CN114862886A - Street tree single tree segmentation method based on MLS point cloud data - Google Patents

Abstract

The invention provides a street tree single tree segmentation method based on MLS point cloud data, which comprises a street tree point cloud obtaining step, a street tree point cloud clustering step, an adhered street tree detection step, an adhered street tree rough segmentation step and a sub segmentation step, and single tree segmentation is completed. The method comprises the steps of dividing a street tree point cloud into street tree clusters by adopting a clustering algorithm, detecting adhered street trees for the street tree clusters, and then carrying out thick-to-thin multi-layer division on the street tree clusters containing a plurality of adhered street trees to obtain a street tree single-tree point cloud; the method has high street tree point cloud identification capability and street tree single tree segmentation precision, and is suitable for complex urban environments containing various ground features.

Description

Street tree single tree segmentation method based on MLS point cloud data

Technical Field

The invention relates to the technical field of street tree single tree segmentation, in particular to street tree single tree segmentation based on MLS point cloud data, and specifically relates to a street tree single tree segmentation method for recognizing and then segmenting street tree point cloud data.

Background

The street tree is used as an important component of an urban ecological system and an urban landscape, and has important ecological and social service functions. The more intensive research on the street trees is the work which needs to be performed urgently by city management departments and digital city construction. A Mobile Laser Scanning (MLS) system is used as an important means for collecting urban close-range three-dimensional space information, and has the capability of rapidly acquiring three-dimensional information of street trees on two sides of a road.

The street tree single tree segmentation refers to identifying and segmenting a single street tree from MLS point cloud data, and is the first step of subsequent research of the street tree, and the existing methods are mainly divided into two types:

and (5) dividing and identifying the edges. And (4) gradually filtering the non-street tree point cloud by adopting an image/point cloud segmentation method according to the appearance characteristics of the street trees and other ground object targets. The method mainly utilizes the priori knowledge to formulate rough segmentation/identification characteristics and rules, and has weak identification capability on the street tree, thereby reducing the accuracy of the single tree segmentation of the street tree.

And identifying and then segmenting. Firstly, a road tree detector generated by a supervised learning algorithm is adopted to identify road tree point cloud from MLS point cloud data, and then algorithms such as clustering and the like are adopted to divide the road tree point cloud into individual road trees. The method has higher street tree identification precision, but the identified street tree single tree segmentation algorithm, particularly the segmentation algorithm of the adhered street tree, needs to be further researched.

Disclosure of Invention

The invention aims to provide a street tree single tree segmentation method for identifying and segmenting street trees, which aims at the problem of street tree adhesion.

The technical scheme of the invention is as follows:

the invention provides a street tree single tree segmentation method based on MLS point cloud data, which comprises the following steps: s1, acquiring the street tree point cloud: acquiring street tree point cloud data through street MLS point cloud data scanned by a mobile laser;

s2, road tree point cloud clustering: clustering the road tree point cloud by adopting a clustering algorithm, and dividing the road tree point cloud into road tree clusters;

s3, detecting the adhered street trees: performing trunk point cloud extraction and trunk point cloud clustering on each street tree cluster, detecting whether the street tree cluster contains a plurality of adhered street trees, executing S4 on the adhered street tree clusters containing the plurality of adhered street trees, and finishing single-tree segmentation on the non-adhered street tree clusters;

s4, coarse segmentation of the adhered street tree: roughly dividing the adhered street tree cluster into single street trees;

s5, fine segmentation of the adhered street trees: and (4) finely dividing the single street trees obtained by the rough division of the S4 to finish the single tree division.

Further, the S1 specifically includes: a street tree detector is used to identify a street tree point cloud from the street MLS point cloud data.

Further, the clustering algorithm adopts a density-based noise application clustering algorithm DBSCAN.

Further, the S2 specifically includes:

taking the (x, y, z) coordinates of the street tree point cloud as input, clustering the street tree point cloud by adopting a clustering algorithm, and dividing the street tree point cloud into street tree clusters; wherein the neighborhood radius ε ₁ 0.5 m; number of minimum neighborhood points N ₁ ＝C ₁ m ₁ ，m ₁ Is the neighborhood radius ε ₁ Average neighborhood point number of all street tree points at 0.5m, C ₁ Is constant and takes 0.02-0.1.

Further, the S3 specifically includes:

s3.1, trunk extraction is carried out on each street tree cluster, and the minimum value z of the point cloud elevation in the current street tree cluster is obtained _min Extracting the elevation value and the minimum elevation value z in the cluster _min Taking the point cloud with the difference smaller than the height difference threshold value as the trunk point cloud;

s3.2, clustering the trunk point cloud by using the (x, y) coordinates of the trunk point cloud as input and adopting a clustering algorithm, and dividing the trunk point cloud into trunk clusters; wherein the neighborhood radius ε ₂ 0.5 m; number of minimum neighborhood points N ₂ ＝1；

S3.3, counting the number of the divided trunk clusters, and if the number is 1, only including a single trunk road tree, completing single tree division; if the number is greater than 1, the street tree cluster comprises a plurality of adhesive street trees, and the individual street trees need to be further divided into individual street trees for execution S4.

Further, the height difference threshold value in S3.1 is 0.4 m.

Further, the S4 specifically includes: and (3) carrying out the following treatment on each adhered street tree cluster:

s4.1, vertically slicing: calculating the mean value of the point cloud three-dimensional coordinates of each trunk cluster, taking the mean value connecting line of adjacent trunk clusters as a normal vector, and vertically slicing the space between the adjacent trunk clusters by the thickness of d, wherein d is 0.01 m;

s4.2, vertical segmentation: counting the number of street tree points contained in the slices, taking the middle plane of the slice with the least number of points as a dividing plane, vertically dividing the adhered street tree into individual street trees, and distributing street tree labels.

Further, the S5 specifically includes:

s5.1, clustering the point cloud of the single-plant street tree by using the (x, y, z) coordinates of the point cloud of the single-plant street tree as input and adopting a clustering algorithm, and dividing the point cloud into sub-clusters; wherein the neighborhood radius ε ₃ 0.2 m; number of minimum neighborhood points N ₃ ＝C ₃ m ₃ ，m ₃ Is the neighborhood radius ε ₃ Average neighborhood point number of all street tree points at 0.2m, C ₃ Is constant and takes the value of 0.2-0.5;

and S5.2, taking the sub-cluster with the largest number of points as the main part of the street tree, reserving the street tree label distributed in the S4.2, and canceling the labels of the other sub-clusters.

S5.3, distributing labels for the unmarked point clouds by using the point clouds with all the distributed street tree labels in the adhered street tree clusters as a training set and adopting a k neighbor classifier;

s5.4, comparing the labels of all points in the adhered street tree cluster with the labels distributed in the S4.2, if the labels are the same, completing the single tree segmentation, otherwise, returning to the S5.1 to execute iterative comparison; in the iteration process, comparing the labels adhered to all the points in the street tree cluster with the labels distributed last time in S5.3, if the labels are the same, completing the single tree segmentation, and ending the iteration.

Further, the number k of the neighbor points is 11.

The invention has the beneficial effects that:

the invention relates to a street tree single tree segmentation method based on MLS point cloud data, which comprises the steps of segmenting a street tree point cloud into street tree clusters by adopting a clustering algorithm, detecting the street tree clusters by adhering the street trees, and then segmenting the street tree clusters containing a plurality of adhered street trees in a multi-layer manner from thick to thin to obtain a street tree single tree point cloud; the method has high street tree point cloud identification capability and street tree single tree segmentation precision, and is suitable for complex urban environments containing various ground features.

In the invention, when a plurality of adhesive street trees are segmented, the trunk is firstly used for rough segmentation, and then the DBSCAN and the k nearest neighbor classifier are combined to finely segment the tree branch tips, thereby effectively improving the segmentation precision.

Additional features and advantages of the invention will be set forth in the detailed description which follows.

Drawings

The above and other objects, features and advantages of the present invention will become more apparent by describing in more detail exemplary embodiments thereof with reference to the attached drawings, in which like reference numerals generally represent like parts throughout.

Fig. 1 shows a flow chart of the street tree splitting process of the present invention.

Fig. 2 shows a schematic diagram of coarse division of the stuck street tree in the present invention.

FIG. 3 is a schematic diagram of street MLS point cloud data in an embodiment of the invention.

Fig. 4 shows a schematic diagram of the road tree point cloud detection and clustering result in the embodiment of the invention.

Fig. 5 is a schematic diagram illustrating a coarse segmentation result of the stuck road tree in the embodiment of the present invention.

Fig. 6 is a diagram illustrating a segmentation result of the stuck street tree in the embodiment of the present invention.

Detailed Description

Preferred embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While the preferred embodiments of the present invention are shown in the drawings, it should be understood that the present invention may be embodied in various forms and should not be limited to the embodiments set forth herein.

As shown in fig. 1, a street tree single tree segmentation method based on MLS point cloud data includes the following steps:

s1, acquiring the street tree point cloud: the method comprises the following steps of obtaining street tree point cloud data through street MLS point cloud data scanned by a mobile laser, and specifically comprises the following steps: a street tree detector is used to identify a street tree point cloud from the street MLS point cloud data.

Wherein: the detection method of the street tree detector adopts a street tree target identification method based on vehicle-mounted 2D LiDAR point cloud data, and the patent number is ZL 201810015134.9; or Weinmann M, Weinmann M, Mallet C, et al. A classification-segmentation frame for the detection of inductive drive in dense MMS point closed data acquisition in street areas [ J ]. removal Sensing,2017,9(03): 277.

S2, road tree point cloud clustering: clustering the road tree point cloud by adopting a clustering algorithm, and dividing the road tree point cloud into road tree clusters; the clustering algorithm adopts a density-based noise application clustering algorithm DBSCAN; the S2 specifically includes:

taking the (x, y, z) coordinates of the street tree point cloud as input, clustering the street tree point cloud by adopting a clustering algorithm, and dividing the street tree point cloud into street tree clusters; wherein the neighborhood radius ε ₁ 0.5 m; number of minimum neighborhood points N ₁ ＝C ₁ m ₁ ，m ₁ Is the neighborhood radius ε ₁ Average neighborhood point number of all street tree points at 0.5m, C ₁ Is constant and takes 0.02-0.1.

S3, detecting the adhered street trees: performing trunk point cloud extraction and trunk point cloud clustering on each street tree cluster, detecting whether the street tree cluster contains a plurality of adhered street trees, executing S4 on the adhered street tree clusters containing the plurality of adhered street trees, and finishing single-tree segmentation on the non-adhered street tree clusters; the method comprises the following specific steps:

s3.1, trunk extraction is carried out on each street tree cluster, and the minimum value z of the point cloud elevation in the current street tree cluster is obtained _min Extracting intra-cluster heightThe minimum value z of the range and the elevation _min Taking the point cloud with the difference smaller than the height difference threshold value as the trunk point cloud; wherein: the height difference threshold value is 0.4 m;

s3.2, clustering the trunk point cloud by using the (x, y) coordinates of the trunk point cloud as input and adopting a clustering algorithm, and dividing the trunk point cloud into trunk clusters; wherein the neighborhood radius ε ₂ 0.5 m; number of minimum neighborhood points N ₂ ＝1；

S3.3, counting the number of the divided trunk clusters, and if the number is 1, only including a single trunk road tree, completing single tree division; if the number is greater than 1, the street tree cluster comprises a plurality of adhesive street trees, and the individual street trees need to be further divided into individual street trees for execution S4.

S4, roughly dividing the adhered street tree: roughly dividing the adhered street tree cluster into single street trees; and (3) carrying out the following treatment on each adhered street tree cluster:

s4.1, vertically slicing: calculating the mean value of the point cloud three-dimensional coordinates of each trunk cluster, taking the mean value connecting line of adjacent trunk clusters as a normal vector, and vertically slicing the space between the adjacent trunk clusters by the thickness of d, wherein d is 0.01 m;

s4.2, vertical segmentation: counting the number of street tree points contained in the slices, taking the middle plane of the slice with the least number of points as a dividing plane, vertically dividing the adhered street tree into individual street trees, and distributing street tree labels.

S5, fine segmentation of the adhered street tree: performing fine segmentation on the single-plant street tree obtained by the rough segmentation of S4, specifically:

s5.1, clustering the point cloud of the single-plant street tree by using the (x, y, z) coordinates of the point cloud of the single-plant street tree as input and adopting a clustering algorithm, and dividing the point cloud into sub-clusters; wherein the neighborhood radius ε ₃ 0.2 m; number of minimum neighborhood points N ₃ ＝C ₃ m ₃ ，m ₃ Is the neighborhood radius ε ₃ Average neighborhood point number of all street tree points at 0.2m, C ₃ Is constant and takes the value of 0.2-0.5;

and S5.2, taking the sub-cluster with the largest number of points as the main part of the street tree, reserving the street tree label distributed in the S4.2, and canceling the labels of the other sub-clusters.

S5.3, taking point clouds adhered to all street tree labels distributed in the street tree cluster as a training set, distributing labels to the unmarked point clouds by adopting a k neighbor classifier, wherein the number k of neighbor points is 11;

s5.4, comparing the labels of all points in the adhered street tree cluster with the labels distributed in the S4.2, if the labels are the same, completing the single tree segmentation, otherwise, returning to the S5.1 to execute iterative comparison; in the iteration process, comparing the labels of all the points in the adhered street tree cluster with the labels distributed last time in S5.3, if the labels are the same, completing the segmentation of the single tree, and ending the iteration.

In the specific implementation:

the experiment uses a ZEB-HORIZONs mobile handheld scanner to collect point cloud data of a section of road with north latitude 32 degrees 04'55.1 "and east longitude 118 degrees 48'58.8", as shown in fig. 3, and the point cloud is colored with z coordinate for easy viewing. The main tree species of the road are tokyo cherry blossom (Cerasus yedonnsis), gingko (Ginkgo biloba), hackberry (Celtis sinensis), Cinnamomum camphora (Cinnamomum camphora) and the like, the height range is 4.6-8.2m, and the crown width range is 2.1-7.4 m. In addition, the house also comprises ground objects such as buildings, driveways, sidewalks, benches, street lamps, bicycles, signboards, pedestrians, cars, shrubs, turf, flower beds and the like.

As shown in fig. 3, is street MLS point cloud data; performing S2 street tree point cloud clustering and S3 adhesive street tree detection to obtain a plurality of adhesive street tree clusters as shown in FIG. 4 for the street tree point cloud detection and clustering results; performing S4 coarse segmentation of the stuck street tree and S5 fine segmentation of the stuck street tree, as shown in fig. 5, where there are a few erroneous segmentations at the tips of the branches of adjacent street trees as a result of the coarse segmentation of the stuck street tree; as shown in fig. 6, the branch tips of the adjacent street trees are accurately segmented as a result of the fine segmentation of the stuck street trees.

Having described embodiments of the present invention, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments.

Claims (9)

1. A street tree single tree segmentation method based on MLS point cloud data is characterized by comprising the following steps:

s1, acquiring the street tree point cloud: acquiring street tree point cloud data through street MLS point cloud data scanned by a mobile laser;

s2, road tree point cloud clustering: clustering the road tree point cloud by adopting a clustering algorithm, and dividing the road tree point cloud into road tree clusters;

s3, detecting the adhered street trees: performing trunk point cloud extraction and trunk point cloud clustering on each street tree cluster, detecting whether the street tree cluster contains a plurality of adhered street trees, executing S4 on the adhered street tree clusters containing the plurality of adhered street trees, and finishing single-tree segmentation on the non-adhered street tree clusters;

s4, coarse segmentation of the adhered street tree: roughly dividing the adhered street tree cluster into single street trees;

s5, fine segmentation of the adhered street trees: and (5) finely dividing the single street tree obtained by the rough division of the S4 to finish the single tree division.

2. The method for street tree single tree segmentation based on MLS point cloud data according to claim 1, wherein S1 specifically is: a street tree detector is used to identify a street tree point cloud from the street MLS point cloud data.

3. The method for street tree single tree segmentation based on MLS point cloud data as claimed in claim 1, wherein the clustering algorithm employs a density-based noise application clustering algorithm DBSCAN.

4. The method for street tree single tree segmentation based on MLS point cloud data according to claim 1, wherein S2 specifically is: taking the (x, y, z) coordinates of the street tree point cloud as input, clustering the street tree point cloud by adopting a clustering algorithm, and dividing the street tree point cloud into street tree clusters; wherein the neighborhood radius ε ₁ 0.5 m; number of minimum neighborhood points N ₁ ＝C ₁ m ₁ ，m ₁ Is the neighborhood radius ε ₁ Average neighborhood point number of all street tree points at 0.5m, C ₁ Is constant and takes 0.02-0.1.

5. The method for street tree single tree segmentation based on MLS point cloud data according to claim 1, wherein S3 specifically is:

s3.1, trunk extraction is carried out on each street tree cluster, and the minimum value z of the point cloud elevation in the current street tree cluster is obtained _min Extracting the elevation value and the minimum elevation value z in the cluster _min Taking the point cloud with the difference smaller than the height difference threshold value as the trunk point cloud;

s3.2, clustering the trunk point cloud by using the (x, y) coordinates of the trunk point cloud as input and adopting a clustering algorithm, and dividing the trunk point cloud into trunk clusters; wherein the neighborhood radius ε ₂ 0.5 m; number of minimum neighborhood points N ₂ ＝1；

S3.3, counting the number of the divided trunk clusters, and if the number is 1, only including a single trunk road tree, completing single tree division; if the number is greater than 1, the street tree cluster comprises a plurality of adhesive street trees, and the individual street trees need to be further divided into individual street trees for execution S4.

6. The method for street tree single tree segmentation based on MLS point cloud data as claimed in claim 5, wherein the height difference threshold of S3.1 is 0.4 m.

7. The method for street tree single tree segmentation based on MLS point cloud data according to claim 1, wherein S4 specifically is: and (3) carrying out the following treatment on each adhered street tree cluster:

s4.1, vertically slicing: calculating the mean value of the point cloud three-dimensional coordinates of each trunk cluster, taking the mean value connecting line of adjacent trunk clusters as a normal vector, and vertically slicing the space between the adjacent trunk clusters by the thickness of d, wherein d is 0.01 m;

s4.2, vertical segmentation: counting the number of street tree points contained in the slices, taking the middle plane of the slice with the least number of points as a dividing plane, vertically dividing the adhered street tree into individual street trees, and distributing street tree labels.

8. The method for street tree single tree segmentation based on MLS point cloud data according to claim 7, wherein S5 specifically is:

s5.1, clustering the point cloud of the single-plant street tree by using the (x, y, z) coordinates of the point cloud of the single-plant street tree as input and adopting a clustering algorithm, and dividing the point cloud into sub-clusters; wherein the neighborhood radius ε ₃ 0.2 m; number of minimum neighborhood points N ₃ ＝C ₃ m ₃ ，m ₃ Is the neighborhood radius ε ₃ Average neighborhood point number of all street tree points at 0.2m, C ₃ Is constant and takes the value of 0.2-0.5;

and S5.2, taking the sub-cluster with the largest number of points as the main part of the street tree, reserving the street tree label distributed in the S4.2, and canceling the labels of the other sub-clusters.

S5.3, distributing labels for the unmarked point clouds by using the point clouds with all the distributed street tree labels in the adhered street tree clusters as a training set and adopting a k neighbor classifier;

s5.4, comparing the labels of all points in the adhered street tree cluster with the labels distributed in the S4.2, if the labels are the same, completing the single tree segmentation, otherwise, returning to the S5.1 to execute iterative comparison; in the iteration process, comparing the labels of all the points in the adhered street tree cluster with the labels distributed last time in S5.3, if the labels are the same, completing the segmentation of the single tree, and ending the iteration.

9. The method as claimed in claim 8, wherein the number of neighboring points k is 11.

Images