CN-121982238-A - Quick extraction method and system for basic mapping geographic entity based on laser three-dimensional scanning

CN121982238ACN 121982238 ACN121982238 ACN 121982238ACN-121982238-A

Abstract

The invention belongs to the technical field of laser mapping, in particular to a method and a system for rapidly extracting a basic mapping geographic entity based on laser three-dimensional scanning, comprising the following steps of S100, performing density self-adaptive voxel downsampling on an original laser point cloud, and calculating geometric, reflection and context characteristics from the original point cloud in a multi-scale neighborhood of each representative point; S200, constructing a hypergraph, generating initial hyperpoints through spectral clustering, iteratively carrying out semantic classification and geometric consistency check, dynamically combining or splitting the hyperpoints to obtain a refined hyperpoint set with stable semantic labels, S300, constructing a subgraph, enabling edge weights to reflect combined geometric and boundary compatibility among the hyperpoints, clustering similar hyperpoints into physical entity examples through calculating a maximum spanning tree and cutting at weak joints, S400, carrying out geometric fitting and regularization on point clouds of each entity example to generate a parameterized model, calculating spatial relations among the entities, and constructing a global semantic topological graph containing complete attributes and relations.

Inventors

WANG LEI
WANG DAN
LU YANG
WANG SHIBO
CUI HUIMIN
HU XIAOXIAO
ZHANG MENGJIAO
Yin Jiuna
YUAN LI
WANG XUELIAN

Assignees

临沭县方正测绘有限公司

Dates

Publication Date: 20260505
Application Date: 20260127

Claims (10)

1. The quick extraction method of the basic mapping geographic entity based on laser three-dimensional scanning is characterized by comprising the following steps of: s100, constructing a point cloud data self-adaptive regularization and multi-scale characteristic field; performing density-adaptive voxelized downsampling on the original laser point cloud to generate a normalized representative point set, and calculating geometric, reflection and context characteristics from the original point cloud in a multi-scale neighborhood of each representative point to form a basic characteristic field penetrating all subsequent processes; S200, hierarchical semantic segmentation based on hypergraph iteration refinement; Constructing a hypergraph by taking the regularized points and the characteristic fields thereof as nodes, generating initial hyperpoints through spectral clustering, iteratively performing semantic classification and geometric consistency check, and dynamically merging or splitting the hyperpoints until convergence to obtain a refined hyperpoint set with stable semantic labels; S300, optimizing entity instantiation of maximum spanning tree cutting based on the graph; Aiming at each semantic category, constructing a subgraph by taking the superpoints of the category as nodes, reflecting the combined geometric and boundary compatibility among the superpoints by edge weights, and clustering the superpoints of the same category into independent physical entity examples by calculating a maximum spanning tree and cutting at weak joints; s400, reconstructing a global semantic topological graph through entity parameterization modeling; Performing geometric fitting and regularization on the point cloud of each entity instance to generate a parameterized model, calculating the spatial relationship among the entities, and constructing a global semantic topological graph containing complete attributes and relationships.
2. The method for rapidly extracting the basic mapping geographic entity based on the laser three-dimensional scanning of claim 1, wherein the detailed steps of S100 are as follows: S101, self-adaptive density voxel downsampling; Acquiring an original laser point cloud , Wherein, the method comprises the steps of, Is a three-dimensional coordinate of which the position is a three-dimensional coordinate, Is the reflection intensity; dividing the point cloud space into an initial voxel grid, and calculating the local point density of each voxel v Voxel size Self-adaptive adjustment according to local point density: Wherein, the The method comprises the steps of setting a preset basic voxel size; is the median of the scene voxel density; is a sensitivity parameter; With one representative point per voxel v Representation, formation of a normalized point cloud ; S102, constructing a multi-scale characteristic field; Representative points for each voxel Three spherical neighborhoods with different physical radiuses are defined in the original point cloud P, namely small scale Mesoscale (mesoscale) Large scale ; At each scale Is a neighborhood of (a) In, calculate geometric features Reflective features And contextual features ; Finally, each representative point Corresponds to a multi-scale feature vector: The characteristic field of the whole scene is recorded as 。
3. The method for rapidly extracting the basic mapping geographic entity based on laser three-dimensional scanning as claimed in claim 2, wherein the method comprises the following steps: the geometric features The calculation method of (1) is as follows: Calculating covariance matrix based on neighborhood point set to obtain eigenvalue From which are extracted: linearity of ; Flatness of plane ; Degree of scattering ; Local height difference ; Normal vector Included angle with zenith direction z ; For the following 、、、、 The values of the geometric feature vector under three scales are obtained respectively, and the cross-scale mean value and standard deviation of the geometric feature vector are calculated to form the geometric feature vector: the reflective feature The calculation method of (1) is as follows: calculating the mean value of the intensity values in the neighborhood And variance of Forming a reflection feature vector: The contextual characteristics The calculation method of (1) is as follows: elevation relative to ground Wherein The digital ground model is obtained through initial filtering; Normalized dot density Wherein As representative points in voxels Centered at a small scale radius Is a set of all original point cloud points in a spherical neighborhood of radius; Forming a contextual feature vector: 。
4. the method for rapidly extracting the basic mapping geographic entity based on the laser three-dimensional scanning of claim 2, wherein the detailed steps of S200 are as follows: S201, hypergraph construction and initial hyperpoint generation; with the regularized point cloud Q as a node, a hypergraph is constructed Node Correlation features The initial similarity between any two nodes u and v is defined as follows: Wherein, the Normalization parameters for the spatial distance; normalization parameters for feature differences; If it is Then, creating superedges between the nodes u and v, performing spectral clustering on the supergraph H, dividing the nodes into K clusters, and enabling each cluster to be called an initial superpoint The aggregate is denoted as ; S202, iterative refinement and semantic marking; For the current super point Performing iteration, and calculating the aggregation characteristics of the method every time: Will be Input to a pre-trained lightweight classifier to obtain a class probability distribution And prediction category 。
5. The method for rapidly extracting the geographic entity from the basic mapping based on the laser three-dimensional scanning as set forth in claim 4, wherein the specific step of S202 further comprises: For two adjacent superpoints And If the same category is satisfied and the feature distance is smaller than the merging threshold, merging into a new superpoint: Wherein, the Is the characteristic distance between two superpoints; Is a merge threshold; If the point is exceeded If the internal feature inconsistency is greater than the splitting threshold, then it is split into two superpoints using the K-Means algorithm: Wherein, the Is a super point Is a measure of internal feature inconsistency; Is a split threshold; after T times of iteration, the final refined super-point set is obtained Each superpoint has a stable semantic tag 。
6. The method for rapidly extracting the basic mapping geographic entity based on the laser three-dimensional scanning of claim 4, wherein the step S300 comprises the following detailed steps: s301, constructing a class specific similarity graph; for each semantic class c, construct an undirected graph : All the super points of the category; Connecting all spatially adjacent pairs of superpoints Weight is given to each edge Representing compatibility scores for the combined superpoints k and l; S302, cutting a maximum spanning tree; For each graph Calculating the maximum spanning tree MaxST, namely the spanning tree with the maximum sum of edge weights; At MaxST, all weights are removed Less than a category-specific threshold Is a side of (2); after the weak edges are removed, maxST is divided into a plurality of connected components, and all the superpoints in each connected component form an independent geographic entity instance 。
7. The method for rapidly extracting the basic mapping geographic entity based on the laser three-dimensional scanning of claim 6, wherein the detailed steps of S400 are as follows: S401, modeling entity geometric parameters; For each entity instance Performing fine modeling by using the corresponding original point cloud subset, wherein each entity Is endowed with a geometric model and physical attributes; S402, constructing a global semantic topological graph; Defining a global topology map : Each entity instance is taken as a node, and the attributes of the entity instance comprise a geometric model, a semantic category and a physical attribute; Representing the spatial relationship between entities.
8. The method for rapidly extracting basic mapping geographic entity based on laser three-dimensional scanning of claim 7, wherein the entity instance The modeling process of (1) includes: building entity, namely fitting a plurality of plane sheets by adopting a RANSAC algorithm, and generating a three-dimensional polygonal surface model with accurate height through plane intersection, contour extraction and regularization; performing cylindrical fitting on the trunk part, outputting the position, the breast diameter and the tree height, and calculating a convex hull or alpha-shape on the crown part to obtain a crown volume model; And (3) extracting the center line of the road point cloud, determining the road surface boundary according to the density edge or the intensity change of the point cloud, and generating the road surface model with the width.
9. The method for rapidly extracting basic mapping geographic entity based on laser three-dimensional scanning of claim 7, wherein the global topological graph is characterized in that The spatial relationship between the intermediate entities includes: calculating a buffer surface of the entity three-dimensional model, and if the intersection area of the two entity buffer surfaces exceeds a threshold value, establishing an adjacent relation; For non-ground entities, searching the nearest ground or building entity below the bottom surface of the non-ground entity, and if the vertical distance is smaller than a threshold value and the mass center projection of the upper entity falls in the projection of the lower entity, establishing a supporting relationship; and establishing the inclusion relationship if the three-dimensional model of one entity completely envelops the model of the other entity through three-dimensional space position analysis.
10. The system is applied to the laser three-dimensional scanning-based rapid basic surveying and mapping geographic entity extraction method as set forth in any one of claims 1 to 9, and is characterized by comprising the following steps: the data normalization and feature calculation module is used for receiving original laser point cloud data, performing self-adaptive density voxel downsampling to generate normalized point clouds, and calculating geometric, reflection and context features in a multi-scale neighborhood for each normalized point to form a multi-scale feature field of a scene; the hierarchical semantic segmentation module is used for constructing a hypergraph based on the regularized point cloud and the characteristic field thereof, and generating a refined hyperpoint set with a stable semantic tag through a spectral clustering and iterative refinement process; The entity instantiation module constructs a class similarity graph taking the superpoints as nodes according to each semantic class, and clusters the superpoints of the semantic classes into independent physical entity instances by calculating the maximum spanning tree and cutting at weak connection positions; The geometric modeling and topology reconstruction module is used for carrying out parameterized geometric modeling on each entity instance, automatically calculating the spatial topological relation among the entities and constructing a global semantic topological graph fusing geometric, semantic and topological information; And the data storage and output module is used for storing and outputting the global semantic topological graph and parameterized models and attribute information of each geographic entity.

Description

Quick extraction method and system for basic mapping geographic entity based on laser three-dimensional scanning Technical Field The invention belongs to the technical field of laser mapping, and particularly relates to a method and a system for rapidly extracting a basic mapping geographic entity based on laser three-dimensional scanning. Background With the increasing popularization and wide application of the three-dimensional laser scanning technology, the method for efficiently and accurately extracting basic geographic entities such as roads, buildings, vegetation and the like from massive point cloud data has become an important technical foundation and key link for constructing national space data infrastructures such as digital twin cities, live-action three-dimensional China and the like. The process not only relates to intelligent processing and automatic identification of large-scale point cloud data, but also directly relates to improvement of national space information service capability and deep promotion of smart city construction. The prior art mainly has the following limitations: (1) The degree of automation is insufficient, a large amount of manual intervention or complex parameter adjustment is relied on, and the method is difficult to be suitable for a large-scale and diversified scene; (2) The entity boundary is fuzzy, namely the segmentation precision is low at complex junctions (such as adhesion between buildings and vegetation) based on a segmentation method of single scale or single feature; (3) The existing method is multi-stop in point cloud classification or instance segmentation, and cannot construct a structural model for describing spatial relationships (such as connection and inclusion) and functional relationships (such as support and attachment) between entities, so that the requirements of advanced GIS analysis and application cannot be directly met. Therefore, the invention provides a method and a system for rapidly extracting a basic mapping geographic entity based on laser three-dimensional scanning. Disclosure of Invention In order to overcome the deficiencies of the prior art, at least one technical problem presented in the background art is solved. The technical scheme adopted for solving the technical problems is that the rapid extraction method of the basic surveying and mapping geographic entity based on laser three-dimensional scanning comprises the following steps: s100, constructing a point cloud data self-adaptive regularization and multi-scale characteristic field; performing density-adaptive voxelized downsampling on the original laser point cloud to generate a normalized representative point set, and calculating geometric, reflection and context characteristics from the original point cloud in a multi-scale neighborhood of each representative point to form a basic characteristic field penetrating all subsequent processes; S200, hierarchical semantic segmentation based on hypergraph iteration refinement; Constructing a hypergraph by taking the regularized points and the characteristic fields thereof as nodes, generating initial hyperpoints through spectral clustering, iteratively performing semantic classification and geometric consistency check, and dynamically merging or splitting the hyperpoints until convergence to obtain a refined hyperpoint set with stable semantic labels; S300, optimizing entity instantiation of maximum spanning tree cutting based on the graph; Aiming at each semantic category, constructing a subgraph by taking the superpoints of the category as nodes, reflecting the combined geometric and boundary compatibility among the superpoints by edge weights, and clustering the superpoints of the same category into independent physical entity examples by calculating a maximum spanning tree and cutting at weak joints; s400, reconstructing a global semantic topological graph through entity parameterization modeling; Performing geometric fitting and regularization on the point cloud of each entity instance to generate a parameterized model, calculating the spatial relationship among the entities, and constructing a global semantic topological graph containing complete attributes and relationships. Preferably, the detailed steps of S100 are: S101, self-adaptive density voxel downsampling; Acquiring an original laser point cloud ,; Dividing the point cloud space into an initial voxel grid, and calculating the local point density of each voxel vVoxel sizeSelf-adaptive adjustment according to local point density: With one representative point per voxel v Representation, formation of a normalized point cloud; S102, constructing a multi-scale characteristic field; Representative points for each voxel Three spherical neighborhoods with different physical radiuses are defined in the original point cloud P, namely small scaleMesoscale (mesoscale)Large scale; At each scaleIs a neighborhood of (a)In, calculate geometric featuresReflective featuresAnd cont