Urban three-dimensional point cloud data lightweight method considering ground feature types
By performing geographic type segmentation and feature factor calculation on urban three-dimensional point cloud data, and combining semantic information for weight assignment and sorting, the problems of low processing efficiency and poor retention of detailed features in large-scale urban point cloud data in the existing technology are solved. It achieves efficient and consistent lightweighting effect.
Patent Information
- Application Number
- CN202410912589.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-09
- Publication Date
- 2024-12-03
AI Technical Summary
The existing three-dimensional point cloud data lightweight method is inefficient when processing large-scale urban point cloud data, and cannot effectively retain the details of land objects, and fail to fully consider the types of land objects and texture color characteristics, resulting in inconsistent simplification effects.
By segmenting the three-dimensional point cloud data in urban areas according to the geographic type, building an octave tree, calculating feature factors such as normal vector difference, curvature difference, texture color difference and number of neighboring points, combining the semantic information of geographic type to give the feature factor personalized weights. , distinguish between feature points and non-feature points, and sort and simplify according to significant values.
It realizes efficient and lightweight data of point clouds in large-scene cities, ensures the integrity of land objects and the consistency of simplified effects, fully retains the detailed characteristics of land objects, and improves processing efficiency.
Smart Images

Figure CN119068109A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of three-dimensional data processing, and in particular relates to a lightweight method for urban three-dimensional point cloud data taking into account the type of land objects. Background Art
[0002] With the rapid development of data acquisition technology, 3D point cloud data has been widely used in many fields such as architectural design, urban planning, industrial manufacturing, etc. due to its comprehensive advantages such as high precision and fast acquisition speed. However, due to the large amount of original 3D point cloud data, high redundancy, and lack of explicit spatial relationship between points, it has caused many obstacles to subsequent editing, processing, analysis, visualization and other applications. Therefore, how to lightweight 3D point cloud data has become a very critical issue.
[0003] At present, there are two main methods for lightweighting 3D point clouds, namely indirect methods and direct methods. The indirect method first converts the point cloud data into triangular mesh data, and then lightweights it based on the existing mature mesh simplification method. Although this type of method can reuse the existing triangular mesh simplification algorithm, it usually takes a lot of time to build a triangular mesh based on a large number of discrete points, resulting in low overall efficiency. The direct method is to directly lightweight the 3D discrete points without any transformation. The main idea of this type of method is to first consider each point in the point cloud and its domain points to construct characteristic factors that can characterize the geometric characteristics of the point cloud object (such as changes in normal vectors, changes in curvature, etc.), and then use weight parameters to synthesize each characteristic factor to obtain a characteristic value, and use this to screen feature points.
[0004] [Reference document 1] Chunyang Ji, Ying Li, Jiahao Fan and Shumei Lan.ANovelSimplification Method for 3D Geometric Point Cloud Based on the Importance ofPoint.IEEE Access.2019.7.
[0005] This paper proposes a detail feature point simplification algorithm based on point importance, which considers four characteristic operators to describe the importance of points: the difference between the normal vector of the sample point and the normal vector of each neighbor point, the projection distance of the neighbor point to the sample tangent plane, the spatial distance between the sample point and each neighborhood point, and the difference between the curvature at the sample point and the curvature at each area. The importance of the sample point can be determined by adding these four characteristic operators. Through the preset threshold, points with importance greater than the threshold are called detail feature points and are retained. For non-detail feature points, this paper further constructs an octree, calculates the difference between the normal vector and the average normal vector of each point in the leaf node, and the difference between the curvature and the average curvature, calculates the sum of the differences, and retains the point with the smallest sum for simplification.
[0006] [Reference 2] Invention patent: A simplified method for high-precision 3D laser measurement of point cloud data for large scenes (CN107341825A)
[0007] This invention patent designs a method for simplifying point cloud data for high-precision three-dimensional laser measurement of large scenes. The method uses a uniform grid method to divide the scattered point cloud into spatial regions, extracts point cloud feature points based on projection residual values, and uses region division and surface variation values to simplify the original point cloud data.
[0008] The main disadvantages of the method proposed in reference document 1 are: 1) It is only applicable to simple 3D models, and the case model studied has a maximum of 48,000 points. Due to the complexity of the algorithm and poor scalability, it cannot process large-scale urban point cloud data (usually tens of millions or even hundreds of millions of points); 2) The texture and color features of the 3D point cloud are not considered; 3) Feature points and non-feature points are considered separately, the calculation is repeated, and the efficiency is low.
[0009] Although the method in reference document 2 can simplify large-scale three-dimensional point clouds with a data volume greater than 100 million, it still has some obvious shortcomings: 1) The uniform grid division area does not take into account the integrity of the ground objects. When the ground objects are divided and distributed in multiple areas, the simplification effect cannot guarantee consistency; 2) Too few feature factors are considered when extracting feature points. Different ground objects have different detail features. This method cannot effectively identify and retain certain detail features. Summary of the invention
[0010] In view of this, the present invention provides a lightweight method for urban three-dimensional point cloud data taking into account the type of land objects, which can overcome the problems and shortcomings of the existing methods and solve the problem of lightweight urban point cloud data in large scenes.
[0011] The technical solution for implementing the present invention is as follows:
[0012] A lightweight method for urban three-dimensional point cloud data considering the type of ground objects includes the following steps:
[0013] Step 1: Segment the urban 3D point cloud data according to the semantic information of the feature type to obtain multi-type feature objects. Construct an octree for any feature type and find the neighbor points within a sphere with a radius of r for each sample point in the point cloud. Assume that there are k neighbor points in the neighborhood of a sample point.
[0014] Step 2: Calculate four characteristic factors of the sample point based on the k neighboring points in the neighborhood: the mean of the normal vector difference of the neighboring points, the mean of the curvature difference, the mean of the texture color difference and the number of neighboring points, and use the linear normalization method to scale the value range of the four characteristic factors to [0, 1];
[0015] Step 3: Perform weighted summation of the four characteristic factors according to the type of land feature, and use it as the significant value s of the sample point p p ; Distinguish feature points and non-feature points in the point cloud according to the size of the saliency value to form a feature point set and a non-feature point set;
[0016] Step 4: traverse each octree node in the point set, sort the sample points in each node from large to small according to the significance value, retain the sample points with larger significance value in the node according to the simplification ratio of the two types of point sets, and delete the sample points with smaller significance value; merge the simplified feature point set and non-feature point set and output the point cloud file in the corresponding format.
[0017] Furthermore, the types of features include terrain, buildings, vegetation, vehicles and other urban components.
[0018] Furthermore, the significant value s p for:
[0019] s p =αX p +βY p +γZ p +δW p (6)
[0020] Where α, β, γ, and δ are the mean values of the normal vector differences X p , mean value of curvature difference Y p , the mean value Z of texture color difference p And the number of neighbor points in the neighborhood W p The weight of .
[0021] Furthermore, a threshold is set, and a point whose significance value is greater than the set threshold is a feature point, and a point whose significance value is not greater than the threshold is a non-feature point.
[0022] Furthermore, the calculation process of the simplified ratio of the feature point set and the simplified ratio of the non-feature point set is as follows:
[0023] Different weights w are set for feature point sets and non-feature point sets according to the actual needs of usersf and w uf , calculate the simplification ratio of different point sets according to the set overall simplification ratio R:
[0024]
[0025] In the formula, w f Represents the weight of the feature point set, w uf represents the weight of the non-feature point set, N is the number of vertices of the simplified point set, N f To simplify the number of vertices in the previous feature point set, N uf is the number of vertices of the non-feature point set before simplification, R is the simplification ratio of the entire 3D point cloud, and R f is the simplified ratio of the feature point set, R uf is the simplified ratio of the non-feature point set.
[0026] Beneficial effects:
[0027] 1) The method of the present invention solves the problem of lightweighting of large-scale urban point cloud data, and uses urban objects of the same type as segmentation units and simplification basic units to ensure the integrity of the objects and the consistency of the simplification effect.
[0028] 2) The present invention couples four characteristic factors, namely, the mean of normal vector difference, the mean of curvature difference, the mean of texture color difference and the number of neighboring points in the neighborhood, to comprehensively judge the characteristic degree of the point.
[0029] 3) The present invention combines the semantic information of the feature type to which each point belongs, and assigns personalized weights to each characteristic factor in different feature types, thereby simplifying the amount of data while fully considering the characteristics of different features and better preserving the features of the features.
[0030] 4) The present invention divides point sets of different types of ground objects into feature point sets and non-feature point sets, and assigns different weights to them, so as to balance retaining the detailed features of different types of ground objects and improving the simplification ratio as much as possible. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION
[0032] The present invention is described in detail below with reference to the accompanying drawings and embodiments.
[0033] The present invention provides a lightweight method for urban 3D point cloud data that takes into account the type of landforms, aiming to solve the problem of lightweight urban point cloud data in large scenes. It uses landforms of the same type as the basic unit for simplification to ensure the integrity of the landforms and the consistency of the simplification effect. It couples the four characteristic factors of normal vector, curvature, texture color and number of neighboring points to comprehensively judge the characteristic degree of the point, simplifying the 3D point cloud of large scenes while fully retaining the detailed characteristics of the landforms. The specific process of the method of the present invention is as follows: Figure 1 As shown:
[0034] Step 1: Segment the urban 3D point cloud data according to the semantic information of the object type to obtain multiple types of objects, such as terrain, vegetation, buildings, vehicles and other urban components. Use existing 3D point cloud segmentation algorithms, such as PointNet++, to process. The following operations will be simplified for specific types of objects.
[0035] Step 2: Build an octree for the segmented object to conveniently and efficiently find the neighbor points within a sphere with a radius of r for each point in the point cloud (called a sample point). Assume that there are k neighbor points in the neighborhood of a sample point.
[0036] Step 3: According to the k neighboring points in the neighborhood, calculate the four characteristic factors of the sample point respectively: the mean of the normal vector difference of the neighboring points, the mean of the curvature difference, the mean of the texture color difference and the number of neighboring points. The specific calculation method is as follows:
[0037] 1) Mean of normal vector difference: After constructing the neighborhood, calculate the covariance matrix C of the neighborhood point set, which reflects the distribution of neighborhood points relative to sample points. The calculation formula of the covariance matrix is as follows:
[0038] (1).
[0039]
[0040] In the formula, p i represents a neighbor point, p is a sample point, and k is the number of neighbor points.
[0041] After calculating the covariance matrix, use eigenvalue decomposition to find the eigenvectors and eigenvalues λ j (j=0, 1, 2), where the smallest eigenvector v0 is the normal vector n of the sample point p. p Then calculate the normal vector n of the sample point p p With neighboring point p j The normal vector n of (1≤j≤k) j The cosine angle of (1≤j≤k) is accumulated and averaged to get the mean value X of the normal vector difference at the sample point. p .
[0042]
[0043] 2) Mean of curvature difference: Calculate the curvature at the sample point p and the curvature at the neighboring point p j These curvature differences are then accumulated and averaged to reflect the sharpness at the sample point p.
[0044]
[0045] In the formula, c p represents the curvature value at the sample point p, c j represents the curvature of the jth neighbor point, Y p Represents the mean curvature difference between the sample point p and its neighboring points.
[0046] 3) Mean of texture color difference: Calculate the texture color (Rp, Gp, Bp) at the sample point p and the neighboring point p j Texture color (R j , G j , B j ) between the sample points (expressed as Euclidean distance), and then accumulate these color Euclidean distance differences and calculate the average to reflect the texture color difference between the sample point and the neighborhood point.
[0047]
[0048] In the formula, Z p That is, the mean value of the texture color difference of sample point p.
[0049] 4) Number of neighbor points in the neighborhood W p : Calculate the number of k-neighborhood points of the sample point p to reflect the sparsity of the points in the neighborhood of the sample point.
[0050] Step 4: Use formula (5) to normalize each characteristic factor calculated in step 3, and use the linear normalization method to scale the value range of the four characteristic factors to the range of [0, 1].
[0051]
[0052] In the formula, E p is the value of various characteristic factors of the normalized previous point p, E min is the minimum value of the characteristic factor, E max is the maximum value of the characteristic factor, E p′ is the value of the characteristic factor of point p after normalization.
[0053] Step 5: Combine the semantic information of the feature type to which each point belongs, assign personalized weights to the characteristic factors of different types of features, and calculate the comprehensive weighted feature value of each point in the point cloud of each type of feature as the saliency value s of the sample point p. p .
[0054] s p =αX p +βY p +γZ p +δW p (6)
[0055] Where α, β, γ, and δ are the mean values of the normal vector differences X p , mean value of curvature difference Y p , the mean value Z of texture color difference p And the number of neighbor points in the neighborhood W p The weight of .
[0056] Step 6: Set a threshold to distinguish feature points from non-feature points in the point cloud, forming a feature point set and a non-feature point set; due to the different importance of the two, different weights w are set for the two point sets based on the actual needs of the user f and w uf , the simplification ratio of different point sets is calculated according to the set overall simplification ratio R. The formula for calculating the simplification ratio of feature point sets and non-feature point sets is:
[0057]
[0058] In the formula, w f Represents the weight of the feature point set, w uf represents the weight of the non-feature point set, N is the number of vertices of the simplified point set, N f To simplify the number of vertices in the previous feature point set, N uf is the number of vertices of the non-feature point set before simplification, R is the simplification ratio of the entire 3D point cloud, and R f is the simplified ratio of the feature point set, R uf is the simplified ratio of the non-feature point set.
[0059] Step 7: According to the calculated simplification ratio, simplify the feature point set and non-feature point set respectively. The specific process of simplification is: traverse each octree node in the point set, sort all the sample points in each node from large to small according to the significance value calculated in step 5, retain the points with larger significance value according to the simplification ratio, and delete the points with smaller significance value.
[0060] Step 8: Merge the simplified eigenvalue point set and the non-feature point set and output the point cloud file in the corresponding format.
[0061] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A lightweight method for urban three-dimensional point cloud data considering the type of ground objects, characterized in that: The following steps are involved: Step 1: Segment the urban 3D point cloud data according to the semantic information of the feature type to obtain multi-type feature objects. Construct an octree for any feature type and find the neighbor points within a sphere with a radius of r for each sample point in the point cloud. Assume that there are k neighbor points in the neighborhood of a sample point. Step 2: Calculate the four characteristic factors of the sample point based on the k neighboring points in the neighborhood: the mean of the normal vector difference of the neighboring points, the mean of the curvature difference, the mean of the texture color difference and the number of neighboring points, and use the linear normalization method to scale the value range of the four characteristic factors to [0,1]; Step 3: Perform weighted summation of the four characteristic factors according to the type of land feature, and use it as the significant value s of the sample point p p ; Distinguish feature points and non-feature points in the point cloud according to the size of the saliency value to form a feature point set and a non-feature point set; Step 4: traverse each octree node in the point set, sort the sample points in each node from large to small according to the significance value, retain the sample points with larger significance value in the node according to the simplification ratio of the two types of point sets, and delete the sample points with smaller significance value; merge the simplified feature point set and non-feature point set and output the point cloud file in the corresponding format.
2. The lightweight method for urban three-dimensional point cloud data according to claim 1, characterized in that: Land feature types include terrain, buildings, vegetation, vehicles and other urban components.
3. The lightweight method for urban three-dimensional point cloud data according to claim 1, characterized in that: Significant value p for: s p =αX p +γY p +γZ p +δW p (6) Where α, β, γ, and δ are the mean values of the normal vector differences X p , mean value of curvature difference Y p , the mean value Z of texture color difference p And the number of neighbor points in the neighborhood W p The weight of .
4. The lightweight method for urban three-dimensional point cloud data according to claim 1, characterized in that: Set a threshold, and the points with a significant value greater than the set threshold are considered feature points, and the points with a significant value not greater than the threshold are considered non-feature points.
5. The lightweight method for urban three-dimensional point cloud data according to claim 1 or 3, characterized in that: The calculation process of the simplified ratio of the feature point set and the simplified ratio of the non-feature point set is as follows: Different weights w are set for feature point sets and non-feature point sets according to the actual needs of users f and w uf , calculate the simplification ratio of different point sets according to the set overall simplification ratio R: In the formula, w f Represents the weight of the feature point set, w uf represents the weight of the non-feature point set, N is the number of vertices of the simplified point set, N f To simplify the number of vertices in the previous feature point set, N uf is the number of vertices of the non-feature point set before simplification, R is the simplification ratio of the entire 3D point cloud, and R f is the simplified ratio of the feature point set, R uf is the simplified ratio of the non-feature point set.
Citation Information
Patent Citations
Simplification method used for point cloud data of large-scene high-precision three-dimension laser measurement
CN107341825A
Cited By
Regional simplification method of building three-dimensional model based on deep learning
CN121330217A
Lightweight BIM model construction method for urban civil engineering quantity calculation
CN121392196A
Lightweight BIM model construction method for urban civil engineering quantity calculation
CN121392196B
