CN113570712B - 3D modeling optimization method based on GCN - Google Patents
3D modeling optimization method based on GCN Download PDFInfo
- Publication number
- CN113570712B CN113570712B CN202111117830.9A CN202111117830A CN113570712B CN 113570712 B CN113570712 B CN 113570712B CN 202111117830 A CN202111117830 A CN 202111117830A CN 113570712 B CN113570712 B CN 113570712B
- Authority
- CN
- China
- Prior art keywords
- node
- modeling
- polygon
- storage file
- entering
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000005457 optimization Methods 0.000 title claims abstract description 22
- 239000011159 matrix material Substances 0.000 claims description 20
- 238000013145 classification model Methods 0.000 claims description 17
- 238000005070 sampling Methods 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 5
- 238000013461 design Methods 0.000 abstract description 17
- 238000010276 construction Methods 0.000 abstract description 7
- 230000000694 effects Effects 0.000 abstract description 4
- 230000000903 blocking effect Effects 0.000 abstract description 3
- 238000004891 communication Methods 0.000 abstract description 3
- 230000006835 compression Effects 0.000 abstract description 3
- 238000007906 compression Methods 0.000 abstract description 3
- 238000013135 deep learning Methods 0.000 abstract description 3
- 238000002474 experimental method Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Processing Or Creating Images (AREA)
Abstract
The invention relates to a 3D modeling optimization method based on GCN, which adopts brand new logic design, judges the vertex state based on a deep learning GCN model, establishes a maximum triangle surface communication domain, regenerates a triangle surface, under the condition of ensuring the minimum change of the overall modeling shape, firstly forming polygonal surfaces by a plurality of triangular surfaces, then converting the polygonal surfaces into the minimum triangular surfaces, thereby realizing the compression and simplification of the ultrahigh-density triangular surface, solving the problems of 'blocking', 'damaged surface' and the like in the 3D modeling process, reducing the final generated surface of the 3D modeling, thereby improving the working efficiency, optimizing the display effect and reducing the dependence of the display equipment, and experiments prove that, the algorithm can optimize and reduce more than 60% of redundant triangular faces in the construction of smart cities, therefore, the design of the invention is very effective for large-scale 3D modeling, particularly 'data city modeling'.
Description
Technical Field
The invention relates to a GCN-based 3D modeling optimization method, and belongs to the technical field of 3D modeling optimization.
Background
In recent years, the construction of smart cities in China is fierce, and information technologies such as artificial intelligence, big data, Internet of things and the like are important supports for the construction of smart cities. The construction of the smart city can not only promote the development of the city, but also help to relieve the large urban diseases and improve the urbanization quality. The three-dimensional reconstruction technology is used as a core construction force of a digital twin system and a propeller of a global smart city project, and under the large environment of day-to-day acquisition facility updating and digital storage explosive growth, intelligent schemes such as focusing depth application and the like are urgently needed to perfect and optimize a reconstruction algorithm scheme.
Three-dimensional reconstruction refers to the establishment of a mathematical model suitable for computer representation and processing of a three-dimensional object, is the basis for processing, operating and analyzing the properties of the three-dimensional object in a computer environment, and is also a key technology for establishing virtual reality expressing an objective world in a computer. The general three-dimensional reconstruction method is to calibrate the camera, i.e. to calculate the relationship between the image coordinate system of the camera and the world coordinate system, and then to reconstruct the three-dimensional information by using the information in a plurality of two-dimensional images.
The existing three-dimensional reconstruction method is to construct triangular surfaces based on point clouds obtained after scanning, and the densities of the triangular surfaces are often very high, so that subsequent storage, transmission, display and reconstruction consume a large amount of time and computer resources. Therefore, it is necessary to provide a three-dimensional reconstruction optimization algorithm and a three-dimensional reconstruction optimization flow to solve the above problems.
Disclosure of Invention
The invention aims to solve the technical problem of providing a 3D modeling optimization method based on GCN, which adopts brand-new logic design, can effectively reduce redundant triangular surfaces in the 3D modeling optimization method under the condition of ensuring the minimum change of the overall modeling shape, reduces the storage capacity and can effectively improve the actual application efficiency of a 3D modeling file.
The invention adopts the following technical scheme for solving the technical problems: the invention designs a 3D modeling optimization method based on GCN, based on a 3D modeling sample storage file in which all nodes respectively correspond to a deletable label or a non-deletable label, and obtaining a 3D modeling node classification model according to the following steps I to II; according to the 3D modeling node classification model, optimizing a target 3D modeling storage file according to the following steps A to C;
step I, acquiring a feature vector group of each order corresponding to a node based on a feature vector group composed of feature values of each preset feature type corresponding to the node, and respectively aiming at each node in a 3D modeling sample storage file, and based on a sampling order K of a preset neighbor node; further acquiring neighbor feature vector groups of each order respectively corresponding to each node in the 3D modeling sample storage file, and then entering the step II;
step II, training a specified classification model based on each node in the 3D modeling sample storage file by taking each order of neighbor feature vector group corresponding to the node as input and a deletable label or a non-deletable label corresponding to the node as output to obtain a 3D modeling node classification model;
step A, according to the method of the step I, obtaining neighbor characteristic vector groups of each order respectively corresponding to each node in a target 3D modeling storage file, and then entering the step B;
b, applying the 3D modeling node classification model to obtain a deletable label or a non-deletable label corresponding to each node in the target 3D modeling storage file, and then entering the step C;
and C, deleting nodes based on each triangular surface in the target 3D modeling storage file according to each node corresponding to the deletable label in the target 3D modeling storage file, executing triangular surface division aiming at polygonal surfaces which are not triangular surfaces, and updating each node and each triangular surface in the target 3D modeling storage file.
As a preferred technical scheme of the invention: the step I comprises the following steps I1 to I5;
step I1., using an array formed by nodes respectively corresponding to two ends of an edge as the representation of the edge, forming an edge set E by obtaining the representation of each edge on all triangular surfaces in the 3D modeling sample storage file, and then entering step I2;
step I2, counting the times of the nodes in the 3D modeling sample storage file appearing in the edge set E respectively, and counting the number N of the nodes in the 3D modeling sample storage file and the value D of the N column and the N row according to the number N of the nodes in the 3D modeling sample storage file and the number N of the nodes in the N row and the N column, wherein N is more than or equal to 1 and less than or equal to NnnThe number of times of appearance of the nth node in the edge set E is equal, the values of the rest positions are 0, a degree matrix D of N x N is constructed, and then the step I3 is carried out;
step I3, based on that p is more than or equal to 1 and less than or equal to N and q is more than or equal to 1 and less than or equal to N, traversing each position in the degree matrix D according to the row sequence number p and the column sequence number q in the degree matrix D where the position is located according to the following formula:
wherein,<vp,vq>representing a connecting line between the p-th node and the q-th node in the 3D modeling sample storage file, and obtaining a value A of the p-th row and the q-th column positionpqThen from each ApqCombining and constructing an N × N adjacent matrix A, and then entering the step I4;
step I4., according to a preset neighbor node sampling order K, regarding each node in the 3D modeling sample storage file, respectively, with the node as a vertex, based on the degree matrix D and the adjacency matrix a, according to the following formula:
min (-) represents taking the minimum function, taking the k-th order neighbor sample as the sample SkOn the basis that K is more than or equal to 1 and less than or equal to K, sequentially executing each K-th order neighbor sampling aiming at the vertex to obtain each neighbor node of which the vertex corresponds to each K-th order neighbor sampling respectively, namely forming each order neighbor node set corresponding to the node; further acquiring each-order neighbor node set corresponding to each node in the 3D modeling sample storage file, and then entering step I5;
i5., aiming at each node in the 3D modeling sample storage file, respectively, obtaining a feature vector group formed by feature values of preset feature types corresponding to each neighbor node in a kth-order neighbor node set corresponding to the node based on that K is more than or equal to 1 and less than or equal to K, obtaining an average feature vector group and a variance feature vector group corresponding to each feature vector group, and then adding the average feature vector group and the variance feature vector group to form the kth-order neighbor feature vector group corresponding to the node; and further acquiring the neighbor feature vector groups of each order respectively corresponding to each node in the 3D modeling sample storage file.
As a preferred technical scheme of the invention: in the step II, a cross entropy loss function is adopted in the process of training the specified classification model.
As a preferred technical scheme of the invention: the step C comprises the following steps C1 to C2;
c1, firstly, respectively aiming at each node corresponding to a deletable label in a target 3D modeling storage file, classifying the node and each node corresponding to the deletable label in each 1-order neighbor node corresponding to the node into the same connected domain, and further obtaining each connected domain; then merging the connected domains containing the same node, updating to obtain each connected domain, only reserving one repeated node in each connected domain, and then entering the step C2;
step C2., based on each triangular surface in the target 3D modeling storage file, executing the following steps C2-1 to C2-5 respectively for the minimum circumscribed polygon corresponding to each connected domain;
step C2-1, initializing the iteration number n as 1, and proceeding to step C2-1-1;
c2-1-1, randomly selecting one node from all nodes corresponding to the deletable labels in the minimum circumscribed polygon, obtaining a polygon formed by triangular faces to which the node belongs, classifying the polygon into a polygon corresponding to the nth iteration, and then entering the step C2-1-2;
c2-1-2, judging whether a node which does not belong to the corresponding deletable label of each polygon corresponding to the nth iteration exists in the minimum circumscribed polygon, if so, entering the step C2-2; otherwise, entering a step C2-4;
c2-2, randomly selecting a node from all nodes which do not belong to the corresponding deletable labels of all the polygons corresponding to the nth iteration in the minimum circumscribed polygon, obtaining the polygon formed by all the triangular surfaces to which the node belongs, classifying the polygon into the polygon corresponding to the nth iteration, and then entering the step C2-3;
c2-3, judging whether the minimum circumscribed polygon has a node which does not belong to the deletable label corresponding to each polygon corresponding to the nth iteration, if so, returning to the step C2-2; otherwise, entering a step C2-4;
step C2-4, respectively aiming at each polygon corresponding to the nth iteration, executing the following steps C2-4-1 to C2-4-3 to realize the division of the triangular surface in the polygon; further realizing the division of the triangular surface in each polygon corresponding to the nth iteration, and then entering the step C2-5;
step C2-4-1, taking the polygon as the polygon to be processed, deleting the central node in the polygon to be processed, updating the polygon to be processed, and entering the step C2-4-2;
c2-4-2, aiming at any vertex on the edge of the polygon to be processed and two adjacent vertices on two sides of the polygon to be processed, connecting the two adjacent vertices on two sides, forming a triangular surface by the three vertices, updating the triangular surface area in the polygon to be processed, and entering the step C2-4-3;
step C2-4-3, judging whether the remaining regions exist in the polygon to be processed except the triangular surface region constructed in the step C2-4-2, if so, taking the remaining regions as the polygon to be processed, and returning to the step C2-4-2; otherwise, ending the operation;
c2-5, judging whether the minimum circumscribed polygon has a node corresponding to the deletable label, if yes, adding 1 for updating aiming at the value of n, and returning to the step C2-1-1; otherwise, the optimization updating of each node and each triangular surface in the target 3D modeling storage file is completed.
Compared with the prior art, the GCN-based 3D modeling optimization method has the following technical effects:
the invention designs a 3D modeling optimization method based on GCN, adopts brand new logic design, judges the vertex state based on a deep learning GCN model, establishes a maximum triangle surface communication domain, regenerates a triangle surface, under the condition of ensuring the minimum change of the overall modeling shape, firstly forming polygonal surfaces by a plurality of triangular surfaces, then converting the polygonal surfaces into the minimum triangular surfaces, thereby realizing the compression and simplification of the ultrahigh-density triangular surface, solving the problems of 'blocking', 'damaged surface' and the like in the 3D modeling process, reducing the final generated surface of the 3D modeling, thereby improving the working efficiency, optimizing the display effect and reducing the dependence of the display equipment, and experiments prove that, the algorithm can optimize and reduce more than 60% of redundant triangular faces in the construction of smart cities, therefore, the design of the invention is very effective for large-scale 3D modeling, particularly 'data city modeling'.
Drawings
FIG. 1 is a schematic flow chart of a GCN-based 3D modeling optimization method designed by the present invention;
FIG. 2 is a schematic diagram of a triangular surface in a 3D modeling storage file in the design application of the present invention;
FIG. 3 is an illustration of the invention in the design of an application of a medium level matrix;
FIG. 4 is a schematic representation of a adjacency matrix in the design application of the present invention;
FIG. 5 is a schematic diagram of neighbor node sampling in the design application of the present invention;
FIG. 6 is a schematic diagram of a minimum circumscribed polygon corresponding to a single circular connected domain in the design application of the present invention;
FIG. 7 is a schematic diagram of the partitioning of polygons in a single circular connected domain in the design application of the present invention;
FIG. 8 is a schematic diagram of the division of the triangular faces in each polygon in a single annular connected domain in the design application of the present invention.
Detailed Description
The following description will explain embodiments of the present invention in further detail with reference to the accompanying drawings.
The invention designs a 3D modeling optimization method based on GCN, based on a 3D modeling sample storage file in which all nodes respectively correspond to a deletable label or a non-deletable label, as shown in figure 1, a 3D modeling node classification model is obtained according to the following steps I to II.
Step I, acquiring a feature vector group of each order corresponding to a node based on a feature vector group composed of feature values of each preset feature type corresponding to the node, and respectively aiming at each node in a 3D modeling sample storage file, and based on a sampling order K of a preset neighbor node; and further acquiring the neighbor feature vector groups of each order respectively corresponding to each node in the 3D modeling sample storage file, and then entering the step II.
In practical applications, the step I includes the following steps I1 to I5.
Step I1. uses an array of nodes corresponding to the ends of the edge as a representation of the edge, such as<vp,vq>That is, the edge of the connection line between the p-th node and the q-th node in the 3D modeling sample storage file, as shown in fig. 2, an edge set E is formed by obtaining representations of edges on all triangular surfaces in the 3D modeling sample storage file, and then step I2 is performed.
Step I2, counting the times of the occurrence of each node in the 3D modeling sample storage file in the edge set E respectively, and storing the number of the nodes of the file based on the 3D modeling sampleThe number N, and 1. ltoreq. n.ltoreq.N, with the value D in the N-th row NnnEqual to the number of times the nth node appears in the edge set E, with the remaining positions having values of 0, construct a degree matrix D of N × N, such as that shown in fig. 3, and then proceed to step I3.
Step I3, based on that p is more than or equal to 1 and less than or equal to N and q is more than or equal to 1 and less than or equal to N, traversing each position in the degree matrix D according to the row sequence number p and the column sequence number q in the degree matrix D where the position is located according to the following formula:
wherein,<vp,vq>representing a connecting line between the p-th node and the q-th node in the 3D modeling sample storage file, and obtaining a value A of the p-th row and the q-th column positionpqThen from each ApqThe N × N adjacency matrix a is constructed by combining them, as shown in fig. 4, and then the process proceeds to step I4.
Step I4., according to a preset neighbor node sampling order K, regarding each node in the 3D modeling sample storage file, respectively, with the node as a vertex, based on the degree matrix D and the adjacency matrix a, according to the following formula:
min (-) represents taking the minimum function, taking the k-th order neighbor sample as the sample SkBased on that K is greater than or equal to 1 and is less than or equal to K, executing each K-th order neighbor sampling in sequence aiming at the vertex to obtain each neighbor node of which the vertex corresponds to each K-th order neighbor sampling respectively, namely forming each order neighbor node set corresponding to the node, such as shown in fig. 5; and further acquiring the neighbor node sets of each order corresponding to each node in the 3D modeling sample storage file, and then entering step I5.
I5., aiming at each node in the 3D modeling sample storage file, respectively, obtaining a feature vector group formed by feature values of preset feature types corresponding to each neighbor node in a kth-order neighbor node set corresponding to the node based on that K is more than or equal to 1 and less than or equal to K, obtaining an average feature vector group and a variance feature vector group corresponding to each feature vector group, and then adding the average feature vector group and the variance feature vector group to form the kth-order neighbor feature vector group corresponding to the node; and further acquiring the neighbor feature vector groups of each order respectively corresponding to each node in the 3D modeling sample storage file.
And II, based on each node in the 3D modeling sample storage file, taking each order of neighbor feature vector group corresponding to the node as input, and taking a deletable label or a non-deletable label corresponding to the node as output, and combining the following cross entropy loss functions:
training a specified classification model to obtain a 3D modeling node classification model; wherein L represents a loss value, y represents a tag value,
representing the full connectivity layer output.
According to the acquisition of the classification model of the 3D modeling node, the target 3D modeling storage file is further optimized according to the following steps A to C as shown in FIG. 1.
And step A, according to the method in the step I, obtaining each-order neighbor feature vector group corresponding to each node in the target 3D modeling storage file respectively, and then entering the step B.
And B, applying the 3D modeling node classification model to obtain the deletable label or the undeletable label corresponding to each node in the target 3D modeling storage file, and then entering the step C.
And C, deleting nodes based on each triangular surface in the target 3D modeling storage file according to each node corresponding to the deletable label in the target 3D modeling storage file, executing triangular surface division aiming at polygonal surfaces which are not triangular surfaces, and updating each node and each triangular surface in the target 3D modeling storage file.
In practical applications, the step C is performed as the following steps C1 to C2.
C1, firstly, respectively aiming at each node corresponding to a deletable label in a target 3D modeling storage file, classifying the node and each node corresponding to the deletable label in each 1-order neighbor node corresponding to the node into the same connected domain, and further obtaining each connected domain; then, the connected domains containing the same node are merged, each connected domain is obtained by updating, only one node in each connected domain is reserved, such as shown in fig. 6, and then the step C2 is performed.
Step C2. performs the following steps C2-1 to C2-5 for the minimum bounding polygon corresponding to each connected domain, such as shown in FIG. 6, respectively, based on each triangular surface in the target 3D modeling storage file.
Step C2-1. initialize the iteration number n equal to 1, and go to step C2-1-1.
And C2-1-1, randomly selecting one node from all nodes corresponding to the deletable labels in the minimum circumscribed polygon, obtaining a polygon formed by triangular surfaces to which the node belongs, classifying the polygon into the polygon corresponding to the nth iteration, and then entering the step C2-1-2.
C2-1-2, judging whether a node which does not belong to the corresponding deletable label of each polygon corresponding to the nth iteration exists in the minimum circumscribed polygon, if so, entering the step C2-2; otherwise, go to step C2-4.
And C2-2, randomly selecting one node from the nodes which do not belong to the corresponding deletable labels of the polygons corresponding to the nth iteration in the minimum circumscribed polygon, obtaining the polygon formed by the triangular surfaces to which the node belongs, classifying the polygon into the polygon corresponding to the nth iteration, such as the polygon shown in FIG. 7, and then entering the step C2-3.
C2-3, judging whether the minimum circumscribed polygon has a node which does not belong to the deletable label corresponding to each polygon corresponding to the nth iteration, if so, returning to the step C2-2; otherwise, go to step C2-4.
Step C2-4, respectively aiming at each polygon corresponding to the nth iteration, executing the following steps C2-4-1 to C2-4-3 to realize the division of the triangular surface in the polygon; and further realizing the division of the triangular surface in each polygon corresponding to the nth iteration, and then entering the step C2-5.
And C2-4-1, taking the polygon as the polygon to be processed, deleting the central node in the polygon to be processed, updating the polygon to be processed, and entering the step C2-4-2.
And C2-4-2, aiming at any vertex on the edge of the polygon to be processed and two adjacent vertices on two sides of the polygon to be processed, connecting the two adjacent vertices on two sides, forming a triangular surface by the three vertices, updating the triangular surface area in the polygon to be processed, and entering the step C2-4-3.
Step C2-4-3, judging whether the remaining regions exist in the polygon to be processed except the triangular surface region constructed in the step C2-4-2, if so, taking the remaining regions as the polygon to be processed, and returning to the step C2-4-2; otherwise, the operation is ended.
C2-5, judging whether the minimum circumscribed polygon has a node corresponding to the deletable label, if yes, adding 1 for updating aiming at the value of n, and returning to the step C2-1-1; otherwise, the optimization updating of each node and each triangular surface in the target 3D modeling storage file is completed, such as shown in FIG. 8.
The technical scheme designs a 3D modeling optimization method based on GCN, adopts brand new logic design, judges the vertex state based on a deep learning GCN model, establishes a maximum triangle surface communication domain and regenerates a triangle surface, under the condition of ensuring the minimum change of the overall modeling shape, a plurality of triangle surfaces form polygon surfaces, and then the polygon surfaces are converted into the minimum triangle surfaces, thereby realizing the compression and simplification of the triangle surfaces with ultrahigh density, solving the problems of 'blocking, damaged surfaces' and the like in the 3D modeling process, reducing the final generated surface of the 3D modeling, further improving the working efficiency, optimizing the display effect and reducing the dependence of display equipment, and experiments prove that the algorithm can optimize more than 60 percent of redundant triangle surfaces in the smart city construction, thereby designing large 3D, especially "data city modeling" is very efficient.
The embodiments of the present invention will be described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.
Claims (4)
1. A3D modeling optimization method based on GCN is characterized in that: based on the 3D modeling sample storage file in which each node respectively corresponds to a deletable label or a non-deletable label, obtaining a 3D modeling node classification model according to the following steps I to II; according to the 3D modeling node classification model, optimizing a target 3D modeling storage file according to the following steps A to C;
step I, acquiring a feature vector group of each order corresponding to a node based on a feature vector group composed of feature values of each preset feature type corresponding to the node, and respectively aiming at each node in a 3D modeling sample storage file, and based on a sampling order K of a preset neighbor node; further acquiring neighbor feature vector groups of each order respectively corresponding to each node in the 3D modeling sample storage file, and then entering the step II;
step II, training a specified classification model based on each node in the 3D modeling sample storage file by taking each order of neighbor feature vector group corresponding to the node as input and a deletable label or a non-deletable label corresponding to the node as output to obtain a 3D modeling node classification model;
step A, according to the method of the step I, obtaining neighbor characteristic vector groups of each order respectively corresponding to each node in a target 3D modeling storage file, and then entering the step B;
b, applying the 3D modeling node classification model to obtain a deletable label or a non-deletable label corresponding to each node in the target 3D modeling storage file, and then entering the step C;
and C, deleting nodes based on each triangular surface in the target 3D modeling storage file according to each node corresponding to the deletable label in the target 3D modeling storage file, executing triangular surface division aiming at polygonal surfaces which are not triangular surfaces, and updating each node and each triangular surface in the target 3D modeling storage file.
2. The GCN-based 3D modeling optimization method of claim 1, wherein: the step I comprises the following steps I1 to I5;
step I1., using an array formed by nodes respectively corresponding to two ends of an edge as the representation of the edge, forming an edge set E by obtaining the representation of each edge on all triangular surfaces in the 3D modeling sample storage file, and then entering step I2;
step I2, counting the times of the nodes in the 3D modeling sample storage file appearing in the edge set E respectively, and counting the number N of the nodes in the 3D modeling sample storage file and the value D of the N column and the N row according to the number N of the nodes in the 3D modeling sample storage file and the number N of the nodes in the N row and the N column, wherein N is more than or equal to 1 and less than or equal to NnnThe number of times of appearance of the nth node in the edge set E is equal, the values of the rest positions are 0, a degree matrix D of N x N is constructed, and then the step I3 is carried out;
step I3, based on that p is more than or equal to 1 and less than or equal to N and q is more than or equal to 1 and less than or equal to N, traversing each position in the degree matrix D according to the row sequence number p and the column sequence number q in the degree matrix D where the position is located according to the following formula:
wherein,<vp,vq>representing a connecting line between the p-th node and the q-th node in the 3D modeling sample storage file, and obtaining a value A of the p-th row and the q-th column positionpqThen from each ApqCombining and constructing an N × N adjacent matrix A, and then entering the step I4;
step I4., according to a preset neighbor node sampling order K, regarding each node in the 3D modeling sample storage file, respectively, with the node as a vertex, based on the degree matrix D and the adjacency matrix a, according to the following formula:
min (-) represents taking the minimum function, taking the k-th order neighbor sample as the sample SkEach of the k-th order neighbor nodes being different from each otherBased on K being more than or equal to 1 and less than or equal to K, sequentially executing each K-th order neighbor sampling aiming at the vertex to obtain each neighbor node of the vertex corresponding to each K-th order neighbor sampling respectively, namely forming each order neighbor node set corresponding to the node; further acquiring each-order neighbor node set corresponding to each node in the 3D modeling sample storage file, and then entering step I5;
i5., aiming at each node in the 3D modeling sample storage file, respectively, obtaining a feature vector group formed by feature values of preset feature types corresponding to each neighbor node in a kth-order neighbor node set corresponding to the node based on that K is more than or equal to 1 and less than or equal to K, obtaining an average feature vector group and a variance feature vector group corresponding to each feature vector group, and then adding the average feature vector group and the variance feature vector group to form the kth-order neighbor feature vector group corresponding to the node; and further acquiring the neighbor feature vector groups of each order respectively corresponding to each node in the 3D modeling sample storage file.
3. The GCN-based 3D modeling optimization method of claim 1, wherein: in the step II, a cross entropy loss function is adopted in the process of training the specified classification model.
4. The GCN-based 3D modeling optimization method of claim 1, wherein: the step C comprises the following steps C1 to C2;
c1, firstly, respectively aiming at each node corresponding to a deletable label in a target 3D modeling storage file, classifying the node and each node corresponding to the deletable label in each 1-order neighbor node corresponding to the node into the same connected domain, and further obtaining each connected domain; then merging the connected domains containing the same node, updating to obtain each connected domain, only reserving one repeated node in each connected domain, and then entering the step C2;
step C2., based on each triangular surface in the target 3D modeling storage file, executing the following steps C2-1 to C2-5 respectively for the minimum circumscribed polygon corresponding to each connected domain;
step C2-1, initializing the iteration number n as 1, and proceeding to step C2-1-1;
c2-1-1, randomly selecting one node from all nodes corresponding to the deletable labels in the minimum circumscribed polygon, obtaining a polygon formed by triangular faces to which the node belongs, classifying the polygon into a polygon corresponding to the nth iteration, and then entering the step C2-1-2;
c2-1-2, judging whether a node which does not belong to the corresponding deletable label of each polygon corresponding to the nth iteration exists in the minimum circumscribed polygon, if so, entering the step C2-2; otherwise, entering a step C2-4;
c2-2, randomly selecting a node from all nodes which do not belong to the corresponding deletable labels of all the polygons corresponding to the nth iteration in the minimum circumscribed polygon, obtaining the polygon formed by all the triangular surfaces to which the node belongs, classifying the polygon into the polygon corresponding to the nth iteration, and then entering the step C2-3;
c2-3, judging whether the minimum circumscribed polygon has a node which does not belong to the deletable label corresponding to each polygon corresponding to the nth iteration, if so, returning to the step C2-2; otherwise, entering a step C2-4;
step C2-4, respectively aiming at each polygon corresponding to the nth iteration, executing the following steps C2-4-1 to C2-4-3 to realize the division of the triangular surface in the polygon; further realizing the division of the triangular surface in each polygon corresponding to the nth iteration, and then entering the step C2-5;
step C2-4-1, taking the polygon as the polygon to be processed, deleting the central node in the polygon to be processed, updating the polygon to be processed, and entering the step C2-4-2;
c2-4-2, aiming at any vertex on the edge of the polygon to be processed and two adjacent vertices on two sides of the polygon to be processed, connecting the two adjacent vertices on two sides, forming a triangular surface by the three vertices, updating the triangular surface area in the polygon to be processed, and entering the step C2-4-3;
step C2-4-3, judging whether the remaining regions exist in the polygon to be processed except the triangular surface region constructed in the step C2-4-2, if so, taking the remaining regions as the polygon to be processed, and returning to the step C2-4-2; otherwise, ending the operation;
c2-5, judging whether the minimum circumscribed polygon has a node corresponding to the deletable label, if yes, adding 1 for updating aiming at the value of n, and returning to the step C2-1-1; otherwise, the optimization updating of each node and each triangular surface in the target 3D modeling storage file is completed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111117830.9A CN113570712B (en) | 2021-09-23 | 2021-09-23 | 3D modeling optimization method based on GCN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111117830.9A CN113570712B (en) | 2021-09-23 | 2021-09-23 | 3D modeling optimization method based on GCN |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113570712A CN113570712A (en) | 2021-10-29 |
| CN113570712B true CN113570712B (en) | 2022-02-15 |
Family
ID=78174126
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111117830.9A Active CN113570712B (en) | 2021-09-23 | 2021-09-23 | 3D modeling optimization method based on GCN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113570712B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114780666B (en) * | 2022-06-23 | 2022-09-27 | 四川见山科技有限责任公司 | Road label optimization method and system in digital twin city |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111402382B (en) * | 2020-03-18 | 2023-04-07 | 东南数字经济发展研究院 | Classification optimization method for improving data rendering efficiency of layered and partitioned three-dimensional model |
| CN111415406B (en) * | 2020-03-18 | 2023-05-12 | 东南数字经济发展研究院 | Hierarchical and blocking three-dimensional model data classification compression method |
| CN112287997A (en) * | 2020-10-27 | 2021-01-29 | 浙江工业大学 | Depth map convolution model defense method based on generative confrontation network |
| CN112749759A (en) * | 2021-01-21 | 2021-05-04 | 西安电子科技大学 | Preprocessing method, system and application of confrontation sample of deep neural network map |
| CN113269228B (en) * | 2021-04-20 | 2022-06-10 | 重庆邮电大学 | Method, device and system for training graph network classification model and electronic equipment |
-
2021
- 2021-09-23 CN CN202111117830.9A patent/CN113570712B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN113570712A (en) | 2021-10-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108038906B (en) | Three-dimensional quadrilateral mesh model reconstruction method based on image | |
| CN115761172A (en) | Single building three-dimensional reconstruction method based on point cloud semantic segmentation and structure fitting | |
| CN112257597B (en) | Semantic segmentation method for point cloud data | |
| CN110751075A (en) | Remote sensing image culture pond detection method based on example segmentation | |
| Fayolle et al. | An evolutionary approach to the extraction of object construction trees from 3D point clouds | |
| CN105405166A (en) | LOD model generation method based on linear quadtree | |
| CN114926602B (en) | Building singleization method and system based on three-dimensional point cloud | |
| CN110674742A (en) | Remote sensing image road extraction method based on DLinkNet | |
| CN111028335A (en) | Point cloud data block surface patch reconstruction method based on deep learning | |
| CN113570712B (en) | 3D modeling optimization method based on GCN | |
| CN113868476A (en) | Octree point cloud preprocessing method based on local density | |
| CN114417951A (en) | Unsupervised machine learning-based automatic subdivision optimization method for ocean unstructured grid | |
| CN110084136A (en) | Context based on super-pixel CRF model optimizes indoor scene semanteme marking method | |
| CN110378047A (en) | A kind of Longspan Bridge topology ambiguity three-dimensional rebuilding method based on computer vision | |
| Hu et al. | Geometric feature enhanced line segment extraction from large-scale point clouds with hierarchical topological optimization | |
| CN113724279A (en) | System, method, equipment and storage medium for automatically dividing traffic cells into road networks | |
| Benes et al. | Urban brush: Intuitive and controllable urban layout editing | |
| Yang et al. | Connectivity-aware Graph: A planar topology for 3D building surface reconstruction | |
| CN114219701A (en) | Dunhuang fresco artistic style conversion method, system, computer equipment and storage medium | |
| CN111402422B (en) | Three-dimensional surface reconstruction method and device and electronic equipment | |
| CN113256793A (en) | Three-dimensional data processing method and system | |
| CN111415406A (en) | Hierarchical and block-divided three-dimensional model data classification compression method | |
| CN114445574B (en) | Method, device and equipment for converting GeoJSON data format into three-dimensional GLB format | |
| CN102142155A (en) | Three-dimensional (3D) terrain model data organization method oriented to network interactive visualization | |
| Falcidieno et al. | Natural surface approximation by constrained stochastic interpolation |
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 |