A 3D visualization method for urban trees based on vehicle-mounted laser point cloud data

Abstract

The present invention belongs to the technical field of vehicle-mounted laser radar point cloud data processing, and specifically relates to a three-dimensional visualization method for urban trees based on vehicle-mounted laser point cloud data. Based on the original point cloud data, after tree segmentation and separation of the crown and trunk, the trunk and crown are modeled separately, and finally a three-dimensional model of the tree in the urban scene is constructed. Compared with the traditional three-dimensional tree modeling method, it is possible to model the trunk and crown separately according to their different morphological characteristics, avoiding the disadvantage of inaccurate expression of the morphological characteristics of single tree modeling, and can simply, quickly and accurately construct a three-dimensional model of urban trees, which can be applied to many fields such as forestry resource surveys, 3D city modeling, and tree parameter extraction.

Description

Technical Field

[0001] The invention belongs to the technical field of vehicle-mounted laser radar point cloud data processing, and in particular relates to a three-dimensional visualization method of urban trees based on vehicle-mounted laser point cloud data. Background Art

[0002] LiDAR is an active remote sensing device that breaks through the traditional single-point measurement method and measures the distance between the sensor and the target object by actively emitting laser pulses. It has the unique advantages of high precision, high resolution and high efficiency, and can especially penetrate the forest vegetation covering the surface to quickly obtain terrain information, providing a new technical means for the extraction of forest structure parameters and the three-dimensional visualization of trees.

[0003] Vegetation models such as trees are an indispensable and important part of urban scene modeling. At the same time, three-dimensional tree models are also very representative natural landscape models in virtual scenes, and have been widely used in virtual tourism, virtual cities, virtual ecological landscapes, etc. Three-dimensional point cloud data can perfectly present the surface structure information, spatial information and model details of objects, and can effectively avoid the problem that traditional image-based three-dimensional reconstruction cannot directly provide spatial information. Constructing a three-dimensional tree model can quantitatively extract vegetation parameters such as tree height and breast diameter, and it is also the basis and key to promoting the digitalization of forestry.

[0004] At present, the commonly used point cloud 3D reconstruction methods mainly include greedy projection triangulation algorithm, marching cube and Poisson surface reconstruction algorithm. The greedy projection triangulation algorithm first projects the point cloud to a plane through the normal, establishes the topological relationship between the point clouds on the two-dimensional plane, and finally obtains a three-dimensional triangular mesh surface model, but this method cannot perform surface smoothing and hole repair at the same time; the marching cube algorithm approximates the isosurface through linear interpolation in the three-dimensional discrete data field, but the reconstruction will contain a large amount of triangular patch data and the information loss is serious; Poisson reconstruction constructs the Poisson equation through smoothing filter indicator function, establishes the relationship between the indicator function gradient field after surface filtering and the surface inward normal vector field, and solves the isosurface, but it cannot capture the local details of the reconstructed surface well. Summary of the invention

[0005] In response to the above-mentioned problems or shortcomings, the present invention provides a three-dimensional tree visualization method based on vehicle-mounted laser point cloud data. Based on the original point cloud data, after tree segmentation and separation of crown and trunk, the trunk and crown are modeled separately, and finally a three-dimensional tree scene in an urban scene is constructed.

[0006] The specific technical solutions of the present invention are as follows:

[0007] Step 1: Denoising and filtering:

[0008] Due to the accuracy of the LiDAR scanning equipment and the influence of the surrounding environment, point cloud data often contains noise points, so the first step is to remove the noise points in the point cloud. Then, data filtering is performed on the denoised point cloud data to obtain ground point and non-ground point data.

[0009] Step 2: Single tree segmentation and trunk and crown extraction:

[0010] The filtered data is normalized. First, the digital elevation model (DEM) is generated using the ground points obtained by filtering, and then the canopy height model (CHM) is obtained by subtracting the DEM height from the non-ground point elevation. The single tree segmentation algorithm based on CHM is used to extract the single tree. Then the cylinder fitting algorithm is used to separate the trunk and crown from the single tree.

[0011] Step 3: Surface reconstruction:

[0012] Use the alpha-shape algorithm to construct three-dimensional models of the trunk and crown respectively. The steps are as follows:

[0013] 3-1. Set distance threshold α

[0014] 3-2. Construct a point set Q = {qi}, (i = 1, 2, ... n) to be modeled. Select a point q1 from it, and the points whose distance to it is less than 2α form a new point set Q1. Select a set of points q2 and q3 from the point set Q1, and find a sphere with a radius of α passing through points q1, q2, q3 and the centers o1 and o2.

[0015] 3-3. Traverse the point set Q1, and find the distance sets L1 and L2 from other points to the sphere centers o1 and o2 in turn. If the distance of one of the sets in L1 and L2 is greater than α, then q1, q2, and q3 are edge contour points, and the three points are connected to form a boundary triangle patch; otherwise, they are not edge contour points, stop traversal, and proceed to the next step;

[0016] 3-4. Select the next set of points in the point set Q1 and judge according to steps (2) and (3) until all points in the point set Q1 are judged. Output a set of triangular facets, and the exposed triangular facets in the set form a local convex hull;

[0017] 3-5. Select the next point in the point set Q and perform judgment according to steps (2), (3), and (4) until all points in the point set Q are judged. Multiple convex hulls reconstruct the surface S(Q).

[0018] Step 4: 3D visualization:

[0019] Combine the trunk and crown models corresponding to individual trees to complete the three-dimensional visualization of urban trees.

[0020] Principles involved in step 3:

[0021] The principle of alpha-shape:

[0022] Assume that there is a point set P, whose alpha-shape is a unique polygon determined by the point set P and the radius parameter α. It can be imagined that there is a circle with a radius of α, which starts rolling from a certain point outside the point set P. If α is large enough, then the circle will not roll into the inside of the point set P, and the trace of the circle rolling is the boundary line of the point set P. If α is very large, the detected boundary line is the convex hull of the point set P. Similarly, the alpha-shape algorithm in three-dimensional space determines the boundary points by making a ball with a radius of α from three points, and creates triangular facets at the obtained boundary points to reconstruct the surface.

[0023] The present invention first performs denoising and filtering on the point cloud to separate ground points and non-ground points from the original vehicle-mounted laser radar point cloud; then uses the digital elevation model obtained from the non-ground points to perform elevation normalization on the non-ground points to obtain a canopy height model; then uses the CHM-based single tree segmentation algorithm to extract single trees, and uses the cylinder fitting algorithm to separate the crown and trunk of the single tree; then uses the alpha-shape algorithm to model the crown and trunk separately, and finally obtains a three-dimensional visualization model of trees in an urban scene. The process is as follows: Figure 1 shown.

[0024] The present invention obtains three-dimensional laser point cloud data of urban trees through a vehicle-mounted laser radar, and according to the different three-dimensional distribution characteristics of the trunk and crown point clouds, performs denoising and filtering, point cloud normalization, single tree segmentation, and trunk and crown separation to complete the three-dimensional visualization of the trees, and establishes a method for extracting trees from urban scenes based on the original laser point cloud and performing three-dimensional visualization.

[0025] In summary, compared with the traditional tree three-dimensional modeling method, the present invention can perform separate modeling according to the different morphological characteristics of the trunk and crown, avoiding the disadvantage of inaccurate expression of the morphological characteristics of single tree modeling, and can simply, quickly and accurately construct a three-dimensional model of urban trees. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic diagram of the process of the present invention;

[0027] Figure 2 This is a schematic diagram of the results of single tree segmentation;

[0028] Figure 3 The results of the separation of trunks and crowns are shown in Figure 1. a.plot1 shows the results of the trunk set; b.plot2 shows the results of the crown set;

[0029] Figure 4This is the alpha-shape schematic diagram;

[0030] Figure 5 These are the three-dimensional visualization results of trees; a.plot1 is the result of a single tree; b.plot2 is the result of all trees. DETAILED DESCRIPTION

[0031] The present invention is further described in detail below through the implementation of a sample plot and in conjunction with the accompanying drawings. The development environment of this example is Microsoft Visual Studio 2019, the external library PCL1.12.0, and the programming language C++.

[0032] A three-dimensional visualization method for urban trees based on vehicle-mounted laser point cloud data includes the following steps:

[0033] Step 1: Take the laser point cloud data of a school vehicle as an example to carry out the three-dimensional visualization of urban trees. The data is obtained using the SSW-2 vehicle-mounted laser scanner. The point cloud includes all ground feature information obtained by the vehicle-mounted system, such as the ground, street trees, buildings, green belts, basic public facilities, etc. Street trees include independent street trees and connected street trees. The point cloud density is an average of 201 points / m2. The acquired point cloud data is denoised using the cloudcompare software, and the denoised point cloud is divided into ground points and non-ground points using the CSF filtering algorithm.

[0034] Step 2: Use non-ground point data to interpolate and generate a DEM model with a resolution of 0.1m. The CHM model is obtained by subtracting the corresponding DEM elevation value from the non-ground point cloud elevation value. A single tree segmentation algorithm based on the CHM model is used to obtain 249 single trees. The trunk and crown are separated using a trunk extraction algorithm based on cylinder fitting. The ring height h is set to 0.1m, the ring fitting error threshold is 0.25, the radius change percentage threshold is 0.3, and the moving step is set to 0.1m.

[0035] Step 3: From the crown set, take out the crown of a single tree each time for surface reconstruction, set the alpha value to 0.35, and the rgb value to 0, 90, 0; from the trunk set, take out the trunk of a single tree each time, set the alpha value to 0.1, and the rgb value to 140, 90, 50, and splice to obtain the 3D model of a single tree. After the crowns and trunks of all single trees are modeled, the 3D visualization model of trees in the urban scene is obtained.

[0036] Step 4, accuracy test. After the above steps, the final tree 3D model is shown in the figure. From the model accuracy test, as shown in the figure, compared with the 3D point cloud, it can be seen that this method accurately extracts the single tree point cloud from the original point cloud and separates the trunk and crown. The 3D tree model restores the morphological characteristics of the tree well. However, due to the low density of the point cloud, there is still room for optimization for more detailed modeling of the interior of the crown.

[0037] This example processes the ground lidar data of the sample site according to the method proposed in the present invention. According to the method steps, the point cloud is de-noised, filtered, and single-tree segmented, and the trunk and crown are separated, all of which achieve good accuracy, indicating that the method has good effects in the three-dimensional visualization of trees in urban scenes.

Claims

1. A three-dimensional visualization method for urban trees based on vehicle-mounted laser point cloud data, comprising the following steps: Step 1: Denoising and filtering: Remove noise from the scene point cloud and separate ground points from non-ground points; Step 2: Single tree segmentation and trunk and crown extraction: The filtered data is normalized to obtain the canopy height model (CHM); the single tree segmentation algorithm based on CHM is used to extract the single tree; then the trunk and crown are separated from the single tree using a method based on cylinder fitting; Step 3: Surface reconstruction: Use the alpha-shape algorithm to construct three-dimensional models of the trunk and crown respectively. The steps are as follows: 3-1. Set distance threshold α 3-2. Construct a point set Q = {qi}, (i = 1, 2, ... n) to be modeled. Select a point q1 from it, and the points whose distance to it is less than 2α form a new point set Q1. Select a set of points q2 and q3 from the point set Q1, and find a sphere with a radius of α passing through points q1, q2, q3 and the centers o1 and o2. 3-3. Traverse the point set Q1, and find the distance sets L1 and L2 from other points to the sphere centers o1 and o2 in turn; if the distance of one of the sets in L1 and L2 is greater than α, then q1, q2, and q3 are edge contour points, and the three points are connected to form a boundary triangle patch; otherwise, they are not edge contour points, stop traversal, and execute the next step; 3-4. Select the next set of points in the point set Q1 and judge according to steps (3-2) and (3-3) until all points in the point set Q1 are judged; output a set of triangular facets, and the exposed triangular facets in the set form a local convex hull; 3-5. Select the next point in the point set Q and judge according to steps (3-2), (3-3), and (3-4) until all points in the point set Q are judged, and multiple convex hulls reconstruct the surface S(Q); Step 4: 3D visualization: Combine the trunk and crown models corresponding to individual trees to complete the three-dimensional visualization of urban trees.

2. The method for three-dimensional visualization of urban trees based on vehicle-mounted laser point cloud data according to claim 1, characterized in that: In step 1, the CSF algorithm is used to separate ground points and non-ground points in the scene point cloud.

Images