Parameter-adaptive three-dimensional point cloud crown volume calculation method

Abstract

The invention discloses a parameter-adaptive three-dimensional point cloud crown volume calculation method, which comprises the following steps of: 1, for input scene three-dimensional point cloud data and a crown label file, judging whether a point cloud is a crown point cloud or not to determine whether the point cloud is a crown point cloud or not so as to extract a crown part; and step 2, circularly traversing each tree crown through tree crown label data, reading the variety of the tree crown, and judging whether the tree crown is a sphere, an ellipsoid or others. And 3, for spherical or ellipsoidal crowns, calculating the volume of each crown one by one by adopting an improved convex hull method, and storing the volume. And 4, for other types of crowns, calculating the volume of each crown one by one by adopting an improved Alpha Shape method, and storing the volume of each crown. And step 5, outputting the volumes of all crowns in the current scene. The problem that a traditional single crown volume calculation algorithm is difficult to adapt to crowns with large form differences is solved.

Description

Technical Field

[0001] The present invention is based on the convex hull method and the Alpha Shape method, combined with the crown category information. For spherical crowns, a fast-forming improved convex hull method is used, and for crowns with dispersed and irregular shapes, a density-adaptive improved AlphaShape method is used. The combination of the two forms a parameter-adaptive three-dimensional point cloud crown volume calculation method. Background Art

[0002] Artificial intelligence and automation technologies have developed rapidly in recent years, with the advancement of mechanization and intelligentization in agricultural scenarios being a current research focus. The emergence of agricultural robots has improved operational efficiency, with their operating mechanisms capable of performing tasks such as crop inspection, fertilization, and harvesting. The tree crown, as the primary site of photosynthesis, not only determines the plant's light energy absorption efficiency but also directly influences its growth rate and biomass accumulation. In applications such as variable-rate spraying, automated pruning, and fruit yield prediction, accurately capturing the geometric structure and volume of the tree crown can significantly improve operational precision and resource utilization efficiency.

[0003] Point cloud data is a three-dimensional structure, usually represented by three-dimensional coordinate values ​​and laser intensity. It can accurately describe the surface geometric characteristics of the tree crown, and can also solve problems such as blurred boundaries and easy occlusion in the image. It has higher stability and stronger robustness. Therefore, the present invention selects point cloud data for tree crown volume calculation.

[0004] Common methods for calculating the volume of tree crowns from three-dimensional point clouds include geometric method, voxel method, convex hull method, spatial triangulated mesh method, and Alpha Shape method. The geometric method fits the tree crown into a geometric body and then calculates the crown volume using the volume formula, but the fitted shape often does not completely match the actual tree crown; the voxel method generates voxels with artificially set side lengths and obtains the crown volume by calculating the sum of all voxel volumes, but point cloud data often reflects the surface information of the crown and does not accurately express the internal pores of the crown, thus resulting in large errors; the convex hull method calculates the crown volume by forming the convex hull of the point cloud surface, which fits spheres, ellipsoids, etc. well, but is not suitable for tree crowns with dispersed and irregular shapes, and the average complexity of the algorithm is O(nlogn); the spatial triangulated mesh method fits the crown shape by forming Delaunay tetrahedronized triangles and then calculates the volume, but the algorithm complexity is or O(n 2 ), which is higher than the convex hull method; the Alpha Shape method simulates a ball with a radius of α rolling on the boundary of the point cloud. By changing the value of the parameter α, a "concave hull" can be formed, thereby reducing the redundant volume of the convex hull method. Summary of the Invention

[0005] This paper proposes a parameter-adaptive method for calculating tree crown volumes from three-dimensional point clouds. Based on the convex hull and alpha shape methods, this algorithm accelerates convex hull formation by finding extreme points to form the initial convex hull. Internal point clouds are then removed to reduce computational effort. The accuracy of the boundary is ensured by dynamically calculating the alpha value based on point cloud density. Depending on the tree species, either the improved convex hull or improved alpha shape method is used.

[0006] The technical solution of the present invention is: a parameter-adaptive crown volume calculation method, comprising the following steps:

[0007] Step 1: For the crown point cloud with a calibrated detection frame, use the spatial index method to extract the crown part inside the detection frame, and remove the remaining ground, trunk and other information to prevent interference with volume calculation.

[0008] Step 2: Loop through each tree crown through the label data, read the crown type, and determine whether its surface is a sphere, ellipsoid, or other.

[0009] Step 3: For spherical or ellipsoidal crowns, use the improved convex hull method to calculate the volume of each crown one by one and save it.

[0010] Step 4: For other types of tree crowns, use the improved Alpha Shape method to calculate the volume of each crown one by one and save it.

[0011] Step 5: Output the volume of all tree crowns in the current scene.

[0012] Furthermore, the step 1 specifically includes:

[0013] Step 1.1: For the input scene 3D point cloud data and tree crown label file, set the input point cloud set containing N points to Each point p contains three coordinates (x, y, z);

[0014] Step 1.2, set the input M label files to T = (t1, t2, ..., t M ), each label file t contains (L, x, y, z, h, w, l, r) information, which respectively represent the category of the detection box (L), the center coordinates of the detection box (x, y, z), the length of the detection box border (h, w, l), and the rotation angle of the detection box (r);

[0015] Step 1.3, determine whether point p is inside the detection box. If so, retain the point cloud as the crown point cloud; otherwise, remove it to extract the crown part.

[0016] Furthermore, the step 2 specifically includes:

[0017] Step 2.1, loop through the categories of the detection boxes in the label file (L), and determine whether the crown shape is a sphere, ellipsoid, or other according to the category, where the judgment standard uses dictionary mapping.

[0018] Furthermore, the step 3 specifically includes:

[0019] In step 3.1, the pre-processed single crown point cloud is layered, and the extreme points of each layer are found. The global convex hull is dynamically generated by inserting the layered convex hull vertices layer by layer.

[0020] In step 3.2, all point clouds within the global convex hull (excluding boundary points) are deleted to reduce the influence of noise and outliers.

[0021] Step 3.3: For the remaining point clouds, construct a new convex hull and calculate the volume of the convex hull.

[0022] Furthermore, the step 4 specifically includes:

[0023] In step 4.1, for the preprocessed single tree crown point cloud, use KNN to generate a neighborhood for each point. Calculate the Euclidean distance between the neighborhood center and the other K points, and take the average of these K distances as the candidate α value. The distance reflects the density of the points.

[0024] Step 4.2: For all candidate α values, sort them and select the median as the final α value. If the number of points is even, take the average of the two middle numbers as the final α.

[0025] In step 4.3, the entire crown is Delaunay tetrahedronized, the circumscribed sphere radius r of each small tetrahedron is calculated, and all tetrahedrons with r ≤ α are selected and spliced ​​to form a closed surface.

[0026] Step 4.4, calculate the volume of the surface.

[0027] The present invention, employing the above technical solution, achieves the following technical benefits: Based on and improving the convex hull method and the AlphaShape method, the present invention integrates tree crown category information to complete three-dimensional point cloud crown volume calculation, resolving the problem that traditional single crown volume calculation algorithms are difficult to adapt to tree crowns with large morphological differences. In the improved convex hull method, an initial convex hull is constructed to delete the point cloud within the crown, thereby accelerating the calculation of the convex hull method. In the improved AlphaShape method, the average distance between the point clouds within each tree is calculated to determine the value of the parameter α. Through these improvements, the algorithm can assign a faster and more appropriate volume calculation method to each tree crown, thereby obtaining more accurate crown volume values. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1This is a flow chart of the parameter-adaptive crown volume calculation method proposed in the present invention (taking osmanthus, thornless holly, and loquat as examples). DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present invention are described clearly and completely below with reference to the accompanying drawings.

[0030] Figure 1 The structure diagram of the parameter-adaptive crown volume calculation method is as follows:

[0031] Step 1: For the input scene 3D point cloud data and tree crown label file, extract the tree crown part within the detection box.

[0032] Set the input point cloud set containing N points to Each point p contains three coordinates (x, y, z). Assume that the input M label files are T=(t1, t2,…, t M ), each label file t contains information such as (L, x, y, z, h, w, l, r), which respectively represent the detection box category (L), the detection box center coordinates (x, y, z), the detection box border length (h, w, l), and the detection box rotation angle (r). By comparing the size of the coordinate values, we determine whether the point p is inside the detection box. If it is, the point cloud is retained as the tree crown point cloud; otherwise, it is discarded.

[0033] Step 2: Loop through each tree crown through the crown label data, read the crown type, and determine whether it is a sphere, ellipsoid, or other.

[0034] The crown type and crown shape are mapped using a dictionary, for example, {“Osmanthus fragrans”: 0, “Horny holly”: 0, “Loquat”: 1}, where 0 represents sphere and ellipsoid, and 1 represents others.

[0035] Step 3: For spherical or ellipsoidal crowns, use the improved convex hull method to calculate the volume of each crown one by one and save it.

[0036] Slice the point cloud. The slice height can be set to 0.1, 0.2, or other values ​​according to the actual situation. Project each slice onto the xy plane to generate a two-dimensional convex hull. Read the three-dimensional coordinates of the two-dimensional convex hull and repeat until all slices are completed. Merge the three-dimensional coordinates of the convex hulls of all slices. Find the extreme points in the six directions of ±x, ±y, and ±z, and construct the initial convex hull with these six points as the boundary. Delete all point clouds within the initial convex hull (excluding boundary points). For the remaining point clouds, construct their three-dimensional convex hull and calculate the volume of the convex hull using the volume attribute in the spatial module.

[0037] Step 4: For other types of tree crowns, use the improved Alpha Shape method to calculate the volume of each crown one by one and save it.

[0038] Perform a KNN on all point clouds, where the K value can be set to 8, 16, or other values ​​based on actual needs. Calculate the Euclidean distance between the neighborhood center and all other points within the neighborhood, and take the average value Q of all distances as the candidate α value for that neighborhood. Sort the Q values ​​of all neighborhoods and select the median. If there is an even number of Q values, take the average of the two middle Q values, and the resulting value is the α value for the canopy. Perform a Delaunay tetrahedronization on the entire canopy, calculate the circumscribed sphere radius r of each small tetrahedron, select all tetrahedrons with r ≤ α, and concatenate them to form a closed surface. The volume of this surface is calculated as the current canopy volume.

[0039] Step 5: Save the volume information of all tree crowns and export them.

[0040] Experimental validation was conducted on a self-developed forest and orchard dataset and compared with traditional algorithms. Because true crown volume values ​​are difficult to obtain, the average value obtained by all algorithms was set as the standard value. The experimental results are shown in Table 1. These results demonstrate that the proposed method can more accurately calculate the crown volumes of different tree species.

[0041] Table 1: Comparison of crown volume calculation results

[0042]

Claims

1. A parameter-adaptive three-dimensional point cloud crown volume calculation method, characterized in that: The following steps are involved: Step 1: For the input scene 3D point cloud data and tree crown label file, determine whether the point cloud is within the detection box to determine whether it is a tree crown point cloud, thereby extracting the tree crown part; Step 2: Loop through each tree crown through the crown label data, read the crown type, and determine whether it is a sphere, ellipsoid, or other; Step 3: For spherical or ellipsoidal crowns, use the improved convex hull method to calculate the volume of each crown one by one and save it; Step 4: For other types of tree crowns, use the improved Alpha Shape method to calculate the volume of each crown one by one and save it; Step 5: Output the volume of all tree crowns in the current scene.

2. A parameter-adaptive three-dimensional point cloud crown volume calculation method according to claim 1, characterized in that: The step 1 specifically includes: Step 1.1: For the input scene 3D point cloud data and tree crown label file, set the input point cloud set containing N points to Each point p contains three coordinates (x, y, z); Step 1.2, set the input M label files to T = (t1, t2, ..., t M ), each label file t contains (L, x, y, z, h, w, l, r) information, which respectively represent the category of the detection box (L), the center coordinates of the detection box (x, y, z), the length of the detection box border (h, w, l), and the rotation angle of the detection box (r); Step 1.3, determine whether point p is inside the detection box. If so, retain the point cloud as the crown point cloud; otherwise, remove it to extract the crown part.

3. The parameter-adaptive three-dimensional point cloud crown volume calculation method according to claim 1 is characterized by: The step 2 specifically includes: Step 2.1, loop through the categories of the detection boxes in the label file (L), and determine whether the crown shape is a sphere, ellipsoid, or other according to the category, where the judgment standard uses dictionary mapping.

4. A parameter-adaptive three-dimensional point cloud crown volume calculation method according to claim 1, characterized in that: The step 3 specifically includes: Step 3.1: Layer the pre-processed single crown point cloud, find the extreme points of each layer, and finally construct the global convex hull by inserting the vertices of the layered convex hull layer by layer; Step 3.2: Delete all point clouds within the global convex hull (excluding boundary points) to reduce the influence of noise and outliers. In step 3.3, for the remaining point cloud, construct a new convex hull and use the volume attribute in the spatial module to obtain the convex hull volume.

5. The parameter-adaptive three-dimensional point cloud crown volume calculation method according to claim 1 is characterized by: The step 4 specifically includes: Step 4.1: For the pre-processed single crown point cloud, use KNN to generate a neighborhood for each point, calculate the Euclidean distance between the neighborhood center and the other K points, and take the average of these K distances as the candidate α value. The distance reflects the density of the points; Step 4.2: For all candidate α values, sort them and select the median as the final α value. If the number of points is even, take the average of the two middle numbers as the final α; Step 4.3: Delaunay tetrahedronize the entire crown, calculate the circumscribed sphere radius r of each small tetrahedron, select all tetrahedrons with r ≤ α, and splice them to form a closed surface; Step 4.4, calculate the volume of the surface.

Images