A forest scene voxel model construction method and system

Abstract

The application discloses a kind of forest scene voxel model construction method and system, the method includes: based on laser radar obtains the point cloud data of reference tree of the region to be measured and the point cloud data of canopy structure;According to the point cloud data of canopy structure, the point cloud data of reference tree is carried out point cloud conversion, and the point cloud conversion data of reference tree is obtained;According to the point cloud conversion data of reference tree, the blank of the point cloud data of canopy structure is filled with reference tree, and high-density point cloud data is generated;High-density point cloud data is voxelized, and forest scene voxel model is constructed.The system includes: data acquisition module, point cloud conversion module, data filling module and construction module.Through using the application, it can be realized in large space range to reconstruct the fine scale three-dimensional structure of forest scene and improve resolution.The application is a kind of forest scene voxel model construction method and system, and can be widely applied to laser radar data three-dimensional modeling technical field.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional modeling technology for lidar data, and in particular to a method and system for constructing a voxel model of a forest scene. Background Technology

[0002] Realistic 3D forest simulations have wide-ranging practical applications, such as forest management, landscape design, computer graphics, and virtual reality. Based on these 3D forest scenes, researchers can control and design experiments to simulate solar radiation and remote sensing signal transmission within the forest canopy, providing a powerful tool for testing inversion algorithms and model assumptions. Existing methods for reconstructing forest scenes based on LiDAR point cloud data either reconstruct forest scenes using ALS point cloud data (which, while covering a larger area, typically has too low a point density to apply tree reconstruction algorithms to create realistic trees, resulting in inaccurate descriptions of tree structure) or reconstruct forest scenes using TLS point cloud data (which can generate high-density point cloud data to reconstruct realistic trees, but is usually limited to small forest plots). Summary of the Invention

[0003] To address the aforementioned technical problems, the present invention aims to provide a method and system for constructing a voxel model of a forest scene, which can reconstruct the fine-scale three-dimensional structure of a forest scene within a large spatial range and improve resolution.

[0004] The first technical solution adopted in this invention is: a method for constructing a voxel model of a forest scene, comprising the following steps:

[0005] Point cloud data of the reference tree and canopy structure of the area to be measured are obtained based on lidar.

[0006] The point cloud data of the reference tree is transformed based on the point cloud data of the canopy structure to obtain the point cloud transformed data of the reference tree;

[0007] Based on the point cloud transformation data of the reference tree, the reference tree fills in the gaps in the point cloud data of the canopy structure, generating high-density point cloud data;

[0008] Voxelization is performed on high-density point cloud data to construct a voxel model of a forest scene.

[0009] Furthermore, the step of acquiring point cloud data of the reference tree and canopy structure of the area to be measured based on lidar specifically includes:

[0010] Ground-based lidar is used to scan the area under test to obtain TLS point cloud data;

[0011] Extract the TLS point cloud data to obtain the point cloud data of the reference tree;

[0012] The area to be measured is scanned using an airborne lidar to obtain ALS point cloud data;

[0013] The ALS point cloud data was segmented using the single-tree segmentation method to obtain the point cloud data of the canopy structure.

[0014] Furthermore, the point cloud data of the reference tree includes the position, height, crown radius, and crown direction of the reference tree; the point cloud data of the canopy structure includes the position, height, crown radius, and crown direction of the prediction tree.

[0015] Furthermore, the step of performing point cloud transformation on the point cloud data of the reference tree based on the point cloud data of the canopy structure to obtain the point cloud transformation data of the reference tree specifically includes:

[0016] The point cloud data of the reference tree is scaled based on the crown radius and height of the predicted tree to obtain the horizontal scaling factor and vertical scaling factor of the reference tree.

[0017] The scaled reference tree is rotated based on the crown direction of the predicted tree to obtain the rotation angle of the reference tree;

[0018] The reference tree is translated based on the position of the prediction tree to obtain the translation amount of the reference tree;

[0019] By integrating the horizontal scaling factor, vertical scaling factor, rotation angle, and translation amount of the reference tree, the point cloud transformation data of the reference tree is obtained.

[0020] Furthermore, the step of voxelizing high-density point cloud data and constructing a voxel model of the forest scene specifically includes:

[0021] High-density point cloud data is classified based on point cloud classification algorithms to obtain point cloud data of tree materials;

[0022] Voxelization is performed on the point cloud data of tree materials to construct a forest scene and generate a voxel model of the forest scene.

[0023] Furthermore, it also includes benchmarking forest scene voxel models using virtual forest scenes, specifically including:

[0024] Based on the RAMI benchmark, a triangular mesh model and a forest scene voxel model are constructed for the virtual forest scene to obtain the PBRT model and VBRT model.

[0025] The images simulated by the PBRT model are used to benchmark the images simulated by the VBRT model.

[0026] The second technical solution adopted in this invention is: a forest scene voxel model construction system, comprising:

[0027] The data acquisition module acquires point cloud data of the reference tree and canopy structure of the area to be measured based on lidar.

[0028] The point cloud conversion module is used to convert the point cloud data of the reference tree to obtain the point cloud data of the prediction tree.

[0029] The data imputation module is used to fill in the gaps in the point cloud data of the canopy structure based on the point cloud data of the prediction tree, and generate high-density point cloud data.

[0030] The module is used to voxelize high-density point cloud data and build a voxel model of a forest scene.

[0031] The beneficial effects of the method and system of this invention are as follows: First, this invention acquires point cloud data of reference trees and canopy structure in the area to be measured based on lidar, which can accurately and quickly measure the point cloud data of most trees in the area to be measured; then, through point cloud transformation of the reference trees, point cloud data of the predicted trees is obtained, and the gaps in the point cloud data of the canopy structure are filled based on the point cloud data of the predicted trees to generate high-density point cloud data, which better solves the problem of low density of point cloud data in the canopy structure; finally, based on voxelization, a voxel model of the forest scene is constructed, which can reconstruct trees without reducing point density; thus, it realizes the reconstruction of the fine-scale three-dimensional structure of the forest scene in a large spatial range and improves the resolution. Attached Figure Description

[0032] Figure 1 This is a flowchart of the steps in a method for constructing a voxel model of a forest scene according to the present invention;

[0033] Figure 2 This is a structural block diagram of a forest scene voxel model construction system according to the present invention;

[0034] Figure 3 This is a flowchart illustrating the construction process of a voxel model of a forest scene according to a specific embodiment of the present invention;

[0035] Figure 4 This is a schematic diagram of a single tree point cloud data according to a specific embodiment of the present invention;

[0036] Figure 5 This is a flowchart illustrating the point cloud data scaling process according to a specific embodiment of the present invention;

[0037] Figure 6 These are comparison images of point cloud data before and after transformation in a specific embodiment of the present invention;

[0038] Figure 7 This is a comparative schematic diagram of different model simulations in red, green and blue bands in a specific embodiment of the present invention.

[0039] Figure 8This is a comparative schematic diagram of different model simulations in the near-infrared, red, and green bands in a specific embodiment of the present invention.

[0040] Figure 9 This is a comparative schematic diagram of different waveband visual views in scenario two of a specific embodiment of the present invention;

[0041] Figure 10 This is a comparative schematic diagram of the histograms of different models of V1 in scenario one of the specific embodiments of the present invention;

[0042] Figure 11 This is a comparative schematic diagram of the histograms of different models of V2 in scenario one of the specific embodiments of the present invention;

[0043] Figure 12 This is a comparative schematic diagram of the histograms of different models of V3 in scenario one of the specific embodiments of the present invention;

[0044] Figure 13 This is a comparative schematic diagram of the histograms of different models of V4 in scenario one of the specific embodiments of the present invention. Detailed Implementation

[0045] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.

[0046] Reference Figure 1 and Figure 3 This invention provides a method for constructing a voxel model of a forest scene, the method comprising the following steps:

[0047] S1. Obtain point cloud data of the reference tree and canopy structure of the area to be measured based on lidar;

[0048] S1.1. Use ground-based lidar to scan the area to be measured and obtain TLS point cloud data;

[0049] S1.2 Extract the TLS point cloud data to obtain the point cloud data of the reference tree;

[0050] Specifically, the point cloud data of the reference tree includes the reference tree's position, height, crown radius, and crown orientation. The definitions of position, height, crown radius, and crown orientation are as follows:

[0051] Given a point cloud of a single tree, the tree's position is defined as its centroid in two-dimensional space, i.e., the average position of all points on the horizontal plane (xy plane), and the tree's height is the height difference between its highest point and the ground. Figure 4The diagram illustrates the derivation of the crown radius and crown direction. First, the point is translated so that the origin of the Cartesian coordinate system coincides with the position of the tree (denoted by O). The xy plane is then mapped by a user-defined value (denoted by d). n (This indicates that) starting from the origin O, the direction is divided equally, and the angle between two adjacent directions is 360° / d. n Degree. For direction i, calculate the distance from O to the farthest point P in that direction. i Length d i The longest length in all directions is defined as the crown radius (i.e., Figure 4 In the context of OP, the azimuth angle of the crown radius (OP) is defined as the crown direction. The accuracy of the crown radius and crown direction is affected by d. n The impact, d n A larger value results in higher precision. However, when the point density is very low, setting a very large d value will reduce accuracy. n The value is meaningless. Here, based on the test, we will use d. n The default value is set to 32.

[0052] S1.3. Use an airborne lidar to scan the area to be measured and obtain ALS point cloud data;

[0053] S1.4. The ALS point cloud data is segmented based on the single-tree segmentation method to obtain the point cloud data of the canopy structure.

[0054] Specifically, the point cloud data of the canopy structure includes the location, height, canopy radius, and canopy orientation of the predicted tree, and the definitions of location, height, canopy radius, and canopy orientation are the same as those in step S1.2.

[0055] S2. Based on the point cloud data of the canopy structure, the point cloud data of the reference tree is transformed to obtain the point cloud transformation data of the reference tree. Since the point cloud of the tree detected by ALS is too sparse to describe the detailed structure of a single tree, the dense point cloud of the reference tree in TLS is transformed to predict the detailed structure of the detected single tree.

[0056] S2.1 Scale the point cloud data of the reference tree according to the crown radius and height of the prediction tree so that the crown radius and height of the reference tree after scaling are equal to the crown radius and height of the prediction tree, and obtain the horizontal scaling factor and vertical scaling factor of the reference tree.

[0057] Specifically, for a single reference tree, the same horizontal scaling factor Sc is applied to both the X and Y coordinates. xy Apply a vertical scaling factor Sc to the Z-coordinate of the point cloud. z Horizontal scaling factor Sc xy and vertical scaling factor Sc z Defined by the following formula:

[0058] Sc xy =ri / r0;

[0059] Sc z =h i / h0;

[0060] Where, r i and h i Let r0 and h0 be the crown radius and height of the prediction tree i, respectively, and r0 and h0 be the crown radius and height of the reference tree, respectively.

[0061] It's important to note that directly multiplying the point coordinates by a scaling factor greater than 1 can result in sparse point cloud data because the pixel spacing is magnified. To overcome this issue, the points are converted to voxels before scaling, and then the horizontal scaling factor Sc defined in the equation above is used. xy and vertical scaling factor Sc z This involves scaling voxels (instead of points), and finally converting the voxels back to points. The detailed process is as follows... Figure 5 As shown, an example is used in two-dimensional space: a voxel with a resolution of 0.02m is created to contain the original point; the 0.02m voxel is multiplied by 2 to generate a new voxel of 0.04m; starting from the lower left corner of the scaled voxel, a new point is iteratively created at a sampling interval of 0.02m, eventually generating 9 points; when converting points to voxels, it is recommended to choose a voxel resolution that is approximately equal to the point spacing. Since the point spacing of the TLS point cloud data at the tree top is approximately 0.02m, 0.02m is used in this embodiment; the sampling interval affects the density of the new points, and it needs to be smaller than the size of the scaled voxel; similarly, this embodiment recommends choosing a sampling interval similar to the point spacing of the original TLS point cloud data, i.e., 0.02m. Figure 6 An example of point cloud transformation is given to illustrate that the transformation method proposed in this embodiment can reconstruct the tree without reducing the point density.

[0062] S2.2 Rotate the scaled reference tree according to the crown direction of the prediction tree so that the crown direction of the rotated reference tree overlaps with the crown direction of the prediction tree, and obtain the rotation angle of the reference tree.

[0063] Specifically, after scaling, the rotation of point cloud data by one angle is defined as:

[0064] angrot=dir i -dir0;

[0065] Where dir i dir0 and dir0 are the crown directions of the prediction tree i and the reference tree, respectively.

[0066] S2.3. Translate the rotated reference tree according to the position of the prediction tree, so that the position of the translated reference tree overlaps with the position of the prediction tree, and obtain the translation amount of the reference tree.

[0067] Reference Figure 6 The images show a comparison of black alder point cloud data before and after transformation. The left image shows the black alder tree before transformation, with a tree height of 19.33m, crown radius of 1.83m, and crown orientation of 180°. The right image shows the black alder tree after transformation, with a tree height of 27.06m, crown radius of 3.56m, crown orientation of 0°, and a horizontal scaling factor of Sc. xy =1.93; Vertical scaling factor Sc z =1.4; rotation angle is angrot=180°.

[0068] S2.4 Integrate the horizontal scaling factor, vertical scaling factor, rotation angle, and translation amount of the reference tree to obtain the point cloud transformation data of the reference tree.

[0069] S3. Based on the point cloud transformation data of the reference tree, fill in the gaps in the point cloud data of the canopy structure using the reference tree to generate high-density point cloud data;

[0070] S4. Voxelize the high-density point cloud data and construct a voxel model of the forest scene.

[0071] S4.1 Classify high-density point cloud data based on point cloud classification algorithm to obtain point cloud data of tree materials;

[0072] Specifically, point clouds are classified into various tree materials, including leaves, trunks, and branches.

[0073] S4.2. Perform voxelization on the point cloud data of tree materials and construct a forest scene to generate a voxel model of the forest scene.

[0074] Specifically, if a stereo pixel contains at least one point, then it is a non-empty voxel, labeled as having a majority of points; high-resolution voxels at the centimeter level, such as 0.02m, are used; in this embodiment, voxels are stored using an octree data structure, thereby reducing memory usage.

[0075] It should be noted that, in order to visualize the voxel-based forest scene, this embodiment uses the voxel-based radiative transfer model VBRT to simulate images from different viewpoints. Unlike other voxel-based radiative transfer models, VBRT uses high-resolution voxels to describe the 3D structure of the forest canopy. Non-empty voxels are filled with solid material, thus eliminating the need to use the turbidity medium assumption to simplify the canopy structure within the voxels.

[0076] As a preferred embodiment of this method, it further includes benchmarking the forest scene voxel model using a virtual forest scene. First, virtual forest data is obtained based on RAMI. Second, a triangular mesh forest scene model and a voxel forest scene model are constructed based on the virtual forest data, and the triangular mesh and voxel scene are visualized using PBRT and VBRT models, respectively. Finally, the image simulated by the VBRT model is used to benchmark the image simulated by the PBRT model.

[0077] Specifically, this embodiment selects a virtual forest dataset based on a triangular mesh to verify the voxel-based modeling method. The virtual forest is simulated using TLS and ALS point cloud data, and the proposed method is used to generate a voxel-based forest scene. To visualize the forest scene, a radiative transfer model is used to simulate multispectral images, including blue, green, red, and near-infrared bands. The triangular mesh-based radiative transfer model PBRT is selected to simulate the triangular mesh-based virtual forest scene image, while the voxel-based radiative transfer model VBRT is selected to simulate the constructed voxel-based virtual forest scene image. It is important to note that since the virtual forest data originates from a triangular mesh, the virtual image simulated by the triangular mesh-based model PBRT is assumed to have zero error and is used as a reference image to benchmark the image simulated by VBRT. Furthermore, the proposed voxel-based modeling method is used to simulate a real forest scene, and images from different perspectives are simulated using VBRT for visualization.

[0078] Scene 1 is a virtual forest scene, with tree species data sourced from RAMI. This scene covers an area of ​​108m × 108m × 30.51m, containing 1017 trees, with a maximum tree height of 30.51m and a leaf area index of 2.58. Tree species include silver birch, black alder, small-leaved linden, and Scots pine. For each species, a reference tree was randomly selected to acquire TLS point cloud data. Ray tracing was used to simulate the TLS point cloud data, employing common scanning settings of the RIEGL VZ-400 scanner. For each laser pulse, the sensor randomly emitted several rays into the scene, each ray emitted within a beam divergence angle range. The first intersection of each ray with the reference tree was recorded as the return point, along with its 3D coordinates and label information. To minimize the impact of occlusion, TLS scans were simulated at multiple locations, including the east, south, west, and north of each reference tree. For each reference tree, the simulated TLS point cloud data from all scan locations were merged together. The simulation of ALS point cloud data is similar to that of TLS simulation, and the scanning parameters of the Optech GEMINI airborne laser topography mapper were used. Individual trees were segmented from the simulated ALS point cloud data using a single-tree segmentation method; 725 out of 1017 trees were detected, a detection rate of 71.29%. Due to segmentation errors, multiple species may exist within the same tree segment; therefore, the species type of an individual tree was determined based on the mode of the species within that segment. The TLS point cloud data and ALS point cloud data were combined using the method described in step S3 above, and the final point cloud was voxelized at a resolution of 0.02m.

[0079] Among them, RAMI (Radiation Transfer Model Intercomparison) provides virtual forest data in various environments.

[0080] Scenario 2 is a real forest scene extracted from the Sierra National Forest in California, located at 37°26′N 119°35′W, with a range of 256m × 256m × 113m and an elevation of 1474m to 1587m. Ground echoes were extracted from the ALS point cloud data, and a digital elevation model (DEM) was generated using inverse distance weighted interpolation with a spatial resolution of 1m. Individual trees were segmented from the ALS point cloud data using a single-tree segmentation method, considering only trees taller than 3m. A total of 1314 trees were identified, with the tallest tree reaching 68.78m. The study area is predominantly coniferous, interspersed with some broadleaf trees. Since there was no TLS point cloud data for individual trees in this area to match the species, no species classification algorithm was run in this example. Instead, three species were randomly selected from the virtual forest dataset above to approximate the species in this scene: Norway maple and black alder approximate broadleaf trees, while Scots pine approximates coniferous trees. 80% of the trees were randomly assigned as Scots pine, 10% as Norway maple, and 10% as black alder. The TLS point cloud data of the reference trees was combined with the real ALS point cloud data to obtain new point cloud data using the proposed method, and then voxelized at a resolution of 0.04m.

[0081] The above scenarios were modeled in 3D and input into a radiative transfer model for visualization. In Scenario 1, a virtual forest scene derived from a triangular mesh was presented, and the PBRT-based model was parameterized, referred to as the perfect input. Conversely, the voxel-based VBRT model was parameterized using ALS and TLS point cloud data, referred to as the imperfect input. Images simulated by PBRT were used to benchmark images simulated by VBRT. In Scenario 2, due to the lack of a realistic 3D forest scene, a voxel-based forest scene was reconstructed using only ALS and TLS point cloud data, and images simulated by VBRT were used to visualize the modeled forest scene.

[0082] When simulating multispectral images of forest scenes, Lambert scattering law was applied to all materials, and non-empty voxels were used to fill solid materials instead of turbid media; the spectral characteristics of different materials for different species were obtained from the RAMI website.

[0083] Based on the above scene construction method, the corresponding simulation image is obtained. In order to more intuitively show the differences between the images simulated by different models, not only are different bands used to visualize the images simulated by different models, but histograms of the corresponding images are also generated.

[0084] In Scene 1, a true-color image is used for visualization simulation using the red, green, and blue bands, such as... Figure 7 As shown, a false-color image is visualized using the near-infrared, red, and green bands. Figure 8 As shown, Figure 7 and Figure 8The images above are PBRT simulations, and the images below are VBRT simulations. Visually, using the same model across different bands to simulate the same scene, the resulting true-color and false-color images appear consistent. Figure 7 or Figure 8 As shown, for scene 1 viewpoint V3, VBRT can observe two missing trees in the lower left corner of the image; however, its simulated image's corresponding histogram is as follows: Figure 10 , Figure 11 , Figure 12 and Figure 13 As shown, the histogram of the image simulated by VBRT matches the histogram of the image simulated by PBRT well.

[0085] In scenario two, such as Figure 9 As shown, the top image is a true-color image simulated by VBRT, and the bottom image is a false-color image simulated by VBRT. Both the true-color and false-color images from different perspectives look very realistic. Unlike the virtual forest scene 1 which uses a planar digital elevation model, this scene uses a real digital elevation model with a resolution of 1m derived from ALS point cloud data. The terrain in the simulated image view V3 also shows a brick-like effect because the resolution of the digital elevation model (1m) is lower than the voxel resolution of the scene (0.04m).

[0086] The statistical data of the simulated images from different models in Scene 1 are shown in the table below:

[0087]

[0088]

[0089] As shown in the table above, the minimum, mean, maximum, and standard deviation values ​​of PBRT and VBRT for each image are very similar.

[0090] like Figure 2 As shown, a voxel model construction system for a forest scene includes:

[0091] The data acquisition module acquires point cloud data of the reference tree and canopy structure of the area to be measured based on lidar.

[0092] The point cloud conversion module is used to convert the point cloud data of the reference tree based on the point cloud data of the canopy structure, so as to obtain the point cloud conversion data of the reference tree.

[0093] The data incomplete module is used to fill in the gaps in the point cloud data of the canopy structure based on the point cloud transformation data of the reference tree, thereby generating high-density point cloud data.

[0094] The module is used to voxelize high-density point cloud data and build a voxel model of a forest scene.

[0095] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0096] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for constructing a forest scene voxel model, characterized in that, The method comprises the following steps: obtaining point cloud data of a reference tree and point cloud data of a canopy structure of a to-be-measured region based on a laser radar; performing point cloud conversion on the point cloud data of the reference tree according to the point cloud data of the canopy structure to obtain point cloud conversion data of the reference tree; filling in the gaps of the point cloud data of the canopy structure with the reference tree according to the point cloud conversion data of the reference tree to generate high-density point cloud data; performing voxelization on the high-density point cloud data to construct a forest scene voxel model. The step of obtaining the point cloud data of the reference tree and the point cloud data of the canopy structure of the to-be-measured region based on the laser radar specifically comprises: scanning the to-be-measured region by using a ground laser radar to obtain TLS point cloud data; extracting the TLS point cloud data to obtain the point cloud data of the reference tree; scanning the to-be-measured region by using an airborne laser radar to obtain ALS point cloud data; segmenting the ALS point cloud data based on a single-tree segmentation method to obtain the point cloud data of the canopy structure. The point cloud data of the reference tree comprises the position, height, crown radius and crown direction of the reference tree; and the point cloud data of the canopy structure comprises the position, height, crown radius and crown direction of a predicted tree. The step of performing point cloud conversion on the point cloud data of the reference tree according to the point cloud data of the canopy structure to obtain the point cloud conversion data of the reference tree specifically comprises: scaling the point cloud data of the reference tree according to the crown radius and height of the predicted tree to obtain a horizontal scaling factor and a vertical scaling factor of the reference tree; rotating the scaled reference tree according to the crown direction of the predicted tree to obtain a rotation angle of the reference tree; translating the rotated reference tree according to the position of the predicted tree to obtain a translation amount of the reference tree; integrating the horizontal scaling factor, the vertical scaling factor, the rotation angle and the translation amount of the reference tree to obtain the point cloud conversion data of the reference tree. horizontal scaling factor Sc xy and vertical scaling factor Sc z defined by the following equation: Sc xy = r i r0; Sc z = h i h0; where r i and h i are the crown radius and height of the predicted tree i, respectively, and r0and h0are the crown radius and height of the reference tree, respectively.

2. The method of claim 1, wherein, The step of performing voxelization on the high-density point cloud data to construct the forest scene voxel model specifically comprises: classifying the high-density point cloud data based on a point cloud classification algorithm to obtain tree material point cloud data; performing voxelization processing on the tree material point cloud data and constructing a forest scene to generate the forest scene voxel model.

3. The method of claim 1, wherein, The method further comprises benchmark testing the forest scene voxel model by using a virtual forest scene, specifically comprising: constructing a triangular mesh model and a forest scene voxel model according to a RAMI benchmark virtual forest scene to obtain a PBRT model and a VBRT model; benchmark testing an image simulated by the VBRT model by using an image simulated by the PBRT model.

4. A forest scene voxel model construction system, characterized by, A device for performing the forest scene voxel model construction method according to claim 1, comprising: a data acquisition module configured to obtain point cloud data of a reference tree and point cloud data of a canopy structure of a to-be-measured region based on a laser radar; a point cloud conversion module configured to perform point cloud conversion on the point cloud data of the reference tree according to the point cloud data of the canopy structure to obtain point cloud conversion data of the reference tree; a data filling module configured to fill in the gaps of the point cloud data of the canopy structure with the reference tree according to the point cloud conversion data of the reference tree to generate high-density point cloud data; a construction module configured to perform voxelization on the high-density point cloud data to construct a forest scene voxel model.

Images