloud occlusion repair method.

[0015] 2. This invention adaptively sizing voxels the LiDAR point cloud data at each height level in the layered LiDAR point cloud data to obtain voxel data of the plot under study. It can dynamically adjust the size and partitioning strategy of the voxels according to the distribution characteristics and density of the point cloud data at different height levels. This not only allows for a more accurate match to the actual distribution of the point cloud data and better capture of the detailed information at different height levels, but also adapts to the density changes of the point cloud data at different height levels. Furthermore, it provides higher quality voxel data for subsequent occlusion repair algorithms, thereby improving the data representation accuracy and algorithm adaptability of this method and effectively enhancing the repair effect of this LiDAR point cloud occlusion repair method.

[0016] 3. This invention, through preprocessing such as data denoising and uniform sampling on LiDAR point cloud data, can effectively remove noise points and optimize data distribution, thereby improving the reliability and uniformity of LiDAR point cloud data. This lays a solid foundation for subsequent complex processing, including adaptive voxelization, dynamic neighborhood analysis, and occlusion repair, thus ensuring the accuracy and adaptability of the final result and improving the repair effect of this LiDAR point cloud occlusion repair method. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the lidar point cloud occlusion repair method based on neighborhood correlation used in an embodiment of the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0019] Example 1: A lidar point cloud occlusion repair method based on neighborhood correlation, such as... Figure 1 As shown, it includes the following steps: Data preparation involves acquiring LiDAR point cloud data of the site under study using an airborne lidar, and then preprocessing the LiDAR point cloud data to obtain preprocessed LiDAR point cloud data. Point cloud data voxels are generated by vertically layering the preprocessed LiDAR point cloud data according to its height to obtain layered LiDAR point cloud data. Adaptive size voxels are then generated for each height layer of the layered LiDAR point cloud data to obtain voxel data of the plot under study. Voxel attribute annotation maps each point in the layered LiDAR point cloud data to the corresponding voxel in the voxel data of the plot under study, and marks whether the voxel contains points, thus obtaining the voxel attribute data of the plot under study. Voxel occlusion repair involves pre-judging voxels in the voxel attribute data of the plot under study to obtain all potentially occluded voxels. Neighborhood correlation analysis is then performed on all potentially occluded voxels, and a dynamic threshold is used to determine whether the voxel is occluded. Based on the judgment results, all occluded voxels are filled and repaired to obtain repaired voxel attribute data, which is then used for LAI inversion.

[0020] It should be noted that voxelization converts point cloud data into a regular voxel grid, which facilitates efficient storage and processing and provides a foundation for subsequent analysis. Voxel attribute annotation, on the other hand, builds upon voxelization by assigning semantic or feature information, such as category and density, to each voxel to support more complex tasks, such as semantic segmentation and object recognition. The two complement each other: voxelization provides structure, and annotation adds meaning, together promoting the in-depth application of point cloud data.

[0021] Specifically, LAI (Leaf Area Index) inversion is the process of estimating the leaf area index of vegetation canopy using remote sensing technology and other methods. This process typically includes the following key stages: First, acquiring relevant data on the vegetation canopy through remote sensing technology (such as optical or lidar remote sensing); next, preprocessing the acquired data, including data denoising and structuring; then, estimating the leaf area index using specific models or algorithms (such as vegetation index models, physical models, or machine learning models); finally, validating and evaluating the accuracy of the inversion results to ensure the reliability and accuracy of the data. This technique is of great significance for assessing plant growth status, ecosystem productivity, and carbon cycling. Voxel data after shading restoration can provide more complete and accurate vegetation structure information for LAI inversion, helping to improve the accuracy and reliability of the inversion results and better reflect the actual growth status of vegetation and ecological environment characteristics.

[0022] The specific steps for preprocessing LiDAR point cloud data to obtain preprocessed LiDAR point cloud data are as follows: Data format conversion: Parses non-standard format point cloud data in LiDAR point cloud data and converts it into a common set of 3D coordinate points; and handles and skips rows with non-standard format in LiDAR point cloud data. For data denoising, a noise threshold is dynamically set based on the local density of points in LiDAR point cloud data. A distance-based filtering method is used to calculate the distance from each point to its nearest neighbor. If the distance exceeds the dynamically set noise threshold, the point is removed. Uniform sampling divides the point cloud space formed by LiDAR point cloud data into a uniform cubic grid, with each grid retaining only one center point.

[0023] The above steps, through preprocessing such as data denoising and uniform sampling, can effectively remove noise points from LiDAR point cloud data and optimize data distribution, thereby improving the reliability and uniformity of LiDAR point cloud data. This lays a solid foundation for subsequent complex processing, including adaptive voxelization, dynamic neighborhood analysis, and occlusion repair, ensuring the accuracy and adaptability of the final result and improving the repair effect of this LiDAR point cloud occlusion repair method.

[0024] The specific steps for vertically layering the preprocessed LiDAR point cloud data according to height to obtain the layered LiDAR point cloud data are as follows: A three-dimensional Cartesian coordinate system is constructed based on the preprocessed LiDAR point cloud data. The x, y, and z axes of this system correspond to the length, width, and height of the preprocessed LiDAR point cloud data, respectively. The height range of the preprocessed LiDAR point cloud data, including its minimum height, is calculated using this three-dimensional Cartesian coordinate system. and maximum height ; The height interval of the point cloud data layer is preset as needed, and the height range of the preprocessed LiDAR point cloud data is divided into several layers based on the preset height interval. The preprocessed LiDAR point cloud data is divided into different layers according to height, resulting in layered LiDAR point cloud data, where each layer contains points within a corresponding height range.

[0025] The specific steps for adaptively scaling the LiDAR point cloud data of each height layer in the layered LiDAR point cloud data to obtain the voxel data of the plot under study are as follows: For each height layer in the layered LiDAR point cloud data, calculate the distance from each point in that height layer to its nearest neighbor, obtain the nearest neighbor distance of each point in that height layer, and take the median of the nearest neighbor distances of all points in that height layer as the voxel size of that height layer. The length, width, and height of the voxel are all set as the voxel dimensions of the height layer, and the horizontal range of the height layer is divided into... The grid is used to obtain voxel data of the plot under study, where , , and among them and These represent the maximum and minimum values ​​of a point within that height layer along the x-axis, respectively. and These represent the maximum and minimum values ​​of a point within that height layer along the y-axis, respectively. This indicates the voxel size of the height layer.

[0026] It should be noted that when determining the voxel size, taking the median of the nearest neighbor distances of all points within that height layer as the voxel size of that height layer can effectively balance the preservation of data details and noise filtering, because the median is not sensitive to outliers and can represent the density of most points. This ensures that the voxel partitioning adapts to the local data distribution and improves the accuracy and adaptability of the voxelization model.

[0027] The above steps adaptively size voxels the LiDAR point cloud data at each height level in the layered LiDAR point cloud data to obtain voxel data of the plot under study. This allows for dynamic adjustment of voxel size and partitioning strategies based on the distribution characteristics and density of point cloud data at different height levels. This not only more accurately matches the actual distribution of the point cloud data and better captures detailed information at different height levels, but also adapts to density variations at different height levels. Furthermore, it provides higher-quality voxel data for subsequent occlusion repair algorithms, thereby improving the data representation accuracy and algorithm adaptability of this method and effectively enhancing the repair effect of this LiDAR point cloud occlusion repair method.

[0028] The specific steps for mapping each point in the layered LiDAR point cloud data to the corresponding voxel in the voxel data of the plot under study, and marking whether the voxel contains a point, to obtain the voxel attribute data of the plot under study are as follows: Determine the voxel partitioning rules, including the length, width, and height of the voxels, find the minimum coordinate value of the layered LiDAR point cloud data, and use it as the starting point for voxel partitioning; For each point in the layered LiDAR point cloud data, the voxel index to which the point belongs is calculated based on the point's coordinates in the three-dimensional rectangular coordinate system and the voxel partitioning rules. The attributes of each voxel are recorded using a hash table or array. If a voxel contains at least one point, it is marked as 1; otherwise, it is marked as 0, thus obtaining the voxel attribute data of the plot under study.

[0029] The specific steps for performing occlusion pre-judgment on the voxel attribute data of the plot under study to obtain all potentially occluded voxels are as follows: The data is processed layer by layer from the top layer of the voxel attribute data of the plot under study to the bottom layer. During this process, for each empty voxel marked as 0 in the voxel attribute data of the plot under study, it is first checked whether the upper voxel of the empty voxel is marked as 0. If the upper voxel is marked as 0, the current empty voxel is skipped. If the upper voxel is marked as 1, it is determined that the current empty voxel may be occluded, and the lower voxel of the empty voxel is checked. Check if the lower voxel of the empty voxel is marked as 0. If the lower voxel is marked as 1, skip the current empty voxel. If the lower voxel is marked as 0, determine that the current empty voxel may be occluded and continue to check other empty voxels to obtain all voxels that may be occluded.

[0030] The specific steps for performing neighborhood correlation analysis on all potentially occluded voxels, determining whether a voxel is occluded based on a dynamic threshold, filling and repairing all occluded voxels based on the determination results, obtaining repaired voxel attribute data, and using it for LAI inversion are as follows: For each voxel that may be occluded, the neighborhood range is dynamically determined based on the point density and fill condition of the current voxel's height layer. When the fill density of this height layer is less than 70%, the neighborhood range is 26 neighborhoods; when the fill density of this height layer is greater than or equal to 70%, the neighborhood range is 8 neighborhoods. Based on the dynamically determined neighborhood range, a neighborhood system is determined for each potentially occluded voxel. The sum of states of the neighborhood system is calculated using a weighted summation method. When the sum of states of the neighborhood system is greater than or equal to the dynamic threshold corresponding to the current voxel, the voxel is determined to be occluded, and its label is filled with 1. When the sum of states of the neighborhood system is less than the dynamic threshold corresponding to the current voxel, the voxel's label is kept at 0. This yields the repaired voxel attribute data. The specific formula for calculating the sum of states of the neighborhood system is as follows: , in Represents the sum of states of the neighborhood system. Represents the first in the neighborhood system The state tag value of each neighboring voxel. Represents the first in the neighborhood system The state weights of each neighboring voxel are assigned hierarchically based on the distance between the current voxel and its neighboring voxels. Indicates the size of the neighborhood.

[0031] It should be noted that the main difference between using 8-neighborhood and 26-neighborhood in neighborhood correlation analysis lies in the scope of analysis and the ability to capture details. 8-neighborhood only considers the 8 adjacent voxels, namely, up and down, left and right, and front and back, which is suitable for analyzing local features and is computationally simple, but may ignore information in the diagonal direction. 26-neighborhood, on the other hand, covers all 26 surrounding voxels, including the diagonal direction, and can more comprehensively reflect the spatial correlation of voxels and capture more complex structural features, but the computational cost is larger. The choice of which neighborhood method to use depends on the specific application requirements and considerations for computational efficiency.

[0032] The specific value of the dynamic threshold needs to be based on the degree of occlusion of the current voxel's height layer, and the specific value of the dynamic threshold also decreases with the height of the current voxel's height layer.

[0033] The above steps involve performing neighborhood correlation analysis on all potentially occluded voxels and determining whether a voxel is occluded based on a dynamic threshold. Based on the determination results, all occluded voxels are filled and repaired. This not only adaptively adjusts the neighborhood range and weights according to the complexity of the occluded area and the data distribution, thus more accurately capturing contextual information and improving repair accuracy, thereby enhancing the algorithm's adaptability to complex occlusion environments and avoiding the information insufficiency or redundancy problems that may arise from fixed neighborhood analysis, but also adjusts the judgment criteria in real time according to the specific conditions of the occluded area, such as height and density, thereby more accurately identifying occluded and missing areas and reducing misjudgments and omissions. This effectively improves the adaptability and accuracy of the occlusion repair algorithm, enhancing the repair effect of this LiDAR point cloud occlusion repair method.

[0034] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A lidar point cloud occlusion repair method based on neighborhood correlation, characterized in that, Includes the following steps: Data preparation involves acquiring LiDAR point cloud data of the site under study using an airborne lidar, and then preprocessing the LiDAR point cloud data to obtain preprocessed LiDAR point cloud data. Point cloud data voxels are generated by vertically layering the preprocessed LiDAR point cloud data according to its height to obtain layered LiDAR point cloud data. Adaptive size voxels are then generated for each height layer of the layered LiDAR point cloud data to obtain voxel data of the plot under study. Voxel attribute annotation maps each point in the layered LiDAR point cloud data to the corresponding voxel in the voxel data of the plot under study, and marks whether the voxel contains points, thus obtaining the voxel attribute data of the plot under study. Voxel occlusion repair involves pre-judging voxels in the voxel attribute data of the plot under study to obtain all potentially occluded voxels. Neighborhood correlation analysis is then performed on all potentially occluded voxels, and a dynamic threshold is used to determine whether the voxel is occluded. Based on the judgment results, all occluded voxels are filled and repaired to obtain repaired voxel attribute data, which is then used for LAI inversion.

2. The lidar point cloud occlusion repair method based on neighborhood correlation according to claim 1, characterized in that, The specific steps for preprocessing LiDAR point cloud data to obtain preprocessed LiDAR point cloud data are as follows: Data format conversion: Parses non-standard format point cloud data in LiDAR point cloud data and converts it into a common set of 3D coordinate points; and handles and skips rows with non-standard format in LiDAR point cloud data. For data denoising, a noise threshold is dynamically set based on the local density of points in LiDAR point cloud data. A distance-based filtering method is used to calculate the distance from each point to its nearest neighbor. If the distance exceeds the dynamically set noise threshold, the point is removed. Uniform sampling divides the point cloud space formed by LiDAR point cloud data into a uniform cubic grid, with each grid retaining only one center point.

3. The lidar point cloud occlusion repair method based on neighborhood correlation according to claim 2, characterized in that, The specific steps for vertically layering the preprocessed LiDAR point cloud data according to height to obtain the layered LiDAR point cloud data are as follows: A three-dimensional Cartesian coordinate system is constructed based on the preprocessed LiDAR point cloud data. The x, y, and z axes of this system correspond to the length, width, and height of the preprocessed LiDAR point cloud data, respectively. The height range of the preprocessed LiDAR point cloud data, including its minimum height, is calculated using this three-dimensional Cartesian coordinate system. and maximum height ; The height interval of the point cloud data layer is preset as needed, and the height range of the preprocessed LiDAR point cloud data is divided into several layers based on the preset height interval. The preprocessed LiDAR point cloud data is divided into different layers according to height, resulting in layered LiDAR point cloud data, where each layer contains points within a corresponding height range.

4. The lidar point cloud occlusion repair method based on neighborhood correlation according to claim 3, characterized in that, The specific steps for adaptively scaling the LiDAR point cloud data of each height layer in the layered LiDAR point cloud data to obtain the voxel data of the plot under study are as follows: For each height layer in the layered LiDAR point cloud data, calculate the distance from each point in that height layer to its nearest neighbor, obtain the nearest neighbor distance of each point in that height layer, and take the median of the nearest neighbor distances of all points in that height layer as the voxel size of that height layer. The length, width, and height of the voxel are all set as the voxel dimensions of the height layer, and the horizontal range of the height layer is divided into... The grid is used to obtain voxel data of the plot under study, where , , and among them and These represent the maximum and minimum values ​​of a point within that height layer along the x-axis, respectively. and These represent the maximum and minimum values ​​of a point within that height layer along the y-axis, respectively. This indicates the voxel size of the height layer.

5. The lidar point cloud occlusion repair method based on neighborhood correlation according to claim 4, characterized in that, The specific steps for mapping each point in the layered LiDAR point cloud data to the corresponding voxel in the voxel data of the plot under study, and marking whether the voxel contains a point, to obtain the voxel attribute data of the plot under study are as follows: Determine the voxel partitioning rules, including the length, width, and height of the voxels, find the minimum coordinate value of the layered LiDAR point cloud data, and use it as the starting point for voxel partitioning; For each point in the layered LiDAR point cloud data, the voxel index to which the point belongs is calculated based on the point's coordinates in the three-dimensional rectangular coordinate system and the voxel partitioning rules. The attributes of each voxel are recorded using a hash table or array. If a voxel contains at least one point, it is marked as 1; otherwise, it is marked as 0, thus obtaining the voxel attribute data of the plot under study.

6. The lidar point cloud occlusion repair method based on neighborhood correlation according to claim 5, characterized in that, The specific steps for performing occlusion pre-judgment on the voxel attribute data of the plot under study to obtain all potentially occluded voxels are as follows: The data is processed layer by layer from the top layer of the voxel attribute data of the plot under study to the bottom layer. During this process, for each empty voxel marked as 0 in the voxel attribute data of the plot under study, it is first checked whether the upper voxel of the empty voxel is marked as 0. If the upper voxel is marked as 0, the current empty voxel is skipped. If the upper voxel is marked as 1, it is determined that the current empty voxel may be occluded, and the lower voxel of the empty voxel is checked. Check if the lower voxel of the empty voxel is marked as 0. If the lower voxel is marked as 1, skip the current empty voxel. If the lower voxel is marked as 0, determine that the current empty voxel may be occluded and continue to check other empty voxels to obtain all voxels that may be occluded.

7. The lidar point cloud occlusion repair method based on neighborhood correlation according to claim 6, characterized in that, The specific steps for performing neighborhood correlation analysis on all potentially occluded voxels, determining whether a voxel is occluded based on a dynamic threshold, filling and repairing all occluded voxels based on the determination results, obtaining repaired voxel attribute data, and using it for LAI inversion are as follows: For each voxel that may be occluded, the neighborhood range is dynamically determined based on the point density and fill condition of the current voxel's height layer. When the fill density of this height layer is less than 70%, the neighborhood range is 26 neighborhoods; when the fill density of this height layer is greater than or equal to 70%, the neighborhood range is 8 neighborhoods. Based on the dynamically determined neighborhood range, a neighborhood system is determined for each potentially occluded voxel. The sum of states of the neighborhood system is calculated using a weighted summation method. When the sum of states of the neighborhood system is greater than or equal to the dynamic threshold corresponding to the current voxel, the voxel is determined to be occluded, and its label is filled with 1. When the sum of states of the neighborhood system is less than the dynamic threshold corresponding to the current voxel, the voxel's label is kept at 0. This yields the repaired voxel attribute data. The specific formula for calculating the sum of states of the neighborhood system is as follows: ， in Represents the sum of states of the neighborhood system. Represents the first in the neighborhood system The state tag value of each neighboring voxel. Represents the first in the neighborhood system The state weights of each neighboring voxel are assigned hierarchically based on the distance between the current voxel and its neighboring voxels. Indicates the size of the neighborhood.

8. The lidar point cloud occlusion repair method based on neighborhood correlation according to claim 7, characterized in that, The specific value of the dynamic threshold needs to be based on the degree of occlusion of the current voxel's height layer, and the specific value of the dynamic threshold also decreases with the height of the current voxel's height layer.

Images