CN110706276B - Skeleton guidance-based three-dimensional model intrinsic symmetry detection method - Google Patents

Abstract

The invention relates to a method for detecting the intrinsic symmetry of a three-dimensional model based on skeleton guidance, belonging to the fields of computer graphics, computer vision and the like. The method comprises the following steps: (1) generating a line framework of the model, and carrying out convex subdivision processing on the model based on the framework; (2) detecting intrinsic symmetry between their associated convex segments based on the similar skeletal limb segments and their subsections; (3) and searching for a similar skeleton structure, and detecting the internal implication symmetry body in the model based on the combination of the external implication symmetry convex bodies. The invention well reduces the calculation complexity of the detection of the intrinsic symmetry, can detect the symmetric bodies which are difficult to find by the existing method, such as small symmetric bodies, sub symmetric bodies embedded in large symmetric bodies and the like, and has the advantages of simplicity, convenience and high efficiency.

Description

Skeleton guidance-based three-dimensional model intrinsic symmetry detection method

Technical Field

The invention belongs to the fields of computer graphics, computer vision and the like, and particularly relates to a three-dimensional model intrinsic symmetry detection method based on skeleton guidance.

Background

Symmetry detection is the finding of transformations that map one part of the three-dimensional model to another without changing shape. There are many methods for symmetry detection with respect to three-dimensional models, and for a thorough understanding of these methods, reference is made to Mitra et al [1 ]. Here, only a brief description will be made, and particularly, description will be mainly made for the detection of intrinsic symmetry.

The transform voting method is a commonly used symmetric detection method. Except for a method of directly searching in a transformation space, most methods are that firstly, the surface of a model is densely sampled, then, the corresponding relation between sampling points is investigated, and possible dominant transformations are detected through voting, and the dominant transformations can reflect the symmetry of the model. Due to the complexity of the transform space, the computation of the correspondence between sample points is often very time consuming. Furthermore, due to voting strategies, these methods are limited to the detection of large-scale symmetries, which are difficult to detect for small-scale or complex symmetries [2 ].

The feature map matching method [3,4] is essentially a scheme of eliminating non-matching geometric objects by using feature matching, and the non-matching geometric objects are aligned by an iterative nearest neighbor method (ICP) [10] to obtain a final detection result. As discussed in [2], these methods are directed to rigid alignment processes, mainly for extrinsic symmetry detection. Furthermore, such methods rely on point/line features extracted from the model surface, which are difficult to represent properly for volumetric features, and therefore, they do not facilitate the detection of symmetric volumes where volumetric features are evident. Typically alignment is done by brute force solution in an exhaustive manner, which is computationally expensive. The feature map is used for symmetry detection, detection calculation can be optimized, but the features can only approximately represent the model, some symmetric bodies can be omitted, and the obtained symmetric bodies are poor in quality.

Spectral clustering methods attempt to find symmetry using statistical methods for clustering. They use a matrix to find the symmetry points of the clusters [6], such as geometric patches or features [2 ]. Since the intrinsic symmetry is a non-rigid transformation and cannot be expressed by a matrix, some methods search for the correspondence by a function mapping method. These methods can avoid noise interference well and improve symmetry detection effect, but they are actually based on sampling points, patches, features and voting strategies. Therefore, they have similar disadvantages to other kinds of methods in terms of detection efficiency, detection effect, detection accuracy, and the like. For example, the method in [6] tends to detect large-scale and fixed-size symmetries, and does not facilitate identifying symmetries with smaller scales. While the method in [2] is difficult to detect large scale symmetries because it does not consider the correlation between features that are too far apart for computational efficiency. Although this method [7] reduces the temporal complexity to O (knlog (n)) (where k is the number of eigenfunctions used and n is the number of vertices of the model), the computational cost is still high. These methods can be viewed as an integration of transform voting (without the need for explicit transform processing) and the feature map matching method. Therefore, they also do not guarantee the generation of a high quality symmetric body, as indicated in the introduction of the transformation voting method and the feature map matching method.

In many practical applications of symmetry detection, the complexity of the computation is an important factor and even a decisive factor affecting the success of the application. Therefore, the method for detecting the symmetry of the optical fiber has very important practical application value.

[1]MITRA N.J.,PAULY M.,WAND M.,CEYLAN D.:Symmetry in 3d geometry:Extraction and applications.Computer Graphics Forum 32,6(2013),1–23.

[2]LI C.,WAND M.,WU X.,SEIDEL H.-P.:Approximate 3d partial symmetry detection using co-occurrence analysis.In Proc.Of International Conference on 3D Vision(3Dv’15)(2015),pp.425–433.

[3]M.Couprie and G.Bertrand.Asymmetric parallel 3d thinning scheme and algorithms based on isthmuses.Pattern Recognition Letters,76:22–31,2016

[4]BERNER A.,BOKELOH M.,WAND M.,SCHILLING A.,SEIDEL H.-P.:A graph-based approach to symmetry detection.In Proc.of the Fifth Eurographics/IEEE VGTC Conference on Point-Based Graphics(SPBG’08)(2008),pp.1–8.

[5]BOKELOH M.,BERNER A.,WAND M.,SEIDEL H.-P.,SCHILLING A.:Symmetry detection using feature lines.Computer Graphics Forum 28,2(2009),697–706.

[6]LIPMAN Y.,CHEN X.,DAUBECHIES I.,FUNKHOUSER T.:Symmetry factored embedding and distance.ACM Transactions on Graphics 29,4(2010),103:1–103:12.

[7]NAGAR R.,RAMAN S.:Fast and accurate intrinsic symmetry detection.In Proceedings of the European Conference on Computer Vision(ECCV)(2018),pp.417–434.

[8]AU O.K.-C.,TAI C.-L.,CHU H.-K.,COHEN-OR D.,LEE T.-Y.:Skeleton extraction by mesh contraction.ACM Transactions on Graphics 27,3(Aug.2008),44:1–44:10.

[9]WANG Y.,XU K.,LI J.,ZHANG H.,SHAMIR A.,LIU L.,CHENG Z.,XIONG Y.:Symmetry hierarchy of man-made objects.In Computer graphics forum(2011),vol.30,Wiley Online Library,pp.287–296.

[10]RUSINKIEWICZ S.,LEVOY M.:Efficient variants of the icp algorithm.In Proc.of 3th International Conference on 3-D Digital Imaging and Modeling Proceedings(2001),pp.145–152.

[11]TIERNY J.,VANDEBORRE J.-P.,DAOUDI M.:3d mesh skeleton extraction using topological and geometrical analyses.In Proc.of 14th Pacific Conference on Computer Graphics and Applications(PG’06)(Tapei,Taiwan,Oct.2006),pp.85–94.6

Disclosure of Invention

The invention solves the problems: the method for detecting the internal implication symmetry of the three-dimensional model based on the skeleton guidance is characterized in that the generated model line skeleton is used as guidance, the model is subjected to convex subdivision processing based on the skeleton, and then the internal implication symmetry body in the model is detected based on the combination of the external implication symmetry convex bodies, so that the method has the advantages of simplicity, convenience and high efficiency.

The technical scheme of the invention is that a three-dimensional model intrinsic symmetry detection method based on skeleton guidance comprises the following steps:

(1) generating a framework of the three-dimensional model, carrying out convex subdivision on the three-dimensional model based on the framework, and establishing a corresponding relation between the framework limb section and the framework node and a convex body of the convex subdivision; the skeleton consists of a plurality of nodes and edges connecting the nodes, wherein the nodes shared by more than two skeleton edges are called skeleton nodes, a skeleton limb is formed from one skeleton node to another skeleton node or a string of edges of one end point of the skeleton, and the end point of the skeleton refers to the node only associated with one edge;

(2) according to the skeleton limb subsections (namely the subsections after the skeleton limb is split) with similar lengths and the skeleton nodes, the extrinsic symmetry between the convex bodies corresponding to the skeleton limb subsections and the skeleton nodes is detected. If the two forms can be superposed through rigid transformation, the two forms are called to be externally symmetrical; two shapes are said to be intrinsically symmetric if they can be made to coincide by non-rigid changes;

(3) searching similar structures in the framework, detecting extrinsic symmetry among all corresponding convex bodies to which the similar structures belong, and searching intrinsic symmetry bodies of the three-dimensional model based on a convex body combination mode;

the method comprises the following steps of (1) generating a framework of the model, carrying out convex dissection on the model based on the framework, and establishing a corresponding relation between a framework limb section and a framework node and a convex body of the convex dissection, and specifically comprises the following steps:

(11) generating a skeleton of the three-dimensional model according to a model mesh contraction method;

(12) initializing and dividing a three-dimensional model according to the corresponding relation information of skeleton limbs to a model grid, and dividing and optimizing according to concave information between the skeleton limbs and skeleton nodes, so that each skeleton node corresponds to a convex-divided body, and each skeleton limb corresponds to a divided body;

(13) carrying out convex dissection on each part of the initialized dissection according to concave information on the three-dimensional model, corresponding the concave position condition during dissection to the skeleton limb, and carrying out sub-dissection on the skeleton limb to form a sub-section of the skeleton limb;

(14) for skeleton limbs with the same length, performing mutual migration treatment on the sub-situations of the skeleton limbs with the same length, so that the skeleton limbs with the same length have similar sub-situations;

(15) and according to the sub-division condition of the skeleton limb, further sub-dividing the respectively associated convex bodies, so that each sub-section of the skeleton limb corresponds to one divided convex body respectively.

In the step (2), according to the similar skeleton limb subsection and the skeleton node, the extrinsic symmetry between the convex bodies corresponding to the similar skeleton limb subsection and the similar skeleton node is detected, and the specific steps are as follows:

(21) finding skeleton limb subsections with equivalent lengths, and detecting whether convex bodies corresponding to the skeleton limb subsections are symmetrical or not;

(22) and detecting whether the convex bodies corresponding to the skeleton nodes are symmetrical.

In the step (3), similar structures are searched in the framework, so that extrinsic symmetry among the corresponding convex bodies to which the structures belong is detected, and then an intrinsic symmetry body of the three-dimensional model is searched based on a convex body combination mode; the method comprises the following specific steps:

(31) finding skeleton limbs with the same length to form a similar skeleton structure;

(32) using each skeleton node as a root node, and gradually associating the connected similar skeleton limbs and other similar tree-shaped skeleton structures to generate similar skeleton structures;

(33) detecting whether the convex bodies corresponding to the similar skeleton structures are symmetrical or not based on the similar skeleton structures; if so, the model parts corresponding to the similar framework structures are intrinsic symmetry, and compared with the prior art, the invention has the advantages that:

(1) the invention well reduces the calculation complexity of the detection of the intrinsic symmetry, can detect the symmetric bodies which are difficult to find by the existing method, such as small symmetric bodies, sub symmetric bodies embedded into large symmetric bodies and the like, and has the advantages of simplicity, convenience and high efficiency.

(2) Based on the extrinsic implication symmetry between the convex split bodies of the model, the symmetric convex bodies are combined by using the similar framework structure of the model, so as to carry out intrinsic implication symmetry detection. Compared with the existing method, the method disclosed by the invention has the advantages that the calculation complexity is well reduced, the defect that a small-scale symmetric body and an embedded symmetric body are difficult to find based on a statistical method is avoided, and a high-quality symmetric body can be obtained. Experimental results show that the method can find more symmetric bodies, the running speed is much faster than that of the existing method, and even the running speed is improved by several orders of magnitude.

Drawings

FIG. 1 is a schematic flow chart of the algorithm of the present invention;

FIG. 2 is a schematic diagram of skeleton structure optimization and symmetry detection of similar skeleton structures in the present invention. Wherein, (a) is an initialized decomposition result graph corresponding to the extracted framework and framework-vertex, (b) is a redundant node graph of the palm framework, (c) is a node graph of the corrected palm framework, and (d) is a node graph corresponding to the palm and the framework;

FIG. 3 is a computational framework diagram of the model subdivision of the present invention. For (a)₁) Initial subdivision (a)₂) Performing optimized subdivision based on the concave information of the model part corresponding to the skeleton nodes and the related skeleton limbs; (b)₁)、(b₂) Subdividing a model part corresponding to the skeleton limb based on concave information; (b)₃)、(b₄) The method comprises the steps of dividing a relevant model part based on the corresponding relation between skeleton limbs;

FIG. 4 is a schematic diagram of the symmetric combination of extrinsic implications and intrinsic implications obtained by the framework of the present invention. (a) The method comprises the following steps of (a) obtaining a model skeleton, (b) obtaining extrinsic symmetry, (c) obtaining a combination schematic diagram of adjacent skeleton limbs and skeleton nodes, and (d) generating a similar skeleton structure schematic diagram by gradually associating similar skeleton limbs;

FIG. 5 is a graph showing the results of comparative experiments between the method of the present invention and other methods. (a) The method is a TSV method, (b) is an SBP method, (c) is an MPISD method, (d) is the method of the invention, (e) is manual labeling data, and (f) is a truth value diagram;

FIG. 6 is a graph of the symmetry results of the method of the present invention for various three-dimensional model tests.

Detailed Description

The present invention will be described in detail below with reference to the accompanying drawings and examples.

As shown in FIG. 1, the method comprises the following specific steps:

(1) generating a curve skeleton of the three-dimensional model and carrying out convex subdivision on the model based on the skeleton: for the symmetry detection task, it is expected that the skeleton not only represents the topology of the model, but also its skeleton limb segments and skeleton nodes can well represent the geometry of the decomposed convex body of the model, thereby facilitating fast calculation. The subdivision calculation steps of the invention are as follows: firstly, based on the corresponding relationship between the skeleton and the model (before, redundant nodes in the skeleton are removed, the skeleton structure is optimized, as shown in (b) and (c) in fig. 2), the model is initially split, and each skeleton limb corresponds to one split part, as shown in (a) in fig. 2; then, for the split parts corresponding to the skeleton nodes and the related skeleton limbs, splitting according to the concave information of the split parts to enable each skeleton node to correspond to a convex split body, as shown in fig. 3; subsequently, for the corresponding portion of the skeleton limb, convex subdivision is performed according to the concave information thereof, and thus the skeleton limb is sub-divided based on the corresponding relationship between the model and the skeleton, as shown in (b) of fig. 3₁) Shown; finally, the skeleton limbs with the same length are subjected to mutual migration processing to further sub-divide the skeleton limbs, and then the associated convex bodies are further sub-divided according to the updated skeleton limb division conditions, as shown in (b) of fig. 3₂)、(b₃) As shown.

(2) Detecting the extrinsic symmetry of the convex body: when the extrinsic symmetry between the subdivided convex bodies is detected, firstly, the similar skeleton limb subsections and skeleton nodes on the skeleton are found, and then whether the corresponding convex bodies are symmetrical or not is checked. Here, two skeletal limb segments are considered similar if they are similar in length; their associated convexities are then examined for extrinsic symmetry. Two skeleton nodes are considered similar if their corresponding convex bodies are similar in volume.

(3) Multi-scale local intrinsic symmetry detection: after the extrinsic symmetry between the subdivision convex bodies is obtained, the intrinsic symmetry bodies of the model are searched through the guidance of similar framework structures and based on a combined mode. For two similar framework structures, if the convex bodies of the two framework structures are respectively and correspondingly externally symmetrical, the corresponding parts of the two framework structures are internally symmetrical. Similar skeleton structure is searched for by gradually increasing from small to large, so that multi-scale internal implication symmetry detection can be carried out.

The steps of the present invention are described separately below.

1. Curve skeleton generation and skeleton-based convex subdivision of model

(1) And (5) generating a skeleton. In consideration of the implementation convenience, the method in [8] is adopted to generate the initial curve skeleton, and then the skeleton is improved to well meet the requirements of people. Due to the detailed influence of the model, the initialized skeleton may have redundant skeleton nodes, as shown in fig. 2 (b), with two connected nodes in the right palm. This can lead to inaccuracies in the skeletal representation of the topology of the model. For this reason, a simple method is designed for quickly searching for the redundant nodes and performing the merging process to eliminate the redundant nodes. This approach mainly takes into account two important factors, namely the distance between nodes and the size of the node counterparts. The processing is based on the Euclidean distance (Ed) between two nodes and the sum (Sr) of the radii of inscribed spheres corresponding to the two nodes in the model. If Ed < σ Sr, then they are considered redundant and the two connection nodes are merged into one connection node, as shown in FIG. 2 (c). A setting of sigma to 1.5 generally gives high quality results.

(2) And (4) subdivision optimization based on convex information. And carrying out initial subdivision on the model by utilizing the corresponding relation between the skeleton limbs, the skeleton nodes and the model, and then optimizing the initial subdivision of the model by two steps. The first step is to optimize by using the concave information on the split parts corresponding to the adjacent skeleton limbs and the skeleton nodes, and the second step is to optimize by using the concave information on the corresponding parts of the skeleton limbs.

In the first step, the cutting lines are found through the concave folds of the model, and the corresponding split parts of each skeleton limb are sub-split. As shown in FIG. 3, for the initially dissected portion of the skeletal limb corresponding to the left arm (a in FIG. 3)₁) Pink), to optimize its subdivision boundary by searching for the appropriate cutting line (a in fig. 3)₂)). The search process for the cutting line is as follows: first, the recessed area of the part is found, since this area may contain suitable cutting lines. To this end, the model is divided into different regions, each region being part of the model that is contracted to a point on the skeletal limb, the boundaries of these regions being called loops, as shown in FIG. 3 (a)₁) Middle line around the arm. In general, the loop length near the boundary of the concave region varies greatly. Therefore, the concave area is found by comparing the length variation between adjacent turns. In the example of fig. 3, one loop is known (as in fig. 3 (a))₁) The white highlight in (1) is much longer than the average length of all the loops and is adjacent to the shorter loop, such a loop being a potential location for a concave area. According to such a loop, a concave region (fig. 3 (a)) is formed extending outward on both sides thereof₂) Arrow in (d) points to a region). Secondly, a suitable cutting line is generated within this recessed area. Find more than ten saddle points in the concave region (in FIG. 3 (a)₂) Medium point), using these saddle points, an approximate cutting plane is found using a least squares method. Then, a point closest to the cutting plane is found on the mesh of the model surface to constitute a cutting line for the subdivision process.

In the second step, the concavity of each coil on the split portion corresponding to the skeleton limb is calculated, then the concavities of the coils are arranged along the skeleton limb, the coil with the smallest local concavity is found, and the portion corresponding to the skeleton limb is convexly split, as shown in fig. 3 (b)₁)、(b₂) As shown. Calculation of the concavity of the coil using the document [11 ]]The specific calculation method of (1) is as follows:

based on the gaussian curvature, the concavity c (l) of the coil 1 is calculated by:

wherein k is₁v and k₂v is the principal curvature at the apex v of the loop line l. Such as [11 ]]The curvature calculation is sensitive to noise. Therefore, it is necessary to perform a filtering process on the concavity of the curve to reduce high frequency noise. For this purpose, as in [11 ]]As suggested, a cut-off frequency f is used_r(typically set to 0.4) a low pass filtering is performed, which is implemented as follows:

wherein f is_C(l)The average Euclidean distance between each vertex v of the loop and the related skeleton nodes is calculated to obtain:

here, d (v) is the euclidean distance between the vertex v and its associated skeleton node, and N is the vertex set of the loop line l. And the filtered loop line concavity C (l) is:

where FT denotes the fourier transform.

(3) And (5) decomposing the correspondence. Model projections corresponding to skeleton limbs of similar length may be in different poses. In order to detect the intrinsic symmetry between them, they need to be further decomposed correspondingly. For two similar skeletal limbs, we first follow the distribution of their loop concavity along the skeletal limb (in fig. 3 (b)₂) Find the local lowest inflection point of such distribution curve (such inflection point corresponds to the segmentation line position of the concave region of the model segmentation), and then find the local lowest inflection point according to the distance between themThe inflection points of the distribution curves of (1) are shifted from each other by the linear correspondence relationship of the distribution curves of (2). Subsequently, the convex body associated with the skeleton limb is further sub-divided according to the inflection point obtained by migration on the skeleton limb. As in (b) of FIG. 3₃，b₄) Although the right leg itself is shown as not bent, it can be split according to the left leg. In such a process, a skeletal limb may be divided into a plurality of sub-segments, one subdivision projection for each sub-segment.

2. Extrinsic symmetry detection of asperities

And (4) detecting whether the two convex bodies corresponding to the similar skeleton limb subsections or the similar skeleton nodes are symmetrical by adopting the method in the item [9 ]. First, a local coordinate system of a convex body is constructed by a principal component analysis method. Next, it is checked by the ICP algorithm [10] whether one convex body can be matched with another convex body. The algorithm samples points on one convex body, then maps the points to another convex body through transformation between convex bodies based on their local coordinate systems, and detects whether the mapped points can be well matched to the convex body after iterative convergence optimization. Since symmetry between the convex bodies is detected, the Root Mean Square Deviation (RMSD) of the mapping points can be used as a similarity measure for fast calculation. When a very low RMSD value is obtained, it is considered that the spur can be matched to another spur. If the two lugs are compatible with each other, they are symmetrical, as shown in FIG. 3 between the lugs in the palm corresponding to the two skeletal nodes.

3. Multi-scale local intrinsic symmetry detection

Along with the gradual small-to-large similar skeleton structure, the multi-scale intrinsic symmetry body can be detected. Considering that symmetry detection is mainly used for model processing, symmetry detection is mainly performed on the shape of a meaningful subdivision instead of searching all possible symmetric combinations. Therefore, similar structures in the skeleton are sought as shown in fig. 4 in the following steps.

(1) First, consider whether two skeletal limbs are similar, regardless of various combinations of skeletal limb sub-segments. Two skeletal limbs are considered similar if they are similar in length.

(2) Adjacent skeleton limbs and skeleton nodes are combined together, each skeleton node is taken as a root node, and the connected similar skeleton limbs and other similar tree-shaped skeleton structures are gradually associated to generate similar skeleton structures. If the convex body volume corresponding to the skeleton limb or the skeleton node is smaller, the skeleton limb or the skeleton node is combined as early as possible. As the skeleton structures gradually change from small to large, we only need to examine structures that extend from similar structures formed earlier when detecting the similarity between them. Two structures are considered similar to each other if they each have symmetrical corresponding nodes and skeletal limbs. When all structures are combined into the skeleton structure of the whole model, the search for similar skeleton structures is stopped.

The following are some experimental data of the present invention:

the experiment was performed on a PC equipped with an Intel i7-8700(3.2GHz) CPU, 16G RAM and a NVIDIA GeForce GTX 1080Ti GPU. The models used in the experiments were all from the pool of Princeton Shape Benchmark and Aim @ Shape models commonly used in the academia.

As a comparison, three representative methods are also implemented for comparison, namely a transformation space voting method (TSV), a skeleton-based point cloud method (SBP), and a multi-scale local symmetry detection Method (MPISD). It can be seen from the results in the table below that without parallel computational acceleration, the method of the present invention is many times faster than these comparative methods, even orders of magnitude faster than MPISD. At the same time, there are usually more symmetric bodies to obtain the model than these methods. Figure 5 shows the symmetry detection results obtained by these several methods for the human body model. It can be seen that none of the three methods of comparison can detect a small-scale symmetric body, and that sub-symmetric bodies embedded in a large-scale symmetric body may be neglected, and the mass of the resulting symmetric body is also poor. The method of the invention has high running speed and can detect more symmetrical bodies with high quality.

FIG. 6 shows the symmetry measurements of the present invention on multiple three-dimensional models, all with high-quality representations of the detected symmetries.

The invention has not been described in detail and is within the skill of the art.

The above description is only a part of the embodiments of the present invention, but the scope of the present invention is not limited thereto. Any changes or substitutions that may be easily made by those skilled in the art within the technical scope of the present disclosure are intended to be included within the scope of the present disclosure.

Claims (4)

1. A three-dimensional model intrinsic symmetry detection method based on skeleton guidance is characterized by comprising the following steps:

(1) generating a framework of the three-dimensional model, carrying out convex subdivision on the three-dimensional model based on the framework, and establishing a corresponding relation between subsections of framework limbs and framework nodes and convex bodies of the convex subdivision; the skeleton consists of a plurality of nodes and edges connecting the nodes, wherein the nodes shared by more than two skeleton edges are called skeleton nodes, a skeleton limb is formed from one skeleton node to another skeleton node or a string of edges of one end point of the skeleton, and the end point of the skeleton refers to the node only associated with one edge;

(2) according to the sub-segments of the skeleton limbs with similar lengths, namely the sub-segments and the skeleton nodes after the skeleton limbs are split, the intrinsic exterior symmetry between the sub-segments of the skeleton limbs and the convex bodies corresponding to the skeleton nodes is detected, if the two bodies can be superposed through rigid transformation, the two bodies are called to be symmetric exterior; two shapes are said to be intrinsically symmetric if they can be made to coincide by non-rigid changes;

(3) and searching similar structures in the framework, detecting the extrinsic symmetry between the corresponding convex bodies to which the similar structures belong, and searching the intrinsic symmetry body of the three-dimensional model based on a convex body combination mode.

2. The method for detecting the built-in symmetry of the three-dimensional model based on the skeleton guidance according to claim 1, characterized in that: the method comprises the following steps of (1) generating a framework of the model, carrying out convex dissection on the model based on the framework, and establishing a corresponding relation between subsegments and framework nodes of the framework limbs and convex bodies of the convex dissection, wherein the specific steps are as follows:

(11) generating a skeleton of the three-dimensional model according to a model mesh contraction method;

(12) carrying out initialization subdivision on the three-dimensional model according to the corresponding relation information of the skeleton limbs to the model mesh, and carrying out subdivision optimization according to the concave information between the skeleton limbs and the skeleton nodes, so that each skeleton node corresponds to one convex subdivision body, and each skeleton limb corresponds to one subdivision body;

(13) carrying out convex dissection on each part of the initialized dissection according to concave information on the three-dimensional model, corresponding the concave position condition during dissection to the skeleton limb, and carrying out sub-dissection on the skeleton limb to form a sub-section of the skeleton limb;

the concrete implementation is as follows: each skeleton limb corresponds to one split part; then, dividing the division parts corresponding to the skeleton nodes and the related skeleton limbs according to the concave information of the division parts, so that each skeleton node corresponds to one convex division body; then, performing convex subdivision on the part corresponding to the skeleton limb according to concave information of the part, and performing sub-subdivision on the skeleton limb based on the corresponding relation between the model and the skeleton to form a sub-section of the skeleton limb;

(14) for skeleton limbs with the same length, performing mutual migration treatment on the sub-situations of the skeleton limbs with the same length, so that the skeleton limbs with the same length have similar sub-situations; the migration treatment process is as follows: for two skeleton limbs with the same length, firstly, finding the local lowest inflection point of the distribution curve according to the distribution condition of the curvature of the loop line along the skeleton limbs, and then, mutually transferring the inflection points according to the linear corresponding relation of the distribution curve between the two skeleton limbs; then, according to the inflection point obtained by the migration on the skeleton limb, further sub-dividing the convex body associated with the skeleton limb to finish the corresponding migration;

(15) according to the sub-division condition of the skeleton limb, the respectively associated convex bodies are further sub-divided, so that each sub-section of the skeleton limb corresponds to one divided convex body respectively.

3. The method for detecting the built-in symmetry of the three-dimensional model based on the skeleton guidance according to claim 1, characterized in that: in the step (2), according to the sub-segments and the skeleton nodes of the similar skeleton limbs, the extrinsic symmetry between the corresponding convex bodies of the two is detected, and the specific steps are as follows:

(21) finding skeleton limb subsections with equivalent lengths, and detecting whether convex bodies corresponding to the skeleton limb subsections are symmetrical or not;

(22) and detecting whether the convex bodies corresponding to the skeleton nodes are symmetrical.

4. The method for detecting the built-in symmetry of the three-dimensional model based on the skeleton guidance according to claim 1, characterized in that: in the step (3), similar structures are searched in the framework, so that extrinsic symmetry among the corresponding convex bodies to which the structures belong is detected, and then an intrinsic symmetry body of the three-dimensional model is searched based on a convex body combination mode; the method comprises the following specific steps:

(31) finding skeleton limbs with the same length to form a similar skeleton structure;

(32) using each skeleton node as a root node, and gradually associating the connected similar skeleton limbs and other similar tree-shaped skeleton structures to generate similar skeleton structures;

(33) detecting whether the convex bodies corresponding to the similar skeleton structures are symmetrical or not based on the similar skeleton structures; if so, the model parts corresponding to the similar skeleton structures are intrinsic symmetry.

Images