CN115018897A

CN115018897A - Method for extracting typical surface feature elements of laser point cloud city

Info

Publication number: CN115018897A
Application number: CN202210724965.XA
Authority: CN
Inventors: 李鹏; 王延霞; 聂士海; 涂晋升; 王梦柯; 宋佳
Original assignee: Chuzhou University
Current assignee: Chuzhou University
Priority date: 2022-06-24
Filing date: 2022-06-24
Publication date: 2022-09-06
Anticipated expiration: 2042-06-24
Also published as: CN115018897B

Abstract

The invention discloses a method for extracting typical surface feature elements of a laser point cloud city, which comprises the following steps: step 1, acquiring urban laser point cloud data; step 2, calculating a characteristic value and a normal vector of each point in the city; step 3, separating ground points and non-ground points; step 4, extracting road element points from the ground points; and 5, extracting building element points and street tree element points from the non-ground points. The invention provides a set of complete urban three-typical surface feature extraction method by aiming at mobile laser point cloud data and utilizing the differences of spatial distribution rules, geometric features and the like of three typical surface feature elements of roads, street trees and buildings, and can provide real-time and accurate basic data and auxiliary decision for urban planning, road change, urban design and the like.

Description

Method for extracting typical surface feature elements of laser point cloud city

Technical Field

The invention relates to the field of laser point cloud data processing methods, in particular to a method for extracting typical surface feature elements of a laser point cloud city.

Background

The method for extracting urban ground feature element information from 3D laser point cloud data is one of the problems concerned by numerous scholars at home and abroad, and at present, researches for extracting urban roads based on the point cloud data mainly focus on ALS data and MLS data. Compared with ALS data, MLS data can acquire a point cloud with higher accuracy and the road occlusion caused by non-road elements is greatly reduced, so that in recent years, extracting urban feature information from MLS data is a very active research subject.

The current research is mainly focused on roads, trees and the like. In the research of road element extraction, methods such as combination with third-party data, projection-based and 3D point-based and road characteristic methods are mainly available. In the method, the road extraction is carried out by adding other information such as road network, depth, vehicle posture and the like in combination with the third-party data. Although the method can successfully extract the road confidence, the method has great dependence on non-point cloud data, and once third-party data is missing or data is wrong, the extraction of road information is not feasible.

The projection-based method mainly uses various attributes (such as height, intensity, pulse width and the like) of MLS data to generate a distance image, and then identifies and extracts a road according to the distance image, and such methods add unnecessary matching errors in the rasterization process, and finally make it difficult to obtain an accurate road boundary.

The method based on road features mainly extracts road elements according to rules such as the shape, the position and the road point features of the road and the characteristics of road edges. On the one hand, these methods have the need to be assisted by other data, and on the other hand, these methods, which are mainly used for chaotic MLS point clouds, have the difficulty of handling complex ground surfaces and lead to high computational complexity.

The urban tree element extraction research mainly focuses on methods based on cluster features, model fitting, region growing and the like. In The study of extracting a single tree based on a cluster feature method, rutzing (2011(rutzing, m., prathast, a.k., Elberink, s., et. tree molding from mobile laser scanning data-sections [ J ]. The photo gram Record,2011,26(135):361 and 372.)) removes large planes such as ground, outer walls, etc. using 3D hough transform and surface growth, while dividing remaining points into clusters, and describes surface roughness using standard deviation in elevation, and then extracts a single tree therefrom according to The ratio of The surface roughness and The point density. The model fitting-based method is mainly used for extracting an object having a specific shape. Such methods may extract a single tree in certain specific situations, but do not perform well in various situations or complex geometries. When using a feature-based approach or a model fitting-based approach, one refinement process is required for each extracted tree cluster.

The region growing method is mainly used for research of 3D point cloud segmentation, extraction of individual trees and the like. The method extracts various trees with high precision and shows excellent performance in a relatively simple environment, however, there are problems in seed selection and growth criteria. Li (2016) (Li, X, string, L, Lin, Q., et al. A new region growing-based segmentation method for high resolution Sensing image [ C ]// Sensing and Remote Sensing. IEEE,2015: 4328-. However, in many MLS data, the trunk point can only acquire the side facing the road (semicircle), and the side facing away cannot be collected. This method is not applicable to the above-mentioned situation because it reduces the horizontal section point cloud of the trunk to a circular shape.

In summary, in the research of extracting the urban typical feature elements by using the moving point cloud data, the existing method has a certain limitation, and the core of the research is only directed at the extraction of a single feature element, and few researches can simultaneously extract a plurality of typical feature elements.

Disclosure of Invention

The invention aims to provide a method for extracting typical surface feature elements of a laser point cloud city, which aims to solve the problem that a plurality of typical surface feature elements cannot be extracted simultaneously by a point cloud data extraction method in the prior art.

In order to achieve the purpose, the technical scheme adopted by the invention is as follows:

a method for extracting typical surface feature elements of a laser point cloud city is used for extracting road element points, street tree element points and building element points from city laser point cloud data and comprises the following steps:

step 1, acquiring urban laser point cloud data;

step 2, calculating a characteristic value and a normal vector of each point in the urban laser point cloud data;

step 3, separating ground points from non-ground points:

selecting one point from the urban laser point cloud data as a seed point, determining a field point of the seed point, and obtaining a ground point from the field point according to the included angle between the seed point and the field point;

continuously traversing and updating the seed points and the field points thereof from the urban laser point cloud data, thereby obtaining all ground points in a region growing traversal mode, separating the ground points from the urban laser point cloud data, and taking the rest points after separating the ground points as non-ground points;

step 4, extracting road element points from the ground points:

calculating the flatness of each ground point according to the characteristic values corresponding to the ground points obtained in the step (3), comparing the flatness of each ground point with a flatness threshold, determining that the ground point is a road key point when the flatness of a certain ground point is greater than or equal to the flatness threshold, and removing the ground point when the flatness of a certain ground point is less than the flatness threshold, thereby obtaining the road key point in the urban laser point cloud data;

step 5, extracting building element points and street tree element points from the non-ground points, wherein the process is as follows:

firstly, clustering and combining non-ground point clusters into a plurality of point clusters by adopting a clustering algorithm;

then based on the normal vector of each point in each point cluster, calculating to obtain an average normal included angle of the point cluster, comparing the average normal included angle of each point cluster with an angle threshold, judging the point cluster to be a regular ground object point cluster when the average normal included angle of a certain point cluster is larger than the angle threshold, and judging the point cluster to be an irregular ground object point cluster when the average normal included angle of a certain point cluster is smaller than or equal to the angle threshold;

identifying each irregular ground object point cluster from bottom to top according to a voxel structure to obtain a street tree element point cluster, wherein points contained in the street tree element point cluster are street tree element points;

and identifying each regular ground point cluster according to a point cluster space range to obtain a building element point cluster, wherein points contained in the building element point cluster are building element points.

In a further step 1, after the urban laser point cloud data is obtained, gross error elimination is carried out on the urban laser point cloud data based on a statistical model so as to eliminate points with gross errors.

In a further step 2, a covariance matrix is constructed for each point, and a feature value and a normal vector of each point are calculated based on the covariance matrix.

In a further step 3, seed points are selected from the urban laser point cloud data based on the curvature value of each point.

Further, when the seed points are updated in step 3, new seed points are obtained from the ground points determined by the previous seed points based on the curvature values.

Further, when each irregular ground object point cluster is identified in the step 5, each layer of the irregular ground object point cluster is subjected to voxelization, each layer of voxelization is subjected to cluster analysis to obtain the boundary of each class of the voxels of each layer, whether the classes of the two adjacent layers are overlapped on the cross section is compared, and if the classes of the two adjacent layers are intersected on the cross section, the two adjacent layers are judged to belong to the same ground object;

and sequentially judging whether other layers above any one layer of voxel point of a certain type are null values or not for each layer belonging to the same ground object according to the sequence from bottom to top, if not, judging that the ground object belonging to each layer is a street tree, and the irregular ground object point cluster in which each layer is located is a street tree element point cluster.

Further, when the street tree element point cluster is obtained, the crown and the trunk of the street tree are judged and obtained based on the area of each layer.

Further, when each regular ground feature point cluster is identified according to the space range of the point cluster, the building element point cluster is judged and obtained based on the height and the maximum direction length of the regular ground feature point cluster.

The invention takes three typical surface feature elements of urban roads, trees and buildings as objects, and adopts the idea of layered stripping to simultaneously extract various surface feature elements, thereby improving the surface feature extraction efficiency and simplifying the technical scheme of the technology.

The invention provides a set of complete urban three-typical surface feature extraction method by using the differences of spatial distribution rules, geometric features and the like of three typical surface feature elements of roads, street trees and buildings aiming at mobile laser point cloud data. The invention of the technology can provide real-time and accurate basic data and auxiliary decision for city planning, road change, city design and the like. Compared with the prior art, the invention has the advantages that:

(1) a method can simultaneously extract a plurality of factors:

the method can simultaneously extract the typical surface feature elements of three cities, namely roads, trees and buildings by only one method, greatly simplifies the flow of extracting the surface feature elements by point cloud, and improves the efficiency of extracting the elements.

(2) The layered stripping can gradually refine the ground object precision:

according to the method, non-target point elimination is taken as a principle, ground objects are finely extracted in a layered stripping mode, real essential points can be reserved to the maximum extent, and the extraction precision is improved.

Drawings

FIG. 1 is a block diagram of a method flow according to an embodiment of the invention.

FIG. 2 is the point cloud data after gross error elimination in the embodiment of the invention

FIG. 3 is a separated ground point in an embodiment of the present invention

FIG. 4 is a non-ground point after separation in an embodiment of the invention

Fig. 5 is a diagram of a road point extraction process in the embodiment of the present invention.

Fig. 6 is a point cloud image after cluster analysis in the embodiment of the present invention.

FIG. 7 is a graph of street trees and other elements based on the mean of normal directions in an embodiment of the present invention.

FIG. 8 is a point cloud voxelization map in an embodiment of the invention.

Fig. 9 is a diagram of a recognition result of a street tree in the embodiment of the present invention.

Fig. 10 is a diagram of a final result of extracting typical feature elements of a city in the embodiment of the invention.

Detailed Description

The invention is further illustrated with reference to the following figures and examples.

As shown in fig. 1, the method for extracting typical feature points of a laser point cloud city according to the present embodiment is used for extracting road element points, road tree element points, and building element points from city laser point cloud data, and includes the following steps:

step 1, acquiring a city laser point cloud data set.

And scanning the city through a laser scanner to obtain a city laser point cloud data set. In the process of laser point cloud data acquisition, the laser scanner may be influenced by the external environment to cause that part of points in the acquired point cloud data set contain small parts of gross errors, and the subsequent processing is influenced by the gross errors, so that the gross errors of the point cloud data set are required to be removed before element extraction, and the points containing the gross errors are removed. In this embodiment, gross error rejection based on a statistical model is adopted, and the formula is as follows:

in the formula (1), P ₀ Is the current point, P, in the point cloud dataset _i Is the current point P ₀ K is P ₀ S is the threshold. P is _i -P ₀ Represents a front point P ₀ And a field point P _i If the variance of the distances between a certain point and the nearest k neighborhood points is greater than the threshold value s, the point is judged to contain gross errors, and the point needs to be removed. The city laser point cloud data with the coarse difference points removed is used in the subsequent steps, and the city laser point cloud data with the coarse difference points removed in the embodiment is shown in fig. 2.

And 2, calculating the characteristic value and the normal vector.

Respectively constructing a covariance matrix in a k neighborhood for each point in the urban laser point cloud data obtained in the step 1, wherein the formula is as follows:

in the formulas (2) and (3), C is a covariance matrix, P _i As the current point of the time-point,

representing the mean of the coordinates of the current point and the k neighborhood points,

is a normal vector, I is an identity matrix, and λ is an eigenvalue. Since the coordinate values of the point cloud data are three-dimensional, the characteristic value λ has three values, respectively λ ₁ 、λ ₂ 、λ ₃ 。

Calculating three eigenvalues lambda of each point based on the covariance matrixes shown in the formulas (2) and (3) ₁ 、λ ₂ 、λ ₃ And normal vector

Wherein λ ₁ >λ ₂ >λ ₃ ，

Is λ ₃ The corresponding vector. The characteristic value and the normal vector are mainly used for further distinguishing the ground feature elements after rough classification.

And 3, separating the ground points and the non-ground points.

The road elements are located in ground points, while the building and road tree elements belong to non-ground points, and the road elements can be primarily separated from the other two elements by separating the ground points from the non-ground points.

This example uses a similar region growing method for separation. First, a seed point is selected, and the formula is as follows:

in the formula (4), K ₀ And H ₀ Respectively representing the curvature and elevation of the initial seed, K _i Is the curvature of the ith point in the data set, n is the number of the point cloud data concentration points of the urban laser, H _j For the elevation of the jth point, m is the number of points that have the smallest curvature at the same time. The curvature is the first factor of seed selection, and only when the point with the minimum curvature value exceeds 1, secondary judgment is carried out by using the minimum elevation, so that the seed point is finally selected.

And (3) constructing a sphere with the radius of r by taking the seed point as a center, wherein all points falling in the sphere are neighborhood points of the seed point. The included angle between the adjacent region point tangent plane and the seed point is the criterion for judging the ground point, and the formula is as follows:

theta in the formula (5) _t Denotes the angle as set forth above, (x) _s ,y _s ,z _s ) Is the spatial coordinate of the seed point, (x) _i ,y _i ,z _i ) The coordinates of the ith neighborhood point representing the seed point, A, B, C is the intercept of the three-dimensional spatial plane equation on the x, y, z coordinate axes.

Equation (6) is used to determine whether a neighborhood point belongs to a ground point, p _i I-th neighborhood point, σ, of seed point _A To determine the threshold value, when theta _t Is smaller than sigma _A And judging that the neighborhood point belongs to the ground point.

The growth of ground points continues by the constant addition of new seeds. The new seed is used to control the process of growing from the original seed to all the surface points. Once a new seed is identified, the k neighborhood points of the new seed can be continually searched to traverse the entire ground point. The judgment conditions of the newly added seed points are as follows:

in the formula (7), K _i Is a point P _i Of curvature of σ _k Is a curvature threshold, C _g Is a road. Only when point P _i Is less than a set threshold value while P _i Also already belonging to the ground point, this point is considered as a new seed.

The above process can separate ground points and non-ground points from the urban laser point cloud data as shown in fig. 2, the ground points separated from fig. 2 are shown in fig. 3, and the non-ground points separated are shown in fig. 4.

And 4, extracting road element points from the ground points.

After the point cloud data is separated into ground points and non-ground points, road tree and building element points can be subdivided on the basis of the point cloud data, wherein the road element points are further extracted from the ground points, and the road tree element points and the building element points are further subdivided at the non-ground points.

Because the distribution of the road element points is approximate to a plane, the road element points can be identified according to the flatness characteristics of each ground point, and the specific formula is as follows:

in the formula (8), λ ₁ 、λ ₂ And λ ₃ And e is a road element point flatness threshold value for the characteristic value of each ground point in the descending order. When point P _i The planeness (i.e. any one ground point) is greater than or equal to a planeness threshold value P _i Then the point P is _i If not, the points are eliminated. And finally, removing outliers from the extracted road element points by using the formula (1) again.

The road element points are extracted from the ground points as shown in fig. 3 through the above-described process, and the extraction process is shown in fig. 5. In the figure, (a) is a ground point, (b) is a road element point after flatness calculation, and (c) is a road element point after outlier elimination.

And 5, extracting the building and street tree element points from the non-ground points.

The building element points are distributed close to the vertical plane, while the street tree element points are relatively scattered and each street tree has the characteristic of aggregation, and the two types of land features are further subdivided according to the phenomenon, and the process is as follows:

(5.1) Cluster analysis

The separated non-ground points are split into a plurality of discontinuous regions, so that the present embodiment combines the continuous regions into the same cluster by using a clustering algorithm, thereby obtaining a plurality of point clusters. Each point cluster is used as a basic unit of the ground object and is distinguished according to the normal difference of different ground object points.

After the non-ground points shown in fig. 4 are subjected to cluster analysis through the above process, the point cloud shown in fig. 6 is obtained.

(5.2) determination of type of feature

The distribution of the element points of the street tree shows aggregative property, other element points such as buildings and the like show vertical face information, the difference of the distribution of the two types of feature points can be distinguished through the normal direction, and the formula is as follows:

in the formulae (9), (10) and (11), θ _i The normal included angle of the ith point in each point cluster is,

and

three vectors, E, normal to the point _θ Is the average angle of each point cluster, and k is the number of points in each point cluster.

Since the normal directions of the ground feature points such as buildings are mostly parallel to the horizontal plane, and the normal directions of the street tree points are scattered and irregular, an angle threshold value delta theta is set, when E is _θ When the point cluster is larger than delta theta, the point cluster is a regular ground object point cluster comprising buildings and poles, and otherwise, the point cluster is an irregular ground object point cluster comprising a street tree. Thus, it is possible to judge and recognize the regular and irregular feature point clusters in this step.

Regular and irregular clusters of feature points as shown in fig. 7 can be identified from fig. 6 through the above-described process.

And (5.3) identifying the element points of the street tree.

The road tree identification is carried out from bottom to top according to the voxel structure for each irregular object point cluster. Firstly, performing voxelization on irregular ground object point clusters, wherein the formula is as follows:

in equation (12), V represents a voxel set, V _i (L, R, C) is the ith voxel, p _ik Is the k point, x, in the ith voxel _ik 、y _ik And z _ik Respectively represent points p _ik The coordinates of (a); l, R, C voxel serial numbers of the current ith voxel in three coordinate axis directions, x _ik 、y _ik、 z _ik Coordinate value, x, representing the current point _min 、y _min And z _min Respectively representing the minimum value of three coordinates in the point data, and Δ v is the length of a unit voxel.

Through the above process, the irregular object point clusters identified in fig. 7 are voxelized as shown in fig. 8. Fig. 8 (a) is a diagram before voxelization and fig. 8 (b) is a diagram after voxelization.

Performing cluster analysis on each layer of voxels of the voxelized irregular object point cluster respectively, solving the corresponding class of each layer and the boundary of each class, and judging whether the class continuously grows or not by comparing whether the classes of two adjacent layers coincide on the cross section, wherein the formula is as follows:

in the formula (13), the first and second groups,

a j-th class representing the i-th layer,

represents the kth class of the (i + 1) th layer,

and

respectively represent the ith layer, the jth class and the (i + 1) th layerVoxels of class k are in the extent of the cross section.

Only when the classes of two adjacent layers intersect in cross section, the two adjacent layers are considered to belong to the same feature.

Judging and identifying each layer belonging to the same ground feature from bottom to top, wherein in the process of judging from bottom to top, if a certain type has a voxel point on the ith layer and a null value suddenly appears on the (i + 1) th layer, the map formed by the type on the ith layer and other layers below the ith layer is possibly a street tree, but only when the (i + k (k >1) th layer j type is not empty, the ground feature is finally judged to be the street tree, the corresponding irregular object point cluster is a street tree element cluster, and the points contained in the street tree element cluster are street tree element points, wherein the formula is as follows:

in the process of judging from bottom to top, the following conditions occur, and the tree is considered to grow from the trunk to the crown:

in the formula (I), the compound is shown in the specification,

and

area of the ith and jth class of i +1 layers, respectively,. DELTA.S _L Is based on the area threshold of the layer.

Since the difference between the cross-sectional areas of the tree crown and the trunk of the street tree is large, when the voxel class satisfying the formula (15) is determined as the tree crown, the merging of the trunks is stopped and the growing of the tree crown is started. The tree crown part element points and the tree trunk part element points of the street tree can be identified in the above mode.

The element points of the road tree identified from the irregular object point cluster by the above process are shown in fig. 9. Fig. 9(a) shows the result of extracting a street tree for a single trunk, and fig. 9(b) shows the result of extracting a street tree with a street lamp post and a billboard mixed in the point cloud data.

(5.4) building element Point identification

The elements similar to buildings in the normal direction in the regular ground point cluster mainly comprise elements such as a billboard, a bus stop, pedestrians and automobiles, and can be finely identified through the space range of the point cluster, and the formula is as follows:

H _c ＞σ _H ∩L _c ＞σ _L (16)

in the formula (16), H _c Height of a cluster of points, σ _H Is a building height threshold, L _c Is the maximum directional length of the cluster of points, σ _L Is the building length threshold.

And (3) judging the regular ground feature point cluster meeting the formula (16) as a building element point cluster, wherein points contained in the building element point cluster are building element points, and otherwise, the regular ground feature point cluster is not the building element point cluster and is discarded.

Finally, the point cloud data obtained by extracting elements from the point cloud data of fig. 2 through the above-described process in the present embodiment is shown in fig. 10.

The embodiments of the present invention are described only for the preferred embodiments of the present invention, and not for the limitation of the concept and scope of the present invention, and various modifications and improvements made to the technical solution of the present invention by those skilled in the art without departing from the design concept of the present invention shall fall into the protection scope of the present invention, and the technical content of the present invention which is claimed is fully set forth in the claims.

Claims

1. A method for extracting typical surface feature elements of a laser point cloud city is used for extracting road element points, street tree element points and building element points from city laser point cloud data, and comprises the following steps:

step 1, acquiring urban laser point cloud data;

step 3, separating ground points from non-ground points:

continuously traversing and updating seed points and field points thereof from the urban laser point cloud data, thereby obtaining all ground points in a region growing traversing mode, separating ground points from the urban laser point cloud data, and taking the rest points after the ground points are separated as non-ground points;

step 4, extracting road element points from the ground points:

firstly, clustering and combining non-ground points into a plurality of point clusters by adopting a clustering algorithm;

then, based on the normal vector of each point in each point cluster, calculating to obtain an average normal included angle of the point cluster, comparing the average normal included angle of each point cluster with an angle threshold, judging that the point cluster is a regular ground object point cluster when the average normal included angle of a certain point cluster is greater than the angle threshold, and judging that the point cluster is an irregular ground object point cluster when the average normal included angle of the certain point cluster is less than or equal to the angle threshold;

2. The method for extracting the typical feature elements in the laser point cloud city as claimed in claim 1, wherein in step 1, after the city laser point cloud data is obtained, the gross error elimination is performed on the city laser point cloud data based on a statistical model to eliminate points with the gross error.

3. The method for extracting the typical feature elements in the laser point cloud city as claimed in claim 1, wherein in step 2, a covariance matrix is constructed for each point, and the eigenvalue and normal vector of each point are calculated based on the covariance matrix.

4. The method as claimed in claim 1, wherein in step 3, a seed point is selected from the city laser point cloud data based on the curvature value of each point.

5. The method as claimed in claim 1, wherein when the seed points are updated in step 3, new seed points are obtained from the ground points determined from the previous seed points based on the curvature values.

6. The method for extracting typical surface feature elements in a laser point cloud city according to claim 1, wherein in the step 5, when each irregular surface feature point cluster is identified, each layer of the irregular surface feature point cluster is voxelized, each layer of voxels is subjected to clustering analysis to obtain the boundary of each class of voxels of each layer, whether the classes of two adjacent layers coincide with each other on the cross section is compared, and if the classes of two adjacent layers intersect with each other on the cross section, the two adjacent layers are judged to belong to the same surface feature;

7. The method for extracting typical feature elements in a laser point cloud city as claimed in claim 6, wherein when obtaining the street tree element point clusters, the crown and trunk of the street tree are judged and obtained based on the area of each layer.

8. The method for extracting typical feature elements in a laser point cloud city according to claim 1, wherein when each regular feature point cluster is identified according to a point cluster space range, the building feature point cluster is judged and obtained based on the height and the maximum direction length of the regular feature point cluster.