CN-114972628-B - Building three-dimensional extraction method based on multispectral laser radar point cloud data

CN114972628BCN 114972628 BCN114972628 BCN 114972628BCN-114972628-B

Abstract

The invention relates to a building three-dimensional extraction method based on multispectral laser radar point cloud data, and belongs to the technical field of multispectral laser radar point clouds. Firstly, extracting dimension characteristics of multispectral laser radar point cloud data, selecting an optimal neighborhood by using shannon entropy theory, extracting geometric characteristics on the optimal neighborhood, dividing the multispectral laser radar point cloud data into three types of lines, planes and discrete by using a support vector machine, extracting multispectral laser radar point cloud data of two-dimensional plane types, and finally extracting multispectral laser radar point cloud building data through height and neighborhood variance filtering. According to the method, the optimal neighborhood is selected for feature extraction, so that the classification accuracy is improved, and the accuracy of multi-spectrum laser radar point cloud data building three-dimensional extraction is improved.

Inventors

WANG QINGWANG
ZHANG ZIFENG
SHEN TAO
SONG JIAN
SHEN SHIQUAN
XIONG HAO

Assignees

昆明理工大学

Dates

Publication Date: 20260505
Application Date: 20220414

Claims (3)

1. A building three-dimensional extraction method based on multispectral laser radar point cloud data is characterized by comprising the following steps of: Step1, screening multispectral laser radar point cloud data according to a spectrum angle, and for each point in the multispectral laser radar point cloud data, marking the point with the spectrum angle smaller than the threshold value into a neighborhood by setting different spectrum angle thresholds, and respectively taking six neighborhoods with the sizes of 10, 20, 50, 100, 150 and 200 points for each point in the multispectral laser radar point cloud as the preparation neighborhood of the point by selecting different spectrum angle thresholds so as to select the optimal neighborhood from the preparation neighborhood in the subsequent Step; step2, constructing a covariance matrix on six prepared neighborhoods selected in Step1 by utilizing three-dimensional coordinates of the point and a neighborhood point thereof for each point of the multispectral laser radar point cloud, performing principal component analysis on the covariance matrix to obtain three characteristic values of the point in the current neighborhood, and calculating dimension characteristics of the point in the current neighborhood according to a formula, wherein the three characteristic values are arranged in descending order of lambda 1 、λ 2 、λ 3 : Wherein a 1D is a one-dimensional linear feature, a 2D is a two-dimensional planar feature, a 3D is a three-dimensional discrete feature, the sum of a 1D 、a 2D 、a 3D is 1, and the dimensional features respectively represent the probability that the point belongs to three different dimensional types; Step3, extracting an optimal neighborhood in the preparation neighborhood according to the shannon entropy theory; Step4, re-extracting geometrical characteristics of the multispectral laser radar point cloud data on the optimal neighborhood, wherein the geometrical characteristics comprise dimension characteristics, surface roughness, plane gradient and intensity roughness; Wherein SR is surface roughness, PS is plane gradient, IR is intensity roughness; step5, inputting geometrical characteristics of the multispectral laser radar point cloud data into a support vector machine for classification, classifying the geometrical characteristics into multispectral laser radar point clouds of three different types, namely one-dimensional linear lines, two-dimensional planar shapes and three-dimensional discrete shapes, extracting multispectral laser radar point cloud data belonging to two-dimensional plane types from the multispectral laser radar point cloud data, and removing ground and non-building planes through height and neighborhood variance filtering to obtain multispectral laser radar point cloud building data.
2. The multi-spectrum laser radar point cloud data building three-dimensional extraction method according to claim 1, wherein Step3 is specifically: On the preparation neighborhoods with different sizes selected for the points of each multispectral laser radar point cloud data in Step1, dimension features are calculated according to the feature values of singular decomposition of the preparation neighborhoods, and entropy values are calculated by using the dimension features of the laser radar point cloud data: Wherein E f represents the calculation of the entropy value using shannon entropy theory, Representing a P-th point neighborhood of size N; Each point on the multispectral laser radar point cloud data is optimized through an optimizer, and the dimensional characteristics of the point are calculated by substituting the formula in Step1 into the optimal neighborhood again; the method comprises the following steps: Wherein the method comprises the steps of And (3) carrying out neighborhood optimization on the P point, and selecting the optimal neighborhood with the minimum entropy value from the prepared neighbors.
3. The method for three-dimensional extraction of multi-spectral lidar point cloud data of claim 1, wherein in Step5, the extracted two-dimensional planar type multi-spectral lidar point cloud data comprises multi-spectral lidar point cloud data distributed in a plane on the ground and on the top surface and other surfaces of the building.

Description

Building three-dimensional extraction method based on multispectral laser radar point cloud data Technical Field The invention relates to a building three-dimensional extraction method based on multispectral laser radar point cloud data, and belongs to the technical field of multispectral laser radar point clouds. Background The automatic extraction of building information in a remote sensing scene is a hot spot problem in remote sensing application research, but the automatic extraction of the building is very difficult due to the influence of a plurality of factors such as different imaging conditions, complex background ground features, various building types and the like. Traditional building extraction based on spectral images is limited by the self limitation of an imaging method, and only two-dimensional plane information of ground objects of a remote sensing scene can be extracted, so that the application of the method in three-dimensional extraction of buildings is limited. However, the traditional LiDAR can only extract three-dimensional space information of a remote sensing scene, lacks spectrum information and still has limitations. The spectrum image and LiDAR data are combined, and two-position spectrum information and three-dimensional space information can be obtained at the same time, but the combined utilization tends to bring about the problems of difficult image alignment, spectrum mismatch, spectrum drift and the like, which brings about a greater difficulty for practical application. With the development of the supercontinuum laser technology, the multispectral LiDAR is applied from research and development, and a multispectral LiDAR system with multispectral information detection capability is used for synchronously acquiring three-dimensional spatial distribution information and spectrum information of a target, so that data basis and information guarantee are provided for three-dimensional fine classification of ground features of a remote sensing scene. The national "civil space infrastructure long-term development planning (2015-2025)" lists multispectral LiDAR in a key direction requiring advanced deployment and development. In the multispectral laser radar remote sensing ground object classification work, three-dimensional space-spectrum integrated information acquisition of an observation target is realized, the problems of mismatch, drift and the like caused by joint utilization of spectrum images and LiDAR data are fundamentally solved, meanwhile, complementary information between the spectrum images and the LiDAR data is considered, a data basis and information guarantee are provided for remote sensing scene building extraction work, and the building three-dimensional extraction precision is greatly improved. Disclosure of Invention The technical problem to be solved by the invention is to provide a building three-dimensional extraction method based on multispectral laser radar point cloud data, which is used for solving the problems of classification misalignment and the like caused by different types of ground features being partitioned into the same neighborhood when the point cloud neighborhood is selected by the traditional method. According to the technical scheme, the building three-dimensional extraction method based on the multispectral laser radar point cloud data comprises the steps of firstly extracting dimension characteristics of the multispectral laser radar point cloud data, selecting an optimal neighborhood by using shannon entropy theory, extracting geometric characteristics on the optimal neighborhood, dividing the multispectral laser radar point cloud data into three types of lines, planes and discretization by using a support vector machine, extracting multispectral laser radar point cloud data of a 2D plane type, and finally extracting multispectral laser radar point cloud building data through height and neighborhood variance filtering. The method comprises the following specific steps: Step1, screening multispectral laser radar point cloud data according to a spectrum angle, and for each point in the multispectral laser radar point cloud data, marking the point with the spectrum angle smaller than a threshold value into a neighborhood by setting different spectrum angle threshold values. Selecting different spectrum angle thresholds, and respectively taking six neighborhoods with the sizes of 10, 20, 50, 100, 150 and 200 points for each point in the multispectral laser radar point cloud as the preparation neighborhoods of the point so as to select the optimal neighborhoods from the preparation neighborhoods in the subsequent steps; step2, constructing a covariance matrix on six preparation neighborhoods selected in Step1 of each point of the multispectral laser radar point cloud by utilizing three-dimensional coordinates of the point and the neighborhood point thereof, performing principal component analysis on the covariance matrix to obtain three ch