A virtual character construction system and method based on multimodal information fusion

Abstract

The present invention discloses a virtual character construction system and method based on multimodal information fusion, which relates to the technical field of virtual character construction, including evaluating the image construction quality of user images in image data; obtaining historical image data of historical sample users in a construction model, obtaining target image data of the user, and analyzing the reference value of the historical sample users to the user's virtual character construction; obtaining target image data, constructing the user's virtual character to obtain virtual character data, obtaining historical virtual character data of target historical sample users, evaluating the degree of fit between the virtual character data and the user, and optimizing the virtual character data to obtain target virtual character data; obtaining user posture data, and rendering the user's virtual character in combination with the user's target virtual character data, and managing the user's constructed virtual character.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual character construction, and in particular to a virtual character construction system and method based on multimodal information fusion. Background Art

[0002] Multimodal information fusion refers to the technology of integrating and processing multiple information from different sensors and different types of data sources. With the continuous development of science and technology, multimodal information fusion technology has been increasingly used in the construction of virtual characters. The construction of virtual characters based on multimodal information fusion mainly includes but is not limited to the following benefits: 1. Enhanced realism. The behavior and communication methods of virtual characters constructed through multimodal information fusion technology more realistically simulate humans, giving users a more realistic experience; 2. Enhanced expressive ability. Different modal information can complement and enhance each other, making the expressions and movements of the constructed virtual characters more comprehensive and delicate.

[0003] In the process of building a virtual character, it is necessary to obtain the user's image data and input the image data into the model to build the user's virtual character. Although the virtual character constructed in this way is fast, there are some problems. First of all, whether the constructed virtual character meets the user's characteristics and the set game style requires manual judgment, and manual judgment is prone to inaccuracies. In addition, once the virtual character does not meet the user's characteristics and the set game style, the model cannot optimize and modify the constructed virtual character, which will not only make the character less realistic, but also lead to insufficient resonance and identification of the user with the character, affecting the user's activity and retention rate, making it impossible for the company to profit, and may even cause serious economic and reputational losses to the company. Summary of the Invention

[0004] The purpose of the present invention is to provide a virtual character construction system and method based on multimodal information fusion to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for constructing a virtual character based on multimodal information fusion, the method comprising:

[0006] Step S100: obtaining user image data, obtaining sample image data in the constructed model, evaluating the image construction quality of the user image in the image data, and obtaining target image data;

[0007] Step S200: Obtain historical image data of historical sample users in the constructed model, obtain target image data of the user, analyze the reference value of the historical sample users to the construction of the user's virtual role, and obtain the target historical sample user;

[0008] Step S300: Acquire target image data, construct a virtual character for the user to obtain virtual character data, obtain historical virtual character data of a target historical sample user, evaluate the degree of fit between the virtual character data and the user, and optimize the virtual character data to obtain target virtual character data;

[0009] Step S400: Acquire the user's posture data, and render the user's virtual character in combination with the user's target virtual character data, and manage the user's constructed virtual character.

[0010] Furthermore, step S100 includes:

[0011] Step S101: Acquire the user's image data in the current cycle, where the image data includes each user image input into the construction model when the user constructs the virtual character;

[0012] Step S102: Acquire sample image data for constructing a model for image evaluation, where the sample image data includes each sample image in the constructed model;

[0013] Step S103: performing grayscale conversion on the user image, constructing a two-dimensional coordinate system with one pixel as a unit length, and obtaining the coordinates of each pixel in the user image;

[0014] Calculate the image sharpness value of each user image of the user, where the image sharpness B of the user's ath user image is a :

[0015] ,

[0016] Among them, K is the preset Laplace convolution kernel; I a (α, β) is the grayscale value of the pixel with abscissa α and ordinate β in the a-th user image; n is the maximum abscissa of the pixel in the a-th user image; m is the maximum ordinate of the pixel in the a-th user image;

[0017] Step S104: Evaluate the image construction quality of each user image in the image data, wherein the image construction quality of the d-th user image in the image data is evaluated. The specific process is:

[0018] Get the average grayscale value μ of the pixel points in the d-th user image d , obtain the image sharpness of each sample image, and calculate the quality evaluation value T of the dth user image d :

[0019] ,

[0020] Where j is the total number of sample images; B´ i is the image sharpness of the i-th sample image; B d is the image sharpness of the d-th user image; μ´ i is the average grayscale value of the pixels in the i-th sample image; γ1 and γ2 are the preset first evaluation coefficient and second evaluation coefficient respectively;

[0021] When T d Less than or equal to the preset quality evaluation threshold T △ , then the quality of the dth user image is judged to be unqualified. When the quality evaluation value T d >T △ ,determine the construction quality of the dth user image as qualified, and record the dth user image as the user’s target user image;

[0022] Step S105: Acquire and aggregate the target user images of the user to obtain the target image data of the user.

[0023] Furthermore, step S200 includes:

[0024] Step S201: acquiring historical image data of historical sample users in the constructed model, wherein the historical sample users are historical users whose constructed virtual characters have been verified and whose virtual character type is the same as that of the user, and the historical image data includes historical user images of the historical sample users;

[0025] Step S202: Construct a two-dimensional coordinate system, perform grayscale conversion on the historical user image to obtain image E(x, y), and perform Gaussian filtering on the historical user image to obtain image R(x, y). The specific filtering formula is:

[0026] ,

[0027] Among them, σ is the standard deviation preset in the construction model;

[0028] Step S203: Obtain each historical sample user in the constructed model, and analyze the reference value of each historical sample user to the construction of the user's virtual role, wherein the reference value of the e-th historical sample user to the construction of the user's virtual role is analyzed. The specific process is:

[0029] Obtain the historical user image of the e-th historical sample user after Gaussian filtering, and use a preset key point detection algorithm to obtain each key joint point in the historical user image of the e-th historical sample user;

[0030] Step S204: Based on the historical user images of the e-th historical sample user, resize and Gaussian filter each target user image in the target image data, and calculate the construction reference value F of the e-th historical sample user to the user e :

[0031] ,

[0032] Among them, δ is the total number of target user images; q is the total number of key joint points; D z ε,ε+1 D is the distance between the εth key joint point and the ε+1th key joint point in the user’s zth target user image; ε,ε+1 is the distance between the εth key joint point and the ε+1th key joint point in the historical user image of the eth historical sample user;

[0033] Step S205: When constructing the reference value F e If the value is greater than the preset construction reference threshold, the e-th historical sample user is determined to have reference value for the user's virtual character construction, and the e-th historical sample user is recorded as the user's target historical sample user;

[0034] The reason why Gaussian filtering is performed on the historical user images in the above steps is that Gaussian filtering can effectively remove random noise in the image, making the image smoother, and Gaussian filtering can ensure the stability of image extraction. By analyzing the image after Gaussian filtering, the accuracy of the reference value constructed by the historical sample user for the user can be guaranteed, thereby providing data support for the optimization of the virtual role constructed by the user in the following text, so that the virtual role constructed by the construction model is more in line with the user itself.

[0035] Furthermore, step S300 includes:

[0036] Step S301: Acquire target image data of the user, construct the user's virtual character using the construction model, obtain virtual character data, the virtual character data is a 3D model of the user's virtual character constructed by the construction model, and obtain a bounding box of the 3D model of the virtual character;

[0037] Obtain the historical virtual character data of the target historical sample user. The historical virtual character data includes the 3D model of the target historical sample user's virtual character. Evaluate the degree of fit between the virtual character data and the user. The specific process is as follows:

[0038] Get the bounding box of the 3D model of the virtual character of each target historical sample user, and calculate the feature fit value U between the virtual character data and the user:

[0039] ,

[0040] Among them, L, W, and H are respectively the length, width, and height of the bounding box of the virtual character's 3D model in the virtual character data; L v , W v , H v are respectively the length, width, and height of the bounding box of the virtual character's 3D model of the user's v-th target historical sample character; p is the total number of each target historical sample user;

[0041] Step S302: When the feature matching value U is greater than the preset feature matching threshold U´, it is determined that the virtual character data fits the user, and the virtual character data is recorded as the target virtual character data. When U < U´, it is determined that the virtual character data does not fit the user. Based on the historical virtual character data of each target historical sample user, various model parameters in the virtual character's 3D model of the virtual character data are adjusted to optimize the user's virtual character's 3D model until U ≥ U´, and the optimized virtual character's 3D model is saved to obtain the target virtual character data.

[0042] Furthermore, step S400 includes:

[0043] Step S401: Obtain the user's pose data, where the pose data includes the video information collected by the camera when the user performs motion capture;

[0044] Step S402: Based on the pose data, combined with the user's target virtual character data, render the user's virtual character, and use image editing software to perform color correction on the user's virtual character and manage the virtual character constructed by the user.

[0045] In order to better implement the above method, a virtual character construction system based on multi-modal information fusion is also proposed. The system includes a target image data module, a reference evaluation module, a data optimization module, and an intelligent management module;

[0046] The target image data module is used to evaluate the image construction quality of the user image in the image data to obtain the target image data;

[0047] The reference evaluation module is used to obtain the user's target image data, analyze the reference value of the historical sample users for the user's virtual character construction, and obtain the target historical sample users;

[0048] The data optimization module is used to evaluate the degree of fit between the virtual character data and the user, and optimize the virtual character data to obtain the target virtual character data;

[0049] The intelligent management module is used to render the user's virtual character and manage the virtual character constructed by the user.

[0050] Furthermore, the target image data module includes a quality evaluation value unit and a target image data unit;

[0051] A quality evaluation value unit, used to calculate a quality evaluation value of a user's user image;

[0052] The target image data unit is used to evaluate the image construction quality of the user image according to the quality evaluation value of the user image to obtain target image data.

[0053] Furthermore, the reference evaluation module includes a reference value construction unit and a reference evaluation unit;

[0054] A construction reference value unit is used to calculate the construction reference value of historical sample users to users;

[0055] The reference analysis unit is used to analyze the reference value of the historical sample users to the user's virtual role construction based on the construction reference value to obtain the target historical sample users.

[0056] Furthermore, the data optimization module includes a virtual character construction unit and a data optimization unit;

[0057] A virtual character building unit, used to build a user's virtual character and obtain virtual character data;

[0058] The data optimization unit is used to optimize the virtual character data to obtain target virtual character data.

[0059] Furthermore, the intelligent management module includes an intelligent management unit;

[0060] The intelligent management unit is used to obtain the user's posture data, and render the user's virtual character in combination with the user's target virtual character data, and manage the user's constructed virtual character.

[0061] Compared with the prior art, the beneficial effects of the present invention are: the present invention realizes the intelligent construction of the user's virtual character, and before the user's virtual character is constructed, the sample image data in the construction model is used to evaluate the image quality of the user's user image, thereby ensuring the image quality of the user image of the virtual character, and optimizing and adjusting the virtual character model constructed by the user through the user's target historical sample users, so that the constructed virtual character model is more suitable for the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a method flow chart of a virtual character construction method based on multimodal information fusion according to the present invention;

[0063] Figure 2It is a module diagram of a virtual character construction system based on multimodal information fusion of the present invention. DETAILED DESCRIPTION

[0064] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0065] Example: Figure 1-Figure 2 As shown, the present invention provides a technical solution, a method for constructing a virtual character based on multimodal information fusion, the method comprising:

[0066] Step S100: obtaining user image data, obtaining sample image data in the constructed model, evaluating the image construction quality of the user image in the image data, and obtaining target image data;

[0067] Wherein, step S100 includes:

[0068] Step S101: Acquire the user's image data in the current cycle, where the image data includes each user image input into the construction model when the user constructs the virtual character;

[0069] Step S102: Acquire sample image data for constructing a model for image evaluation, where the sample image data includes each sample image in the constructed model;

[0070] Step S103: performing grayscale conversion on the user image, constructing a two-dimensional coordinate system with one pixel as a unit length, and obtaining the coordinates of each pixel in the user image;

[0071] Calculate the image sharpness value of each user image of the user, where the image sharpness B of the user's ath user image is a :

[0072] ,

[0073] Among them, K is the preset Laplace convolution kernel; I a (α, β) is the grayscale value of the pixel with abscissa α and ordinate β in the a-th user image; n is the maximum abscissa of the pixel in the a-th user image; m is the maximum ordinate of the pixel in the a-th user image;

[0074] Step S104: Evaluate the image construction quality of each user image in the image data, wherein the image construction quality of the d-th user image in the image data is evaluated. The specific process is:

[0075] Get the average grayscale value μ of the pixel points in the d-th user image d , obtain the image sharpness of each sample image, and calculate the quality evaluation value T of the dth user image d :

[0076] ,

[0077] Where j is the total number of sample images; B´ i is the image sharpness of the i-th sample image; B d is the image sharpness of the d-th user image; μ´ i is the average grayscale value of the pixels in the i-th sample image; γ1 and γ2 are the preset first evaluation coefficient and second evaluation coefficient respectively;

[0078] When T d Less than or equal to the preset quality evaluation threshold T △ , then the construction quality of the dth user image is judged to be unqualified. When the quality evaluation value T d >T △ ,determine the construction quality of the dth user image as qualified, and record the dth user image as the user’s target user image;

[0079] Step S105: Acquire and aggregate the target user images of the user to obtain the target image data of the user;

[0080] Step S200: Obtain historical image data of historical sample users in the constructed model, obtain target image data of the user, analyze the reference value of the historical sample users to the construction of the user's virtual role, and obtain the target historical sample user;

[0081] Wherein, step S200 includes:

[0082] Step S201: acquiring historical image data of historical sample users in the constructed model, wherein the historical sample users are historical users whose constructed virtual characters have been verified and whose virtual character type is the same as that of the user, and the historical image data includes historical user images of the historical sample users;

[0083] Step S202: Construct a two-dimensional coordinate system, perform grayscale conversion on the historical user image to obtain image E(x, y), and perform Gaussian filtering on the historical user image to obtain image R(x, y). The specific filtering formula is:

[0084] ,

[0085] Among them, σ is the standard deviation preset in the construction model;

[0086] Step S203: Obtain each historical sample user in the constructed model, and analyze the reference value of each historical sample user to the construction of the user's virtual role, wherein the reference value of the e-th historical sample user to the construction of the user's virtual role is analyzed. The specific process is:

[0087] Obtain the historical user image of the e-th historical sample user after Gaussian filtering, and use a preset key point detection algorithm to obtain each key joint point in the historical user image of the e-th historical sample user;

[0088] Step S204: Based on the historical user images of the e-th historical sample user, resize and Gaussian filter each target user image in the target image data, and calculate the construction reference value F of the e-th historical sample user to the user e :

[0089] ,

[0090] Among them, δ is the total number of target user images; q is the total number of key joint points; D z ε,ε+1 D is the distance between the εth key joint point and the ε+1th key joint point in the user’s zth target user image; ε,ε+1 is the distance between the εth key joint point and the ε+1th key joint point in the historical user image of the eth historical sample user;

[0091] Step S205: When constructing the reference value F e If the value is greater than the preset construction reference threshold, the e-th historical sample user is determined to have reference value for the user's virtual character construction, and the e-th historical sample user is recorded as the user's target historical sample user;

[0092] Step S300: Acquire target image data, construct a virtual character for the user to obtain virtual character data, obtain historical virtual character data of a target historical sample user, evaluate the degree of fit between the virtual character data and the user, and optimize the virtual character data to obtain target virtual character data;

[0093] Wherein, step S300 includes:

[0094] Step S301: Obtain the target image data of the user, use the construction model to construct the user's virtual character, and obtain virtual character data. The virtual character data is the 3D model of the user's virtual character constructed by the construction model. Obtain the bounding box of the virtual character 3D model;

[0095] Obtain the historical virtual character data of the target historical sample users of the user. The historical virtual character data includes the 3D model of the virtual character of the target historical sample users. Evaluate the degree of fit between the virtual character data and the user. The specific process is as follows:

[0096] Obtain the bounding boxes of the 3D models of the virtual characters of each target historical sample user of the user, and calculate the feature fit value U between the virtual character data and the user:

[0097] ,

[0098] where L, W, and H are the length, width, and height of the bounding box of the 3D model of the virtual character in the virtual character data; L v , W v , H v are respectively the length, width, and height of the bounding box of the 3D model of the virtual character of the v-th target historical sample character of the user; p is the total number of all target historical sample users;

[0099] For example, the length, width, and height L, W, and H of the bounding box of the 3D model of the virtual character in the virtual character data are 10, 4, and 8 respectively; p is 2; the length, width, and height L1, W1, and H1 of the bounding box of the 3D model of the virtual character of the first target historical sample character of the user are 12, 8, and 10 respectively; the length, width, and height L2, W2, and H2 of the bounding box of the 3D model of the virtual character of the second target historical sample character of the user are 14, 10, and 6 respectively;

[0100] ,

[0101] Step S302: When the feature fit value U is greater than the preset feature fit threshold U´, determine that the virtual character data fits the user, and record the virtual character data as the target virtual character data. When U < U´, determine that the virtual character data does not fit the user. Based on the historical virtual character data of each target historical sample user, adjust the various model parameters in the 3D model of the virtual character of the virtual character data, optimize the 3D model of the user's virtual character until U ≥ U´, and save the optimized 3D model of the virtual character to obtain the target virtual character data;

[0102] For example, the various model parameters include geometric parameters, transformation parameters, etc.;

[0103] Step S400: obtaining the user's posture data, and combining it with the user's target virtual character data, rendering the user's virtual character, and managing the user's constructed virtual character;

[0104] Wherein, step S400 includes:

[0105] Step S401: Acquire user's posture data, where the posture data includes video information collected by a camera when the user is performing motion capture;

[0106] Step S402: Based on the posture data and in combination with the user's target virtual character data, the user's virtual character is rendered, and the user's virtual character is color-corrected using image editing software, and the user's constructed virtual character is managed;

[0107] In order to better implement the above method, a virtual character construction system based on multimodal information fusion is also proposed. The system includes a target image data module, a reference evaluation module, a data optimization module, and an intelligent management module;

[0108] a target image data module, configured to evaluate the image construction quality of the user image in the image data to obtain target image data;

[0109] The reference evaluation module is used to obtain the target image data of the user, analyze the reference value of historical sample users to the construction of the user's virtual role, and obtain the target historical sample users;

[0110] A data optimization module is used to evaluate the degree of compatibility between the virtual character data and the user, and to optimize the virtual character data to obtain target virtual character data;

[0111] An intelligent management module is used to render the user's virtual character and manage the user's constructed virtual character;

[0112] Wherein, the target image data module includes a quality evaluation value unit and a target image data unit;

[0113] A quality evaluation value unit, configured to calculate a quality evaluation value of a user's user image;

[0114] a target image data unit, configured to evaluate the image construction quality of the user image according to the quality evaluation value of the user image to obtain target image data;

[0115] Among them, the reference evaluation module includes a reference value construction unit and a reference evaluation unit;

[0116] A construction reference value unit is used to calculate the construction reference value of historical sample users to users;

[0117] A reference analysis unit, configured to analyze the reference value of the historical sample users to the user's virtual role construction based on the construction reference value, and obtain the target historical sample users;

[0118] Among them, the data optimization module includes a virtual character construction unit and a data optimization unit;

[0119] A virtual character building unit, used to build a user's virtual character and obtain virtual character data;

[0120] A data optimization unit, used for optimizing the virtual character data to obtain target virtual character data;

[0121] Wherein, the intelligent management module includes an intelligent management unit;

[0122] The intelligent management unit is used to obtain the user's posture data, and render the user's virtual character in combination with the user's target virtual character data, and manage the user's constructed virtual character.

[0123] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A method for constructing a virtual character based on multimodal information fusion, characterized in that: The method comprises: Step S100: obtaining user image data, obtaining sample image data in the constructed model, evaluating the image construction quality of the user image in the image data, and obtaining target image data; Step S200: Acquire historical image data of historical sample users in the constructed model, acquire target image data of the user, analyze the reference value of the historical sample users to the construction of the user's virtual role, and obtain the target historical sample user; Step S300: Acquire the target image data, construct a virtual character for the user to obtain virtual character data, obtain historical virtual character data of the target historical sample user, evaluate the degree of fit between the virtual character data and the user, and optimize the virtual character data to obtain target virtual character data; Step S400: acquiring the user's posture data, and rendering the user's virtual character in combination with the user's target virtual character data, and managing the user's constructed virtual character; The step S100 includes: Step S101: acquiring image data of a user in a current cycle, wherein the image data includes each user image input into a construction model when the user constructs a virtual character; Step S102: Acquire sample image data for constructing a model for image evaluation, wherein the sample image data includes each sample image in the constructed model; Step S103: performing grayscale conversion on the user image of the user, constructing a two-dimensional coordinate system with one pixel as a unit length, and obtaining the coordinates of each pixel in the user image; Calculate the image sharpness value of each user image of the user, wherein the image sharpness B of the a-th user image of the user is a : ， Among them, K is the preset Laplace convolution kernel; I a (α, β) is the grayscale value of the pixel with abscissa α and ordinate β in the a-th user image; n is the maximum abscissa of the pixel in the a-th user image; m is the maximum ordinate of the pixel in the a-th user image; Step S104: Evaluate the image construction quality of each user image in the image data, wherein the image construction quality of the d-th user image in the image data is evaluated, and the specific process is as follows: Get the average value μ of the grayscale value of the pixels in the d-th user image d , obtain the image sharpness of each sample image, and calculate the quality evaluation value T of the d-th user image d : ， Wherein, j is the total number of sample images; B´ i is the image sharpness of the i-th sample image; B d is the image sharpness of the d-th user image; μ´ i is the average grayscale value of the pixels in the i-th sample image; γ1 and γ2 are respectively the preset first evaluation coefficient and the second evaluation coefficient; When the T d Less than or equal to the preset quality evaluation threshold T △ , then the quality of the d-th user image is determined to be unqualified. When the quality evaluation value T d >T △ , determining that the construction quality of the d-th user image is qualified, and recording the d-th user image as the target user image of the user; Step S105: Acquire and aggregate target user images of the user to obtain target image data of the user.

2. The method for constructing a virtual character based on multimodal information fusion according to claim 1, characterized in that: The step S200 includes: Step S201: Acquiring historical image data of historical sample users in the constructed model, wherein the historical sample users are historical users whose constructed virtual characters have been verified and whose virtual character type is the same as that of the user, and the historical image data includes historical user images of the historical sample users; Step S202: Construct a two-dimensional coordinate system, perform grayscale conversion on the historical user image to obtain image E(x, y), and perform Gaussian filtering on the historical user image to obtain image R(x, y). The specific filtering formula is: ， Among them, σ is the standard deviation preset in the construction model; Step S203: Obtain each historical sample user in the constructed model, and analyze the reference value of each historical sample user to the construction of the user's virtual role, wherein the reference value of the e-th historical sample user to the construction of the user's virtual role is analyzed. The specific process is: Obtain the historical user image of the e-th historical sample user after Gaussian filtering, and use a preset key point detection algorithm to obtain each key joint point in the historical user image of the e-th historical sample user; Step S204: Based on the historical user images of the e-th historical sample user, resize and Gaussian filter each target user image in the target image data, and calculate the construction reference value F of the e-th historical sample user for the user e : ， Wherein, δ is the total number of target user images; q is the total number of key joint points; D z ε,ε+1 D is the distance between the εth key joint point and the ε+1th key joint point in the zth target user image of the user; ε,ε+1 is the distance between the εth key joint point and the ε+1th key joint point in the historical user image of the eth historical sample user; Step S205: When the reference value F is constructed e If the value of the e-th historical sample user is greater than a preset construction reference threshold, it is determined that the e-th historical sample user has reference value for the construction of the user's virtual role, and the e-th historical sample user is recorded as the target historical sample user of the user.

3. The method for constructing a virtual character based on multimodal information fusion according to claim 2, characterized in that: The step S300 includes: Step S301: Obtain the target image data of the user, use a construction model to construct the virtual character of the user, and obtain virtual character data. The virtual character data is the 3D model of the virtual character of the user constructed by the construction model, and obtain the bounding box of the 3D model of the virtual character; Obtain the historical virtual character data of the target historical sample user of the user. The historical virtual character data includes the 3D model of the virtual character of the target historical sample user, and evaluate the degree of fit between the virtual character data and the user. The specific process is as follows: Obtain the bounding boxes of the 3D models of the virtual characters of each target historical sample user of the user, and calculate the feature fit value U between the virtual character data and the user: ， Wherein, L, W, and H are respectively the length, width, and height of the bounding box of the 3D model of the virtual character in the virtual character data; L v 、W v 、H v are the length, width and height of the bounding box of the 3D model of the virtual character of the vth target historical sample character of the user; p is the total number of the target historical sample users; Step S302: When the feature fit value U is greater than the preset feature fit threshold U´, determine that the virtual character data fits the user, and record the virtual character data as the target virtual character data. When U < U´, determine that the virtual character data does not fit the user, and based on the historical virtual character data of each target historical sample user, adjust the various model parameters in the 3D model of the virtual character of the virtual character data to optimize the 3D model of the virtual character of the user until U ≥ U´, and save the optimized 3D model of the virtual character to obtain the target virtual character data.

4. The method for constructing a virtual character based on multimodal information fusion according to claim 3, characterized in that: The step S400 includes: Step S401: Obtain the pose data of the user. The pose data includes the video information collected by the camera when the user performs motion capture; Step S402: Based on the pose data, combine the target virtual character data of the user to render the virtual character of the user, and use image editing software to perform color correction on the virtual character of the user and manage the constructed virtual character of the user.

5. A virtual character construction system based on multimodal information fusion, used to execute a virtual character construction method based on multimodal information fusion according to any one of claims 1 to 4, characterized in that: The system includes a target image data module, a reference evaluation module, a data optimization module, and an intelligent management module; The target image data module is used to evaluate the image construction quality of the user image in the image data to obtain target image data; The reference evaluation module is used to obtain the target image data of the user, analyze the reference value of the historical sample user for the construction of the virtual character of the user, and obtain the target historical sample user; The data optimization module is used to evaluate the degree of fit between the virtual character data and the user, and optimize the virtual character data to obtain the target virtual character data; The intelligent management module is used to render the virtual character of the user and manage the constructed virtual character of the user.

6. A virtual character construction system based on multimodal information fusion according to claim 5, characterized in that: The target image data module includes a quality evaluation value unit and a target image data unit; The quality evaluation value unit is used to calculate the quality evaluation value of the user image of the user; The target image data unit is used to evaluate the image construction quality of the user image according to the quality evaluation value of the user image to obtain target image data.

7. The virtual character construction system based on multimodal information fusion according to claim 5, characterized in that: The reference evaluation module includes a reference value construction unit and a reference evaluation unit; The construction reference value unit is used to calculate the construction reference value of the historical sample user to the user; The reference analysis unit is used to analyze the reference value of the historical sample users to the user's virtual role construction based on the construction reference value to obtain the target historical sample users.

8. The virtual character construction system based on multimodal information fusion according to claim 5, characterized in that: The data optimization module includes a virtual character construction unit and a data optimization unit; The virtual character construction unit is used to construct the virtual character of the user and obtain virtual character data; The data optimization unit is used to optimize the virtual character data to obtain target virtual character data.

9. The virtual character construction system based on multimodal information fusion according to claim 5, characterized in that: The intelligent management module includes an intelligent management unit; The intelligent management unit is used to acquire the user's posture data, and render the user's virtual character in combination with the user's target virtual character data, and manage the user's constructed virtual character.

Images