Forest open-ground lidar point cloud registration method

Abstract

This invention provides a method for registering forest point clouds using airborne and ground-based lidar. The method includes: extracting tree points from airborne and ground-based lidar point cloud data of a forest area, and projecting these tree points onto a two-dimensional plane to obtain corresponding airborne and ground-based canopy binary maps; enhancing both the airborne and ground-based canopy binary maps, and calculating corresponding airborne and ground-based canopy contour points based on the pixel connectivity between each pixel and its corresponding neighbor in the enhanced canopy binary maps; matching the airborne and ground-based canopy contour points, and spatially registering the airborne and ground-based lidar point cloud data based on the matching results to obtain the forest point cloud registration result. This method achieves cross-dimensional fusion calculation of horizontal and vertical translations, shortening the registration time and improving the accuracy of constructing a three-dimensional forest structure.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional reconstruction technology, and in particular to a method for registering point clouds of forest airspace lidar. Background Technology

[0002] Accurately describing the three-dimensional structure of forests is the scientific basis for conducting forest ecology research, exploring the self-organization and feedback mechanisms of forest ecosystems, and is also an urgent need to propose scientific and effective forest management strategies and balance the contradiction between social development and sustainable use of ecosystems.

[0003] In related technologies, a high-precision three-dimensional forest structure is constructed using ground-based lidar, while airborne lidar rapidly scans the forest scene from the air, providing information on the horizontal distribution and vertical structure of the forest canopy surface. External equipment such as GPS and artificial targets are used to assist in stitching together the data from these two platforms, thereby constructing the three-dimensional forest structure information. However, in dense forest areas with severe canopy shading, the stitching operation using auxiliary equipment is difficult, and existing automated stitching methods struggle to accurately associate tree features, resulting in low stitching accuracy or even failure, thus leading to inaccurate three-dimensional forest structures. Summary of the Invention

[0004] This invention provides a forest air-ground lidar point cloud registration method to solve the defects of existing technologies in constructing three-dimensional structures of dense forest areas with severe canopy shading, where the accuracy of point cloud stitching using auxiliary equipment or existing automated stitching methods is low, resulting in inaccurate three-dimensional forest structures.

[0005] This invention provides a method for forest air-to-ground lidar point cloud registration, comprising: Tree points are extracted from airborne lidar point cloud data and ground-based lidar point cloud data in the forest area, and the tree points are projected onto a two-dimensional plane to obtain the corresponding airborne canopy binary map and ground-based canopy binary map. The airborne canopy binary image and the ground canopy binary image are enhanced respectively, and the corresponding airborne canopy contour points and ground canopy contour points are calculated according to the pixel connectivity between each pixel point and its corresponding neighboring points in the enhanced canopy binary image. The airborne canopy contour points and the ground-based canopy contour points are matched, and the airborne lidar point cloud data and the ground-based lidar point cloud data are spatially registered based on the matching results to obtain the forest point cloud registration results.

[0006] According to the present invention, a forest airborne lidar point cloud registration method is provided, wherein the calculation of corresponding airborne canopy contour points and ground-based canopy contour points is performed based on the pixel connectivity between each pixel point and its corresponding neighboring points in the enhanced canopy binary image, including: Obtain the n-neighborhood points of each target pixel in the enhanced canopy binary image, and calculate the average value of the product of the target pixel and the pixel values ​​of the n neighboring points to obtain the pixel connectivity index; n≥2, and n is an integer; If the pixel connectivity index is greater than 0 and less than 1, the target pixel is determined to be the corresponding airborne canopy contour point or ground-based canopy contour point.

[0007] According to a forest airborne lidar point cloud registration method provided by the present invention, before matching the airborne canopy contour points and the ground-based canopy contour points, the method further includes: Calculate the covariance matrix of each canopy contour point within its preset neighborhood radius; Calculate the response value of each canopy contour point based on the determinant and trace of the covariance matrix; The airborne canopy binary map and the ground-based canopy binary map are each divided into a preset number of grids, and the canopy contour point with the largest corresponding response value in each grid is determined as the key corner point.

[0008] According to the present invention, a forest airborne lidar point cloud registration method is provided, wherein matching the airborne canopy contour points and the ground-based canopy contour points includes: Construct a set of foundation point pairs based on the key corner points corresponding to the binary map of the foundation canopy; Search for airborne point pairs that are equidistant from each corresponding point pair in the ground point pair set among the key corner points of the airborne canopy binary map; construct equidistant corresponding point pairs based on each corresponding point pair in the ground point pair set and the airborne point pairs. Based on the planar Euclidean transformation and the equidistant corresponding point pairs, the rotation angle and pixel translation vector from the ground canopy binary image to the airborne canopy binary image are calculated respectively.

[0009] According to the present invention, a forest airborne lidar point cloud registration method, after calculating the rotation angle and pixel translation vector from the ground-based canopy binary image to the airborne canopy binary image based on the plane Euclidean transformation and the equidistant corresponding point pairs, the method further includes: The binary image of the foundation canopy is transformed according to the rotation angle and the pixel translation vector, and the transformed binary image of the foundation canopy is divided into multiple grids. Calculate the mean value of the difference between the center point of each grid and the pixel value of the corresponding point of the center point in the airborne canopy binary image; The overlap of each pair of equidistant corresponding points is determined based on the mean value, and the pair of equidistant corresponding points with the highest overlap is selected as the optimal corresponding point pair. The rotation angle and pixel translation vector corresponding to the optimal corresponding point pair are used as the matching result.

[0010] According to the present invention, a forest airborne lidar point cloud registration method is provided, wherein the spatial registration of the airborne lidar point cloud data and the ground-based lidar point cloud data based on the matching result includes: Based on the rotation angle in the matching result, calculate the horizontal rotation matrix between the airborne lidar point cloud data and the ground-based lidar point cloud data; Based on the coordinates of the three-dimensional mapping points in the airborne lidar point cloud data and the ground-based lidar point cloud data respectively, calculate the horizontal translation vector. Search for ground mapping points with the same horizontal coordinates as the optimal corresponding point pair from the ground-based lidar point cloud data, and calculate the vertical translation vector based on the coordinate difference of the ground mapping points in the vertical direction. The airborne lidar point cloud data and the ground-based lidar point cloud data are registered according to the horizontal rotation matrix, the horizontal translation vector, and the vertical translation vector. The registered airborne lidar point cloud data and the ground-based lidar point cloud data are stitched together using the iterative nearest neighbor algorithm to obtain the stitched forest point cloud data.

[0011] The forest airborne and ground-based lidar point cloud registration method provided by this invention extracts tree points from airborne and ground-based lidar point cloud data of forest areas and maps them to a two-dimensional plane to obtain corresponding airborne and ground-based canopy binary maps. This transforms three-dimensional point cloud registration into two-dimensional image matching, reducing computational complexity. By enhancing the airborne and ground-based canopy binary maps respectively and utilizing the pixel connectivity between pixels and neighboring points to calculate the corresponding canopy contour points, the method overcomes the defect that individual tree features are easily occluded and fail to be extracted in dense forest areas, thus improving the robustness of canopy overall geometric feature extraction. Spatial registration of airborne and ground-based lidar point cloud data is achieved through the matching results of airborne and ground-based canopy contour points, realizing cross-dimensional fusion calculation of horizontal and vertical translation, shortening the registration time, and improving the accuracy of constructing the three-dimensional structure of the forest. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0013] Figure 1 This is one of the flowcharts illustrating the forest air-ground lidar point cloud registration method provided by the present invention.

[0014] Figure 2 This is a schematic diagram of the canopy binary image and preprocessing results provided by the present invention.

[0015] Figure 3 This is a schematic diagram of the canopy contour line and key corner point extraction results provided by the present invention.

[0016] Figure 4 This is a schematic diagram of the key corner point matching and overlap analysis mechanism provided by the present invention.

[0017] Figure 5 This is the second flowchart illustrating the forest air-ground lidar point cloud registration method provided by the present invention.

[0018] Figure 6 This is a schematic diagram of the coarse registration and fine stitching results of the three-dimensional point cloud provided by the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] The following is combined with Figures 1-3 The present invention describes a forest air-ground lidar point cloud registration method.

[0021] Figure 1 This is one of the flowcharts illustrating the forest air-to-ground lidar point cloud registration method provided by the present invention, such as... Figure 1 As shown, the method includes the following: Step 110: Extract tree points from the airborne lidar point cloud data and the ground-based lidar point cloud data of the forest area, and project the tree points onto a two-dimensional plane to obtain the corresponding airborne canopy binary map and the ground-based canopy binary map.

[0022] In this step, airborne lidar point cloud data can be acquired by airborne radar scanning the forest area from above. For example, airborne radar includes, but is not limited to, a hexacopter drone equipped with a RIEGL miniVUX-1UAV scanner, which scans at a certain flight altitude (e.g., 50 meters).

[0023] In this step, ground-based lidar point cloud data can be obtained by scanning the forest area from the bottom up using a ground-based lidar mounted under the forest canopy. For example, a 360-degree scan can be performed using a RIEGL VZ-1000 ground-based lidar with the tripod leveled.

[0024] In this step, tree points can be obtained by filtering the above point cloud data to remove ground undulation points. For example, ground points can be separated using a cloth-simulated ground filtering algorithm, and point cloud points above a certain height (such as 6 meters) can be extracted.

[0025] In this step, the three-dimensional coordinates can be vertically mapped to the XY horizontal plane at a preset resolution, and the tree points can be projected onto a two-dimensional plane.

[0026] For example, by horizontally rasterizing the tree points above 6 meters with a resolution of 0.1 meters, the corresponding canopy binary map can be obtained.

[0027] In this step, the airborne canopy binary map and the ground-based canopy binary map can be generated by assigning specific binary pixels to the raster. For example, if there are canopy projection points in the grid, the pixel value is set to 1, and if there are no canopy points, it is set to 0.

[0028] In this embodiment, by obtaining the three-dimensional structural information of the forest from both aerial and ground-based perspectives, and then performing height cropping and vertical projection, a corresponding canopy binary image is obtained. This transforms complex three-dimensional spatial calculations into low-computing-power two-dimensional image processing.

[0029] For example, in a survey scenario of a dense birch broad-leaved forest plot (standing density of about 150 trees / hectare) measuring 40m × 30m in Forest Farm A, the operators obtained data from drones and ground-based lidar at the same time. The system (the main body of the method execution) automatically ran a cloth simulation algorithm to remove ground points and extract the three-dimensional point cloud of the tree canopy above 6m. Subsequently, the system discarded the Z-axis elevation information, mapped the three-dimensional point cloud onto a two-dimensional grid at a resolution of 0.1m, and marked points with leaves as 1 (white) and gaps as 0 (black), thus quickly obtaining two two-dimensional canopy binary maps.

[0030] Step 120: Enhance the airborne canopy binary image and the ground canopy binary image respectively, and calculate the corresponding airborne canopy contour points and ground canopy contour points according to the pixel connectivity between each pixel point and its corresponding neighboring points in the enhanced canopy binary image.

[0031] In this step, the enhancement process can be achieved by structural repair of the original binary image through morphological and filtering algorithms.

[0032] For example, dilation and erosion closing operations with a kernel scale of 5 pixels are used to fill discrete holes in the image, and median filtering is used to eliminate sharp gear-like edges.

[0033] In this step, the pixel connectivity feature can be obtained by calculating the mathematical state relationship between the central target pixel and its surrounding neighboring pixels. For example, the four neighboring points above, below, left, and right of the target pixel can be obtained, and the average value of the product of the target pixel and the pixel values ​​of these four points can be calculated to obtain the pixel connectivity index.

[0034] In this step, pixels whose connectivity is within a specific threshold range in the airborne canopy binary image and the ground canopy binary image can be screened to determine the corresponding airborne canopy contour points and ground canopy contour points. For example, when the calculated connectivity index value is greater than 0 and less than 1, the point is determined to be an edge contour point.

[0035] In this embodiment, morphological restoration is performed on the dimensionality-reduced binary image, and the canopy edge is extracted based on the connectivity between local pixels to eliminate the interference of internal holes caused by the sparse point cloud.

[0036] Figure 2 This is a schematic diagram of the canopy binary image and preprocessing results provided by the present invention. Figure 2 In the illustrated embodiment, noise appears inside the generated binary image because the lidar beam cannot penetrate all the gaps between leaves. For the canopy binary images obtained from UAV lidar point cloud data and ground-based lidar point cloud data, the black holes inside the canopy are first filled using dilation and erosion algorithms, and then the canopy edges are smoothed using median filtering, completing the preprocessing of the canopy binary image. Subsequently, the system checks each pixel in the image: if a point itself is a canopy (value 1), and its four points above, below, left, and right are also canopies (values ​​all 1), its connectivity average is 1, indicating that the pixel is inside the canopy; if there are 0s (background) around it, its connectivity average will be between 0 and 1. The system selects all pixels between 0 and 1, thus obtaining the overall canopy outline from the UAV and ground-based perspectives.

[0037] Step 130: Match the airborne canopy contour points and the ground canopy contour points, and perform spatial registration of the airborne lidar point cloud data and the ground lidar point cloud data based on the matching results to obtain the forest point cloud registration results.

[0038] In this step, the matching can be achieved by finding the spatial correspondence between the geometric features of the airborne and ground-based systems in a two-dimensional plane.

[0039] For example, by constructing equidistant key corner point pairs for initial screening and combining them with gridded overlap analysis for verification, the optimal rotation angle and pixel translation vector for two-dimensional matching can be calculated.

[0040] In this step, the spatial registration described above can be achieved by decoupling the matching parameters calculated in the two-dimensional plane and back-mapping them to the three-dimensional coordinate system.

[0041] For example, a Z-axis rotation matrix can be generated using a two-dimensional rotation angle, and the horizontal X and Y axis translations can be inferred using two-dimensional translation points. The vertical Z-axis translation can be calculated by extracting the elevation difference of ground points at the same coordinate position.

[0042] In this embodiment, the corresponding forest point cloud registration result is obtained after position transformation by integrating the three-dimensional translation matrix. For example, the obtained coarse registration result is used to perform fine stitching using the Iterative Closest Point (ICP) algorithm, and finally the registered forest point cloud data is obtained.

[0043] The forest airborne and ground-based lidar point cloud registration method provided in this invention extracts tree points from airborne and ground-based lidar point cloud data of forest areas and maps them to a two-dimensional plane to obtain corresponding airborne and ground-based canopy binary maps. This transforms three-dimensional point cloud registration into two-dimensional image matching, reducing computational complexity. By enhancing the airborne and ground-based canopy binary maps and utilizing pixel connectivity between pixels and neighboring pixels to calculate the corresponding canopy contour points, the method overcomes the defect that individual tree features are easily occluded and fail to be extracted in dense forest areas, improving the robustness of canopy overall geometric feature extraction. Spatial registration of airborne and ground-based lidar point cloud data is achieved through the matching results of airborne and ground-based canopy contour points, realizing cross-dimensional fusion calculation of horizontal and vertical translation, shortening registration time, and improving the accuracy of constructing the three-dimensional forest structure.

[0044] In some embodiments, based on the pixel connectivity between each pixel and its corresponding neighboring pixels in the enhanced canopy binary image, the corresponding airborne canopy contour points and ground-based canopy contour points are calculated respectively, including: obtaining the n neighboring points of each target pixel in the enhanced canopy binary image, and calculating the average value of the product of the target pixel and the pixel values ​​of the n neighboring points to obtain the pixel connectivity index; n≥2, and n is an integer; when the pixel connectivity index is greater than 0 and less than 1, the target pixel is determined as the corresponding airborne canopy contour point or ground-based canopy contour point.

[0045] In this embodiment, n can be 4 or 8, such as obtaining the top, bottom, left and right 4-neighborhoods of the target pixel or the surrounding 8-neighborhoods.

[0046] In this embodiment, since pixel values ​​are either 0 or 1, a connectivity index of 1 indicates that the point and its surroundings are all tree canopy (internal point); a value of 0 indicates that the surroundings are all background (noise or external point); only when the value is between 0 and 1 does it mean that the point is located exactly on the boundary line between the tree canopy and the open ground. In this embodiment, the canopy contour line is determined based on the pixel neighborhood connectivity, the geometric covariance matrix of the contour line is calculated, and response values ​​are constructed to quantitatively describe the key corner points of the contour.

[0047] Specifically, based on canopy pixels Its n (e.g., 4) neighboring points connectivity Determine the outline: ; ; ; in, Represents a pixel value (0 or 1). If it equals 0, then... As noise; if An equality of 1 indicates that all four neighboring points are tree points. Points within the canopy; if , indicating that at least one of the four neighboring points is a background point, then These are the canopy contour points.

[0048] The forest air-ground lidar point cloud registration method provided in this invention constructs a connectivity mathematical model based on the mean of local pixel products, accurately defines the inner and outer boundaries of tree canopies in a discrete binary grid, and replaces complex traditional edge detection operators with lower computing power cost, thereby extracting continuous geometric contour lines. This method provides algorithmic support for the rapid processing of dimensionality reduction features of large-scale forest point clouds.

[0049] In some embodiments, before matching the airborne canopy contour points and the ground-based canopy contour points, the method further includes: calculating the covariance matrix of each canopy contour point within its preset neighborhood radius; calculating the response value of each canopy contour point based on the determinant and trace of the covariance matrix; dividing the airborne canopy binary map and the ground-based canopy binary map into a preset number of grids, and determining the canopy contour point with the largest corresponding response value in each grid as the key corner point.

[0050] In this embodiment, the covariance matrix can be calculated by statistically analyzing the spatial distribution variance and covariance of the contour point coordinates within the local window.

[0051] For example, taking a certain contour point as the center, select a circular or square neighborhood with a radius of 5 pixels, and calculate the 2×2 covariance matrix of the X and Y coordinates of all contour points within this range to reflect the degree of geometric change of the local edge.

[0052] Specifically, calculate the covariance matrix of each contour point. : ; in, Represents the variance between variables. This can be expressed by the following formula: ; in, Indicates the neighborhood radius (e.g., set to 5 pixels). and This represents the mean of the sample parameters. For the first i The coordinates of the neighboring pixels.

[0053] In this embodiment, the saliency of corner features can be quantified by the algebraic properties of the matrix. For example, the response value can be obtained by subtracting the square of the trace of the matrix (representing the sum of eigenvalues) under preset weights from the determinant of the covariance matrix (representing the product of eigenvalues). K , K The larger the value, the more obvious the contour curvature or intersection at that point.

[0054] Specifically, the response value is determined by the following formula: ; in, The determinant of the covariance matrix. Let T be the trace of the covariance matrix, and let T be the trace of the covariance matrix. Both are expressed by the following formulas: ; ; .

[0055] In this embodiment, key corner points can be determined by spatial partitioning combined with local nonmaximum suppression. For example, the entire image can be divided into several uniform grids (e.g., 64 grids for airborne images and 36 grids for ground-based images), and the unique highest response value point in each independent grid can be selected as the final feature point for matching.

[0056] Figure 3 This is a schematic diagram of the canopy contour line and key corner point extraction results provided by the present invention. Figure 3 In the embodiment shown, for the canopy binary map corresponding to either UAV lidar data or ground-based lidar data, the contour lines are extracted by utilizing the pixel connectivity between each pixel point and its corresponding neighboring points in the canopy binary map, and the canopy contour point with the largest response value in the grid corresponding to each contour line is taken as the key corner point.

[0057] The forest air-ground lidar point cloud registration method provided in this invention constructs a quantitative corner response model by calculating the covariance matrix and introduces a gridded local extremum screening mechanism. This method can extract features with significant geometric representativeness and uniform spatial distribution from continuous and irregular tree canopy outlines, thereby preventing feature point clustering and filtering out false corner points. This improves the anti-interference ability and computational efficiency of subsequent point cloud matching algorithms in complex forest occlusion environments.

[0058] In some embodiments, matching airborne canopy contour points and ground-based canopy contour points includes: constructing a ground-based point pair set based on the key corner points corresponding to the ground-based canopy binary image; searching for airborne point pairs in the key corner points of the airborne canopy binary image that are equidistant from each corresponding point pair in the ground-based point pair set; constructing equidistant corresponding point pairs based on each corresponding point pair in the ground-based point pair set and the airborne point pairs; and calculating the rotation angle and pixel translation vector from the ground-based canopy binary image to the airborne canopy binary image based on the plane Euclidean transformation and the equidistant corresponding point pairs.

[0059] In this embodiment, it can be achieved by selecting two points from the feature points of the foundation image and connecting them. For example, in order to ensure the stability of subsequent calculations, two points with a spacing greater than half the width of the binary image of the foundation crown are selected from the key corner points of the foundation to form a pair, thus obtaining a set of foundation point pairs.

[0060] In this embodiment, by calculating the Euclidean distance between the ground point pairs, and then finding two points with the same distance value in the corner point combination of the airborne image, these two sets of points are bound one-to-one to obtain the corresponding equidistant point pairs. This method utilizes the geometric property of invariant distance in rigid body transformation to quickly eliminate a large number of incorrect matching combinations and reduce the search space.

[0061] In this embodiment, the pixel coordinates of the equidistant corresponding point pairs with established mapping relationship are substituted into a planar Euclidean transformation matrix containing rotation angle and translation vector to solve the equation, which is used to obtain the relative spatial pose relationship of the two dimensionality-reduced images on the two-dimensional plane, and to provide horizontal transformation parameters for the final three-dimensional spatial registration.

[0062] Specifically, for the key corner points corresponding to the UAV and the ground base, point pair sets with a spacing greater than half the width of the ground base canopy binary map are constructed respectively. Equal-distance airborne and ground base point pairs are mapped one-to-one. Based on planar Euclidean transformation, the rotation and translation from the ground base canopy binary map to the airborne canopy binary map are calculated, specifically using the following formula: ; in, These are key corner points in the binary map of the foundation canopy. For point At the corresponding corner points in the UAV canopy binary map, This represents the angle between two images rotated counterclockwise. This is the pixel translation vector.

[0063] In this embodiment, the unknown term can be calculated based on a set of corresponding point pairs. and .

[0064] The forest airborne lidar point cloud registration method provided in this invention constructs a ground point pair set by using key corner points corresponding to the ground canopy binary image, and searches for airborne point pairs that are equidistant from each corresponding point pair in the ground point pair set to construct equidistant corresponding point pairs. Finally, based on the planar Euclidean transformation and the equidistant corresponding point pairs, the rotation angle and pixel translation vector from the ground canopy binary image to the airborne canopy binary image are calculated respectively. This method introduces rigid body equidistant constraints to replace the global brute-force exhaustive search, realizing the rapid estimation and solution of two-dimensional matching parameters and improving the computational efficiency of image registration.

[0065] In some embodiments, after calculating the rotation angle and pixel translation vector from the ground-based canopy binary image to the airborne canopy binary image based on the planar Euclidean transformation and equidistant corresponding point pairs, the method further includes: transforming the ground-based canopy binary image based on the rotation angle and pixel translation vector, and dividing the transformed ground-based canopy binary image into multiple grids; calculating the mean value of the difference between the center point of each grid and the corresponding pixel value of the center point in the airborne canopy binary image; determining the overlap degree of each set of equidistant corresponding point pairs based on the mean value, selecting the set of equidistant corresponding point pairs with the highest overlap degree as the optimal corresponding point pair, and using the rotation angle and pixel translation vector corresponding to the optimal corresponding point pair as the matching result.

[0066] In this embodiment, a set of parameters can be used to rotate and translate the ground image into a digital matrix, and then the image can be uniformly divided into a preset number of uniform grids (such as 15×15) to provide representative sampling points covering the entire area and prevent evaluation deviation caused by local tree canopy occlusion.

[0067] In this embodiment, the pixel values ​​(either 0 or 1) of the center point of the ground grid and the same point in the airborne image are extracted respectively, the absolute difference between the two is calculated and divided by the total number of grids to obtain the mean value.

[0068] In this embodiment, the overlap index is constructed by subtracting the average difference from 1 (the closer the index is to 1, the higher the degree of matching). The pair with the highest overlap value is selected from all candidate equidistant point pairs to eliminate the erroneous matching parameters caused by pseudo equidistant points in the forest, so as to ensure the uniqueness and accuracy of the two-dimensional registration result.

[0069] Specifically, based on the rotation and translation obtained for each pair of corresponding points, the binary image of the ground canopy is transformed and divided into a 15×15 grid. The overlap between the two binary images of the canopy is calculated by the difference between the pixel value of the center point of the grid and the corresponding point in the UAV canopy binary image described in step S2. The specific calculation method is as follows: ; in, The pixel value of the grid center point in the binary map of the foundation canopy. for The pixel value of the corresponding point in the UAV canopy binary image. Indicates the grid number; if A value approaching 1 indicates a higher accuracy of the corresponding point pair; when... A smaller value indicates low overlap, which may lead to incorrect matching.

[0070] In this embodiment, the corresponding point pair with the highest overlap can be taken as the optimal corresponding point pair, and the calculated rotation and translation can be taken as the final matching result.

[0071] The forest airborne lidar point cloud registration method provided in this invention transforms the ground canopy binary image and divides it into grids using rotation angles and pixel translation vectors. It calculates the mean difference between the center point of each grid and the corresponding pixel value in the airborne canopy binary image. Then, based on the mean, it determines the overlap of each pair of equidistant corresponding points and selects the pair with the highest overlap as the optimal pair. Finally, it uses the rotation angle and pixel translation vector corresponding to the optimal pair as the matching result. This global gridded overlap verification mechanism can solve mismatches during local feature extraction and ensures the high robustness of the dimensionality reduction matching result.

[0072] In some embodiments, spatial registration is performed on airborne lidar point cloud data and ground-based lidar point cloud data based on the matching results, including: (1) Based on the rotation angle in the matching result, the horizontal rotation matrix between the computer-borne lidar point cloud data and the ground-based lidar point cloud data.

[0073] In this embodiment, it can be constructed using the planar rotation angle obtained by two-dimensional matching. For example, the two-dimensional rotation angle can be substituted into the three-dimensional rigid body transformation formula to generate a 3×3 or 4×4 matrix rotating around the Z-axis, which is used to adjust the relative orientation of the ground-based point cloud and the airborne point cloud in the horizontal space.

[0074] Specifically, using the rotation angle Calculate the rotation matrix It can be expressed by the following formula: (2) Calculate the horizontal translation vector based on the coordinates of the three-dimensional mapping points of the optimal corresponding point pairs in the airborne lidar point cloud data and the ground-based lidar point cloud data.

[0075] In this embodiment, it can be achieved by inverse mapping of two-dimensional feature points to three-dimensional space. For example, the X and Y coordinates of the optimal two-dimensional corner points in the original three-dimensional tree point cloud are extracted, and the algebraic difference between the two is calculated to establish the absolute displacement parameters of the two sets of point clouds on the horizontal plane.

[0076] (3) Search for ground mapping points with the same horizontal coordinates as the optimal corresponding point pair from the ground-based lidar point cloud data, and calculate the vertical translation vector based on the coordinate difference of the ground mapping points in the vertical direction.

[0077] In this embodiment, in order to address the problem that relying solely on tree canopy cannot accurately align terrain height, this embodiment can calculate the vertical translation vector by comparing elevations across data layers.

[0078] For example, using the XY coordinates of the aforementioned corner points, the Z-axis elevation value of the corresponding position can be retrieved from the ground-based lidar point cloud data and subtracted to obtain the corresponding vertical translation vector, which is used to fill in the missing height dimension of the two-dimensional image.

[0079] (4) Register the airborne lidar point cloud data with the ground-based lidar point cloud data according to the horizontal rotation matrix, the horizontal translation vector and the vertical translation vector.

[0080] In this embodiment, the translation in three-dimensional space is obtained based on the translation vectors in the horizontal and vertical directions, and is expressed as follows: in, This represents the 3D mapping of the optimal corresponding point pair in the binary canopy map to the UAV lidar point cloud. The optimal corresponding point pair is the 3D mapping point in the ground-based lidar point cloud.

[0081] Next, based on the rotation matrix Translation vector Complete airborne lidar point cloud data and ground-based lidar point cloud data Coarse registration between them is specifically expressed as: (5) The registered airborne lidar point cloud data and the ground-based lidar point cloud data are stitched together using the iterative nearest neighbor algorithm to obtain the stitched forest point cloud data.

[0082] In this embodiment, based on coarse registration, the iterative nearest neighbor algorithm is used to continuously find the closest point pairs in the two sets of point clouds and minimize their sum of squared distances to eliminate minor residual errors.

[0083] Specifically, to evaluate the splicing effect, this embodiment uses the average error ( ), root mean square error ( The evaluation of coarse registration and fine stitching accuracy is specifically expressed as follows: in, This represents the sample points in the point cloud data of the airborne lidar. express The corresponding point in the ground-based lidar point cloud data.

[0084] According to experimental data, the average error and root mean square error of coarse registration of 3D point clouds are both 0.11 meters. After fine stitching, the average error and root mean square error are reduced to 0.07 meters. Furthermore, the coarse registration operation time is only 0.722 seconds, which is a significant improvement compared to traditional registration methods.

[0085] Figure 4 This is a schematic diagram of the key corner matching and overlap analysis mechanism provided by the present invention. Figure 4 In the illustrated embodiment, during the key corner point matching process, the left side represents the two-dimensional outline of the canopy from the perspective of the UAV (airborne), with the dots on the outline representing the airborne key corner points extracted by covariance matrix response value calculation. The dashed lines represent the corresponding ground point pair sets. The right side represents the two-dimensional outline of the canopy from the perspective of the ground, with the dots on the outline representing the ground key corner points extracted by covariance matrix response value calculation. The dashed lines represent the corresponding airborne point pair sets. The two solid lines represent the one-to-one mapping and connection of the ground point pair sets and the airborne point pair sets, indicating the matching relationship between the two types of points. During the overlap analysis process, the entire ground-canopy binary map is divided into a 10×10 grid. The numbers within the grid represent the pixel values ​​of the corresponding center points, with 1 representing the canopy and 0 representing the background open space. The gray shaded areas represent the canopy distribution range in the image.

[0086] The forest air-ground lidar point cloud registration method provided in this invention decouples and upgrades the two-dimensional image matching results, uses the canopy to solve horizontal displacement and the ground to solve vertical displacement, thereby effectively improving the efficiency and accuracy of three-dimensional spatial fusion of lidar data across platforms in complex forest areas without artificial targets.

[0087] In addition, the forest airborne lidar point cloud registration method proposed in this embodiment includes point cloud filtering, binary image generation, feature extraction, image matching, and three-dimensional spatial registration. The entire process requires no manual intervention or external auxiliary equipment (such as artificial targets), realizing fully automatic stitching of airborne and ground-based lidar point clouds, which improves the practicality and universality of the method.

[0088] Figure 5 This is the second flowchart illustrating the forest air-to-ground lidar point cloud registration method provided by the present invention. Figure 5 In the illustrated embodiment, the method further includes the following steps: S1. Obtain point clouds from airborne and ground-based lidar in the forest plots, perform ground filtering on the point clouds, project non-ground points onto the horizontal plane, and generate a binary map of the forest canopy. S2. Perform dilation and erosion closure operations on the binary image of the canopy to fill the holes in the binary image, and smooth the edges of the canopy shape by filtering. S3. Determine the canopy contour line based on the pixel neighborhood connectivity, calculate the geometric covariance matrix of the contour line, and construct response values ​​to quantitatively describe the key corner points of the contour. S4. Construct point pair sets in the airborne and ground-based canopy binary maps respectively, search for two pairs of points with equal distances, calculate the transformation matrix, introduce overlap discrimination, and complete the binary map matching. S5. Combine the rotation matrix obtained by matching and the horizontal spatial distance of the corresponding point pairs to complete the horizontal registration of the 3D point cloud, and complete the vertical registration based on the ground point position. On this basis, perform fine stitching of airborne and ground-based point clouds, and evaluate the stitching accuracy.

[0089] Figure 6 This is a schematic diagram of the coarse registration and fine stitching results of the 3D point cloud provided by the present invention. Figure 6 In the illustrated embodiment, combining the horizontal rotation and translation obtained from image matching with the vertical translation obtained based on ground points enables coarse registration of the LiDAR point cloud in three-dimensional space, such as... Figure 6 As shown in the image on the left (gray dots represent ground-based lidar point clouds, and black dots represent airborne lidar point clouds), this embodiment further utilizes the iterative nearest neighbor algorithm to perform fine stitching on the coarsely registered point cloud data, which can improve the point cloud stitching accuracy.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for registering forest airspace lidar point clouds, characterized in that, include: Tree points are extracted from airborne lidar point cloud data and ground-based lidar point cloud data in the forest area, and the tree points are projected onto a two-dimensional plane to obtain the corresponding airborne canopy binary map and ground-based canopy binary map. The airborne canopy binary image and the ground canopy binary image are enhanced respectively, and the corresponding airborne canopy contour points and ground canopy contour points are calculated according to the pixel connectivity between each pixel point and its corresponding neighboring points in the enhanced canopy binary image. The airborne canopy contour points and the ground-based canopy contour points are matched, and the airborne lidar point cloud data and the ground-based lidar point cloud data are spatially registered based on the matching results to obtain the forest point cloud registration results.

2. The forest air-to-ground lidar point cloud registration method according to claim 1, characterized in that, The step of calculating the corresponding airborne canopy contour points and ground-based canopy contour points based on the pixel connectivity between each pixel and its corresponding neighboring pixels in the enhanced canopy binary image includes: Obtain the n-neighborhood points of each target pixel in the enhanced canopy binary image, and calculate the average value of the product of the target pixel and the pixel values ​​of the n neighboring points to obtain the pixel connectivity index; n≥2, and n is an integer; If the pixel connectivity index is greater than 0 and less than 1, the target pixel is determined to be the corresponding airborne canopy contour point or ground-based canopy contour point.

3. The forest air-to-ground lidar point cloud registration method according to claim 1, characterized in that, Before matching the airborne canopy contour points and the ground-based canopy contour points, the method further includes: Calculate the covariance matrix of each canopy contour point within its preset neighborhood radius; Calculate the response value of each canopy contour point based on the determinant and trace of the covariance matrix; The airborne canopy binary map and the ground-based canopy binary map are each divided into a preset number of grids, and the canopy contour point with the largest corresponding response value in each grid is determined as the key corner point.

4. The forest air-to-ground lidar point cloud registration method according to claim 3, characterized in that, The matching of the airborne canopy contour points and the ground-based canopy contour points includes: Construct a set of foundation point pairs based on the key corner points corresponding to the binary map of the foundation canopy; Search for airborne point pairs that are equidistant from each corresponding point pair in the ground point pair set among the key corner points of the airborne canopy binary map; construct equidistant corresponding point pairs based on each corresponding point pair in the ground point pair set and the airborne point pairs. Based on the planar Euclidean transformation and the equidistant corresponding point pairs, the rotation angle and pixel translation vector from the ground canopy binary image to the airborne canopy binary image are calculated respectively.

5. The forest air-to-ground lidar point cloud registration method according to claim 4, characterized in that, After calculating the rotation angle and pixel translation vector from the ground-based canopy binary image to the airborne canopy binary image based on the plane Euclidean transformation and the equidistant corresponding point pairs, the method further includes: The binary image of the foundation canopy is transformed according to the rotation angle and the pixel translation vector, and the transformed binary image of the foundation canopy is divided into multiple grids. Calculate the mean value of the difference between the center point of each grid and the pixel value of the corresponding point of the center point in the airborne canopy binary image; The overlap of each pair of equidistant corresponding points is determined based on the mean value, and the pair of equidistant corresponding points with the highest overlap is selected as the optimal corresponding point pair. The rotation angle and pixel translation vector corresponding to the optimal corresponding point pair are used as the matching result.

6. The forest air-to-ground lidar point cloud registration method according to claim 5, characterized in that, The spatial registration of the airborne lidar point cloud data and the ground-based lidar point cloud data based on the matching results includes: Based on the rotation angle in the matching result, calculate the horizontal rotation matrix between the airborne lidar point cloud data and the ground-based lidar point cloud data; Based on the coordinates of the three-dimensional mapping points in the airborne lidar point cloud data and the ground-based lidar point cloud data respectively, calculate the horizontal translation vector. Search for ground mapping points with the same horizontal coordinates as the optimal corresponding point pair from the ground-based lidar point cloud data, and calculate the vertical translation vector based on the coordinate difference of the ground mapping points in the vertical direction. The airborne lidar point cloud data and the ground-based lidar point cloud data are registered according to the horizontal rotation matrix, the horizontal translation vector, and the vertical translation vector. The registered airborne lidar point cloud data and the ground-based lidar point cloud data are stitched together using the iterative nearest neighbor algorithm to obtain the stitched forest point cloud data.

Images