CN112669462A - Model processing method and system suitable for scanning point cloud - Google Patents
Model processing method and system suitable for scanning point cloud Download PDFInfo
- Publication number
- CN112669462A CN112669462A CN202110064218.3A CN202110064218A CN112669462A CN 112669462 A CN112669462 A CN 112669462A CN 202110064218 A CN202110064218 A CN 202110064218A CN 112669462 A CN112669462 A CN 112669462A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- data
- point
- cloud data
- sub
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 17
- 230000004927 fusion Effects 0.000 claims abstract description 15
- 238000012545 processing Methods 0.000 claims abstract description 14
- 238000000034 method Methods 0.000 claims description 37
- 239000013598 vector Substances 0.000 claims description 26
- 239000002356 single layer Substances 0.000 claims description 23
- 230000008569 process Effects 0.000 claims description 14
- 238000001914 filtration Methods 0.000 claims description 12
- 239000010410 layer Substances 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 11
- 238000007781 pre-processing Methods 0.000 claims description 8
- 230000002146 bilateral effect Effects 0.000 claims description 7
- 238000004321 preservation Methods 0.000 claims description 5
- 230000011218 segmentation Effects 0.000 claims description 4
- 238000005192 partition Methods 0.000 claims description 3
- 238000000513 principal component analysis Methods 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 230000002093 peripheral effect Effects 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 abstract description 14
- 238000003860 storage Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 7
- 230000007547 defect Effects 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Images
Landscapes
- Image Generation (AREA)
Abstract
The invention discloses a model processing method and a system suitable for scanning point clouds, which are used for rapidly fusing a plurality of dense point clouds into a single dense point cloud through rapid fusion of the plurality of scanning point clouds, thereby eliminating the overlapping of models and improving the processing efficiency of the models; by quickly simplifying a single dense point cloud, the data volume of the model is greatly reduced on the premise of protecting the details of the model, and the subsequent processing time and the storage pressure of the model are reduced; the point cloud is converted into a grid model through a point cloud rapid triangulation technology, and the details and the characteristics of the model are not damaged; through the mesh denoising with detail protection, the surface smoothness of the model is improved, and the loss of details is reduced. The invention can quickly convert a plurality of scanning point clouds into a grid model with low data volume, high precision and smooth surface on the premise of keeping the precision and detail characteristics. The data volume of the final grid model is effectively reduced, and the reconstruction efficiency of the grid model of the scanning data is greatly improved.
Description
Technical Field
The invention relates to the technical field of model processing, in particular to a model processing method and system suitable for scanning point cloud.
Background
In recent years, the three-dimensional optical measurement technology is rapidly developed in the industry of industrial detection and reverse solution by virtue of the advantages of non-contact, high efficiency, full-field measurement and the like. With the improvement of the efficiency and the precision of the detection and reverse process in industrial production, the scanning point cloud on the surface of the part is quickly converted into a high-precision three-dimensional model to become a technical core point.
In the scanning process, the measuring equipment acquires surface information of parts from multiple angles and reconstructs point clouds, so that the data volume of an original point cloud model obtained by scanning is huge, and a large amount of repeated and redundant information and noise exist. Therefore, it is important to rapidly process a plurality of scanning point clouds into a complete grid model with low data volume, low noise and high precision. The model processing comprises key steps of point cloud fusion, point cloud simplification, meshing, mesh model denoising and the like.
The existing model processing technology can obtain a better processing result for point clouds with small data volume, low redundancy and low noise, but is difficult to be suitable for processing industrial scanning point clouds. On the one hand, industrial scan point clouds are typically numerous, data-intensive due to the need for integrity and detail expression, and contain a large amount of redundant data due to the large degree of overlap. The existing processes of fusion, simplification, gridding and the like neglect the problems of efficiency and data copying and storage because of pursuing the regularity, uniformity and the like of grids. For low data volume point clouds, the problems caused by time consumption of an algorithm and overlarge data are not necessarily obvious, but for a plurality of scanning point clouds with large data volume, the problems of space and time are particularly obvious. On the other hand, the actual scanning process is influenced by the environment, the noise of the scanning point cloud is more, and the existing grid denoising method such as laplacian fairing and various improved algorithms thereof, bilateral filtering and various anisotropic filtering algorithms is difficult to separate model features from the high-noise point cloud, so that the model details are blurred while the noise is removed.
For scanning point clouds with large data volume, redundancy and noise, the existing model processing method has obvious defects in the aspects of efficiency, data storage space and feature preserving denoising, so that a model processing method and a system for the scanning point clouds are needed.
Disclosure of Invention
The invention aims to make up for the defects of the prior art and provides a model processing method and system suitable for scanning point cloud. The core of the technical scheme of the invention is that a uniform space structure is constructed for a plurality of pieces of scanning data, and then key technologies such as point cloud fusion, simplification, gridding and the like are all carried out in the space structure, so that the whole process from a plurality of point cloud models to a single grid model of the scanning data is integrated, and the problems of low efficiency caused by step-by-step processing of each step, overlarge space complexity caused by data copying and the like are avoided.
The invention is realized by the following technical scheme:
a model processing method suitable for scanning point cloud specifically comprises the following steps:
s1, preprocessing the initial point cloud data, establishing a topological relation for the preprocessed point cloud data by using a kdTere data structure, and movably fusing multi-layer point cloud data into single-layer point cloud data;
s2, inputting the fused single-layer point cloud data, and performing multiple sub-cluster division on the point cloud data by adopting a segmented hierarchical clustering method to obtain point cloud data based on curvature simplification;
s3, traversing all vertexes of the simplified point cloud data, searching respective adjacent points through kdTere, and rebuilding a point cloud model into a triangular mesh model through Delaunay triangulation of the vertexes and the adjacent points;
and S4, adopting a feature preserving fairing method to realize feature preserving simplification of the triangular mesh model.
The point cloud data preprocessing method in step S1 includes:
inputting n pieces of ordered point cloud data, and adding the attribute of whether the point cloud data is a boundary point or not;
traversing each point cloud data, setting the data point as a boundary point when holes exist in eight neighborhoods of the data point, and otherwise, setting the data point as a non-boundary point;
counting the average point distance d of each point cloud dataiThe average dot spacing of all data points is
And adding an ID attribute to each data point, wherein the ID number of the ith point cloud data is i, and then storing all the point cloud data into the same container.
Step S1
Searching all neighborhood data points in a fixed radius range of each central data point by using the kdTere;
when the ID attributes of all neighborhood data points are the same as the attributes of the central data point, the neighborhood data points are single-layer point cloud data;
if the field data point ID attribute is different from the center data point attribute, multilayer point cloud data exists in the area, and the multilayer point cloud data is fused into single-layer point cloud data.
Step S1 describes moving and fusing the multi-layer point cloud data into single-layer point cloud data, which includes the following steps:
extracting all data points with different ID attributes from the central data point in the multi-layer point cloud data area;
finding out the nearest point to the central point in the neighborhood data points of each ID attribute, and calculating the centroid coordinates of the nearest point and the central data point;
moving the central data point to a centroid coordinate position, and marking whether the preserved attribute of all data points in the radius area is different from the ID attribute of the central point as non-preserved;
and traversing all the data points in sequence, and finally extracting all the data points with the preserved attribute as preserved data, namely the single-layer point cloud data after fusion.
The specific process of step S2 is as follows:
before each time of sub-cluster division and segmentation of point cloud data, the point cloud data is firstly dividedPerforming principal component analysis on the whole to obtain a main direction of the point cloud data, and dividing along a plane which passes through a central point and has the normal direction as the main direction; one point cloud data is divided into two sub-clusters, and the farthest distance d from the data point in the two sub-clusters to the center of the sub-cluster is calculated respectivelytWhen d istGreater than a predetermined threshold dmaxThe same partitioning operation is recursively performed on both sub-clusters until the maximum distance dtLess than a predetermined threshold dmaxCurvature analysis of the sub-clusters is started.
The curvature analysis is carried out on the sub-clusters, and the specific method is as follows: carrying out covariance analysis on all data points in the well-divided sub-clusters to obtain characteristic values of lambda respectively1、λ2、λ3(λ1＞λ2＞λ3) The corresponding curvature of the sub-cluster is
When the sub-cluster ctGreater than a set curvature threshold cmaxContinuing to divide the sub-clusters until the curvature of the sub-clusters is smaller than a set threshold value, and stopping dividing; calculating the distance from all data points in the sub-cluster to the center of the sub-cluster, and finding out the data point closest to the center of the sub-cluster; traversing all data points in the sub-cluster, removing boundary points and data points closest to the center of the sub-cluster, and setting whether the attribute of other data points is reserved as non-reserved; and traversing all the data points, and reserving whether the reserved attribute is reserved for the reserved data point, namely the point cloud data based on the curvature simplification.
The specific process of step S3 is as follows:
traversing all vertexes of the simplified point cloud data, searching respective adjacent points, performing covariance analysis on the adjacent points to fit a plane of a local area of the vertexes, projecting each point onto the plane to obtain a two-dimensional plane point set, performing plane Delaunay triangulation on the two-dimensional point set, and obtaining a 1-ring connection mode of the vertexes; and combining the connection relations of all the vertexes and the peripheral adjacent points to obtain the triangular mesh model.
The specific process of step S4 is as follows:
carrying out bilateral filtering adjustment based on area and centroid distance on a triangular patch normal vector in the triangular mesh model to obtain the normal vector of each patch; and then moving the vertex position of the mesh to enable the normal vector of the neighborhood triangular patch to be close to the normal vector of the patch, and finally realizing the feature preservation simplification of the triangular mesh model.
A model processing system adapted for scanning point clouds: comprises the following steps:
a single-layer point cloud data fusion module: preprocessing initial point cloud data, establishing a topological relation for the preprocessed point cloud data by using a kdTere data structure, and movably fusing multi-layer point cloud data into single-layer point cloud data;
single-layer point cloud data curvature simplifying module: performing multiple sub-cluster division on the fused single-layer point cloud data by adopting a partition type hierarchical clustering method to obtain point cloud data which is simplified based on curvature;
a point cloud model reconstruction module: traversing all vertexes of the simplified point cloud data, searching respective adjacent points, and reconstructing a point cloud model into a triangular mesh model through Delaunay triangulation of the vertexes and the kdTere neighborhoods thereof;
the feature preservation simplification module: and a feature preserving fairing method is adopted to realize feature preserving simplification of the triangular mesh model.
And preprocessing the initial point cloud data, namely adding the attribute of whether the input n pieces of ordered point cloud data are boundary points.
The invention has the advantages that: the invention is a model processing method suitable for scanning point clouds with large data volume, large redundancy and high noise, and a plurality of dense point clouds are quickly fused into a single dense point cloud through the quick fusion of the plurality of scanning point clouds, so that the overlapping of models is eliminated, and the processing efficiency of the models is improved; by quickly simplifying a single dense point cloud, the data volume of the model is greatly reduced on the premise of protecting the details of the model, and the subsequent processing time and the storage pressure of the model are reduced; the point cloud is converted into a grid model through a point cloud rapid triangulation technology, and the details and the characteristics of the model are not damaged; through the mesh denoising with detail protection, the surface smoothness of the model is improved, and the loss of details is reduced.
The invention can quickly convert a plurality of scanning point clouds into a grid model with low data volume, high precision and smooth surface on the premise of keeping the precision and detail characteristics. The data volume of the final grid model is effectively reduced, and the reconstruction efficiency of the grid model of the scanning data is greatly improved.
Drawings
FIG. 1 is a flow chart illustrating the overall steps of the present invention.
Fig. 2 is a schematic diagram of boundary extraction.
FIG. 3 is a schematic diagram of multi-layer point cloud fusion.
FIG. 4 is a diagram illustrating hierarchical clustering sub-cluster partitioning.
Fig. 5a and 5b are simplified diagrams based on curvature data.
Fig. 6 is a partial plan projection view.
FIGS. 7a, 7b, and 7c are schematic views of Delou triangulation in a local plane.
Fig. 8a, 8b, and 8c are schematic diagrams illustrating hole filling and gridding results.
Fig. 9a and 9b are schematic diagrams of bilateral filtering of grid patch normal vectors.
Fig. 10a, 10b, and 10c are schematic diagrams of mesh model after mesh vertex adjustment and denoising.
Detailed Description
The core technology provided by the invention comprises the following four points:
(1) multi-scanning point cloud fast fusion technology
Aiming at the problem that a large number of overlapping areas exist in point cloud data, the multi-view dense point cloud rapid fusion technology is provided. By establishing a point cloud space topology based on a k neighborhood, an overlapping area is automatically, quickly and robustly identified, and mobile fusion is carried out, so that the uniqueness of point cloud appearance expression is ensured.
(2) Detail protection fast point cloud simplifying technology
Aiming at point clouds containing redundant data or dense point clouds with huge number of points, the rapid point cloud simplification technology for feature preservation is provided. And the point number of the flat area is rapidly reduced by utilizing the point cloud space distribution characteristic. The method greatly reduces the number of point clouds, and ensures enough high curvature characteristics so as to improve the efficiency and the precision of subsequent registration.
(3) Quick triangulation technology for point cloud model
Aiming at the defects that the traditional gridding algorithm is low in efficiency or can change point coordinates, the rapid gridding technology with lossless point cloud precision is provided. And rapidly reconstructing the point cloud model into a mesh model through Delaunay triangulation of the vertexes and k neighborhoods thereof.
(4) Grid model feature-preserving fairing technology
Aiming at the problem of large surface noise after scanning point cloud gridding, the feature preserving fairing technology of a scanning grid model is provided. On the basis of keeping model edges, angles and other detailed characteristics, model surface noise is greatly reduced through two-step filtering of coarse filtering and fine filtering; and the sharpening effect is achieved by changing the topological connection relation of the vertex at the sharp feature.
The specific embodiment is as follows:
FIG. 1 is a schematic flow chart of the present invention for scanning data from a plurality of point clouds to a complete grid. The specific process is as follows: traversing each point cloud data, setting the data point as a boundary point when holes exist in eight neighborhoods of the data point, otherwise setting the data point as a non-boundary point
Step S01, inputting n pieces of ordered point cloud data, adding boundary point attributes to the point cloud data, traversing each piece of point cloud data, setting the point boundary point attributes as boundary points when holes exist in eight neighborhoods of the data points, otherwise setting the points as non-boundary points, as shown in fig. 2, all the solid points are marked as boundary points because holes exist in the eight neighborhoods. Counting the average point distance d of each point cloud dataiThe average dot spacing of all data points is
And adding an ID attribute to each data point, wherein the ID number of the ith point cloud data is i. And then storing all the point cloud data into the same container, and starting the data fusion operation of a plurality of point clouds.
And step S02, applying a kdTere data structure to establish a topological relation for the point cloud data in the container, firstly adding an attribute of whether to remain for each data point, initializing all the data points to be remained, then traversing each point cloud data in sequence to perform point cloud fusion, searching all neighborhood data points in a fixed radius range of each central data point by using the kdTere, and when the ID attributes of all the neighborhood data points are the same as the attributes of the central data points, indicating that the point is single-layer point cloud data without processing operation. If the attribute of the neighborhood data point ID is different from the attribute of the central data point, the existence of multi-layer point cloud data is indicated, and the multi-layer point cloud data is fused into single-layer point cloud.
Step S03: in the multi-layer point cloud data area, extracting all data points with different ID attributes from the data points of the middle data point, finding out the point which is closest to the central data point in the data points with each ID attribute, calculating the centroid coordinates of the closest point and the central data point, moving the central data point to the centroid coordinate position, and marking whether the reserved attributes of all the data points with different ID attributes from the central data point in the radius area are not reserved. And traversing all the data points in sequence, and finally extracting all the data points with the preserved attribute as preserved, namely the fused point cloud data, wherein the effect is shown in fig. 3.
Step S04: and performing sub-cluster division on the point cloud data by adopting a partition type hierarchical clustering method on the fused single-layer point cloud data. Before each division, the point cloud is subjected to principal component analysis on the whole point cloud to obtain the principal direction of the point cloud data, and the point cloud is divided along a plane passing through the central point and taking the normal direction as the principal direction. One point cloud data is divided into two sub-clusters, and the farthest distance d from the data point in the two sub-clusters to the center of the sub-cluster is calculated respectivelytWhen d istGreater than a predetermined threshold dmaxThe same segmentation operation is recursively performed on both sub-clusters, and the segmentation plane graph lines drawn in the graph taper down as the graph is recursively segmented, as shown in fig. 4. Up to dtLess than a predetermined threshold dmaxCurvature analysis of the sub-clusters is started.
Step S05: performing covariance analysis on all data points in the divided sub-cluster, wherein the sub-cluster has M data points and the centroid p of the M data points, and the covariance matrix M can be expressed as
Performing eigenvalue decomposition on the covariance matrix, wherein the obtained eigenvalues are respectively lambda1、λ2、λ3(λ1＞λ2＞λ3). The corresponding curvature of the sub-cluster is
When the sub-cluster ctGreater than a set curvature threshold cmaxAnd continuing to divide the sub-clusters until the curvature of the sub-clusters is smaller than a set threshold value, and stopping dividing. And calculating the distance from all data points in the sub-cluster to the center of the sub-cluster, and finding out the data point closest to the center of the sub-cluster. And traversing all data points in the sub-cluster, removing the boundary point and the data point closest to the center of the sub-cluster, and setting the attribute of whether other data points are reserved as non-reserved. And traversing all the data points, and reserving whether the reserved attribute is reserved for the reserved data point, namely the data result after the curvature-based simplification. The point cloud data shown in fig. 5(a) is subjected to data reduction, and the reduction result is shown in fig. 5 (b).
Step S06: traversing all vertexes in the point cloud, searching M three-dimensional points in the neighborhood sub-cluster of any vertex p, and then obtaining the characteristic value lambda of the sub-cluster covariance matrix M through covariance analysis (consistent with the step S05)1、λ2、λ3(λ1＞λ2＞λ3). Then, a local plane S is constructed, which is defined by lambda3The corresponding feature vector is a normal vector and passes through the centroid of the sub-cluster data points. The points within the sub-cluster are projected onto the plane S along the normal vector direction of the plane S, thereby converting the m three-dimensional points into m two-dimensional plane points, as shown in fig. 6.
Step S07: constructing a two-dimensional coordinate system on the plane S, wherein the origin of the coordinate system is a projection point o of a vertex p, and the X axis points to the nearest adjacent point p1Projected point v of1. All the neighboring pointsThe projection points are connected with the projection point of the vertex p, and the sorting from small to large is carried out according to the included angle theta (theta is more than or equal to 0 degree and less than or equal to 360 degrees) between the respective connecting line and the X axis. Then, the plane triangle delta ov is obtained firstly1v2Putting the triangle into a triangular patch set F, and then acquiring a triangle delta ov according to a counterclockwise sequence2v3And judging the characteristics of the two triangles, namely judging the point v3Whether or not it falls in Δ ov1v2In the circumscribed circle. If point v3Not in the circumscribed circle, Δ ov2v3Put into the triangular patch set F, as shown in fig. 7 (a); if point v3In the circumscribed circle, the point v is3From this time the planar point set is removed, which in turn takes the triangle Δ ov2v4And the empty circle judgment is also performed until a second triangle is added to the triangular patch set F, as shown in fig. 7 (b). And then, continuously judging whether the next triangle is in the circumscribed circle of the previous triangle in the anticlockwise sequence, repeating the step until all the triangles are judged, and obtaining a triangle set F taking the o point as the center at the moment. None of the triangle vertices in the set are within the circumscribed circle of another triangle, i.e., a delaunay triangulation result is formed, as shown in fig. 7 (c).
Step S08: mapping the connection relation of the two-dimensional vertexes into the three-dimensional point cloud according to the projection relation, namely a two-dimensional triangle delta ov in a plane2v3Corresponding to a spatial triangle Δ pp in space2p3Thereby obtaining the connection relationship between the vertex p and the vertex of the neighborhood ring 1. And after the 1-ring connection relation of all the vertexes in the point cloud is determined, the connection relation among the vertexes of the whole point cloud model is obtained. There may be a special case in this process as shown in fig. 8 (a). When local triangulation is performed with a as the vertex, Δ abv is generated2And Δ av7b, and when local triangulation is performed with b as the vertex, Δ v is generated7bv2I.e. at points a, v during gridding7，b，v2Two different connection modes are generated inside the formed quadrangle. For this case, Δ abv will be2、Δav7b and Δ v7bv2All are deleted and the holes av are left7bv2. The vertices are then directly connected to close the hole, thus merging the connection of points a and b, as shown in fig. 8 (b). This operation is performed for all points in the entire model to obtain a complete mesh model. Fig. 8(c) shows the result of gridding fig. 5 (b).
Step S09: and calculating the mass centers and normal vectors of all triangular patches in the mesh model. Traversing each triangular patch f in the mesh model, and searching for the patch f sharing one edge with the triangular patch fi. Statistics of f using bilateral filteringiWeight w to f normal vectori。
wi＝AiBi
Wherein,
liis a triangle fiThe distance from the centroid to the centroid of the triangle f; n isiIs a triangle fiThe normal vector of (a); n is the normal vector of the triangle f; sigma1The standard deviation of the centroid distance distribution; sigma2And standard deviation of normal vector included angle distribution. Through multiple iterations (generally set to 5 times), the direction change of the normal vector of each patch in the grid is smooth, and due to the protection characteristics of bilateral filtering opposite side and angle features, the normal vectors of the feature parts still keep the original direction well. Fig. 9(a) and (b) show normal vector directions of patches before and after filtering, respectively.
Step S10: and adjusting the coordinates of the grid vertexes according to the ideal normal vector. Traversing all vertexes of the mesh, and searching the neighborhood triangular patch f of each vertex piObtaining ideal normal vector n of the triangular patchesiAnd the edge starting from the vertex p among the three edges of the triangular patch, as shown in fig. 10 (a). According to ideal normal vector n of each triangular patchiComputing an ideal normal vector n for a vertex pp
Wherein, TiIs a triangle fiThe area of (d); k is the number of neighborhood triangular patches. Then calculating the movement vector of the vertex
Where m is the number of edges starting from the vertex p. Therefore, the new coordinate p 'of the vertex p after adjustment can be expressed as p' ═ p + Vp. After multiple iterations (generally set to 5 times), the whole grid model tends to be smooth, and the characteristics of edges, corners and the like are retained because normal vectors are protected in bilateral filtering. Fig. 10(b) and (c) show the grids before and after the noise, respectively.
The above description is merely a detailed description of specific embodiments of the present invention, but the present invention is not limited thereto. Various alterations, modifications and improvements will occur to those skilled in the art without departing from the spirit and scope of the invention.
Claims (10)
1. A model processing method suitable for scanning point cloud is characterized by comprising the following steps:
s1, preprocessing the initial point cloud data, establishing a topological relation for the preprocessed point cloud data by using a kdTere data structure, and movably fusing multi-layer point cloud data into single-layer point cloud data;
s2, inputting the fused single-layer point cloud data, and performing multiple sub-cluster division on the point cloud data by adopting a segmented hierarchical clustering method to obtain point cloud data based on curvature simplification;
s3, traversing all vertexes of the simplified point cloud data, searching respective adjacent points through kdTere, and rebuilding a point cloud model into a triangular mesh model through Delaunay triangulation of the vertexes and the adjacent points;
and S4, adopting a feature preserving fairing method to realize feature preserving simplification of the triangular mesh model.
2. The method of claim 1, wherein the model processing method is applied to the scanning point cloud and comprises the following steps: the point cloud data preprocessing method in step S1 includes:
inputting n pieces of ordered point cloud data, and adding the attribute of whether the point cloud data is a boundary point or not;
traversing each point cloud data, setting the data point as a boundary point when holes exist in eight neighborhoods of the data point, and otherwise, setting the data point as a non-boundary point;
counting the average point distance d of each point cloud dataiThe average dot spacing of all data points is
And adding an ID attribute to each data point, wherein the ID number of the ith point cloud data is i, and then storing all the point cloud data into the same container.
3. The method of claim 2, wherein the model processing method is applied to the scanning point cloud and comprises the following steps: step S1, the method for establishing a topological relationship for the preprocessed point cloud data using the kdTree data structure includes:
searching all neighborhood data points in a fixed radius range of each central data point by using the kdTere;
when the ID attributes of all neighborhood data points are the same as the attributes of the central data point, the neighborhood data points are single-layer point cloud data;
if the field data point ID attribute is different from the center data point attribute, multilayer point cloud data exists in the area, and the multilayer point cloud data is fused into single-layer point cloud data.
4. The method of claim 3, wherein the model processing method is applied to the scanning point cloud and comprises the following steps: step S1 describes moving and fusing the multi-layer point cloud data into single-layer point cloud data, which includes the following steps:
extracting all data points with different ID attributes from the central data point in the multi-layer point cloud data area;
finding out the nearest point to the central point in the neighborhood data points of each ID attribute, and calculating the centroid coordinates of the nearest point and the central data point;
moving the central data point to a centroid coordinate position, and marking whether the preserved attribute of all data points in the radius area is different from the ID attribute of the central point as non-preserved;
and traversing all the data points in sequence, and finally extracting all the data points with the preserved attribute as preserved data, namely the single-layer point cloud data after fusion.
5. The method of claim 4, wherein the model processing method is applied to the scanning point cloud and comprises the following steps: the specific process of step S2 is as follows:
before each time of sub-cluster division and segmentation of point cloud data, performing principal component analysis on the whole point cloud data to obtain a point cloud data principal direction, and segmenting along a plane passing through a central point and taking the normal direction as the principal direction;
one point cloud data is divided into two sub-clusters, and the farthest distance d from the data point in the two sub-clusters to the center of the sub-cluster is calculated respectivelytWhen d istGreater than a predetermined threshold dmaxThe same partitioning operation is recursively performed on both sub-clusters until the maximum distance dtLess than a predetermined threshold dmaxCurvature analysis of the sub-clusters is started.
6. The method of claim 5, wherein the model processing method is applied to the scanning point cloud and comprises the following steps: the curvature analysis is carried out on the sub-clusters, and the specific method is as follows:
carrying out covariance analysis on all data points in the well-divided sub-clusters to obtain characteristic values of lambda respectively1、λ2、λ3，λ1＞λ2＞λ3The corresponding curvature of the sub-cluster is
When the sub-cluster ctGreater than a set curvature threshold cmaxContinuing to divide the sub-clusters until the curvature of the sub-clusters is smaller than a set threshold value, and stopping dividing;
calculating the distance from all data points in the sub-cluster to the center of the sub-cluster, and finding out the data point closest to the center of the sub-cluster;
traversing all data points in the sub-cluster, removing boundary points and data points closest to the center of the sub-cluster, and setting whether the attribute of other data points is reserved as non-reserved;
and traversing all the data points, and reserving whether the reserved attribute is reserved for the reserved data point, namely the point cloud data based on the curvature simplification.
7. The method of claim 6, wherein the model processing method is applied to the scanning point cloud: the specific process of step S3 is as follows:
traversing all vertexes of the simplified point cloud data, searching respective adjacent points, performing covariance analysis on the adjacent points to fit a plane of a local area of the vertexes, projecting each point onto the plane to obtain a two-dimensional plane point set, performing plane Delaunay triangulation on the two-dimensional point set, and obtaining a 1-ring connection mode of the vertexes;
and combining the connection relations of all the vertexes and the peripheral adjacent points to obtain the triangular mesh model.
8. The method of claim 7, wherein the model processing method is applied to the scanning point cloud and comprises the following steps: the specific process of step S4 is as follows:
carrying out bilateral filtering adjustment based on area and centroid distance on a triangular patch normal vector in the triangular mesh model to obtain the normal vector of each patch;
and moving the vertex position of the mesh to enable the normal vector of the neighborhood triangular patch to be close to the normal vector of the patch, and finally realizing the feature preservation simplification of the triangular mesh model.
9. A model processing system adapted for scanning point clouds, comprising: comprises the following steps:
a single-layer point cloud data fusion module: preprocessing initial point cloud data, establishing a topological relation for the preprocessed point cloud data by using a kdTere data structure, and movably fusing multi-layer point cloud data into single-layer point cloud data;
single-layer point cloud data curvature simplifying module: performing multiple sub-cluster division on the fused single-layer point cloud data by adopting a partition type hierarchical clustering method to obtain point cloud data which is simplified based on curvature;
a point cloud model reconstruction module: traversing all vertexes of the simplified point cloud data, searching respective adjacent points, and reconstructing a point cloud model into a triangular mesh model through Delaunay triangulation of the vertexes and the kdTere neighborhoods thereof;
the feature preservation simplification module: and a feature preserving fairing method is adopted to realize feature preserving simplification of the triangular mesh model.
10. The model processing system for scanning point cloud of claim 9, wherein: and preprocessing the initial point cloud data, namely adding the attribute of whether the input n pieces of ordered point cloud data are boundary points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110064218.3A CN112669462B (en) | 2021-01-18 | 2021-01-18 | Model processing method and system suitable for scanning point cloud |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110064218.3A CN112669462B (en) | 2021-01-18 | 2021-01-18 | Model processing method and system suitable for scanning point cloud |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112669462A true CN112669462A (en) | 2021-04-16 |
| CN112669462B CN112669462B (en) | 2023-02-28 |
Family
ID=75415525
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110064218.3A Active CN112669462B (en) | 2021-01-18 | 2021-01-18 | Model processing method and system suitable for scanning point cloud |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112669462B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113344992A (en) * | 2021-05-31 | 2021-09-03 | 山东大学 | Global point cloud registration method, system, storage medium and equipment |
| CN115578429A (en) * | 2022-11-21 | 2023-01-06 | 西北工业大学 | Point cloud data-based mold online precision detection method |
| CN116030022A (en) * | 2023-02-01 | 2023-04-28 | 天津风霖物联网科技有限公司 | Quality detection system and method for building prefabricated part |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2674913A1 (en) * | 2012-06-14 | 2013-12-18 | Softkinetic Software | Three-dimensional object modelling fitting & tracking. |
| CN107749079A (en) * | 2017-09-25 | 2018-03-02 | 北京航空航天大学 | A kind of quality evaluation of point cloud and unmanned plane method for planning track towards unmanned plane scan rebuilding |
| CN110363861A (en) * | 2019-07-14 | 2019-10-22 | 南京林业大学 | Field crop three-dimensional reconstruction method based on laser radar point cloud |
| CN110889861A (en) * | 2019-11-15 | 2020-03-17 | 广州供电局有限公司 | Point cloud extraction method for electric power tower |
| US20200111251A1 (en) * | 2018-10-08 | 2020-04-09 | The Hong Kong Polytechnic University | 3d indoor modeling method, system and device based on point cloud data |
| CN111260683A (en) * | 2020-01-09 | 2020-06-09 | 合肥工业大学 | Target detection and tracking method and device for three-dimensional point cloud data |
| CN111986115A (en) * | 2020-08-22 | 2020-11-24 | 王程 | Accurate elimination method for laser point cloud noise and redundant data |
| US20200388044A1 (en) * | 2019-06-06 | 2020-12-10 | Zhejiang University | Plant point cloud acquisition, registration and optimization method based on tof camera |
-
2021
- 2021-01-18 CN CN202110064218.3A patent/CN112669462B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2674913A1 (en) * | 2012-06-14 | 2013-12-18 | Softkinetic Software | Three-dimensional object modelling fitting & tracking. |
| CN107749079A (en) * | 2017-09-25 | 2018-03-02 | 北京航空航天大学 | A kind of quality evaluation of point cloud and unmanned plane method for planning track towards unmanned plane scan rebuilding |
| US20200111251A1 (en) * | 2018-10-08 | 2020-04-09 | The Hong Kong Polytechnic University | 3d indoor modeling method, system and device based on point cloud data |
| US20200388044A1 (en) * | 2019-06-06 | 2020-12-10 | Zhejiang University | Plant point cloud acquisition, registration and optimization method based on tof camera |
| CN110363861A (en) * | 2019-07-14 | 2019-10-22 | 南京林业大学 | Field crop three-dimensional reconstruction method based on laser radar point cloud |
| CN110889861A (en) * | 2019-11-15 | 2020-03-17 | 广州供电局有限公司 | Point cloud extraction method for electric power tower |
| CN111260683A (en) * | 2020-01-09 | 2020-06-09 | 合肥工业大学 | Target detection and tracking method and device for three-dimensional point cloud data |
| CN111986115A (en) * | 2020-08-22 | 2020-11-24 | 王程 | Accurate elimination method for laser point cloud noise and redundant data |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113344992A (en) * | 2021-05-31 | 2021-09-03 | 山东大学 | Global point cloud registration method, system, storage medium and equipment |
| CN113344992B (en) * | 2021-05-31 | 2022-06-28 | 山东大学 | Global point cloud registration method, system, storage medium and equipment |
| CN115578429A (en) * | 2022-11-21 | 2023-01-06 | 西北工业大学 | Point cloud data-based mold online precision detection method |
| CN116030022A (en) * | 2023-02-01 | 2023-04-28 | 天津风霖物联网科技有限公司 | Quality detection system and method for building prefabricated part |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112669462B (en) | 2023-02-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112669462B (en) | Model processing method and system suitable for scanning point cloud | |
| CN107123164B (en) | Three-dimensional reconstruction method and system for keeping sharp features | |
| Rutishauser et al. | Merging range images of arbitrarily shaped objects | |
| Woo et al. | A new segmentation method for point cloud data | |
| CN107680168B (en) | Grid simplifying method based on plane fitting in three-dimensional reconstruction | |
| CN111696210A (en) | Point cloud reconstruction method and system based on three-dimensional point cloud data characteristic lightweight | |
| Lozes et al. | Partial difference operators on weighted graphs for image processing on surfaces and point clouds | |
| Mencl et al. | Graph-based surface reconstruction using structures in scattered point sets | |
| CN110807781A (en) | Point cloud simplification method capable of retaining details and boundary features | |
| KR100512760B1 (en) | Method for generating 3d mesh from 3d points by using shrink-wrapping scheme of boundary cells | |
| Wang et al. | Restricted delaunay triangulation for explicit surface reconstruction | |
| CN112270746A (en) | Aluminum alloy 3D printing point cloud simplification algorithm based on neighborhood covariance characteristic parameter threshold | |
| Grimm et al. | Shape classification and normal estimation for non-uniformly sampled, noisy point data | |
| El Sayed et al. | An efficient simplification method for point cloud based on salient regions detection | |
| CN113111741A (en) | Assembly state identification method based on three-dimensional feature points | |
| CN115147433A (en) | Point cloud registration method | |
| Li | A voxel graph-based resampling approach for the aerial laser scanning of urban buildings | |
| CN113469907B (en) | Data simplification method and system based on blade profile characteristics | |
| Salvador et al. | From silhouettes to 3d points to mesh: towards free viewpoint video | |
| Wang et al. | A feature preserved mesh simplification algorithm | |
| CN113763563A (en) | Three-dimensional point cloud geometric grid structure generation method based on plane recognition | |
| CN114820505A (en) | Non-contact measurement method of dynamic target | |
| Vierjahn et al. | Surface-reconstructing growing neural gas: A method for online construction of textured triangle meshes | |
| CN116977592B (en) | Three-dimensional structured reconstruction method, device and computer medium based on winding number | |
| Annuth et al. | Smart growing cells: Supervising unsupervised learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |