A three-dimensional model retrieval method and system based on spatial distribution projection images

Abstract

The present application relates to the field of computer aided design, in particular to a kind of three-dimensional model retrieval method and system based on spatial distribution projection image;The system includes processing module and output module, processing module is used to obtain the spatial distribution projection image corresponding to each three-dimensional model by the way of projecting the three-dimensional point cloud corresponding to three-dimensional model to typical surface;Again, the descriptor of three-dimensional model is obtained by spatial distribution projection image, finally the similarity of three-dimensional model is judged by descriptor;Output module is used to output retrieval result according to similarity;The method is realized by system, based on the typical surface of three-dimensional model with rotation and scaling invariance obtains the spatial distribution projection image corresponding to part surface, then the retrieval of three-dimensional model is realized by similarity comparison, reduces the complexity of comparison, solves the problem that the description data is complex due to view direction uncertainty in the existing three-dimensional model retrieval method, improves retrieval efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of computer-aided design, in particular to a three-dimensional model retrieval method and system based on spatial distribution projection images. BACKGROUND

[0002] In recent years, with the continuous development of computer-aided design technology, a large number of three-dimensional models have been accumulated in various manufacturing industries such as aviation, aerospace, automobile, machinery and other enterprises. From the perspective of manufacturing information, three-dimensional models are widely associated with relevant information upstream and downstream of the product life cycle, including process, processing and maintenance, etc. These information contains important experience and knowledge. Therefore, how to effectively retrieve reusable three-dimensional models is an important means to improve product development efficiency.

[0003] In existing research and application, using mature two-dimensional image processing technology to realize three-dimensional shape analysis is an important technical means, and some typical three-dimensional model similarity analysis methods have appeared.

[0004] In the document "Funkhouser T, Min P, Kazhdan M, et al. A search engine for 3D models[J]. ACM Transactions on Graphics, 2003, 22(1): 83-105.", a three-dimensional model retrieval method based on view projection is proposed. The method projects three-dimensional models onto 13 views and obtains a set of two-dimensional projection views. Through distance transformation, the views are converted into gray images and discretized to a set of concentric circles. Fourier transform is used to obtain the descriptors of these views, and three-dimensional model retrieval is realized by calculating the Fourier descriptors. In order to obtain rotation and scaling invariance, the method needs to extract multiple projection views from the model, which usually requires a large amount of calculation time, resulting in that the method is not conducive to efficient retrieval in a large model library.

[0005] In the Chinese invention patent with patent publication number CN109543054A, a three-dimensional model retrieval method considering feature dimension reduction is proposed. The method constructs a feature library of three-dimensional model multi-view projection images, calculates the eigenvalues and eigenvectors of the feature library using singular value decomposition algorithm, selects representative views of each three-dimensional object to reduce the number of views, realizes the reduction of feature dimension while selecting representative views, and calculates the distance between two three-dimensional models on this basis. However, single projection image is easy to be affected by the loss of detail information caused by interference, covering and shielding inside three-dimensional model, and then affects the retrieval accuracy.

[0006] The researchers found that the existing three-dimensional model retrieval method has the problem of complex description data caused by uncertain view direction, and therefore carried out targeted research. SUMMARY

[0007] In order to solve the problem of complex description data caused by uncertain view direction in the existing three-dimensional model retrieval method, the application provides a three-dimensional model retrieval method and system based on spatial distribution projection image.

[0008] The spatial distribution projection image is abbreviated as SDPI. The spatial distribution projection image is the basis for realizing the application. The specific technical solutions of the application are described in detail below.

[0009] In a first aspect, the application discloses a three-dimensional model retrieval method based on spatial distribution projection image.

[0010] The method obtains the spatial distribution projection image corresponding to each three-dimensional model by projecting the three-dimensional point cloud corresponding to the three-dimensional model onto the typical surface; then obtains the descriptor of the three-dimensional model through the spatial distribution projection image, and finally judges the similarity of the three-dimensional model through the descriptor, so as to realize the retrieval of the three-dimensional model.

[0011] The typical surface refers to a reference surface in the three-dimensional model which is invariant to rotation and scaling.

[0012] It should be noted that the three-dimensional model retrieval method disclosed in the application comprises a three-dimensional model shape representation and similarity measurement method based on spatial distribution projection image.

[0013] In a second aspect, the application discloses a three-dimensional model retrieval system based on spatial distribution projection image, which is used to implement the method of the first aspect.

[0014] The system comprises a processing module and an output module.

[0015] The processing module is used to obtain the spatial distribution projection image corresponding to each three-dimensional model by projecting the three-dimensional point cloud corresponding to the three-dimensional model onto the typical surface; then obtain the descriptor of the three-dimensional model through the spatial distribution projection image, and finally judge the similarity of the three-dimensional model through the descriptor.

[0016] The output module is used to output the retrieval result according to the similarity.

[0017] In a third aspect, the present application discloses a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method in the first aspect.

[0018] Compared with the prior art, the present application has the following advantages and beneficial effects.

[0019] (1) The three-dimensional model retrieval method based on spatial distribution projection image disclosed by the present application can obtain a unique spatial distribution projection image SDPI for each three-dimensional model, avoiding the defect that multiple view images need to be obtained in the prior art, and further reducing the complexity of shape comparison.

[0020] (2) The three-dimensional model retrieval method based on spatial distribution projection image disclosed by the present application overcomes the occlusion and coverage phenomenon in the three-dimensional model based on the shape representation method of SDPI, and avoids the information loss in the process of converting from three-dimensional to two-dimensional.

[0021] (3) The three-dimensional model retrieval method based on spatial distribution projection image disclosed by the present application realizes the rapid retrieval feedback of the three-dimensional model based on the similarity analysis method. BRIEF DESCRIPTION OF DRAWINGS

[0022] The present application is further illustrated in combination with the following drawings and examples, and all the conceptual innovations of the present application should be regarded as the disclosed content and the protection scope of the present application.

[0023] Figure 1 It is a general flow chart of the three-dimensional model similarity analysis method of the present application.

[0024] Figure 2 It is an example of the three-dimensional model of the motor in example 4 of the present application.

[0025] Figure 3 It is the three-dimensional point cloud corresponding to the three-dimensional model of the motor in example 4 of the present application.

[0026] Figure 4 It is the SDPI corresponding to the three-dimensional model of the motor in example 4 of the present application.

[0027] Figure 5 It is part of the model data set used in example 4 of the present application.

[0028] Figure 6 It is the input when the gasket is taken as the query object in example 4 of the present application.

[0029] Figure 7 It is the calculation result when the gasket is taken as the query object in example 4 of the present application. DETAILED DESCRIPTION

[0030] Example 1

[0031] The embodiment discloses a three-dimensional model retrieval method based on a space distribution projection image, acquires the space distribution projection image corresponding to each three-dimensional model by projecting the three-dimensional point cloud corresponding to the three-dimensional model to a typical surface, acquires the descriptor of the three-dimensional model through the space distribution projection image, and finally judges the similarity of the three-dimensional model through the descriptor, so that the retrieval of the three-dimensional model is realized, wherein the typical surface refers to a reference surface with rotation and scaling invariance in the three-dimensional model.

[0032] The core content of the method includes three points: one is to acquire the typical surface with rotation and scaling invariance in the three-dimensional model; two is to acquire the space distribution projection image; and three is to retrieve the three-dimensional model according to the similarity of the SDPI.

[0033] Firstly, random sampling is performed on the surface of the three-dimensional model, the spatial shape information of the three-dimensional model is converted into a three-dimensional point cloud, the eigenvalue and eigenvector of the point cloud are calculated, and the typical surface with rotation and scaling invariance in the three-dimensional model is obtained; wherein the three-dimensional model in the direction of the typical surface presents the most detailed information.

[0034] Then, the point cloud is projected to the typical surface, and the point cloud is converted into a space distribution projection image, namely SDPI, according to the distribution density of the points on the typical surface; the image overcomes the occlusion and coverage phenomenon in the three-dimensional model.

[0035] Finally, the local shape descriptor of the SDPI is calculated by using the SIFT algorithm, the retrieval of the three-dimensional model is realized by solving the minimum distance between two local shape descriptors.

[0036] The embodiment also discloses a three-dimensional model retrieval system based on a space distribution projection image, which is used to implement the method.

[0037] The system comprises a processing module and an output module.

[0038] The processing module is used for acquiring the space distribution projection image corresponding to each three-dimensional model by projecting the three-dimensional point cloud corresponding to the three-dimensional model to a typical surface, acquiring the descriptor of the three-dimensional model through the space distribution projection image, and finally judging the similarity of the three-dimensional model through the descriptor.

[0039] The output module is used for outputting the retrieval result according to the similarity.

[0040] As Figure 1As shown, the three-dimensional model retrieval system based on spatial distribution projection image runs as follows: all three-dimensional models in the model library are traversed to obtain the descriptors corresponding to all three-dimensional models in the model library; the descriptor corresponding to the target three-dimensional model is obtained after receiving the user input target three-dimensional model; the similarity between the descriptors corresponding to the three-dimensional models in the model library and the target three-dimensional model is calculated, the similarity values are sorted from large to small, and the result is output.

[0041] Embodiment 2

[0042] This embodiment is further optimized on the basis of the above-mentioned embodiment 1.

[0043] In a specific embodiment, the acquisition method of the typical surface is as follows: according to the three-dimensional point cloud of the three-dimensional model, a 3× m matrix M is constructed, and three eigenvalues and corresponding eigenvectors of the matrix M are calculated, and then the typical surface is determined by two eigenvectors in the three eigenvectors.

[0044] In a specific embodiment, the acquisition method of the spatial distribution projection image corresponding to each three-dimensional model is as follows: the three-dimensional point cloud N is projected onto the typical surface to obtain a two-dimensional point cloud N , a score matrix k of k × k is constructed according to the image resolution H , and is initialized, and a score rule is also set for each point in the typical surface; then all points in the two-dimensional point cloud N are traversed, the score matrix H is valued according to the score rule, and the score of each element in the valued score matrix r is corrected by a perception radius H to obtain a corrected score matrix ; then the pixel value is calculated according to the corrected score, and finally the score matrix H is converted into a spatial distribution projection image.

[0045] In a specific embodiment, the acquisition method of the three-dimensional point cloud N is as follows: by randomly selecting triangular patches on the three-dimensional model, three random points are obtained, and the point cloud containing all random points is taken as the three-dimensional point cloud n of the three-dimensional model. N n > 3.

[0046] ​In a specific embodiment, obtaining the descriptor of the three-dimensional model through the spatial distribution projection image specifically refers to: after obtaining the spatial distribution projection image corresponding to each three-dimensional model, extracting the local shape descriptor from the spatial distribution projection image Descriptor as a three-dimensional model .

[0047] In one embodiment, the SIFT algorithm is used to extract local shape descriptors from spatially distributed projection images. LF .

[0048] In a specific embodiment, judging the similarity of three-dimensional models by descriptors specifically refers to: according to the local shape descriptors corresponding to each three-dimensional model, LF The local features of each 3D model are obtained, and the similarity between any two 3D models is calculated based on the number of local features and the distance between the local features; the larger the similarity value, the more similar the two 3D models are.

[0049] In a specific embodiment, the retrieval of the three-dimensional model specifically refers to: according to the local shape descriptor corresponding to each three-dimensional model, LF Calculate the similarity between the 3D models in the model library and the target 3D model, sort the 3D models in the model library from large to small according to the similarity value, and generate the retrieval results.

[0050] The other parts of this embodiment are the same as those of the above embodiment, so they will not be described in detail.

[0051] Example 3:

[0052] This embodiment describes in detail a three-dimensional model retrieval method based on spatially distributed projection images on the basis of the above embodiments, which specifically includes steps S1, S2, and S3 that are performed in sequence.

[0053] Step S1: converting the three-dimensional model into a point cloud by performing random sampling on each surface of the three-dimensional model.

[0054] Step S2: calculate the typical face direction of each 3D model based on the 3D point cloud, and convert the spatial distribution information of the 3D point cloud into SDPI.

[0055] Step S3: retrieval of the three-dimensional model using the SDPI representation method.

[0056] The step S1 specifically includes step S11, step S12, and step S13.

[0057] Step S11: randomly select a triangle on the 3D model. The probability of each triangle being selected is s / S ,in s is the area of ​​the triangle, SThe surface area of the three-dimensional model.

[0058] Step S12, the three vertices of the selected triangular facet are respectively denoted as p 1, p 2 and p 3, two random numbers a and b are generated, and a random point P is obtained from the triangular facet: p

[0059] .

[0060] Step S13, steps S11 and S12 are repeated n times to obtain n random points p , and the three-dimensional model is converted into a three-dimensional point cloud containing n random points: N

[0061] ,

[0062] wherein x x, y y and z z are the three-dimensional coordinate values of each point, and the subscript is the number of each point.

[0063] The step S2 specifically comprises steps S21, S22, S23, S24, S25, S26, S27, S28 and S29.

[0064] Step S21, a set of the first m points in the three-dimensional point cloud is obtained, 3≤ m ≤ n , and a 3× m matrix M is constructed:

[0065] ,

[0066] wherein x x, y y and z z are the three-dimensional coordinate values of each point, and the subscript is the number of each point.

[0067] Step S22, three eigenvalues M 1, d 2 and d 3 of the matrix d and corresponding eigenvectors are calculated; wherein d 1≥ d 2≥ d 3.​3.

[0068] Step S23, according to the feature vector determine the typical surface, project the three-dimensional point cloud N to the plane, get the corresponding two-dimensional point cloud N’ :

[0069] .

[0070] Step S24, set the image resolution to k , initialize k × k two-dimensional zero matrix as the score matrix H , each element of the matrix corresponds to a pixel of the image, for each point in the two-dimensional point cloud N’ ( x , y ), the score of each element is calculated by the following method h ab :

[0071]

[0072]

[0073] wherein, h ab is the score of the element in the H th row and the a th column of the score matrix b ;

[0074] min x , min y respectively represent the minimum value in the X N’ Y direction of the two-dimensional point cloud 、 ;

[0075] max x , max y respectively represent the maximum value in the X N’ Y direction of the two-dimensional point cloud 、 ;

[0076] d x , d y respectively represent the length of each image pixel X 、 Y direction.

[0077] Step S25, traverse all points in the two-dimensional point cloud N’ , according to the above score rule to score matrixH assigning values;

[0078] Step S26, setting the perception radius as r , and H correcting the score of each element in the score matrix to obtain a corrected score matrix :

[0079]

[0080] wherein, is the score of the element in the i-th row and the j-th column of the corrected score matrix , i is the score of the element in the i-th row and the j-th column of the score matrix j , is the score of the element in the i-th row and the j-th column of the score matrix H ; l m

[0081] Step S27, converting the corrected score of each element in the corrected score matrix to a pixel value of the image, and finally forming the SDPI:

[0082]

[0083] wherein, c i,j represents the pixel value of the pixel corresponding to each element in the score matrix H ;

[0084] max H , min H respectively represent the maximum value and the minimum value of all elements in the score matrix H ;

[0085] is an operator, representing the rounding function.

[0086] Step S28, extracting the local shape descriptor of the SDPI by using the SIFT algorithm, as the descriptor of the three-dimensional model.

[0087] The step S3 specifically comprises steps S31-S34.

[0088] Step S31, constructing a model library. For each three-dimensional model in the model library, repeating steps S1 and S2 to establish the local shape descriptor corresponding to each three-dimensional model.

[0089] Step S32, inputting a target three-dimensional model q to be compared by a user, and constructing the local shape descriptor of the target three-dimensional model q by using steps S1 and S2, denoted as​​LF q .

[0090] Step S33, calculate the similarity between all 3D models in the model library and the target 3D model. s The similarity between a 3D model and the target 3D model sim s for:

[0091]

[0092] in, sim s is similarity;

[0093] LF s For the s The local shape descriptor of a 3D model, 1≤ s ≤ t ; t is the number of 3D models in the model library;

[0094] | LF s | indicates the s The number of local features in the local shape descriptor of a 3D model;

[0095] lf s,o Indicates the s The local shape descriptor of the 3D model o local features;

[0096] LF q Target 3D model q Local shape descriptor of ;

[0097] | LF q | represents the target 3D model q The number of local features in the local shape descriptor;

[0098] lf q,u Represents the target 3D model q The local shape descriptor of u local features;

[0099] d (·) represents the Euclidean function that calculates the distance between two local features;

[0100] Indicated by LF s Towards LFq Calculate the directed average distance;

[0101] Indicated by LF q Towards LF s Calculate the directed average distance;

[0102] The numerical value of similarity is also called similarity measure value.

[0103] Step S34: sort the three-dimensional models in descending order according to the similarity measurement value and output them.

[0104] The higher the order of the 3D model, the higher the similarity. The output result is the retrieval result of the 3D model.

[0105] The other parts of this embodiment are the same as those of the above embodiment, so they will not be described in detail.

[0106] Example 4:

[0107] This embodiment is based on the third embodiment, and takes the three-dimensional model Prt of a motor commonly used in the machinery industry as an example, and describes steps S1, S2, and S3 in detail. Figure 2 shown.

[0108] Step S1, random sampling is performed on the surface of the motor three-dimensional model Prt, and the motor three-dimensional model Prt is converted into a 3D point cloud N ,like Figure 3 shown.

[0109] Step S2, using 3D point cloud N Calculate typical faces and transform the 3D point cloud N The spatial distribution information of the surface is converted into a spatial distribution projection map under a typical surface, SDPI.

[0110] Step S21, obtain 3D point cloud N The first 20,000 points in the set are used to construct a 3×2000 matrix M :

[0111] ;

[0112] Step S22, calculate the matrix M The three eigenvalues ​​are 2.6276×10 3 , 0.9437×10 3 , 0.9122×10 3, the eigenvectors corresponding to the three eigenvalues [0.9998, -0.0195, 0.0043], [0.0175, 0.9587, 0.2840], [-0.0096, -0.2838, 0.9588].

[0113] Step S23, projecting the three-dimensional point cloud to the plane determined by the eigenvectors [0.9998, -0.0195, 0.0043] and [0.0175, 0.9587, 0.2840] corresponding to the two largest eigenvalues of the covariance matrix of the three-dimensional point cloud, a two-dimensional point cloud is obtained. N N’ .

[0114] Step S24, setting the image resolution to 200, initializing a 200x200 score matrix H , and setting a score rule for each point in the two-dimensional point cloud N’ . x y

[0115] Step S25, traversing all points in the two-dimensional point cloud N’ , and assigning values to the score matrix H according to the score rule.

[0116] Step S26, setting the perception radius to 3, and correcting the scores of each element in the assigned score matrix H to obtain a corrected score matrix .

[0117] Step S27, calculating the gray value corresponding to each element according to the corrected score matrix , and finally converting the corrected score matrix to an SDPI. At this time, the SDPI corresponding to the motor model Prt is shown as Figure 4 .

[0118] Step S28, extracting the local shape descriptor of the SDPI as the descriptor of the three-dimensional model using the SIFT algorithm, and obtaining a total of 28 local shape descriptors .

[0119] Step S3, implementing three-dimensional model retrieval using the SDPI representation method, including steps S31, S32, S33, and S34.

[0120] Step S31, constructing a model library as shown in Figure 5 , and repeating steps S1 and S2 for each three-dimensional model in the model library to establish the descriptor corresponding to each three-dimensional model.

[0121] Step S32, using the descriptor of the three-dimensional model as a query to search the model library, and obtaining a total of 10 three-dimensional models Figure 6 ​​​The gasket model shown in "input" is the target three-dimensional model to be compared q The shape information of the target three-dimensional model is described by using step S1 and step S2, denoted as q LF q .

[0122] Step S33, the similarity of each three-dimensional model in the model library and the target three-dimensional model is calculated. q

[0123] Step S34, each three-dimensional model is sorted according to the similarity measure value from large to small and then output. The final search result with the gasket model as the target is shown as "calculation result" in Figure 7 .

[0124] The other parts of this embodiment are the same as the above-described embodiments, and thus will not be described again.

[0125] Embodiment 5:

[0126] This embodiment discloses a computer readable storage medium, which stores a computer program. The program is executed by a processor to implement the method of any one of embodiments 1 to 4.

[0127] The above is only a preferred embodiment of the present application, and does not limit the present application in any form. Although the embodiments of the present application have been shown and described above, it should be understood that the above-described embodiments are exemplary and cannot be understood as limiting the present application. Those skilled in the art can make changes, modifications, replacements and variations to the above-described embodiments without departing from the principles and purposes of the present application within the scope of the present application.​​

Claims

1. A three-dimensional model retrieval method based on spatially distributed projection images, characterized in that: By projecting the 3D point cloud corresponding to the 3D model onto a typical surface, the spatial distribution projection image corresponding to each 3D model is obtained. The descriptors of the 3D model are then obtained from the spatial distribution projection image. Finally, the similarity of the 3D models is judged by the descriptors, thereby achieving 3D model retrieval. Among them, the typical surface refers to the reference surface in the 3D model that is invariant to rotation and scaling. It includes: Step S1, converting the 3D model into a point cloud by randomly sampling each surface of the 3D model; Step S2, calculating the typical face direction of each 3D model based on the 3D point cloud, and converting the spatial distribution information of the 3D point cloud into SDPI; Step S3, using the SDPI representation method to retrieve the three-dimensional model; The step S2 specifically includes: Step S21, obtaining a set of first m points in the three-dimensional point cloud, 3≤m≤n, and constructing a 3×m matrix M; ； Among them, x, y and z are the three-dimensional coordinate values ​​of each point, and the subscript is the number of each point; Step S22, calculate the three eigenvalues ​​d1, d2, d3 of the matrix M and the corresponding eigenvectors ; Among them, d1≥d2≥d3; Step S23, according to the feature vector Determine the typical surface, project the 3D point cloud N onto the plane, and obtain the corresponding 2D point cloud N'; ； Step S24: Set the image resolution to k, initialize a k×k two-dimensional zero matrix as the score matrix H, where each element of the matrix corresponds to a pixel of the image. For each point (x, y) in the two-dimensional point cloud N', calculate the score h of each element using the following method: ab ; ； Among them, h ab is the score of the element in row a and column b in the score matrix H; min x 、min y Respectively represent the minimum value in the X and Y directions of the two-dimensional point cloud N'; max x 、max y Respectively represent the maximum values ​​in the X and Y directions of the two-dimensional point cloud N'; d x d y Respectively represent the length of each image pixel in the X and Y directions; Step S25, traverse all points in the two-dimensional point cloud N' and assign values ​​to the score matrix H according to the above-mentioned scoring rules; Step S26, setting the perception radius to r, and correcting the score of each element in H to obtain a corrected score matrix H'; ； in, To correct the score of the element in row i and column j in the score matrix H', is the score of the lth row and mth column in the score matrix H; Step S27, converting the corrected score of each element in the corrected score matrix H' into the pixel value of the image, and finally forming the SDPI: ； Among them, c i,j Indicates the pixel value of each element corresponding to the pixel in the score matrix H; max H 、min H Respectively represent the maximum and minimum values ​​of all elements in the score matrix H; Is an operator, indicating the rounding function; Step S28, using the SIFT algorithm to extract the local shape descriptor of the SDPI as the descriptor of the three-dimensional model; The step S3 specifically includes: Step S31, building a model library, repeating steps S1 and S2 for each 3D model in the model library to establish a local shape descriptor corresponding to each 3D model; In step S32, the user inputs the target 3D model q to be compared, and uses steps S1 and S2 to construct a local shape descriptor for the target 3D model q, which is recorded as LF q ; Step S33, calculating the similarity between all three-dimensional models in the model library and the target three-dimensional model, the similarity between the s-th three-dimensional model in the model library and the target three-dimensional model is sim s for: ； Among them, sim s is similarity; LF s is the local shape descriptor of the sth 3D model, 1≤s≤t; t is the number of 3D models in the model library; |LF s | represents the number of local features in the local shape descriptor of the s-th 3D model; lf s,o LF represents the oth local feature in the local shape descriptor of the sth 3D model; q is the local shape descriptor of the target 3D model q; |LF q | represents the number of local features in the local shape descriptor of the target 3D model q; lf q,u represents the u-th local feature in the local shape descriptor of the target 3D model q; d(·) represents the Euclidean function for calculating the distance between two local features; Indicates that LF s Xiang LF q Calculate the directed average distance; Indicates that LF q Xiang LF s The calculated directed average distance; the numerical value of similarity is also called the similarity measure value; Step S34: sort the three-dimensional models in descending order according to the similarity measurement value and output them.

2. The three-dimensional model retrieval method based on spatially distributed projection images according to claim 1, characterized in that: The typical surface is obtained by constructing a 3×m matrix M based on the 3D point cloud of the 3D model, calculating three eigenvalues ​​and corresponding eigenvectors of the matrix M, and then determining the typical surface through two of the three eigenvectors.

3. The three-dimensional model retrieval method based on spatially distributed projection images according to claim 1, characterized in that: The method of obtaining the spatial distribution projection image corresponding to each three-dimensional model is as follows: projecting the three-dimensional point cloud N onto the typical surface to obtain the two-dimensional point cloud N', constructing and initializing the k×k score matrix H according to the image resolution k, and setting the scoring rule for each point in the typical surface; then traversing all points in the two-dimensional point cloud N', assigning values ​​to the score matrix H according to the scoring rule, and correcting the score of each element in the assigned score matrix H by the perception radius r; then calculating the pixel value according to the score, and finally converting the score matrix H into a spatial distribution projection image.

4. The three-dimensional model retrieval method based on spatially distributed projection images according to claim 3, characterized in that: The three-dimensional point cloud N is obtained by randomly selecting triangular facets on the three-dimensional model to obtain n random points, and using the point cloud containing all the random points as the three-dimensional point cloud N of the three-dimensional model; n>3.

5. The three-dimensional model retrieval method based on spatially distributed projection images according to claim 1, characterized in that: Acquiring the descriptor of the three-dimensional model through the spatially distributed projection image specifically means: after acquiring the spatially distributed projection image corresponding to each three-dimensional model, extracting the local shape descriptor from the spatially distributed projection image as the descriptor of the three-dimensional model.

6. The three-dimensional model retrieval method based on spatially distributed projection images according to claim 5, characterized in that: The local shape descriptor LF is extracted from the spatially distributed projection image using the SIFT algorithm.

7. The three-dimensional model retrieval method based on spatially distributed projection images according to claim 1, characterized in that: Determining the similarity of three-dimensional models through descriptors specifically means: obtaining the local features of each three-dimensional model based on the local shape descriptor LF corresponding to each three-dimensional model, and calculating the similarity of any two three-dimensional models by the number of local features and the distance between the local features; the larger the similarity value, the more similar the two three-dimensional models are.

8. The three-dimensional model retrieval method based on spatially distributed projection images according to claim 7, characterized in that: The retrieval of three-dimensional models specifically refers to: calculating the similarity between the three-dimensional models in the model library and the target three-dimensional model based on the local shape descriptors LF corresponding to each three-dimensional model, sorting the three-dimensional models in the model library from large to small according to the similarity values, and generating retrieval results.

9. A three-dimensional model retrieval method based on spatially distributed projection images according to claim 7 or 8, characterized in that: The distance between local features is calculated using the Euclidean function.

10. A three-dimensional model retrieval system based on spatially distributed projection images, used to implement the three-dimensional model retrieval method based on spatially distributed projection images according to any one of claims 1 to 9, characterized in that: The system includes a processing module and an output module; the processing module is used to obtain spatially distributed projection images corresponding to each three-dimensional model by projecting the three-dimensional point cloud corresponding to the three-dimensional model onto a typical surface; then obtain descriptors of the three-dimensional model through the spatially distributed projection images, and finally judge the similarity of the three-dimensional models through the descriptors; the output module is used to output retrieval results based on the similarity.

Images