A system and method for automatically reviewing virtual characters in the metaverse

Through the automatic review system, the review rule database and feature matching technology are used to solve the problems of low efficiency and high risk of misjudgment of virtual role review in the metaverse, and accurate and efficient virtual role review is achieved.

CN118193777BActive Publication Date: 2024-11-01SHENZHEN INHERENT COLOR DIGITAL TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410392955.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-02
Publication Date
2024-11-01
Estimated Expiration
2044-04-02

AI Technical Summary

Technical Problem

The existing problems of low efficiency, easy omission and high risk of misjudgment in the metaverse through manual review.

Method used

Design an automatic review system. By building a review rule database, importing a virtual role to be reviewed, traversing each rule in the database to determine the model location, generating a target picture, and matching it with the content of the review rule to determine whether it is Meet the requirements.

Benefits of technology

It realizes accurate, efficient and comprehensive review of virtual characters in the metaverse, reduces the risk of misjudgment and improves the review efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118193777B_ABST
    Figure CN118193777B_ABST
Patent Text Reader

Abstract

It is based on the above problems that the present invention proposes a system and method for automatically reviewing virtual characters in the metaverse. The system and method construct a review rule database, import the model to be reviewed of the virtual character, and traverse each review rule in the review rule database to perform the following steps: determine the model part corresponding to the review rule content, the model part includes one or more local parts of the virtual character model, generate a target image of the model to be reviewed corresponding to the model part, perform feature matching between the target image and the review rule content to obtain the matching degree between the target image and the review rule content, the review rule content is the text content or graphic content of the corresponding review rule, and judge whether the model to be reviewed meets the requirements of the corresponding review rule based on the matching result, so as to realize accurate and efficient all-round review of the model of the virtual character in the metaverse.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of metaverse technology, and in particular to a system and method for automatically reviewing virtual characters in the metaverse. Background Art

[0002] The Metaverse is a virtual, digital world built with virtual reality technology. People can use virtual character models to represent themselves in the Metaverse to explore the virtual environment and interact with other virtual characters in real time. The Metaverse is widely used in social entertainment, online office, online education, online shopping, and telemedicine. Its rise and development has greatly enriched the way and content of people's social, entertainment, life, and work. In order to cater to people's diverse preferences and needs, virtual characters in the Metaverse can be designed as very diverse three-dimensional images. In order to provide users with a civilized and orderly Metaverse environment and ensure the safe and stable operation of the Metaverse environment, it is necessary to conduct a comprehensive review of virtual characters in the Metaverse in accordance with relevant laws and regulations. However, the number of virtual characters in the Metaverse is extremely large, and there are various regulations on the requirements for virtual character images. The existing manual review method has low efficiency, is easy to miss, and has a high risk of misjudgment. Summary of the invention

[0003] Based on the above problems, the present invention proposes a system and method for automatically reviewing virtual characters in the Metaverse, which can realize accurate and efficient all-round review of the models of virtual characters in the Metaverse.

[0004] In view of this, a first aspect of the present invention proposes a system for automatically reviewing virtual characters in a metaverse, comprising a review rule database for storing review rule content and a review server connected to the review rule database for reviewing virtual character models in the metaverse according to the review rule content, wherein the review rule database comprises a text rule library for storing text content in the review rule content and a picture material library for storing picture materials in the review rule content, and the review server is configured to:

[0005] Build a review rules database;

[0006] Import the model of the virtual character to be reviewed;

[0007] Traverse each review rule in the review rule database to perform the following steps:

[0008] Determining a model part corresponding to the content of the review rule, wherein the model part includes one or more local parts of the virtual character model;

[0009] Generate a target image of the model to be reviewed corresponding to the model part;

[0010] Performing feature matching between the target image and the review rule content to obtain a matching degree between the target image and the review rule content, wherein the review rule content is a text content or graphic content corresponding to the review rule;

[0011] Determine whether the model to be reviewed meets the requirements of the corresponding review rules based on the matching results.

[0012] A second aspect of the present invention provides a method for automatically reviewing a virtual character in a metaverse, comprising:

[0013] Constructing a review rule database, wherein the review rule database includes a text rule library and a picture material library;

[0014] Import the model of the virtual character to be reviewed;

[0015] Traverse each review rule in the review rule database to perform the following steps:

[0016] Determining a model part corresponding to the content of the review rule, wherein the model part includes one or more local parts of the virtual character model;

[0017] Generate a target image of the model to be reviewed corresponding to the model part;

[0018] Performing feature matching between the target image and the review rule content to obtain a matching degree between the target image and the review rule content, wherein the review rule content is a text content or graphic content corresponding to the review rule;

[0019] Determine whether the model to be reviewed meets the requirements of the corresponding review rules based on the matching results.

[0020] Preferably, the step of generating a target image of the model to be reviewed corresponding to the model part specifically includes:

[0021] Generate a local model of the model to be reviewed corresponding to the model part;

[0022] Determine a first target image generation viewing angle of the local model, where the first target image generation viewing angle is a viewing angle at which the local model has a maximum visible area;

[0023] Rotating the local model to a posture corresponding to the viewing angle of the first target image;

[0024] Generate a projection image of the local model along the sight direction of an observer outside the screen at the first target image generation viewing angle;

[0025] The projection image is determined as a target image of the model to be reviewed corresponding to the model part.

[0026] Preferably, the step of determining the first target image generation viewing angle of the local model specifically includes:

[0027] Determine a reference pose of the local model, wherein the reference pose is the pose of the local model corresponding to the front view of the model to be examined when the local model is located on the model to be examined in a natural state;

[0028] Construct a three-dimensional space coordinate system;

[0029] Get the pre-configured n sets of rotation angles Where i is a positive integer between 1 and n, ω i is the azimuth angle in the i-th group of rotation angles, is the altitude angle in the i-th group of rotation angles;

[0030] Based on the rotation angle Construct n first target planes in the three-dimensional space coordinate system, and the azimuth angle of the plane normal of the i-th first target plane in the three-dimensional space coordinate system is ω i , the altitude angle is

[0031] Determine the azimuth angle of n plane normals as ω j , the altitude angle is The second target plane, Within the angle range, the local model has a maximum projection area on the second target plane, where j is a positive integer between 1 and n;

[0032] Determine a third target plane having the largest projection area among the n second target planes;

[0033] The plane normal direction of the third target plane is determined as the first target image generation viewing angle.

[0034] Preferably, the azimuth angles of n plane normals are determined as ω j , the altitude angle is The steps of obtaining the second target plane specifically include:

[0035] Configure the rotation step size Δθ used to perform the rotation iterations, where

[0036] The altitude angle and azimuth angle of the plane normal constructed for the i-th first target plane are The four adjacent planes of

[0037] Calculate the projection area s of the local model onto the first target plane i0 and the projection areas s projected onto the adjacent planesi1 、s i2 、s i3 、s i4 ;

[0038] When i0 Not satisfied i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ), the adjacent plane with the largest projection area is determined as the first target plane;

[0039] Repeat the above steps to construct the plane normal for the i-th first target plane. The altitude angle and azimuth angle are The four adjacent planes of the local model are projected onto the first target plane by calculating the projection area s i0 and the projection areas s projected onto the adjacent planes i1 、s i2 、s i3 、s i4 The steps, and when s i0 Not satisfied i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ), the step of determining the adjacent plane with the largest projection area as the first target plane until s i0 Satisfy si0≥max(s i1 ,s i2 ,s i3 ,s i4 );

[0040] When i0 Satisfaction i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ), the first target plane is determined as the second target plane.

[0041] Preferably, the step of performing feature matching between the target image and the review rule content to obtain the matching degree between the target image and the review rule content specifically includes:

[0042] Inputting the target image into a pre-trained image description model to generate a first description content of the target image, wherein the first description content is a text content describing the image, type, state and / or action of each object contained in the target image;

[0043] When the review rule content is text content, feature matching is performed between the first description content of the target image and the text content in the review rule content;

[0044] Obtaining a degree of matching between the target image and the review rule content based on a feature matching between the first description content of the target image and the text content in the review rule content;

[0045] When the review rule content is graphic content, feature matching is performed on the first description content of the target image with the text content in the review rule content, and feature matching is performed on the target image with the image material in the review rule content;

[0046] The degree of matching between the target image and the review rule content is obtained based on the feature matching between the first description content of the target image and the text content in the review rule content, and the feature matching between the target image and the image material in the review rule content.

[0047] Preferably, the step of performing feature matching between the first description content of the target image and the text content in the review rule content specifically includes:

[0048] Generate a first text vector corresponding to the first description content of the target image;

[0049] Generate or obtain a second text vector corresponding to the text content in the review rule content;

[0050] The distance between the first text vector and the second text vector in the vector space is calculated.

[0051] Preferably, the step of generating a first text vector corresponding to the first description content of the target image specifically includes:

[0052] Performing sentence segmentation on the first description content of the target image to obtain a plurality of first description sentences;

[0053] Generate a first text vector corresponding to each first description sentence;

[0054] The step of calculating the distance between the first text vector and the second text vector in the vector space specifically includes:

[0055] Calculate the distance d between the first text vector corresponding to each first description sentence and the second text vector k1 , where k1 is a positive integer between 1 and l1, and l1 is the number of first description sentences obtained by segmenting the first description content of the target image;

[0056] Calculate a first average distance between the first text vector and the second text vector:

[0057]

[0058] The step of obtaining the matching degree between the target image and the review rule content based on the feature matching between the first description content of the target image and the text content in the review rule content specifically includes:

[0059] The first average distance The degree of match between the target image and the content of the review rule is determined.

[0060] Preferably, the step of performing feature matching between the target image and the image material in the review rule content specifically includes:

[0061] Inputting the picture material in the review rule content into a pre-trained picture description model to generate a second description content of the picture material;

[0062] Segmenting the second description content of the picture material to obtain a plurality of second description sentences;

[0063] Generate a third text vector corresponding to each second description sentence;

[0064] Calculate the distance d between the first text vector corresponding to each first description sentence and the third text vector corresponding to each second description sentence k1,k2 , where k2 is a positive integer between 1 and l2, and l2 is the number of second description sentences obtained by segmenting the second description content of the picture material;

[0065] Calculate a second average distance between the first text vector and the third text vector:

[0066]

[0067] Preferably, the step of obtaining the degree of matching between the target image and the review rule content based on the feature matching between the first description content of the target image and the text content in the review rule content, and the feature matching between the target image and the image material in the review rule content specifically includes:

[0068] The first average distance The second average distance between the first text vector and the third text vector sum The degree of match between the target image and the content of the review rule is determined.

[0069] It is based on the above problems that the present invention proposes a system and method for automatically reviewing virtual characters in the metaverse. The system and method construct a review rule database, import the model to be reviewed of the virtual character, and traverse each review rule in the review rule database to perform the following steps: determine the model part corresponding to the review rule content, the model part includes one or more local parts of the virtual character model, generate a target image of the model to be reviewed corresponding to the model part, perform feature matching between the target image and the review rule content to obtain the matching degree between the target image and the review rule content, the review rule content is the text content or graphic content of the corresponding review rule, and judge whether the model to be reviewed meets the requirements of the corresponding review rule based on the matching result, so as to realize accurate and efficient all-round review of the model of the virtual character in the metaverse. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 is a schematic diagram of a system for automatically reviewing virtual characters of a metaverse provided by an embodiment of the present invention;

[0071] Figure 2 It is a flowchart of a method for automatically reviewing virtual characters in the metaverse provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0072] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0073] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0074] In the description of the present invention, the term "multiple" refers to two or more. Unless otherwise clearly defined, the orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention. The terms "connection", "installation", "fixation", etc. should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. In addition, the terms "first", "second", etc. are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first", "second", etc. can explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, the meaning of "multiple" is two or more.

[0075] In the description of this specification, the description of the terms "one embodiment", "some implementations", "specific embodiments", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0076] The following describes a system and method for automatically reviewing virtual characters in the metaverse according to some embodiments of the present invention with reference to the accompanying drawings.

[0077] like Figure 1 As shown, the first aspect of the present invention proposes a system for automatically reviewing virtual characters in the metaverse, including a review rule database for storing review rule content and a review server connected to the review rule database for reviewing virtual character models in the metaverse according to the review rule content, wherein the review rule database includes a text rule library for storing text content in the review rule content and a picture material library for storing picture materials in the review rule content.

[0078] like Figure 2 As shown, the review server is configured as follows:

[0079] Build a review rules database;

[0080] Import the model of the virtual character to be reviewed;

[0081] Traverse each review rule in the review rule database to perform the following steps:

[0082] Determining a model part corresponding to the content of the review rule, wherein the model part includes one or more local parts of the virtual character model;

[0083] Generate a target image of the model to be reviewed corresponding to the model part;

[0084] Performing feature matching between the target image and the review rule content to obtain a matching degree between the target image and the review rule content, wherein the review rule content is a text content or graphic content corresponding to the review rule;

[0085] Determine whether the model to be reviewed meets the requirements of the corresponding review rules based on the matching results.

[0086] Specifically, in the step of constructing a review rule database, the review rules for virtual characters in the metaverse can be formulated based on the review experience of the reviewers, or can be generated based on the recognition of the review specification text. The review rule content in the review rule database is presented in the form of text content or graphic content. When the review rule content is text content, it is stored in the text rule library of the review rule database; when the review rule content is graphic content, the review rule content includes associated stored text content and image materials, and the text content part is stored in the text rule library of the review rule database, and the image material part is stored in the image material library of the review rule database. In the technical scheme of the above-mentioned implementation method, the model to be reviewed of each virtual character is sequentially imported and matched with the content of each review rule in the review rule data to determine whether the model to be reviewed of the virtual character meets the requirements of the corresponding review rule.

[0087] For any review rule, when the review rule content is to limit a specific model part, the model part it limits will usually be reflected in its text content and / or image material. The same review rule can limit multiple model parts at the same time. When the review rule is not to limit a specific model part, it limits the virtual character model as a whole. In the step of determining the model part corresponding to the review rule content, one or more model parts corresponding to the review rule content can be entered at the same time as the review rule content; it is also possible to extract the corresponding model part by performing semantic recognition on the corresponding text content, or extract the corresponding model part by performing image recognition on the corresponding image material.

[0088] In the technical solutions of some embodiments of the present invention, before the step of generating a target image of the model to be reviewed corresponding to the model part, the review server is configured to:

[0089] Determine whether the review rule content has a corresponding model part, for example, determine whether the model part field corresponding to the review rule content in the review rule database is a null value. When the model part field corresponding to the review rule content is a null value, it is determined that the review rule content does not have a corresponding model part, otherwise it is determined that the review rule content has a corresponding model part;

[0090] When the review rule content has a corresponding model part, the target image includes one or more partial images of the model to be reviewed corresponding to the model part;

[0091] When there is no corresponding model part in the review rule content, the target image is a picture of the model to be reviewed that covers the entire model to be reviewed at a specific viewing angle.

[0092] Preferably, in the step of generating a target image of the model to be reviewed corresponding to the model part, the review server is configured to:

[0093] Generate a local model of the model to be reviewed corresponding to the model part;

[0094] Determine a first target image generation viewing angle of the local model, where the first target image generation viewing angle is a viewing angle at which the local model has a maximum visible area;

[0095] Rotating the local model to a posture corresponding to the viewing angle of the first target image;

[0096] Generate a projection image of the local model along the sight direction of an observer outside the screen at the first target image generation viewing angle;

[0097] The projection image is determined as a target image of the model to be reviewed corresponding to the model part.

[0098] Specifically, the step of generating a local model of the model to be reviewed corresponding to the model part is to extract the local model corresponding to the model part from the model to be reviewed as an independent model, and the local model does not include parts of the model to be reviewed other than the model part.

[0099] The first target image generation viewing angle is the angle of the local model's posture relative to the line of sight of the observer outside the screen when generating the target image, and the direction perpendicular to the screen is usually taken as the line of sight of the observer outside the screen. In the technical solution of the above-mentioned implementation, the step of determining the first target image generation viewing angle of the local model is specifically to determine the specific posture of the local model relative to the line of sight of the observer outside the screen, so that the local model has the largest visible area in the line of sight of the observer outside the screen.

[0100] In the step of generating a projection image of the local model along the sight direction of an observer outside the screen under the first target image generation viewing angle, the review server is configured to:

[0101] Determine the projection surface of the local model by a plane perpendicular to the sight line direction of the observer outside the screen;

[0102] The projection image is obtained by projecting the visible surface of the local model onto the projection surface when the local model is in a position corresponding to the viewing angle generated by the first target image. The visible surface of the local model is the surface that can be observed by the observer outside the screen when the local model is in a position corresponding to the viewing angle generated by the first target image.

[0103] In the technical solutions of some embodiments of the present invention, after the step of determining the model part corresponding to the review rule content, the review server is configured to:

[0104] When the review rule content does not have a corresponding model part, determining a second target image generation perspective of the model to be reviewed, the second target image generation perspective being a frontal perspective of the model to be reviewed;

[0105] Rotating the model to be reviewed to a posture corresponding to the viewing angle generated by the second target image;

[0106] Generate a projection image of the model to be reviewed along the line of sight of an observer outside the screen at the second target image generation viewing angle;

[0107] The projection image is determined as the target image of the model to be reviewed.

[0108] Preferably, in the step of determining the first target image generation viewing angle of the local model, the review server is configured to:

[0109] Determine a reference pose of the local model, wherein the reference pose is the pose of the local model corresponding to the front view of the model to be examined when the local model is located on the model to be examined in a natural state;

[0110] Construct a three-dimensional space coordinate system;

[0111] Get the pre-configured n sets of rotation angles Where i is a positive integer between 1 and n, ω i is the azimuth angle in the i-th group of rotation angles, is the altitude angle in the i-th group of rotation angles;

[0112] Based on the rotation angle Construct n first target planes in the three-dimensional space coordinate system, and the azimuth angle of the plane normal of the i-th first target plane in the three-dimensional space coordinate system is ω i , the altitude angle is

[0113] Determine the azimuth angle of n plane normals as ω j , the altitude angle is The second target plane, Within the angle range, the local model has a maximum projection area on the second target plane, where j is a positive integer between 1 and n;

[0114] Determine a third target plane having the largest projection area among the n second target planes;

[0115] The plane normal direction of the third target plane is determined as the first target image generation viewing angle.

[0116] Specifically, the pose of a stereo model represents the position and posture of the stereo model, which includes the position of the stereo model, the relative position relationship between the various components, and the rotation state of the stereo model, wherein the rotation state of the stereo model is determined relative to the line of sight of the observer outside the screen, and is usually represented by the rotation angle of the stereo model. The rotation angle of the stereo model is a relative value, so it is necessary to define a reference pose, and use the reference angle corresponding to the reference pose as a reference value to evaluate the size of the rotation angle of the stereo model after rotation.

[0117] In the technical solution of the above-mentioned embodiment, the position of the local model that satisfies the following two conditions is determined as the reference position of the local model: 1) the local model is located on the model to be reviewed in a natural state; 2) the rotational perspective of the model to be reviewed is the frontal perspective of the model to be reviewed. The local model is located on the model to be reviewed in a natural state, which means that the limbs of the virtual character are in a natural and relaxed state in an upright posture. For a three-dimensional model of a virtual character, especially a three-dimensional model of a humanoid virtual character, its frontal perspective refers to the rotational perspective in which the front of the three-dimensional model of the virtual character faces the screen, that is, the direction opposite to the line of sight of the observer outside the screen.

[0118] In the technical solution of the above-mentioned implementation, the three-dimensional space coordinate system refers to a spatial rectangular coordinate system, and the three coordinate axes constituting the three-dimensional space coordinate system are mutually orthogonal. Exemplarily, in the step of constructing the three-dimensional space coordinate system, the vertical direction parallel to the plane where the screen is located is determined as the first coordinate axis direction of the three-dimensional space coordinate system, the horizontal direction parallel to the plane where the screen is located is determined as the second coordinate axis direction of the three-dimensional space coordinate system, and the direction perpendicular to the screen is determined as the third coordinate axis direction of the three-dimensional space coordinate system.

[0119] In the technical solution of the above-mentioned implementation mode, Indicates that the three-dimensional space coordinate system is rotated by The plane normal is the spatial range of the central axis, and the rotation angle is The second target plane corresponding to the plane normal of is the plane with the largest projection area of ​​the local model within the space range, and the rotation angle and rotation angle There is a one-to-one correspondence.

[0120] In the technical solutions of the above embodiments, the value of n is positively correlated with the complexity of the surface structure of the local model, that is, the more complex the surface structure of the local model is, the larger the value of n is configured. In the technical solutions of some embodiments of the present invention, the azimuth of the n plane normals is determined as ω j , the altitude angle is The steps of determining the second target plane can be processed synchronously by constructing n processing threads to increase the speed of determining the third target plane.

[0121] Preferably, when determining the azimuth angles of n plane normals as ω j , the altitude angle is In the steps of the second target plane, the review server is configured to:

[0122] Configure the rotation step size Δθ used to perform the rotation iterations, where

[0123] The altitude angle and azimuth angle of the plane normal constructed for the i-th first target plane are The four adjacent planes of

[0124] Calculate the projection area s of the local model onto the first target plane i0 and the projection areas s projected onto the adjacent planes i1 、s i2 、s i3 、s i4 ;

[0125] Wheni0 Not satisfied i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ), the adjacent plane with the largest projection area is determined as the first target plane;

[0126] Repeat the above steps to construct the plane normal for the i-th first target plane. The altitude angle and azimuth angle are The four adjacent planes of the local model are projected onto the first target plane by calculating the projection area s i0 and the projection areas s projected onto the adjacent planes i1 、s i2 、s i3 、s i4 The steps, and when s i0 Not satisfied i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ), the step of determining the adjacent plane with the largest projection area as the first target plane until s i0 Satisfaction i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 );

[0127] When i0 Satisfaction i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ), the first target plane is determined as the second target plane.

[0128] In the technical solution of the above embodiment, the rotation step Δθ is the step size of the azimuth angle and the altitude angle in each rotation when performing the rotation iteration. The rotation step Δθ should be much smaller than So that the handler can be rotated at The spatial range in which the plane normal of the central axis is the center axis has a sufficient number of iteration steps, so that the second target plane with the largest projection area in the spatial range can be more accurately located. Since the value of n is related to the complexity of the surface structure of the local model, the rotation step Δθ also needs to be dynamically determined according to the complexity of the surface structure of the local model. Specifically, after determining the value of n, according to The size of Δθ determines the rotation step length.

[0129] Preferably, in the step of configuring a rotation step size Δθ for performing rotation iterations, the review server is configured to:

[0130] Get the pre-configured minimum rotation step size Δθ0 and the default number of iteration steps m;

[0131] when When the rotation step length Δθ is configured as

[0132] when , the rotation step Δθ is configured as the minimum rotation step Δθ0.

[0133] Preferably, in the step of performing feature matching between the target image and the review rule content to obtain a matching degree between the target image and the review rule content, the review server is configured as follows:

[0134] Inputting the target image into a pre-trained image description model to generate a first description content of the target image, wherein the first description content is a text content describing the image, type, state and / or action of each object contained in the target image;

[0135] When the review rule content is text content, feature matching is performed between the first description content of the target image and the text content in the review rule content;

[0136] Obtaining a degree of matching between the target image and the review rule content based on a feature matching between the first description content of the target image and the text content in the review rule content;

[0137] When the review rule content is graphic content, feature matching is performed on the first description content of the target image with the text content in the review rule content, and feature matching is performed on the target image with the image material in the review rule content;

[0138] The degree of matching between the target image and the review rule content is obtained based on the feature matching between the first description content of the target image and the text content in the review rule content, and the feature matching between the target image and the image material in the review rule content.

[0139] Specifically, the image description model is a deep visual language model trained using a supervised training model based on a large number of images and their corresponding first description contents. It can recognize the image, type, state, action, etc. of various objects in the input image, and output the first description content of the corresponding image using natural language description.

[0140] In the technical solution of the above-mentioned implementation, when the review rule content is text content, the first description content of the target image is used to perform feature matching with the text content in the review rule content, and the feature matching result between the target image and the review rule content is determined based on the text feature matching between the two file contents. When the review rule content is graphic content, it is necessary to perform feature matching on the first description content of the target image and the text content in the review rule content at the same time, and to perform feature matching on the target image and the picture material in the review rule, and to determine the feature matching result between the target image and the review rule content based on the two feature matching situations.

[0141] Preferably, in the step of performing feature matching between the first description content of the target image and the text content in the review rule content, the review server is configured to:

[0142] Generate a first text vector corresponding to the first description content of the target image;

[0143] Generate or obtain a second text vector corresponding to the text content in the review rule content;

[0144] The distance between the first text vector and the second text vector in the vector space is calculated.

[0145] In the technical scheme of the above-mentioned implementation mode, the step of obtaining the matching degree between the target image and the review rule content based on the feature matching between the first description content of the target image and the text content in the review rule content is specifically to determine the distance between the first text vector and the second text vector in the vector space as the matching degree between the first description content of the target image and the text content in the review rule content.

[0146] In the technical solutions of some embodiments of the present invention, each time the step of feature matching the first description content of the target image with the text content in the review rule content is performed, a second text vector corresponding to the text content in the review rule content is generated in real time. In the technical solutions of other embodiments of the present invention, the review rule database also includes a text content vector library, and in the step of constructing the review rule database, the review server is configured as follows:

[0147] Vectorizing the text content in the review rule content to generate a second text vector corresponding to the text content in the review rule content;

[0148] The second text vector is written into the text content vector library.

[0149] In the step of calculating the distance between the first text vector and the second text vector in the vector space, the distance between the first text vector and the second text vector in the vector space may be the Euclidean distance or the Manhattan distance between the two vectors.

[0150] Preferably, in the step of generating a first text vector corresponding to the first description content of the target image, the review server is configured to:

[0151] Performing sentence segmentation on the first description content of the target image to obtain a plurality of first description sentences;

[0152] Generate a first text vector corresponding to each first description sentence;

[0153] The step of calculating the distance between the first text vector and the second text vector in the vector space specifically includes:

[0154] Calculate the distance d between the first text vector corresponding to each first description sentence and the second text vector k1 , where k1 is a positive integer between 1 and l1, and l1 is the number of first description sentences obtained by segmenting the first description content of the target image;

[0155] Calculate a first average distance between the first text vector and the second text vector:

[0156]

[0157] The step of obtaining the matching degree between the target image and the review rule content based on the feature matching between the first description content of the target image and the text content in the review rule content specifically includes:

[0158] The first average distance The degree of match between the target image and the content of the review rule is determined.

[0159] Specifically, the first description content of the target image is usually a paragraph consisting of multiple sentences or a text block consisting of multiple paragraphs, and the text content of the review rule is a rule statement that limits a certain specific situation of the three-dimensional model of the virtual character as a whole or in part, so the text content of the review rule is usually a short sentence. In the technical scheme of the above-mentioned implementation mode, the first description content of the target image is first divided into multiple first description sentences through sentence segmentation processing, and then these first description sentences are vectorized to obtain multiple first text vectors. After calculating the distance between each first text vector and the second text vector respectively, the matching degree between the target image and the content of the review rule is obtained based on the average value of the distance between these first text vectors and the second text vector.

[0160] Preferably, in the step of performing feature matching between the target image and the image material in the review rule content, the review server is configured to:

[0161] Inputting the picture material in the review rule content into a pre-trained picture description model to generate a second description content of the picture material;

[0162] Segmenting the second description content of the picture material to obtain a plurality of second description sentences;

[0163] Generate a third text vector corresponding to each second description sentence;

[0164] Calculate the distance d between the first text vector corresponding to each first description sentence and the third text vector corresponding to each second description sentence k1,k2 , where k2 is a positive integer between 1 and l2, and l2 is the number of second description sentences obtained by segmenting the second description content of the picture material;

[0165] Calculate a second average distance between the first text vector and the third text vector:

[0166]

[0167] In the technical solution of the above implementation, the picture description model is used to generate a second description content corresponding to the picture material, and then the second description content is feature matched with the first description content to achieve feature matching between the target picture and the picture material.

[0168] Preferably, in the step of obtaining the degree of matching between the target image and the review rule content based on the feature matching between the first description content of the target image and the text content in the review rule content, and the feature matching between the target image and the image material in the review rule content, the review server is configured as follows:

[0169] The first average distance The second average distance between the first text vector and the third text vector sum The degree of match between the target image and the content of the review rule is determined.

[0170] In the technical solutions of other embodiments of the present invention, the target image and the image material may be directly matched by graphic features. In these embodiments, in the step of matching the target image with the image material in the review rule content, the review server is configured as follows:

[0171] Using a feature extraction algorithm to extract features from the target image and the image material in the review rule content, to obtain one or more first feature descriptors of the target image and one or more second feature descriptors of the image material in the review rule content;

[0172] Using a nearest neighbor matching algorithm to match the first feature descriptor and the second feature descriptor, to obtain a matching degree between each first feature descriptor and each second feature descriptor;

[0173] According to the matching degree between the first feature descriptor and the second feature descriptor, it is determined whether the target image contains a graphic that is identical or similar to the image material in the review rule content.

[0174] In the technical solution of the above-mentioned implementation, the feature extraction algorithm may be SIFT (Scale-Invariant Feature Transform), SURF (Speeded Up Robust Features), ORB (Oriented FAST and Rotated BRIEF), etc. The feature descriptor is a numerical representation of the image features used to represent the key points in the image. The key points are usually located at the corners, edges or local areas with rich textures in the image. The feature descriptor is calculated based on the image information around the key points.

[0175] In the technical solutions of other embodiments of the present invention, before the step of judging whether the target image contains a graphic that is identical or similar to the image material in the review rule content based on the matching degree of the first feature descriptor and the second feature descriptor, it also includes screening and verifying the matching results of the first feature descriptor and the second feature descriptor based on a preset matching degree threshold to remove erroneous matches and improve the accuracy of judging identical or similar graphics.

[0176] like Figure 2 As shown, the second aspect of the present invention proposes a method for automatically reviewing a virtual character in a metaverse, comprising:

[0177] Constructing a review rule database, wherein the review rule database includes a text rule library and a picture material library;

[0178] Import the model of the virtual character to be reviewed;

[0179] Traverse each review rule in the review rule database to perform the following steps:

[0180] Determining a model part corresponding to the content of the review rule, wherein the model part includes one or more local parts of the virtual character model;

[0181] Generate a target image of the model to be reviewed corresponding to the model part;

[0182] Performing feature matching between the target image and the review rule content to obtain a matching degree between the target image and the review rule content, wherein the review rule content is a text content or graphic content corresponding to the review rule;

[0183] Determine whether the model to be reviewed meets the requirements of the corresponding review rules based on the matching results.

[0184] Specifically, in the step of constructing a review rule database, the review rules for virtual characters in the metaverse can be formulated based on the review experience of the reviewers, or can be generated based on the recognition of the review specification text. The review rule content in the review rule database is presented in the form of text content or graphic content. When the review rule content is text content, it is stored in the text rule library of the review rule database; when the review rule content is graphic content, the review rule content includes associated stored text content and image materials, and the text content part is stored in the text rule library of the review rule database, and the image material part is stored in the image material library of the review rule database. In the technical scheme of the above-mentioned implementation method, the model to be reviewed of each virtual character is sequentially imported and matched with the content of each review rule in the review rule data to determine whether the model to be reviewed of the virtual character meets the requirements of the corresponding review rule.

[0185] For any review rule, when the review rule content is to limit a specific model part, the model part it limits will usually be reflected in its text content and / or image material. The same review rule can limit multiple model parts at the same time. When the review rule is not to limit a specific model part, it limits the virtual character model as a whole. In the step of determining the model part corresponding to the review rule content, one or more model parts corresponding to the review rule content can be entered at the same time as the review rule content; it is also possible to extract the corresponding model part by performing semantic recognition on the corresponding text content, or extract the corresponding model part by performing image recognition on the corresponding image material.

[0186] In the technical solutions of some embodiments of the present invention, before the step of generating the target image of the model to be reviewed corresponding to the model part, the step further includes:

[0187] Determine whether the review rule content has a corresponding model part, for example, determine whether the model part field corresponding to the review rule content in the review rule database is a null value. When the model part field corresponding to the review rule content is a null value, it is determined that the review rule content does not have a corresponding model part, otherwise it is determined that the review rule content has a corresponding model part;

[0188] When the review rule content has a corresponding model part, the target image includes one or more partial images of the model to be reviewed corresponding to the model part;

[0189] When there is no corresponding model part in the review rule content, the target image is a picture of the model to be reviewed that covers the entire model to be reviewed at a specific viewing angle.

[0190] Preferably, the step of generating a target image of the model to be reviewed corresponding to the model part specifically includes:

[0191] Generate a local model of the model to be reviewed corresponding to the model part;

[0192] Determine a first target image generation viewing angle of the local model, where the first target image generation viewing angle is a viewing angle at which the local model has a maximum visible area;

[0193] Rotating the local model to a posture corresponding to the viewing angle of the first target image;

[0194] Generate a projection image of the local model along the sight direction of an observer outside the screen at the first target image generation viewing angle;

[0195] The projection image is determined as a target image of the model to be reviewed corresponding to the model part.

[0196] Specifically, the step of generating a local model of the model to be reviewed corresponding to the model part is to extract the local model corresponding to the model part from the model to be reviewed as an independent model, and the local model does not include parts of the model to be reviewed other than the model part.

[0197] The first target image generation viewing angle is the angle of the local model's posture relative to the line of sight of the observer outside the screen when generating the target image, and the direction perpendicular to the screen is usually taken as the line of sight of the observer outside the screen. In the technical solution of the above-mentioned implementation, the step of determining the first target image generation viewing angle of the local model is specifically to determine the specific posture of the local model relative to the line of sight of the observer outside the screen, so that the local model has the largest visible area in the line of sight of the observer outside the screen.

[0198] The step of generating a projection image of the local model along the line of sight of an observer outside the screen at the first target image generation viewing angle specifically includes:

[0199] Determine the projection surface of the local model by a plane perpendicular to the sight line direction of the observer outside the screen;

[0200] The projection image is obtained by projecting the visible surface of the local model onto the projection surface when the local model is in a position corresponding to the viewing angle generated by the first target image. The visible surface of the local model is the surface that can be observed by the observer outside the screen when the local model is in a position corresponding to the viewing angle generated by the first target image.

[0201] In the technical solutions of some embodiments of the present invention, after the step of determining the model part corresponding to the content of the review rule, the following steps are further included:

[0202] When the review rule content does not have a corresponding model part, determining a second target image generation perspective of the model to be reviewed, the second target image generation perspective being a frontal perspective of the model to be reviewed;

[0203] Rotating the model to be reviewed to a posture corresponding to the viewing angle generated by the second target image;

[0204] Generate a projection image of the model to be reviewed along the line of sight of an observer outside the screen at the second target image generation viewing angle;

[0205] The projection image is determined as the target image of the model to be reviewed.

[0206] Preferably, the step of determining the first target image generation viewing angle of the local model specifically includes:

[0207] Determine a reference pose of the local model, wherein the reference pose is the pose of the local model corresponding to the front view of the model to be examined when the local model is located on the model to be examined in a natural state;

[0208] Construct a three-dimensional space coordinate system;

[0209] Get the pre-configured n sets of rotation angles Where i is a positive integer between 1 and n, ω i is the azimuth angle in the i-th group of rotation angles, is the altitude angle in the i-th group of rotation angles;

[0210] Based on the rotation angle Construct n first target planes in the three-dimensional space coordinate system, and the azimuth angle of the plane normal of the i-th first target plane in the three-dimensional space coordinate system is ω i , the altitude angle is

[0211] Determine the azimuth angle of n plane normals as ω j , the altitude angle is The second target plane, Within the angle range, the local model has a maximum projection area on the second target plane, where j is a positive integer between 1 and n;

[0212] Determine a third target plane having the largest projection area among the n second target planes;

[0213] The plane normal direction of the third target plane is determined as the first target image generation viewing angle.

[0214] Specifically, the pose of a stereo model represents the position and posture of the stereo model, which includes the position of the stereo model, the relative position relationship between the various components, and the rotation state of the stereo model, wherein the rotation state of the stereo model is determined relative to the line of sight of the observer outside the screen, and is usually represented by the rotation angle of the stereo model. The rotation angle of the stereo model is a relative value, so it is necessary to define a reference pose, and use the reference angle corresponding to the reference pose as a reference value to evaluate the size of the rotation angle of the stereo model after rotation.

[0215] In the technical solution of the above-mentioned embodiment, the position of the local model that satisfies the following two conditions is determined as the reference position of the local model: 1) the local model is located on the model to be reviewed in a natural state; 2) the rotational perspective of the model to be reviewed is the frontal perspective of the model to be reviewed. The local model is located on the model to be reviewed in a natural state, which means that the limbs of the virtual character are in a natural and relaxed state in an upright posture. For a three-dimensional model of a virtual character, especially a three-dimensional model of a humanoid virtual character, its frontal perspective refers to the rotational perspective in which the front of the three-dimensional model of the virtual character faces the screen, that is, the direction opposite to the line of sight of the observer outside the screen.

[0216] In the technical solution of the above-mentioned implementation, the three-dimensional space coordinate system refers to a spatial rectangular coordinate system, and the three coordinate axes constituting the three-dimensional space coordinate system are mutually orthogonal. Exemplarily, in the step of constructing the three-dimensional space coordinate system, the vertical direction parallel to the plane where the screen is located is determined as the first coordinate axis direction of the three-dimensional space coordinate system, the horizontal direction parallel to the plane where the screen is located is determined as the second coordinate axis direction of the three-dimensional space coordinate system, and the direction perpendicular to the screen is determined as the third coordinate axis direction of the three-dimensional space coordinate system.

[0217] In the technical solution of the above-mentioned implementation mode, Indicates that the three-dimensional space coordinate system is rotated by The plane normal is the spatial range of the central axis, and the rotation angle is The second target plane corresponding to the plane normal of is the plane with the largest projection area of ​​the local model within the space range, and the rotation angle and rotation angle There is a one-to-one correspondence.

[0218] In the technical solutions of the above embodiments, the value of n is positively correlated with the complexity of the surface structure of the local model, that is, the more complex the surface structure of the local model is, the larger the value of n is configured. In the technical solutions of some embodiments of the present invention, the azimuth of the n plane normals is determined as ω j , the altitude angle is The steps of determining the second target plane can be processed synchronously by constructing n processing threads to increase the speed of determining the third target plane.

[0219] Preferably, the azimuth angles of n plane normals are determined as ω j , the altitude angle is The steps of obtaining the second target plane specifically include:

[0220] Configure the rotation step size Δθ used to perform the rotation iterations, where

[0221] The altitude angle and azimuth angle of the plane normal constructed for the i-th first target plane are The four adjacent planes of

[0222] Calculate the projection area s of the local model onto the first target plane i0 and the projection areas s projected onto the adjacent planes i1 、s i2 、s i3 、s i4 ;

[0223] When i0 Not satisfied i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ), the adjacent plane with the largest projection area is determined as the first target plane;

[0224] Repeat the above steps to construct the plane normal for the i-th first target plane. The altitude angle and azimuth angle are The four adjacent planes of the local model are projected onto the first target plane by calculating the projection area s i0and the projection areas s projected onto the adjacent planes i1 、s i2 、s i3 、s i4 The steps, and when s i0 Not satisfied i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ), the step of determining the adjacent plane with the largest projection area as the first target plane until s i0 Satisfaction i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 );

[0225] When i0 Satisfaction i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ), the first target plane is determined as the second target plane.

[0226] In the technical solution of the above embodiment, the rotation step Δθ is the step size of the azimuth angle and the altitude angle in each rotation when performing the rotation iteration. The rotation step Δθ should be much smaller than So that the handler can be rotated at The spatial range in which the plane normal of the central axis is the center axis has a sufficient number of iteration steps, so that the second target plane with the largest projection area in the spatial range can be more accurately located. Since the value of n is related to the complexity of the surface structure of the local model, the rotation step Δθ also needs to be dynamically determined according to the complexity of the surface structure of the local model. Specifically, after determining the value of n, according to The size of Δθ determines the rotation step length.

[0227] Preferably, the step of configuring a rotation step length Δθ for performing rotation iteration specifically comprises:

[0228] Get the pre-configured minimum rotation step size Δθ0 and the default number of iteration steps m;

[0229] when When the rotation step length Δθ is configured as

[0230] when , the rotation step Δθ is configured as the minimum rotation step Δθ0.

[0231] Preferably, the step of performing feature matching between the target image and the review rule content to obtain the matching degree between the target image and the review rule content specifically includes:

[0232] Inputting the target image into a pre-trained image description model to generate a first description content of the target image, wherein the first description content is a text content describing the image, type, state and / or action of each object contained in the target image;

[0233] When the review rule content is text content, feature matching is performed between the first description content of the target image and the text content in the review rule content;

[0234] Obtaining a degree of matching between the target image and the review rule content based on a feature matching between the first description content of the target image and the text content in the review rule content;

[0235] When the review rule content is graphic content, feature matching is performed on the first description content of the target image with the text content in the review rule content, and feature matching is performed on the target image with the image material in the review rule content;

[0236] The degree of matching between the target image and the review rule content is obtained based on the feature matching between the first description content of the target image and the text content in the review rule content, and the feature matching between the target image and the image material in the review rule content.

[0237] Specifically, the image description model is a deep visual language model trained using a supervised training model based on a large number of images and their corresponding first description contents. It can recognize the image, type, state, action, etc. of various objects in the input image, and output the first description content of the corresponding image using natural language description.

[0238] In the technical solution of the above-mentioned implementation, when the review rule content is text content, the first description content of the target image is used to perform feature matching with the text content in the review rule content, and the feature matching result between the target image and the review rule content is determined based on the text feature matching between the two file contents. When the review rule content is graphic content, it is necessary to perform feature matching on the first description content of the target image and the text content in the review rule content at the same time, and to perform feature matching on the target image and the picture material in the review rule, and to determine the feature matching result between the target image and the review rule content based on the two feature matching situations.

[0239] Preferably, the step of performing feature matching between the first description content of the target image and the text content in the review rule content specifically includes:

[0240] Generate a first text vector corresponding to the first description content of the target image;

[0241] Generate or obtain a second text vector corresponding to the text content in the review rule content;

[0242] The distance between the first text vector and the second text vector in the vector space is calculated.

[0243] In the technical scheme of the above-mentioned implementation mode, the step of obtaining the matching degree between the target image and the review rule content based on the feature matching between the first description content of the target image and the text content in the review rule content is specifically to determine the distance between the first text vector and the second text vector in the vector space as the matching degree between the first description content of the target image and the text content in the review rule content.

[0244] In the technical solutions of some embodiments of the present invention, each time the step of feature matching the first description content of the target image with the text content in the review rule content is performed, a second text vector corresponding to the text content in the review rule content is generated in real time. In the technical solutions of other embodiments of the present invention, the review rule database also includes a text content vector library, and the step of constructing the review rule database includes:

[0245] Vectorizing the text content in the review rule content to generate a second text vector corresponding to the text content in the review rule content;

[0246] The second text vector is written into the text content vector library.

[0247] In the step of calculating the distance between the first text vector and the second text vector in the vector space, the distance between the first text vector and the second text vector in the vector space may be the Euclidean distance or the Manhattan distance between the two vectors.

[0248] Preferably, the step of generating a first text vector corresponding to the first description content of the target image specifically includes:

[0249] Performing sentence segmentation on the first description content of the target image to obtain a plurality of first description sentences;

[0250] Generate a first text vector corresponding to each first description sentence;

[0251] The step of calculating the distance between the first text vector and the second text vector in the vector space specifically includes:

[0252] Calculate the distance d between the first text vector corresponding to each first description sentence and the second text vector k1 , where k1 is a positive integer between 1 and l1, and l1 is the number of first description sentences obtained by segmenting the first description content of the target image;

[0253] Calculate a first average distance between the first text vector and the second text vector:

[0254]

[0255] The step of obtaining the matching degree between the target image and the review rule content based on the feature matching between the first description content of the target image and the text content in the review rule content specifically includes:

[0256] The first average distance The degree of match between the target image and the content of the review rule is determined.

[0257] Specifically, the first description content of the target image is usually a paragraph consisting of multiple sentences or a text block consisting of multiple paragraphs, and the text content of the review rule is a rule statement that limits a certain specific situation of the three-dimensional model of the virtual character as a whole or in part, so the text content of the review rule is usually a short sentence. In the technical scheme of the above-mentioned implementation mode, the first description content of the target image is first divided into multiple first description sentences through sentence segmentation processing, and then these first description sentences are vectorized to obtain multiple first text vectors. After calculating the distance between each first text vector and the second text vector respectively, the matching degree between the target image and the content of the review rule is obtained based on the average value of the distance between these first text vectors and the second text vector.

[0258] Preferably, the step of performing feature matching between the target image and the image material in the review rule content specifically includes:

[0259] Inputting the picture material in the review rule content into a pre-trained picture description model to generate a second description content of the picture material;

[0260] Segmenting the second description content of the picture material to obtain a plurality of second description sentences;

[0261] Generate a third text vector corresponding to each second description sentence;

[0262] Calculate the distance d between the first text vector corresponding to each first description sentence and the third text vector corresponding to each second description sentence k1,k2 , where k2 is a positive integer between 1 and l2, and l2 is the number of second description sentences obtained by segmenting the second description content of the picture material;

[0263] Calculate a second average distance between the first text vector and the third text vector:

[0264]

[0265] In the technical solution of the above implementation, the picture description model is used to generate a second description content corresponding to the picture material, and then the second description content is feature matched with the first description content to achieve feature matching between the target picture and the picture material.

[0266] Preferably, the step of obtaining the degree of matching between the target image and the review rule content based on the feature matching between the first description content of the target image and the text content in the review rule content, and the feature matching between the target image and the image material in the review rule content specifically includes:

[0267] The first average distance The second average distance between the first text vector and the third text vector sum The degree of match between the target image and the content of the review rule is determined.

[0268] In the technical solutions of other embodiments of the present invention, the target image and the image material may be directly matched by graphic features. In these embodiments, the step of matching the target image with the image material in the review rule content specifically includes:

[0269] Using a feature extraction algorithm to extract features from the target image and the image material in the review rule content, to obtain one or more first feature descriptors of the target image and one or more second feature descriptors of the image material in the review rule content;

[0270] Using a nearest neighbor matching algorithm to match the first feature descriptor and the second feature descriptor, to obtain a matching degree between each first feature descriptor and each second feature descriptor;

[0271] According to the matching degree between the first feature descriptor and the second feature descriptor, it is determined whether the target image contains a graphic that is identical or similar to the image material in the review rule content.

[0272] In the technical solution of the above-mentioned implementation, the feature extraction algorithm may be SIFT (Scale-Invariant Feature Transform), SURF (Speeded Up Robust Features), ORB (Oriented FAST and Rotated BRIEF), etc. The feature descriptor is a numerical representation of the image features used to represent the key points in the image. The key points are usually located at the corners, edges or local areas with rich textures in the image. The feature descriptor is calculated based on the image information around the key points.

[0273] In the technical solutions of other embodiments of the present invention, before the step of judging whether the target image contains a graphic that is identical or similar to the image material in the review rule content based on the matching degree of the first feature descriptor and the second feature descriptor, it also includes screening and verifying the matching results of the first feature descriptor and the second feature descriptor based on a preset matching degree threshold to remove erroneous matches and improve the accuracy of judging identical or similar graphics.

[0274] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0275] According to the embodiments of the present invention as described above, these embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and changes can be made based on the above description. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can make good use of the present invention and the modified use based on the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for automatically reviewing virtual characters in the metaverse, characterized in that: include: Constructing a review rule database, wherein the review rule database includes a text rule library and a picture material library; Import the model of the virtual character to be reviewed; Traverse each review rule in the review rule database to perform the following steps: Determining a model part corresponding to the content of the review rule, wherein the model part includes one or more local parts of the virtual character model; Generate a target image of the model to be reviewed corresponding to the model part; Performing feature matching between the target image and the review rule content to obtain a matching degree between the target image and the review rule content, wherein the review rule content is a text content or graphic content corresponding to the review rule; Determine whether the model to be reviewed meets the requirements of the corresponding review rules based on the matching results; The step of generating a target image of the model to be reviewed corresponding to the model part specifically includes: Generate a local model of the model to be reviewed corresponding to the model part; Determine a first target image generation viewing angle of the local model, where the first target image generation viewing angle is a viewing angle at which the local model has a maximum visible area; Rotating the local model to a posture corresponding to the viewing angle of the first target image; Generate a projection image of the local model along the sight direction of an observer outside the screen at the first target image generation viewing angle; Determine the projection image as a target image of the model to be reviewed corresponding to the model part; The step of determining the first target image generation viewing angle of the local model specifically includes: Determine a reference pose of the local model, wherein the reference pose is the pose of the local model corresponding to the front view of the model to be examined when the local model is located on the model to be examined in a natural state; Construct a three-dimensional space coordinate system; Get the pre-configured n sets of rotation angles Where i is a positive integer between 1 and n, ω i is the azimuth angle in the i-th group of rotation angles, is the altitude angle in the i-th group of rotation angles; Based on the rotation angle Construct n first target planes in the three-dimensional space coordinate system, and the azimuth angle of the plane normal of the i-th first target plane in the three-dimensional space coordinate system is ω i , the altitude angle is Determine the azimuth angle of n plane normals as ω j , the altitude angle is The second target plane, Within the angle range, the local model has a maximum projection area on the second target plane, where j is a positive integer between 1 and n; Determine a third target plane having the largest projection area among the n second target planes; Determining the plane normal direction of the third target plane as the first target image generation viewing angle; Determine the azimuth angle of n plane normals as ω j , the altitude angle is The steps of obtaining the second target plane specifically include: Configure the rotation step size Δθ used to perform the rotation iterations, where The altitude angle and azimuth angle of the plane normal constructed for the i-th first target plane are The four adjacent planes of Calculate the projection area s of the local model onto the first target plane i0 and the projection areas s projected onto the adjacent planes i1 、s i2 、s i3 、s i4 ; When i0 Not satisfied i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ), the adjacent plane with the largest projection area is determined as the first target plane; Repeat the above steps to construct the plane normal for the i-th first target plane. The altitude angle and azimuth angle are The four adjacent planes of the local model are projected onto the first target plane by calculating the projection area s i0 and the projection areas s projected onto the adjacent planes i1 、s i2 、s i3 、s i4 The steps, and when s i0 Not satisfied i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ), the step of determining the adjacent plane with the largest projection area as the first target plane until s i0 Satisfaction i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ); When i0 Satisfaction i0 ≥max(s i1 ,s i2 ,s i3 ,s i4 ), the first target plane is determined as the second target plane.

2. The method for automatically reviewing virtual characters in the metaverse according to claim 1, characterized in that: The step of performing feature matching between the target image and the review rule content to obtain the matching degree between the target image and the review rule content specifically includes: Inputting the target image into a pre-trained image description model to generate a first description content of the target image, wherein the first description content is a text content describing the image, type, state and / or action of each object contained in the target image; When the review rule content is text content, feature matching is performed between the first description content of the target image and the text content in the review rule content; Obtaining a degree of matching between the target image and the review rule content based on a feature matching between the first description content of the target image and the text content in the review rule content; When the review rule content is graphic content, feature matching is performed on the first description content of the target image with the text content in the review rule content, and feature matching is performed on the target image with the image material in the review rule content; The degree of matching between the target image and the review rule content is obtained based on the feature matching between the first description content of the target image and the text content in the review rule content, and the feature matching between the target image and the image material in the review rule content.

3. The method for automatically reviewing virtual characters in the metaverse according to claim 2, characterized in that: The step of performing feature matching between the first description content of the target image and the text content in the review rule content specifically includes: Generate a first text vector corresponding to the first description content of the target image; Generate or obtain a second text vector corresponding to the text content in the review rule content; The distance between the first text vector and the second text vector in the vector space is calculated.

4. The method for automatically reviewing virtual characters in the metaverse according to claim 3, characterized in that: The step of generating a first text vector corresponding to the first description content of the target image specifically includes: Performing sentence segmentation on the first description content of the target image to obtain a plurality of first description sentences; Generate a first text vector corresponding to each first description sentence; The step of calculating the distance between the first text vector and the second text vector in the vector space specifically includes: Calculate the distance d between the first text vector corresponding to each first description sentence and the second text vector k1 , where k1 is a positive integer between 1 and l1, and l1 is the number of first description sentences obtained by segmenting the first description content of the target image; Calculate a first average distance between the first text vector and the second text vector: The step of obtaining the matching degree between the target image and the review rule content based on the feature matching between the first description content of the target image and the text content in the review rule content specifically includes: The first average distance The degree of match between the target image and the content of the review rule is determined.

5. The method for automatically reviewing virtual characters in the metaverse according to claim 4, characterized in that: The step of matching the target image with the image material in the review rule content specifically includes: Inputting the picture material in the review rule content into a pre-trained picture description model to generate a second description content of the picture material; Segmenting the second description content of the picture material to obtain a plurality of second description sentences; Generate a third text vector corresponding to each second description sentence; Calculate the distance d between the first text vector corresponding to each first description sentence and the third text vector corresponding to each second description sentence k1,k2 , where k2 is a positive integer between 1 and l2, and l2 is the number of second description sentences obtained by segmenting the second description content of the picture material; Calculate a second average distance between the first text vector and the third text vector:

6. The method for automatically reviewing virtual characters in the metaverse according to claim 5, characterized in that: The step of obtaining the degree of matching between the target image and the review rule content based on the feature matching between the first description content of the target image and the text content in the review rule content, and the feature matching between the target image and the image material in the review rule content specifically includes: The first average distance The second average distance between the first text vector and the third text vector sum The degree of match between the target image and the content of the review rule is determined.

7. A system for automatically reviewing virtual characters in the metaverse, characterized in that: It includes a review rule database for storing review rule content and a review server connected to the review rule database for reviewing virtual character models in the metaverse according to the review rule content, the review rule database includes a text rule library for storing text content in the review rule content and a picture material library for storing picture materials in the review rule content, and the review server is configured to implement the method as described in any one of claims 1-6.