WO2015149302A1

WO2015149302A1 - Method for rebuilding tree model on the basis of point cloud and data driving

Info

Publication number: WO2015149302A1
Application number: PCT/CN2014/074634
Authority: WO
Inventors: 张晓鹏; 李红军; 郭建伟; 代明睿; 刘佳
Original assignee: 中国科学院自动化研究所
Priority date: 2014-04-02
Filing date: 2014-04-02
Publication date: 2015-10-08

Abstract

Disclosed is a method for rebuilding a tree model on the basis of a point cloud and data driving. The method comprises the following steps: tree point cloud data are acquired and preprocessed, and classification representation of the tree model is defined; a cylinder moving method is provided and used for extracting main branch framework points from the tree point cloud data, and a branch and leaf separation process is carried out; crown feature points are extracted from the tree point cloud data; a classification ion flow method is provided and used for structuralizing the main branch framework points and the crown feature points; the complete tree model is obtained by means of rebuilding on the basis of the structuralized framework points and radiuses of all the branches. A solution is provided for rebuilding the complete tree model in the three-dimensional point cloud data, the obtained rebuilt model and an original point cloud have a high degree of fit, and the good rebuilding result can be obtained on models which are severely blocked and are complex in form.

Description

FIELD OF THE INVENTION The present invention relates to the field of cross-technology of plant modeling and computer graphics processing, and relates to physical point measurement using a three-dimensional laser scanner to obtain tree point cloud data, in particular, a scan from A method for reconstructing a complete 3D tree model from 3D point cloud data.

BACKGROUND OF THE INVENTION Plants play an important role in our daily lives, both to regulate ecological balance, to green the environment, and to clean the air. Accurate reconstruction of plant models can be applied in many areas, such as directing agricultural forestry production, protecting endangered old trees, or providing virtual environments in digital cinema and entertainment games. At present, the techniques of plant modeling can be roughly divided into four categories: rule-based methods, geometric analysis-based methods, sketch-based methods, and tree-based methods.

In recent years, due to the rapid development of digital means, the reconstruction method based on tree digitization has gained more and more attention in plant reconstruction. At present, tree photographs, point clouds, etc. obtained by tree digitization are mainly used as input data, and some Knowledge and rules are obtained to obtain a plant model similar to the input data. In addition to obtaining better visual effects, the accuracy of the reconstruction model has gradually become one of the goals of reconstruction. Among them, the photo-based reconstruction method has the advantages of convenient data collection and good visual effect of reconstructing the model, but the preparation work is too much, the manual interaction is large, and the reconstruction model and the real model can only remain at a certain angle or a certain angle, and It does not reflect the true form of the plant. The method based on 3D scanning point cloud takes the 3D scan data of the tree as input, and the geometric information is rich, the precision is high, and the model with much higher precision than the photo can be obtained. Cheng2007 (Z. Cheng, X. Zhang and B. Chen, "Simple Reconstruction of Tree Branches From a Single Range Image," Journal of Computer Science and Technology, vol. 22, no. 6, pp. 846C858, 2007 ) In the method of reconstructing the point cloud data of the tree, the crown portion is separated from the trunk portion by manual segmentation, and for the trunk portion, the branch is first detected by the depth of the scanned image. The separation between the two, and then the cylindrical point to obtain the skeleton point and the radius corresponding to the skeleton point, the method can obtain relatively accurate skeleton position and branch radius, but can not reconstruct the twigs and crown. Neubert2007 (Neubert, T. Franken and O. Deussen, Approximate Image Based Tree-Modeling Using Particle Flows, ACM Transactions on Graphics (TOG). ACM, 26(3): 88, 2007. ) Using particle flow methods for data In the driven tree modeling, they first extract the directional field from the two orthogonally photographed trees, and then use the particle flow method to connect the crown to the main branch along the directional field, so that the reconstructed model is related to the input photo. Match. However, this method uses the motion of a single layer particle stream and is therefore only applicable to secondary structures. The present invention extends the method by using a hierarchical particle flow method to extract a multi-level structure of trees, and the present invention reconstructs a model from point cloud data. Livny2010 (Y. Livny, F. Yan, M. Olson, B. Chen, H. Zhang and J. El-Sana, "Automatic Reconstruction of Tree Skeletal Structures from Point Clouds," ACM Trans. Graph, vol.29, no .151, pp. 1C8, 2010. ) A method based on global fitting optimization is proposed to extract the skeleton structure of point cloud data. This method is more robust, but it still cannot reconstruct the master accurately in the case of severe occlusion. The skeleton of the branches and crowns. Livny2011 (Y. Livny, S. Pirk, Z. Cheng, F. Yan, O. Deussen, D. Cohen-Or and B. Chen. Texture-lobes for Tree Modelling, ACM Transactions on Graphics (Proceedings of SIGGRAPH 2011), Vol.30, no.4, 2011.) A method for representing trees based on splinters is proposed. The method can be used for reconstruction of trees or for re-presenting existing tree models. The present invention has been carried out on this method. The expansion, taking into account the information of the canopy and the visible main branch, and filling the canopy and the main branch by the method of hierarchical particle flow.

SUMMARY OF THE INVENTION The present invention provides a tree model reconstruction method based on point cloud and data driving, so as to solve the problem that the existing tree reconstruction method is not accurate, and can only process a relatively simple model, which is difficult for a model with severe occlusion and complex shape. The disadvantage of obtaining better reconstruction results.

To achieve the above object, the present invention provides a point cloud and data driven tree model reconstruction method, the method comprising the following steps: Step S1, obtaining tree point cloud data, pre-processing it, and defining a hierarchical representation of the tree model;

Step S2: extracting a main branch skeleton point and a radius thereof from the tree point cloud data, and performing branch and leaf separation processing;

Step S3: extracting a crown feature point from the tree point cloud data;

Step S4, structuring the main branch skeleton point;

Step S5: structuring the crown feature points and calculating the radius of each branch; Step S6, reconstructing a complete tree model according to the skeleton points and radii of all the branches that have been structured.

The invention adopts the technology of computer graphics processing to reconstruct a complete tree model from the scanned three-dimensional tree point cloud data. The invention automatically and accurately in a complex point cloud through comprehensive analysis of the local geometrical relationship of the point cloud and the spatial position relationship. In the data, the skeleton point of the main branch is calculated and the radius is obtained, and the crown shape is completely restored. A large number of fine branches are simulated in accordance with the biological characteristics, so that the reconstruction model obtains a higher realism on an accurate basis.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a flow chart of a method for reconstructing a point cloud and a data driven tree model according to the present invention; FIG. 2 is a view showing a point cloud data of a white pine tree obtained according to an embodiment of the present invention; FIG. Tree grading indicates a schematic diagram;

4 is a schematic view of a cylinder search space in accordance with an embodiment of the present invention;

5 is a flow chart of a cylinder movement search method according to an embodiment of the present invention;

6 is a main branch skeleton point extracted by a fractional ion current according to an embodiment of the present invention; FIG. 7 is a schematic diagram showing the result of branch and leaf separation according to an embodiment of the present invention;

8 is a schematic diagram showing the result of extracting the feature points of the canopy according to an embodiment of the present invention; FIG. 9 is a schematic diagram showing the structure of the main branches and the crowns according to an embodiment of the present invention; FIG. 10 is a reconstruction according to an embodiment of the present invention. a complete white pine tree model; Figure 11 is a comparison of the results of the present invention with the prior art in the case of input point cloud data occlusion; Figure 12 is a graphical representation of the results obtained by reconstructing a scanned forest point cloud using the present invention.

DETAILED DESCRIPTION OF THE INVENTION In order to make the objects, the technical solutions and the advantages of the present invention more comprehensible, the present invention will be described in detail below with reference to the accompanying drawings.

1 is a flow chart of a point cloud and data driven tree model reconstruction method according to the present invention. As shown in FIG. 1, the method of the present invention includes the following steps: Sl, tree point cloud data acquisition and preprocessing, and defining a tree model. Graded representation; S2, main branch skeleton point extraction and branch and leaf separation; S3, canopy feature point extraction; S4, main branch skeleton point structure; S5, canopy feature point structure; S6, complete model reconstruction.

Each of the above steps will be described in more detail below.

Step S1: acquiring tree point cloud data, pre-processing the same, and defining a hierarchical representation of the tree model;

To achieve real plant 3D modeling, we must first obtain the original model of the plant. At present, there are two main methods for obtaining the original model of the plant, which are to obtain the image model by camera and the 3D point cloud model by laser scanner. The invention utilizes a three-dimensional laser scanner (eg

Cyrax), tools such as visual photography capture tree point cloud data, point cloud analysis techniques to preserve point clouds on single and multiple trees that need to be reconstructed, and to remove points from other objects, ie preprocessing. 2 is a white pine tree scanning point cloud data acquired according to an embodiment of the invention.

In addition, according to the knowledge of botany, the present invention classifies the shape of the tree: main branches, twigs and leaves, wherein the main branches contain 2 to 4 branches, and the twigs contain 2-5 branches, the specific series and Tree species and tree age are related. Generally speaking, there are many grades of tree age. The number of main branches is NL, the number of twigs is NS, and twigs and leaves are combined to form a canopy. 3 is a schematic diagram showing hierarchical representation of trees according to an embodiment of the present invention, taking 5 levels as an example, wherein FIG. 3A shows the main structure of trees, FIG. 3B shows elements of level 5, and FIG. 3C shows trees of level 5. model.

Step S2: extracting a main branch skeleton point and a radius thereof from the tree point cloud data, and performing branch and leaf separation processing; The first step is to extract the main branch skeleton point from the point cloud data of the tree, and use the extracted main branch skeleton point to realize the separation of the branches and leaves, and then complete the structure of the main branch and the canopy in the subsequent steps.

The step S2 includes the following steps:

Step S2.1: calculating a local geometric quantity at each point p of the point cloud;

The local geometric quantity at each point P of the point cloud includes a normal direction Q^, a principal curvature (» and k ₂ (p) (fc ₁ (p) < /c ₂ (p)), and a main direction corresponding to the main curvature 3⁄4) and 0).

The step S2.1 further includes the following steps:

Step S2.1.1, establishing a kd tree of the entire point cloud data;

In computational geometry, the kd tree is one of the fastest data structures that have been proven to find neighbors. The method of building a kd tree is not described here.

Step S2.1.2, for each point p in the point cloud data, use the kd tree to find multiple of them, such as 15 or 30 neighbors, assuming that the neighbors are from the same plane, using the least squares method Combine this plane, using the normal vector of this plane as the normal direction n of point p (

Step S2.1.3, establish a local coordinate system at point p, and fit a quadric surface (u, v) = c ₀ u ² + 2c uv + c ₂ v ² , where (u, v, h( u, i;)) is a point in the space, c. , c , c ₂ are surface coefficients. After obtaining the coefficient of the quadric surface, the principal curvature / and fc ₂ (p) at the point P and the main directions 3⁄4 and 3⁄40?) are calculated using the quadric surface. ;

Step S2.1.4. Define the radius of the curvature circle at point p as r(p) = l/k ₂ (p), and the center of curvature is b(p) = p- r(p) (p). , with 3⁄43⁄4? Approximate to the axial direction of the branch at p, r(p) approximates the radius of the branch at that point, and b(p) is estimated to be the position of the trunk axis at that point, also known as the branch skeleton point.

Step S2.2, defining a cylinder fitting the p-shaped branch shape based on the local geometric quantity at each point p of the point cloud, and searching for a potential branch (ie, moving the cylinder method) by using the movement of the cylinder, thereby extracting Obtaining the main branch skeleton point and its radius;

As shown in Fig. 4, a cylinder at a point p is defined as S(q) = S = (b, r, 3⁄4H), where b, r is the local geometric quantity calculated in the step S2.1: The center of curvature, the radius of curvature and the main direction represent the center of the bottom of S, r is the radius of the bottom circle, 3 is the axial direction, and H is the height. Degree is a fixed value. The extended cylinder of s is denoted by T(S) = α ^, ζ ), where 入 is a certain value, such as 1.3, TS) defines the search space for finding potential main branch skeleton points, which will be located in TS) The scanned data points are recorded as a set s).

5 is a flow chart of a cylinder movement search method according to an embodiment of the present invention. As shown in FIG. 5, the cylinder movement search method includes the following steps:

Step S2.2.1. Construct the initial cylinder: Select a point p from the point cloud data. , construct the initial cylinder S(p ₀ ) = S( ¹ ) = (b( ¹ ), r( ¹ ) ( ¹ ), H( ¹ )), where b (D represents the initial cylinder S (the bottom surface of D) Center, r (D is the initial cylinder S (D of the bottom circle radius of D, 3 (D represents the initial cylinder S (the axial direction, H (D represents the initial cylinder S (D height, initial cylinder S (expansion) The cylinder is T(S(D), TCS^) = ω, λιΓ ⁽¹⁾ , ά ⁽¹⁾ , 2Η ω), and (S(D) represents all points in T(S ⁽¹⁾ );

Step S2.2.2, first determine whether any point q in the set Sd)) is valid, that is, if the cylinders S(p.) and S(q) satisfy: 1) b(q) 6 S(p.) 2) r(q)/r(p ₀ ) - 1.0; 3) The angle between ϊ and ϊ(ρ.) is small, where b(q) represents the center of curvature of the cylinder S(q), r (q) represents the radius of curvature of the cylinder S(q), r(p.) represents the radius of curvature of the cylinder S(p.), represents the axial direction of the cylinder S(q), and ϊ(ρ.) represents the cylinder In the axial direction of S(p.), the point q is referred to as a valid point; if the number of effective points in the set Sd)) is greater than the first threshold N _x (set to 6 in an embodiment of the present invention), the initial The cylinder S (the successor cylinder S( ² ) of D exists, and its parameters bG), rG), ² ) are the average of the center of curvature, the radius of curvature and the main direction of all points in the set SW), respectively, and then S All points in ( are marked as processed and are no longer accessed in subsequent steps;

Step S2.2.3, pay attention to the initial cylinder S (D has an axially opposite associated cylinder S(-1) = (b(i), r(D, - SW'Hd)), S(D and S (-D is axially opposite, so look for the successor cylinder S(_ ² ) from S(-D) according to the method of step S2.2.2;

Ho step S2.2.4, the S ⁽²⁾ or S - continues until the search does not find any subsequent cylinder ⁽²⁾ on the basis of the number of cylinders if found N ₂ than the second threshold value (in the embodiment of the present invention, a In the example set to 6), the cylinders form a main branch segment, and the center of the circle is recorded as the main branch skeleton point; Step S2.2.5, reselect an unprocessed point in the point cloud data and repeat the above four steps until all points are marked as processed, and an ordered cylindrical sequence is obtained, which is Moving the sequence of the cylinder, we also get an ordered skeleton point sequence T, = ( 1 = 1 η }, where each (representing a section of the main branch,) represents the first section of the main branch, indicating that the j-th branch contains the skeleton point Number of.

Figure 6 is a schematic diagram of the main branch skeleton points extracted in accordance with an embodiment of the present invention. In Figure 6, these skeleton points are not completely connected, and each skeleton point is represented by its corresponding cylinder (including the center and the radius of the mouth).

Step S2. 3. Using the extracted main branch skeleton points, the point cloud data is divided into two parts: a point cloud on the main branch and a point cloud on the crown to realize separation of the branches and leaves;

In this step, the point cloud data is divided into a point cloud on the main branch and a point cloud on the crown. The point cloud located in the column search area is considered to be a point cloud located on the main branch, and other point clouds are considered It is a point cloud located in the canopy. Since the search radius of the cylindrical search area is λ^Ο^, where rp) is the radius of the cylinder, which is considered to be the radius of the main branch at the position, in an embodiment of the present invention, the input is ^, so the search area of the cylinder is more realistic. The range of branches is large, and while the point clouds on the main branches are all located in the search area, some of the point clouds on the crown are enveloped. But this does not adversely affect the accuracy of the final result, because the result of the separation of the main branch from the canopy point cloud makes us pay more attention to the part of the point cloud located in the canopy, a small amount of missing will not affect the shape and periphery of the canopy point cloud The contours have an effect. Figure 7 shows the results of the separation of the branches of the Pinus bungeana according to an embodiment of the present invention. In Figure 7, the point cloud (the portion of the gray scale) on the main branch is recorded as ^ ^, the point cloud on the crown (the gray scale is bright) Part of it).

Step S3, extracting a crown feature point from the tree point cloud data;

In an embodiment of the present invention, a multi-scale method is constructed to extract a crown feature point in the tree model. In the method, each scale corresponds to a first-level representation of the tree, that is, the NS scales Ω ₂ , ..., n _NS corresponds to the NS-level twig representation of the tree, and ni

For each scale Ω, divide the bounding box of the crown point cloud into several uniform and equal sizes (the size of the cube. When a cube meets the following two conditions, it is effective, that is, the center of gravity of the 1⁄2 is the crown A feature point on the scale: 1) The number of points belonging to ^ in 1⁄2 cannot be less than the third threshold N ₃ , in the present invention In one embodiment, N _{3 is} set to 5; 2) 1⁄2 has no points belonging to ^ ^. The list of valid cubes corresponding to the scale is ^: = . Calculating the center of gravity of all 1⁄2 in each scale is calculated, then all the centers of gravity ^ are considered to be all the feature points of the canopy on the scale Ω. FIG. 8 is a schematic diagram showing the results of extracting feature points of a canopy according to an embodiment of the invention. In the figure, points with brighter gray scales (shown by small balls) are feature points, and these feature points are evenly distributed in space, and can be better. The ground shows the shape of the canopy.

Step S4: structuring the main branch skeleton point;

The main branch skeleton points obtained by the foregoing steps do not have a complete connection relationship, that is, the skeleton points inside each of the obtained main branches have a connection relationship, but there is no connection information between the main branches and the main branches, and three of the crowns are The feature points of the scale are also completely discrete. Therefore, in an embodiment of the present invention, the skeleton points are effectively and accurately structured by using a three-dimensional hierarchical particle flow motion method to construct a hierarchical branch.

For the main branch part, first set a root node G (you can select the position of the tree model close to the ground) to guide the particle flow motion, and then to the ordered list of ordered skeleton points τ, - = {c , h = 1 , .. Each branch in η places the particle ^ with its lower end point as the initial position of the particle, and finally uses the remaining main branch skeleton point as the attraction to guide the particle to run in space, find a suitable connectable point, and use the particle The trajectory is used as a supplement to the main branch to connect the scattered branches to achieve the structuring of all the main branch points. Among them, each particle in the process of searching for connectable points, before tapping its own connectable points, is constantly performing a cone-like hierarchical search. The purpose of the cone-shaped hierarchical search is to find the branches that can be connected. As follows: Let the representative particle ρ / in the speed direction of 1 ,, ρ / record at the position of 1 ,, that is, the trajectory point of the particle running, for the particle located at ρ / position, ρ / as the apex, as the axis, Take ι. High, to. Establish a cone for the angle between the busbar and the axis

F. If there is a branch skeleton point other than (in this cone), select the point with the smallest angle among these skeleton points in this area as the "connectable point" of W, if there is no search in the cone Connect the point, and then use ρ/ as the apex, and then as the axis, with h! = h _up + ι. _Be high, with 6^ = θ _ηρ + 6>. The angle between the busbar and the axis, where 1 ^ is the height Increase the amount, for the increase of the angle, then create a new cone, in which to find the connectable point again, if still not found and < l _max , ^ < ^n _ax , then go ahead Expand the height and angle, repeat the above search process, where / 1⁄2 ^ and ^ are the maximum height and angle. After searching for the connectable point g _attaeh , the cone-shaped area search will not be performed during the subsequent movement, and the g _atta record is taken as the attraction point of ?^.

Step S5: structuring the crown feature points and calculating the radius of each branch; after fully connecting the main branch skeleton points, in the step, the corresponding tree first-level canopy feature points are used as the starting point of the particle flow To the main branch skeleton point, the trajectory of the particle running is used as the skeleton point of the fine branch; the crown feature points corresponding to other levels of the tree are moved as the starting point of the particle flow to the upper primary skeleton point, and the trajectory of the particle still remains. As the skeleton point of the level, the structuring of all the feature points is realized; finally, the radius of each branch is estimated according to the radius of the main branch.

The structure of the canopy feature points has a great similarity with the structure of the main branch skeleton points. First, set the canopy attraction point X (for example, you can select one third of the tree height above the root node G), and then use NS respectively. The crown feature points of the scales are the particles at the beginning of the particle flow, and the NS scale particles are divided into NS batches to move toward the main branch skeleton point under the guidance of the main branch skeleton point and the attraction point X, so as to find a suitable one. The connection point, and finally all run to the main branch skeleton point, the trajectory of the particle running as the skeleton point of the twig. The particles at the i scale all run to the main branch skeleton point to form the (NL+l) level branch, and the particles at the Ω ₂ scale all run to the NL+l) branching skeleton point (NL+2). The branching, and so on, the particles at the final scale all run to form twigs. This process completes the complete structuring of the canopy feature points and the main branch skeleton points. 9 is a schematic diagram showing the results of structuring of the branches of the main branches and the crowns according to an embodiment of the present invention.

Since the main factor of the tree reconstruction effect is the radius of the trunk of the visible part of the main branch, the radius of the twig does not have a large influence on the application of the model. Therefore, the present invention sets the radius of the root of the twig to a fixed multiple of the radius of the parent branch. For example, 0.7 times, and the tip radius of the twig is set to be a fixed σ. In an embodiment of the invention, σ is set to 0.2 cm. The radius of the intermediate skeleton point of the twig is obtained by linear fitting.

Step S6: reconstructing a complete tree model according to the skeleton points and radii of all the branches that have been structured. In this step, firstly, using the skeleton structure and radius of all the branches that have been structured, the branches are fitted with cylinders of different radii (triangular mesh model), and the three-dimensional mesh model of the trees is reconstructed; then texture synthesis is adopted. The method adds a suitable texture to the mesh model (the texture addition method belongs to the texture processing method commonly used in the prior art, which will not be described here), and then according to the size of the input point cloud data and the type of the leaf (broad or coniferous) The statistical estimation method is used to determine the leaf information such as the number of leaves required, the length and width of the leaves, and different leaves are added to the end of the three-dimensional mesh model twig, thereby realizing the complete reconstruction of the tree three-dimensional model.

Figure 10 is a complete three-dimensional tree model reconstructed from scanned white-skin tree point cloud data using the method of the present invention, which includes the main branches, canopies, and leaves. As can be seen from Figure 10, the reconstructed tree model is very consistent with the shape of the input point cloud, which is reflected in two aspects: 1) The point cloud on the main branch is in good agreement with the reconstructed model, which illustrates this The main branch reconstruction of the invention is accurate in both the position and the radius of the skeleton point; 2) The crown portion completely fills the entire canopy space, and accurately reduces the concave and convex features of the crown. In addition, for the occluded point cloud data shown in FIG. 11A, the method of Livny 2010 (shown in FIG. 11B) extracts the branches affected by the input point cloud density, and for the densely packed areas, the branches grow easily along the canopy. The surface crawls, and the method of the present invention (as shown in Fig. 11C) can accurately calculate the position of the main branch and the canopy feature points, and the accuracy of calculating the branch radius is higher, and the final result also has a good aesthetic appearance.

Furthermore, the method of the invention can also be easily applied to model reconstruction aspects of trees. The difference between the model reconstruction method of the forest point cloud data and the reconstruction method of the single tree is that the root node G and the canopy attraction point X are first set, and the rest are similar. The present invention establishes two sectional planes parallel to the ground plane from the ground plane at a height of 1/4 and 1/3 of the ground plane in space, and the skeleton point between the section planes is called an alternative point attraction point. During the movement of the particle flow structured by the feature points, each point automatically searches for the closest point to its own Euclidean distance from the candidate attraction points during the motion as its own attraction point, so as to attract the point at the ground plane. The upper projection point is its own root node. After the root node and the attraction point search method are determined, the feature point is structured by the particle flow motion. Fig. 12A is input forest point cloud data, and Fig. 12B is a forest model obtained by reconstruction using the method of the present invention. The method of the invention automatically and accurately calculates the skeleton points of the branches in the complex tree point cloud model and obtains the accurate radius by comprehensively analyzing the local geometrical relationship of the point cloud and the spatial position relationship, and changes the manual interaction in the previous trunk extraction algorithm. More, the workload is large, and the results are not accurate enough. At the same time, the method of grading particle flow motion automatically realizes the connection of the main branch skeleton points and the comprehensive characterization of the canopy feature points and the skeleton skeleton points, which completely restores the crown shape. And it is in line with the biological characteristics to simulate a large number of twigs, so that the reconstruction model obtains a higher sense of reality on an accurate basis.

The above experimental results and the method based on point cloud and data-driven tree model reconstruction can be applied to agricultural and forestry measurement, tree protection, digital entertainment and virtual scene simulation applications, and have high practical application value.

The specific embodiments of the present invention have been described in detail with reference to the preferred embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and scope of the present invention are intended to be included within the scope of the present invention.

Claims

Rights request

1. A point cloud and data-driven tree model reconstruction method, characterized in that the method includes the following steps:

Step S1: Obtain tree point cloud data, preprocess it, and define a hierarchical representation of the tree model;

Step S2 extracts the main branch skeleton points and their radius from the tree point cloud data, and performs branch and leaf separation processing;

Step S3: Extract crown feature points from the tree point cloud data;

Step S4: Structure the main branch skeleton points;

Step S5: Structure the crown feature points and calculate the radius of each branch; Step S6: Reconstruct the complete tree model based on the skeleton points and radii of all structured branches.

2. The method according to claim 1, characterized in that, in the step S1, a three-dimensional laser scanner and a visual photography tool are used to obtain the tree point cloud data, and the individual sums that need to be reconstructed are retained based on the point cloud analysis technology. Point clouds on multiple trees, removing points from other objects.

3. The method according to claim 1, characterized in that, in the step S1, the shape of the tree is divided into main branches, twigs and leaves, wherein the number of grades of the main branches is recorded as NL, and the number of grades of the twigs is NL. Numbered as NS, the twigs and leaves form the crown.

4. The method according to claim 1, characterized in that, in the step S2, the main branch skeleton point and its radius are extracted from the tree point cloud data using a moving cylinder method, and the step S2 is performed. One step includes the following steps:

Step S2.1. Calculate the local geometric quantities at each point p of the point cloud: normal direction p principal curvature ^^^ and ^ ), where kp)≤k ₂ (p), and the principal direction corresponding to the principal curvature 5 (ρ )and

¾P) ;

Step S2.2: Define a cylinder that fits the branch shape at point p based on the local geometric quantity at each point p in the point cloud. Use the movement of the cylinder to search for potential branches, thereby extracting the main branch skeleton points and its radius;

Step S2.3: Use the extracted main branch skeleton points to divide the point cloud data into main branches. The point cloud on the tree and the point cloud on the tree crown realize the separation of branches and leaves.

5. The method according to claim 4, characterized in that step S2.1 further includes the following steps:

Step S2.1.1. Establish a kd tree of the entire point cloud data;

Step S2.1.2. For each point p in the point cloud data, use kd tree to find its multiple neighboring points. Assume that these neighboring points come from the same plane, use the least squares method to fit the plane, and use this The normal vector of the plane is used as the normal direction of point p);

Step S2.1.3. Establish a local coordinate system at point p, and obtain a quadratic surface by fitting. Use this quadratic surface to calculate the principal curvature ^ at point p! and ^ and the main direction 5¾ρ) Step S2.1.4. Define the radius of the curvature circle at point p as r(p) = l/k p), and the center of the curvature circle as b(p) = p-r(p)n(p). Use ¾ represents the axial direction of the branch at p, r(p) represents the radius of the branch at this point, and b(p) represents the position of the trunk axis at this point, that is, the branch skeleton point.

6. The method according to claim 4, characterized in that step S2.2 further includes the following steps:

Step S2.2. Select any point p from the point cloud data. , construct the initial cylinder S(p ₀ ) = = ( ^,τ^,ά^, ^), where b« represents the center of the bottom surface of the initial cylinder _S C0, r(D represents the bottom surface of the initial cylinder S(D The radius of the circle, ^ ¹ ) represents the axial direction of the initial cylinder S(, H(D represents the height of the initial cylinder S(D), and the expanded cylinder of the initial cylinder S(D) is recorded as T(S ⁽¹⁾ ),

, where, is a certain value, and the set (S ⁽¹⁾ ) represents all points in T (S ⁽¹⁾ );

Step S2.2.2. Determine whether any point q in the set Sd)) is valid. If the number of valid points in the set Sd)) is greater than the first threshold, it is considered that the successor cylinder S of the initial cylinder S ( exists, and Its parameters b( ² ), r( ² ) ( ² ) are respectively the average of the curvature center, curvature radius and main direction of all points in the set (S«), and all points in the initial cylinder S( are marked as Processed;

Step S2.2.3. Starting from the companion cylinder _s (- 1) ₌ ω, _Γ ω, — 3ω, _Η ω) which is opposite to the axial direction of the initial cylinder S ( ), follow step _S 2.2.2 to find the successor cylinder. s( ^-2 ) _; Step S2.2.4. Continue the search based on the successor cylinder S( ² ) or its companion cylinder S( ^-2 ) until no successor cylinder is found. If the number of found cylinders is greater than the second threshold N ₂ , then these cylinders form a main branch segment, and the center of the circle is used as the main branch skeleton point; Step S2.2.5, re-select an unprocessed point in the point cloud data and repeat the above four steps until all The points are all marked as processed. At this time, an ordered cylinder sequence and an ordered skeleton point sequence η = {C[, h = 1, _nj } are obtained, where each c represents a section of the main branch, and j represents the first branch. Segment main branch, nj represents the number of skeleton points included in the _jth segment main branch.

7. The method according to claim 6, characterized in that in step S2.2.2, if cylinders S(p.) and S(q) satisfy: l ) b(q) 6 S(p ₀ ); 2) r(q)/r(p ₀ ) - 1.0; 3) The angle between (q) and ^(p.) is small, where S(q) represents the cylinder at the point, b (q) represents the center of curvature of cylinder S(q), r(q) represents the radius of curvature of cylinder S(q), r(p.) represents the radius of curvature of cylinder S(p.), (q) represents The axial direction of cylinder S(q), (ρ.) represents the axial direction of cylinder S(p.), then point q is a valid point.

8. The method according to claim 4, characterized in that, in step 2.3, the point cloud located in the cylindrical search area is considered to be a point cloud located on the main branch, and the other point clouds are considered to be points located on the crown of the tree. cloud.

9. The method according to claim 1, characterized in that, in the step S3, a multi-scale method is used to extract the crown feature points, wherein each scale corresponds to a first-level representation of the tree, NS scales Ω ₂ , ..., Ω _N5 respectively correspond to the NS-level twig representation of trees, and.丄〉. ? 〉…〉. ^; For each scale, divide the bounding box of the crown point cloud into a number of uniform and equal-sized cubes {V The center of gravity bi of the effective ¼ is considered to be a characteristic point of the tree crown on the scale; Calculate the center of gravity of all effective ¼ at each scale b Obtain all characteristic points of the tree crown on the scale.

10. The method according to claim 9, characterized in that when a cube voxel ¼ satisfies the following two conditions, it is said that ¼ is effective:

1) The number of points belonging to A in ¼ is not less than the third threshold N _3;

2) There is no point belonging to _b in ¼.

11. The method according to claim 1, characterized in that, in the steps S4 and S5, a three-dimensional hierarchical particle flow motion method is used to perform the main branch skeleton points and crown feature points. Row structuring, gp:

1) Comprehensively connect the main branch skeleton points;

2) Use the corresponding first-level crown feature point of the tree as the starting point of the particle flow to move to the skeleton point of the main branch, and the trajectory of the particle running as the skeleton point of the thin branch at that level;

3) Use the canopy feature points corresponding to other levels of the tree as the starting point of the particle flow to move to the main branch skeleton point of the previous level, and the trajectory of the particle running is still used as the skeleton point of that level;

4) Finally, the radius of each branch is estimated based on the radius of the main branch.

12. The method according to claim 1, characterized in that, in step S6, the structure and radius of all structured branches are first used to simulate the branches using cylinders of triangular mesh models with different radii. Combined, reconstruct the three-dimensional mesh model of the tree; then add appropriate textures to the mesh model; finally, according to the scale of the input point cloud data and the type of leaves, determine the required leaf information, and add different leaves to all The three-dimensional mesh model describes the ends of the twigs, and finally achieves a complete reconstruction of the three-dimensional tree model.