CN104318622B - A Triangular Mesh Modeling Method for Inhomogeneous 3D Point Cloud Data in Indoor Scenes - Google Patents

Abstract

The invention relates to a triangular mesh modeling method, in particular to a triangular mesh modeling method of indoor scene non-uniform three-dimensional point cloud data, which comprises the following steps: step 1, acquiring nonuniform three-dimensional point cloud data of an indoor scene: step 2, projection of a neighborhood tangent plane: step 3, optimizing the projection neighborhood N': step 4, obtaining the Delaunay adjacent edge of the given point p: and 5, finishing triangular mesh modeling: and repeating the steps 2-4, repeating the algorithm for each point, and then completing the triangular mesh modeling of the whole three-dimensional point cloud. According to the method, the sector area is uniformly divided and each direction is selected in a balanced mode, the adjacent points of the given point are well selected in each direction, the distribution of the adjacent points in each direction is more balanced, and accurate modeling of the non-uniform three-dimensional point cloud data of the indoor scene is achieved. Meanwhile, the selection of each nearest neighbor point also effectively simplifies the neighborhood of a given point, so that the method has lower running time and higher modeling efficiency.

Description

A Triangular Mesh Modeling Method for Inhomogeneous 3D Point Cloud Data in Indoor Scenes

technical field

The present invention relates to a triangular mesh modeling method, in particular to a triangular mesh modeling method for non-uniform three-dimensional point cloud data of an indoor scene.

Background technique

With the development of 3D scanning ranging technology, 3D point cloud data is widely used in reverse engineering, industrial inspection, autonomous navigation and other fields. 3D point cloud data processing technology has played a vital role as the basis for realizing the above applications. In the 3D point cloud data processing technology, the triangular mesh modeling of 3D point cloud data is a very key technology. Since the indoor environment is a structured scene, the triangular mesh modeling technology is especially suitable for the 3D modeling of the indoor scene. It can not only describe the indoor scene vividly, but also lay a good foundation for the classification and target recognition of the indoor scene. . The introduction of an excellent triangular mesh modeling method can greatly improve the actual application situation and improve application performance. When obtaining 3D point cloud data of indoor scenes, the working characteristics of the laser ranging equipment and the sudden change of the indoor environment structure can easily cause the instability of the scanning line spacing, which makes the distribution of 3D point cloud data extremely unstable. Uniformity brings great difficulties to triangular mesh modeling of indoor scenes. Triangular mesh modeling of 3D point cloud data has always been a research hotspot in the field of 3D point cloud data processing, and its modeling methods can be roughly divided into two categories: modeling methods based on Delaunay triangulation and modeling methods based on region growth. Generally speaking, although the modeling method based on Delaunay triangulation has good running results, it requires a large number of calculations, so that the execution efficiency of the algorithm is low and the modeling speed is slow; the modeling method of region growth has good The operation efficiency and the modeling speed are fast, but sometimes the modeling effect is not ideal. For the triangular mesh modeling of non-uniform 3D point cloud data, there are rare reports at home and abroad, and the existing triangular mesh modeling technology is no longer applicable. For example, the well-known ball-pivoting algorithm (Ball-Pivoting Algorithm) needs to run multiple times according to different sizes of balls to process uneven 3D point cloud data, and sometimes the results are not ideal; another example, based on 2D Delaunay triangulation The surface reconstruction method focuses on the construction of the Delaunay adjacent edges of the sampling points in the tangent plane, and back-projects them into the three-dimensional space to form a triangular mesh model. This method runs faster, but the sampling conditions There are strict restrictions on the point cloud distribution and cannot handle non-uniform 3D point cloud data. In each local triangulation, it is difficult to obtain accurate Delaunay adjacent edges, and the overall modeling effect is poor.

Contents of the invention

In order to overcome the deficiencies in the prior art, the object of the present invention is to provide a triangular mesh modeling method for non-uniform 3D point cloud data of indoor scenes. This method is aimed at an indoor scene. First, the laser scanning rangefinder is used to obtain the non-uniform 3D point cloud data of the indoor scene, which is essentially a non-uniform point set in the 3D space. Then the point set is obtained by a certain surface modeling method. Constructed into a triangular mesh topology to accurately describe real indoor scenes. This method solves the problems of low modeling quality, inability to describe the actual scene, and deviation from the real topology due to the uneven distribution of point cloud data, and also has a faster modeling speed.

In order to achieve the purpose of the above invention and solve the problems in the prior art, the technical solution adopted by the present invention is: a triangular mesh modeling method for non-uniform three-dimensional point cloud data of indoor scenes, comprising the following steps:

Step 1. Acquire non-uniform 3D point cloud data of indoor scenes: Obtain indoor scene information as non-uniform 3D point cloud data through laser sensors;

Step 2. Neighborhood tangent plane projection: select a given point p=(x, y, z), calculate the average distance d of the 3D point cloud, set 5d as the neighborhood radius, and obtain the neighborhood N={p of the given point _i =(x _i , y _i , z _i )|1≤i≤k}, where: p _i is the neighbor point, i is the serial number of the neighbor point, k is the number of neighbor points, passing through the neighborhood around point p N, calculate the normal vector n of the point p; through the normal vector n, construct the tangent plane T at point p, and project the adjacent point p _i onto the tangent plane T, and record the projected point set as the projected neighborhood N ';

Step 3. Optimizing the projected neighborhood N′: Through the uniform division of the fan-shaped area and the balanced selection in all directions, further optimize the projected neighborhood N′ of the given point p, so that the given point p has the nearest projected neighbors in all directions , record the optimized point set as the optimized neighborhood N″;

Step 4. Obtain the Delaunay adjacent edge of the given point p: use the two-dimensional Delaunay method to carry out triangular mesh modeling on the given point p and its optimized neighborhood N″, and reversely map the obtained two-dimensional Delaunay triangular mesh to the three-dimensional adjacent domain space, and extract and store the 3D Delaunay adjacent edges connected to a given point p;

Step 5, complete triangular mesh modeling: repeat steps 2-4, repeat the above algorithm for each point, and then complete the triangular mesh modeling of the entire 3D point cloud.

The step 2 neighborhood tangent plane projection specifically includes the following sub-steps:

Step (a), calculate the average distance d of the three-dimensional point cloud, set 5d as the neighborhood radius, and extract the neighborhood N of the given point p;

Step (b), through the formula

Find the covariance matrix M of the neighborhood N, where: p _i is the neighbor point, i is the serial number of the neighbor point, k is the number of neighbor points, and T is the vector transposition symbol, which transposes the column vector into a row vector;

Step (c), obtaining the eigenvalues λ ₁ , λ ₂ , λ ₃ of M (λ ₁ <λ ₂ <λ ₃ ), and the corresponding eigenvectors v ₁ , v ₂ , v ₃ ;

Step (d), unitize the eigenvector v1 corresponding to the minimum eigenvalue λ1, namely obtain the normal vector _{n of} the given point p;

Step (e), project each point in the neighborhood N onto the tangent plane T corresponding to the normal vector n, and record the projected point set as the projected neighborhood N'.

The step 3 optimizes the projected neighborhood N', specifically including the following sub-steps:

Step (a), in the tangent plane T, take the given point p as the origin, construct a straight line passing through the origin p, and make it rotate clockwise from the horizontal position for one circle, divide a fan-shaped area at an interval of 22.5 degrees, and finally form 16 fan-shaped areas centered on the origin p;

Step (b), placing the projection neighborhood N' in the tangent plane T of point p into the fan-shaped area divided in step (a);

Step (c), extracting the points closest to the origin p in each fan-shaped area, these points constitute an optimized point set, which is recorded as the optimized neighborhood N″.

The step 4 obtains the Delaunay adjacent edge of the given point p, which specifically includes the following sub-steps:

Step (a), using the divide and conquer method, divide the optimized neighborhood N" into several small areas according to the x coordinates, and the number of points in each small area is not more than 3. For each small area, it is required can guarantee to meet the Delaunay criterion;

Step (b), integrating adjacent small areas into a larger area meeting the Delaunay criterion;

Step (c), repeat step (b), and then merge each larger area layer by layer until all larger areas are merged into a whole, so far, a single local Delaunay triangulation is completed; according to the uniqueness of Delaunay triangulation Criterion, the only Delaunay triangulation result can be obtained;

Step (d), extracting the edge connected to the given point p from the constructed two-dimensional Delaunay triangulation result, mapping it back to the three-dimensional neighborhood space, and storing it, the stored edge is the edge of the given point p Three-dimensional Delaunay edges.

The beneficial effect of the present invention is: a triangular mesh modeling method for indoor scene non-uniform three-dimensional point cloud data, comprising the following steps: Step 1, obtaining indoor scene non-uniform three-dimensional point cloud data: using a laser sensor to obtain indoor scene information, As non-uniform 3D point cloud data; step 2, neighborhood tangent plane projection: select a given point p=(x, y, z), calculate the average distance d of the 3D point cloud, set 5d as the neighborhood radius, and obtain the given point The neighborhood of a fixed point N={p _i =( _xi ,y _i , _zi )|1≤i≤k}, where: p _i is the neighbor point, i is the serial number of the neighbor point, and k is the number of neighbor points , through the neighborhood N around the point p, calculate the normal vector n of the point p; through the normal vector n, construct the tangent plane T at the point p, and project the neighboring point p _i onto the tangent plane T, remember the projection The point set of is the projected neighborhood N′; step 3, optimize the projected neighborhood N′: through the uniform division of the fan-shaped area and the balanced selection in all directions, further optimize the projected neighborhood N′ of the given point p, so that the given point p is in There are the nearest projected neighbors in each direction, and the optimized point set is recorded as the optimized neighborhood N″; step 4, obtain the Delaunay adjacent edge of the given point p: use the two-dimensional Delaunay method to calculate the given point p and its optimized neighborhood N″ carry out triangular mesh modeling, reversely map the obtained two-dimensional Delaunay triangular mesh to the three-dimensional neighborhood space, and extract and store the three-dimensional Delaunay adjacent edges connected to the given point p; step 5, complete the triangular mesh construction Modeling: Repeat steps 2-4, repeat the above algorithm for each point, and then complete the triangular mesh modeling of the entire 3D point cloud. Compared with the prior art, the present invention utilizes the uniform division of the fan-shaped area and the balanced selection in all directions, and better selects the neighbors of a given point in each direction, so that the distribution of the neighborhood in each direction is more balanced, and the realization of Accurate Modeling of Inhomogeneous 3D Point Cloud Data for Indoor Scenes. At the same time, the selection of the nearest neighbor points in each direction also effectively simplifies the neighborhood of a given point, making this method have lower running time and higher modeling efficiency.

Description of drawings

Fig. 1 is a flow chart of the present invention.

Figure 2 is a schematic diagram of the steps of the present invention.

In the figure: (a) is the neighborhood tangent plane projection graph, (b) is the optimized graph for the projected neighborhood N′, (c) is the Delaunay neighbor graph for obtaining a given point p, (d) is the graph under the tangent plane Delaunay modeling diagram.

Fig. 3 is a schematic diagram of local plane triangular mesh modeling without using the present invention.

Fig. 4 is a diagram of the triangular mesh modeling results of the non-uniform 3D point cloud data of the indoor scene 1.

In the figure: (a) is the overall rendering, (b) and (c) are the partial details.

Fig. 5 is a triangular mesh modeling result diagram of indoor scene 2 non-uniform 3D point cloud data;

In the figure: (a) is the overall rendering, (b) and (c) are the partial details.

detailed description

The present invention will be further described below in conjunction with accompanying drawing.

As shown in Figure 1, a triangular mesh modeling method for non-uniform 3D point cloud data of indoor scenes comprises the following steps: Step 1, obtaining non-uniform 3D point cloud data of indoor scenes: obtaining indoor scene information through a laser sensor, As non-uniform 3D point cloud data; step 2, neighborhood tangent plane projection: select a given point p=(x, y, z), calculate the average distance d of the 3D point cloud, set 5d as the neighborhood radius, and obtain the given point The neighborhood of a fixed point N={p _i =( _xi ,y _i , _zi )|1≤i≤k}, where: p _i is the neighbor point, i is the serial number of the neighbor point, and k is the number of neighbor points , through the neighborhood N around the point p, calculate the normal vector n of the point p; through the normal vector n, construct the tangent plane T at the point p, and project the neighboring point p _i onto the tangent plane T, remember the projection The point set of is the projected neighborhood N′; step 3, optimize the projected neighborhood N′: through the uniform division of the fan-shaped area and the balanced selection in all directions, further optimize the projected neighborhood N′ of the given point p, so that the given point p is in There are the nearest projected neighbors in each direction, and the optimized point set is recorded as the optimized neighborhood N″; step 4, obtain the Delaunay adjacent edge of the given point p: use the two-dimensional Delaunay method to calculate the given point p and its optimized neighborhood N″ carry out triangular mesh modeling, reversely map the obtained two-dimensional Delaunay triangular mesh to the three-dimensional neighborhood space, and extract and store the three-dimensional Delaunay adjacent edges connected to the given point p; step 5, complete the triangular mesh construction Modeling: Repeat steps 2-4, repeat the above algorithm for each point, and then complete the triangular mesh modeling of the entire 3D point cloud.

The step 2 neighborhood tangent plane projection specifically includes the following sub-steps:

Step (a), calculate the average distance d of the three-dimensional point cloud, set 5d as the neighborhood radius, and extract the neighborhood N of the given point p;

Step (b), through the formula

Find the covariance matrix M of the neighborhood N, where: p _i is the neighbor point, i is the serial number of the neighbor point, k is the number of neighbor points, and T is the vector transposition symbol, which transposes the column vector into a row vector;

Step (c), obtaining the eigenvalues λ ₁ , λ ₂ , λ ₃ of M (λ ₁ <λ ₂ <λ ₃ ), and the corresponding eigenvectors v ₁ , v ₂ , v ₃ ;

Step (d), unitize the eigenvector v1 corresponding to the minimum eigenvalue λ1, namely obtain the normal vector _{n of} the given point p;

Step (e), each point in the neighborhood N is projected onto the tangent plane T corresponding to the normal vector n, and the projected point set is recorded as the projected neighborhood N'; as shown in Figure 2 (a) shown.

The step 3 optimizes the projected neighborhood N', specifically including the following sub-steps:

Step (a), in the tangent plane T, take the given point p as the origin, construct a straight line passing through the origin p, and make it rotate clockwise from the horizontal position for one circle, divide a fan-shaped area at an interval of 22.5 degrees, and finally form 16 fan-shaped areas centered on the origin p;

Step (b), placing the projection neighborhood N' in the tangent plane T of point p into the fan-shaped area divided in step (a);

Step (c), extracting the points closest to the origin p in each fan-shaped area, these points constitute the optimized point set, which is recorded as the optimized neighborhood N"; as shown in Figure 2 (b).

The step 4 obtains the Delaunay adjacent edge of the given point p, which specifically includes the following sub-steps:

Step (a), using the divide and conquer method, divide the optimized neighborhood N" into several small areas according to the x coordinates, and the number of points in each small area is not more than 3. For each small area, it is required can guarantee to meet the Delaunay criterion;

Step (b), integrating adjacent small areas into a larger area meeting the Delaunay criterion;

Step (c), repeat step (b), and then merge each larger area layer by layer until all larger areas are merged into a whole, so far, a single local Delaunay triangulation is completed; according to the uniqueness of Delaunay triangulation Criterion, the only Delaunay triangulation result can be obtained;

Step (d), extracting the edge connected to the given point p from the constructed two-dimensional Delaunay triangulation result, mapping it back to the three-dimensional neighborhood space, and storing it, the stored edge is the edge of the given point p Three-dimensional Delaunay edge; as shown in Figure 2 (c). Then repeat the operation, repeat steps 2-4 for each point in the original point cloud data, and obtain the final tangent plane modeling as shown in Figure 2 (d). All stored edges are displayed according to the triangulation results, and the triangular mesh modeling results shown in Fig. 4 and Fig. 5 can be obtained.

To sum up, the present invention does not directly model the discrete points in the 3D point cloud with triangular meshes, but first performs neighborhood tangent plane projection on them, and then uses fan-shaped area division and isotropic balance selection to optimize the given points. The fixed-point projection neighborhood makes the distribution in all directions more balanced, and finally uses the balanced optimization neighborhood to realize the establishment of the local triangular mesh model. This not only solves the problem of accurate modeling of non-uniform 3D point cloud data in indoor scenes, but also avoids the generation of large-area holes in the modeling process. As shown in Figure 3, it truly describes the topology of indoor scenes. Existence, which makes the number of discrete points drop sharply during local neighborhood modeling, avoids the huge operating overhead caused by modeling a large number of data points in the local area, greatly improves the operating efficiency of modeling, and reduces the time required for modeling. operation hours.

Claims (4)

1. a triangular mesh modeling method of indoor scene heterogeneous three-dimensional point cloud data, is characterized in that comprising the following steps: Step 1. Acquire non-uniform 3D point cloud data of indoor scenes: Obtain indoor scene information as non-uniform 3D point cloud data through laser sensors; Step 2. Neighborhood tangent plane projection: select a given point p=(x,y,z), calculate the average distance d of the 3D point cloud, set 5d as the neighborhood radius, and obtain the neighborhood N={p of the given point _i ＝(x _i , y _i , z _i )|1≤i≤k}, where: p _i is the neighbor point, i is the serial number of the neighbor point, k is the number of neighbor points, passing through the neighborhood around point p N, calculate the normal vector n of the point p; through the normal vector n, construct the tangent plane T at point p, and project the adjacent point p _i onto the tangent plane T, and record the projected point set as the projected neighborhood N '; Step 3. Optimizing the projected neighborhood N′: Through the uniform division of the fan-shaped area and the balanced selection in all directions, further optimize the projected neighborhood N′ of the given point p, so that the given point p has the nearest projected neighbors in all directions , record the optimized point set as the optimized neighborhood N″; Step 4. Obtain the Delaunay adjacent edge of the given point p: use the two-dimensional Delaunay method to carry out triangular mesh modeling on the given point p and its optimized neighborhood N″, and reversely map the obtained two-dimensional Delaunay triangular mesh to the three-dimensional adjacent domain space, and extract and store the 3D Delaunay adjacent edges connected to a given point p; Step 5, complete triangular mesh modeling: repeat steps 2-4, repeat the above algorithm for each point, and then complete the triangular mesh modeling of the entire 3D point cloud. 2. according to the described triangular mesh modeling method of a kind of indoor scene heterogeneous three-dimensional point cloud data of claim 1, it is characterized in that: described step 2 neighborhood tangent plane projection, specifically comprises the following substeps: Step (a), calculate the average distance d of the three-dimensional point cloud, set 5d as the neighborhood radius, and extract the neighborhood N of the given point p; Step (b), through the formula $\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$ Find the covariance matrix M of the neighborhood N, where: p _i is the neighbor point, i is the serial number of the neighbor point, k is the number of neighbor points, and T is the vector transposition symbol, which transposes the column vector into a row vector; Step (c), obtain the eigenvalues λ ₁ , λ ₂ , λ ₃ of M, where λ ₁ <λ ₂ <λ ₃ , and the corresponding eigenvectors v ₁ , v ₂ , v ₃ ; Step (d), unitize the eigenvector v1 corresponding to the minimum eigenvalue λ1, namely obtain the normal vector _{n of} the given point p; Step (e), project each point in the neighborhood N onto the tangent plane T corresponding to the normal vector n, and record the projected point set as the projected neighborhood N'. 3. The triangular mesh modeling method of a kind of indoor scene heterogeneous three-dimensional point cloud data according to claim 1, is characterized in that: described step 3 optimizes projection neighborhood N ', specifically comprises the following substeps: Step (a), in the tangent plane T, take the given point p as the origin, construct a straight line passing through the origin p, and make it rotate clockwise from the horizontal position for one circle, divide a fan-shaped area at an interval of 22.5 degrees, and finally form 16 fan-shaped areas centered on the origin p; Step (b), placing the projection neighborhood N' in the tangent plane T of point p into the fan-shaped area divided in step (a); Step (c), extracting the points closest to the origin p in each fan-shaped area, these points constitute an optimized point set, which is recorded as the optimized neighborhood N″. 4. according to the triangular mesh modeling method of a kind of indoor scene heterogeneous three-dimensional point cloud data according to claim 1, it is characterized in that: described step 4 obtains the Delaunay adjacent edge of given point p point, specifically comprises the following substeps: Step (a), using the divide and conquer method, divide the optimized neighborhood N" into several small areas according to the x coordinates, and the number of points in each small area is not more than 3. For each small area, it is required can guarantee to meet the Delaunay criterion; Step (b), integrating adjacent small areas into a larger area meeting the Delaunay criterion; Step (c), repeat step (b), and then merge each larger area layer by layer until all larger areas are merged into a whole, so far, a single local Delaunay triangulation is completed; according to the uniqueness of Delaunay triangulation Criterion, the only Delaunay triangulation result can be obtained; Step (d), extracting the edge connected to the given point p from the constructed two-dimensional Delaunay triangulation result, mapping it back to the three-dimensional neighborhood space, and storing it, the stored edge is the edge of the given point p Three-dimensional Delaunay sides.