Vent travel element universe virtual character cooperation system and method based on large space interaction

Abstract

The invention relates to the technical field of computer systems and human-computer interaction, in particular to a text travel element universe virtual role collaboration system and method based on large-space interaction. The method specifically comprises the following steps: tracking a user spatial position in real time through a multi-source positioning technology, and carrying out identity authentication and equipment legality verification by fusing multi-modal features; constructing a virtual-real fused semantic point cloud structural body, binding role interaction nodes, dynamically generating a task collaboration graph by using a graph neural network, and driving multiple roles to collaboratively execute subtasks; calling a graph attention network and a GPT module to deduce personalized plot branches in combination with user multi-modal input and a historical track, and rendering an immersive interactive scene in real time; spatial preferences and content interests are mined by adopting DBSCAN clustering and depth feature coding, a dynamic portrait is constructed, and a task path and content presentation are optimized through a weighted recommendation algorithm. According to the method, cloud rendering, edge calculation and cross-module cooperative control are integrated, and the interactivity and immersion experience of the text travel element universe are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of computer systems and human-computer interaction technology, and in particular to a cultural and tourism metaverse virtual character collaboration system and method based on large-space interaction. Background Art

[0002] With the acceleration of the digital transformation of the cultural and tourism industry, metaverse technology has gradually become the core means of constructing virtual-reality fusion scenes. Especially in the field of cultural tourism, users' demand for immersive interactive experience and multi-role collaboration is increasing. In existing technologies, virtual role collaboration systems mostly rely on static task allocation and a single interaction mode, which is difficult to adapt to the dynamic task decomposition and real-time collaboration needs in large-space scenes. In addition, the user identity authentication and device security verification mechanisms are still imperfect, resulting in an urgent need to improve the interaction efficiency and personalized experience in multi-user environments. At the same time, traditional solutions have technical bottlenecks in the dynamic generation of cultural content, user behavior-driven plot evolution, and real-time synchronization across terminals, which limits the immersion and resource allocation efficiency of virtual-reality fusion scenes.

[0003] The Chinese invention application with publication number CN119474452A discloses a cultural and tourism content recommendation system based on the metaverse, including a cultural and tourism content recommendation platform, a classification item matching unit, a refined matching unit, a content division unit, and a video sorting unit; the cultural and tourism content of the metaverse is rationally recommended in a progressive manner, which helps to improve the matching degree of the cultural and tourism content recommendation and the user's viewing effect, and performs video division feedback analysis on the cultural and tourism content classification items that the user is interested in through information feedback, so as to understand the user's tendency to play the video in the main recommendation item or priority recommendation item from a historical perspective, so as to further divide and match the played video, and conduct in-depth playback order information processing and recommendation analysis on the scenic spot popularity information, which helps to make rational cultural and tourism content recommendations based on the user's interest tendencies and the popularity of the scenic spot itself in the played video content.

[0004] In order to meet the technical requirements of high-precision positioning, multimodal interaction, intelligent collaboration and safe control in the cultural and tourism metaverse, it is urgent to build a full-link solution that integrates large-space perception, dynamic task allocation, AI-driven content generation and personalized recommendations. Summary of the Invention

[0005] The purpose of the present invention is to address the problems existing in the background technology and propose a cultural and tourism metaverse virtual character collaboration system and method based on large-space interaction.

[0006] The technical solution of the present invention is a method for coordinating virtual characters in a cultural and tourism metaverse based on large-space interaction, which specifically includes the following implementation steps:

[0007] S1. Identify user identity and verify device legitimacy through multimodal data, build an initial state vector based on spatial positioning and posture information, automatically match role labels and update the collaborative space map, and perform user role initialization and dynamic synchronization with the virtual environment;

[0008] S2. Generate a semantic point cloud structure by fusing real-time perception data with a semantic graph. Construct a behavior-driven graph by binding semantic nodes to character states. Generate a renderable model using voxel reconstruction and meshing algorithms. Perform real-time rendering of the spatial model, synchronize its state, and perform periodic broadcasting.

[0009] S3: Dynamically perceive the role task capabilities through graph neural networks and generate a set of participating tasks. Combined with the task decomposition algorithm, the macro goal is broken down into subtasks and then assigned to multiple role nodes to build a task collaboration graph, monitoring the execution status and global feedback updates in real time.

[0010] S4: Construct user interaction states through multimodal input recognition and time series fusion, call the cultural tourism plot graph, and infer personalized plot branches based on the graph attention network and GPT module to generate node states. Finally, convert the environmental parameters and character behavior model into a rendering instruction set through the rendering task scheduling function.

[0011] S5. By collecting user behavior trajectories and multimodal data, using DBSCAN clustering and deep feature coding to construct spatial preference and content interest vectors, extracting task completion efficiency and emotional characteristics to form personalized portraits, combining the AI ​​recommendation algorithm of cosine similarity and geographic proximity to dynamically optimize task paths and interaction methods, and updating user preference weights through a feedback mechanism to complete the improvement of the adaptive experience of the cultural and tourism metaverse.

[0012] Preferably, the identification process of identifying a user through multimodal data is as follows:

[0013] Collect user multimodal features, including face image I f , voice sample V s 、Device ID(D i ) and the user enters account credentials A k , calculate the user identity matching degree P u :

[0014]

[0015] in, Represents the user's real-time face image I f With registered image Similarity score of Represents the user voice sample V s Register sample with the system The matching degree; sim() represents the cosine similarity function; Indicates device ID and registration The degree of match between Indicates that the user enters account credentials A k and registration account credentials The matching degree between them; Bool() represents the judgment function, which outputs 1 if the condition is met, otherwise it outputs 0; α, β, γ, η represent weight coefficients, α+β+γ+η=1.

[0016] Preferably, the synchronization process of the spatial model real-time rendering and state synchronization is:

[0017] S31, collect real-time perception data {scene point cloud P cloud and structural features T struct}, fusion of predefined semantic graph K scene With real-time perception data, generate semantic point cloud structure S env :

[0018] S env =f map (P cloud ,T struct ,K scene );

[0019] Among them, f map () represents the fusion function, which generates the structure through the deep semantic network DGCNN;

[0020] S32, according to the character state R0 and the spatial structure S env , bind the virtual character to the specific semantic node in the virtual environment to form a behavior-driven graph G act :G act =f bind (R0,S env );

[0021] Among them, f bind () represents the semantic matching function; G act Represents a behavior-driven diagram;

[0022] The behavior graph can be defined as a set of triples as follows:

[0023] G act ={(R i ,s j ,a k )|R i ∈R on ,s j ∈S env ,a k ∈ActionSet};

[0024] Among them, R iRepresents a role entity; s j Represents a spatial semantic node; a k Indicates interactive action; R on Represents the set of all currently online roles; ActionSet represents the preset interaction set;

[0025] S33, perform real-time rendering and state synchronization of the spatial model: convert the semantic point cloud structure S env and behavior-driven graph G act Mapped to renderable model M render , and define the set of all online roles R on :{(R i )|i=0,1,..,n}, construct the synchronization state matrix S sync (t):

[0026]

[0027] Wherein, i represents the sequence number; represents the identity ID of the i-th character at time t; Represents the position coordinates of the i-th character at time t in three-dimensional space; represents the orientation of the i-th character at time t; represents the action performed by the i-th character at time t.

[0028] Preferably, the renderable model M render The generation process is:

[0029] S41. Perform semantic parsing and spatial stratification of point cloud data through label-driven voxel reconstruction method to construct semantic voxel structure.

[0030] S42, using Marching Cubes mesh reconstruction algorithm to generate topological geometry M geo , and perform differentiated reconstruction strategies for different structures according to semantic labels;

[0031] S43, combining semantic labels to map materials and textures, forming a variety of appearances that meet the requirements of realistic rendering; then according to the character behavior graph G act Dynamically mount interactive event points to bind the virtual environment to user behavior logic to achieve controllable interactivity;

[0032] S44. Build a multi-level LOD model based on point cloud information entropy to achieve automatic switching between near and far views and resource optimization, thereby outputting a structurally complete, behavior-driven, and highly efficient 3D visual model M. render .

[0033] Preferably, the task collaboration graph construction process is as follows:

[0034] S51. By modeling each character's collaboration capabilities and current spatial state, a dynamic task perception mechanism is implemented, and the types of tasks that the character can participate in are inferred. This means dynamically identifying the set of tasks that each character can participate in within the current semantic space.

[0035] Among them, R i Indicates the current i-th role entity; f task () represents the task perception function, which generates the task perception vector based on the graph neural network GAT; Indicates the current role's perceptible task list; G act Represents a behavior-driven diagram; S env Represents a semantic point cloud structure;

[0036] S52. Decompose complex goals into multiple subtasks according to the current space mission requirements k =f decomp (Task env ) and automatically distribute them to multiple role nodes to form a collaborative execution graph;

[0037] That is, building a task collaboration map based on the collaborative capabilities of multiple roles:

[0038]

[0039] Among them, Task env Represents the original macro task set triggered in the current virtual scene; f decomp () represents the task decomposition function; Task k Represents a subtask, the kth decomposed execution unit task; G coop represents a collaborative graph; Indicates the jth available task Task k virtual characters; Representing a role The set of tasks that can be perceived or completed.

[0040] Preferably, the node status generation process is as follows:

[0041] S61, call the embedded cultural tourism plot map G plot , containing event node V plot With conditional edge E cond :G plot =(V plot ,E cond );

[0042] S62. Combined with historical trajectory and task status Extract the user's current plot location node C curr ;

[0043] S63, call the plot reasoning function to generate the next node state

[0044]

[0045] Among them, f story () represents the GAT+GPT fine-tuning module based on the graph attention network, which infers the plot branch according to the current state vector and task conditions;

[0046] S64. Output plot node status

[0047] Preferably, the user preference weight update process is as follows:

[0048] S71, behavior trajectory collection and multi-dimensional feature modeling, define each user's behavior trajectory in the virtual space as a temporal event sequence T (u) :T (u) ={(t i ,p i ,a i ,r i ,s i ,e i )|i=1,2,...,N};

[0049] Among them, t i Indicates the timestamp of the behavior; p i Indicates the user's position in three-dimensional space; a i Indicates the behavior type; r i Indicates the result of behavior; i Indicates the duration of the behavior; e i represents the emotional state; N represents the total number of behaviors;

[0050] Through the deep behavioral feature encoder f enc , convert the behavior sequence into an embedding vector B (u) :B (u) =[b1,b2,...,b i ,...,b N ],b i =f enc (t i ,p i ,a i ,r i ,s i ,e i );

[0051] Among them, b i Represents the embedding representation of the i-th behavior;

[0052] S72. Build a user interest preference profile and output a set of user-frequently visited areas Content type weight vector and the task preference vector parameter

[0053] S73, recommend the next task, interactive role, virtual route to the user, and output the user preference R (u) :

[0054]

[0055] Among them, r represents the recommended item; sim() represents the cosine similarity function; r content Represents the cultural content vector corresponding to the recommended item; r task represents the task type vector corresponding to the recommended item; λ1, λ2, and λ3 represent the weighted coefficients of the recommendation factor; geoscore() represents the geographical proximity score between the recommended item and the user's hotspot area; R all Represents the set of all recommended items;

[0056] S74. The user's actual feedback on the recommended content will be used to update the interest weight:

[0057] in, Represents the content preference vector for the next recommendation period t+1; represents the content preference vector at the current time t; Δω represents the recommendation feedback correction vector; η represents the learning rate.

[0058] Preferably, the process of constructing a user interest preference profile is as follows:

[0059] S81, hotspot area identification: using DBSCAN clustering algorithm to cluster trajectory coordinates {p i} Perform spatial density analysis:

[0060] Among them, ε represents the maximum neighborhood distance; MinPts represents the minimum number of cluster points; Indicates the set of areas frequently visited by users;

[0061] S82. Constructing content type weight vector

[0062]

[0063] Among them, w k Represents the user's interest weight in the k-th cultural content; represents the number of interactions with the k-th cultural content; represents the total duration of stay on the k-th type of cultural content; represents the proportion of positive emotions during the interaction with the k-th cultural content; α', β', and λ' represent weight coefficients; K represents the total number of different types of text content;

[0064] S83. Extract the completion efficiency and interest level of different types of tasks to form task preference vector parameters

[0065] Where M represents the total number of task types; t j represents the user's preference score for the j-th task.

[0066] The technical solution of the present invention is a cultural and tourism metaverse virtual character collaboration system based on large-space interaction, which is used to execute a cultural and tourism metaverse virtual character collaboration method based on large-space interaction, including:

[0067] Large space perception module, used to achieve high-precision position tracking of multiple people in large spaces using multi-source positioning technology;

[0068] A virtual character intelligent collaboration module, which is used to conduct multi-level interactions between tourist characters, tour guide characters, and local cultural NPCs through an AI-driven virtual character control system;

[0069] The cultural content generation and integration module is used to dynamically model elements in real cultural tourism scenes into interactive virtual cultural scenes based on digital twin and semantic parsing technology;

[0070] The rendering and data synchronization module is used to utilize cloud computing and edge rendering to achieve rendering consistency and real-time interactive response when multiple people enter the large-scale metaverse at the same time;

[0071] User behavior analysis and recommendation module, used to analyze user preference behavior paths and generate personalized recommendation tasks;

[0072] The cross-module collaborative control and security management module, the control center of the system, coordinates data exchange and process control between various functional modules, and reviews user device permissions based on the user device authentication procedure to ensure the security of system data and the privacy of user interactions.

[0073] Preferably, the review process for reviewing user device permissions based on the user device authentication procedure is as follows:

[0074] S101, the user device obtains its own device ID, i.e. D i , calculate the application parameter AP=H(D i )∈G2;

[0075] Calculate device connection parameter LS = (AP) s ∈G2, {LS, D i}Transmit to the cross-module collaborative control and security management module and apply for device access application;

[0076] Where H is a predefined hash function, H: {0,1} * →G2; G2 is a predefined prime number p-order cyclic group, the generator is g2; s is a predefined application permission, s∈Z p ; Z p is an integer group of order p;

[0077] S102, when the cross-module collaborative control and security management module receives the user equipment access application, it receives {LS, D i}, calculate the link verification parameter LV=H(D i );

[0078] Calculate the first-level link parameter LPⅠ=e(AA,LV);

[0079] Calculate the secondary link parameter LPⅡ=e(g1,LS);

[0080] Where g1 is the generator of the predefined prime p-order cyclic group G1; AA represents the audit authority parameter, AA = (g1) s ∈G1; e is a predefined bilinear mapping, e:G1×G2→G T ; G T is a prime number p-order cyclic group;

[0081] If LPⅠ=LPⅡ, the user device is allowed to connect to the system, which ensures the legitimacy of the device; otherwise, it is rejected.

[0082] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0083] The present invention designs a virtual character collaboration system and method for the cultural and tourism metaverse based on large-space interaction. Through multimodal identity authentication and bilinear mapping device verification technology, the security of the cultural and tourism metaverse system and the legitimacy of user identity are significantly improved, and unauthorized device access is effectively prevented; based on large-space positioning and semantic point cloud dynamic modeling technology, high-precision real-time synchronization of virtual and real fusion scenes is achieved, and the dynamic task collaboration graph of the graph neural network and the AI ​​plot generation engine are combined to improve the efficiency of multi-role collaboration and adaptive evolution of plot branches; through the construction of user portraits through spatiotemporal behavior analysis and deep feature coding, combined with the weighted recommendation algorithm to dynamically optimize task paths and content presentation, the accuracy of personalized experience and user immersion are enhanced; and based on the integration of cloud rendering and edge computing technology, efficient resource allocation and rendering consistency in multi-user scenarios are guaranteed, taking into account the real-time and stability of the system, and promoting the overall improvement of the interactivity and cultural communication efficiency of the cultural and tourism metaverse. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 This is a flowchart of a method for collaborative virtual character collaboration in the cultural tourism metaverse based on large-space interaction proposed by the present invention;

[0085] Figure 2 This is a system architecture diagram of a cultural and tourism metaverse virtual character collaboration system based on large-space interaction proposed by the present invention. DETAILED DESCRIPTION

[0086] Example 1, as Figure 1 As shown, the present invention proposes a method for coordinating virtual characters in a cultural and tourism metaverse based on large-space interaction, and its specific implementation steps are as follows:

[0087] S1. When receiving a user entry request, automatically identify the user's identity attributes (including but not limited to tourists, tour guides, and system AI characters), and establish a positioning data model of the user in a large space through the device worn by the user (including but not limited to AR / VR glasses and mobile terminals), and obtain the user's spatial position and orientation in real time. The specific implementation process is as follows:

[0088] S11, collect user multimodal features through camera, microphone and user terminal, including face image I f , voice sample V s 、Device ID(D i ) and the user enters account credentials A k , calculate the user identity matching degree P u :

[0089]

[0090] in, Represents the user's real-time face image I f With registered image Similarity score of Represents the user voice sample V s Register sample with the system sim() represents the similarity function, and this embodiment adopts the cosine similarity function; Indicates device ID and registration The degree of match between Indicates that the user enters account credentials A k and registration account credentials The matching degree between them; Bool() represents the judgment function, which outputs 1 if the condition is met, otherwise it outputs 0; α, β, γ, η represent weight coefficients, which are used to adjust the importance of each recognition dimension, α+β+γ+η=1;

[0091] Extract the set trusted identity threshold T id , if Pu ≧T id ,determine the user's identity is legitimate, proceed to the next step, and perform device authentication based on the user device authentication procedure;

[0092] S12. If the user device is authenticated, obtain the initial position value (x0, y0, z0) and the attitude angle information and the local coordinate transformation trajectory matrix T slam (t), at this time t=0, that is, T slam (0);

[0093] Among them, (x0, y0, z0) represents the three-dimensional coordinate position of the user at initialization; φ0 represents the roll angle (roll), which represents the rotation around the X axis; ψ0 represents the pitch angle (pitch), which represents the rotation around the Y axis; ω0 represents the yaw angle (yaw), which represents the rotation around the Z axis; T slam (0) represents the spatial transformation matrix when visual SLAM is initialized;

[0094] Construct the initial vector of the spatial state S0: S0=[x0,y0,z0,φ0,ψ0,ω0,T slam (0)];

[0095] S13, automatically matching user role label U according to the recognition result tag , and get the initialization role state R0;

[0096] S14. After completing the role creation, the user access signal is released and the collaborative space status map is updated:

[0097] G t =G t-1 ∪{R0,S0,T role};

[0098] Among them, G t represents the collaborative space map after the user accesses, that is, the overall collaborative state of the virtual space at the current time t; G t-1 represents the spatial collaborative state map before the user accesses, that is, the spatial collaborative state map at time t-1; T role Represents the character's initial task set (including but not limited to tour routes and interaction goals).

[0099] S2. Dynamically construct an interactive, renderable, and collaborative virtual space replica based on the real-world space structure, user space status, and role tags, and synchronize multiple user views, status, and behavior models in real time. Specifically:

[0100] S21, collect real-time perception data {scene point cloud P cloud and structural features T struct}, fusion of predefined semantic graph Kscene With real-time perception data, generate semantic point cloud structure S env :

[0101] S env =f map (P cloud ,T struct ,K scene );

[0102] Among them, f map () represents the fusion function, which generates the structure through the deep semantic network (DGCNN);

[0103] S22, according to the character state R0 and the spatial structure S env , bind the virtual character to the specific semantic node in the virtual environment to form a behavior-driven graph G act :G act =f bind (R0,S env );

[0104] Among them, f bind () represents the semantic matching function, which matches the interactive area according to the role identity (including but not limited to tour guide → explanation point, tourist → AR ticket entrance); G act Represents the behavior-driven graph, i.e., the role-scene interaction graph, which describes the relationship topology of all users and objects in the current virtual space;

[0105] The behavior graph can be defined as a set of triples as follows:

[0106] G act ={(R i ,s j ,a k )|R i ∈R on ,s j ∈S env ,a k ∈ActionSet};

[0107] Among them, R i Represents a role entity; s j Represents a spatial semantic node; a k Indicates interactive action; R on Represents the set of all currently online roles; ActionSet represents the preset interaction set;

[0108] S23: Perform real-time rendering and state synchronization of the spatial model, specifically:

[0109] S2301, the semantic point cloud structure S env and behavior-driven graph G actMapped to renderable model M render First, the semantic parsing and spatial stratification of point cloud data are performed by label-driven voxel reconstruction method to construct semantic voxel structure; secondly, the Marching Cubes grid reconstruction algorithm is used to generate the topological geometric structure M geo , and implements differentiated reconstruction strategies for different structures (such as ground, buildings, cultural relics, etc.) according to semantic labels; then combines semantic labels to map materials and textures to form diverse appearances that meet the requirements of realistic rendering; then based on the character behavior graph G act Dynamically mount interactive event points to bind the virtual environment to user behavior logic to achieve controllable interactivity; finally, build a multi-level LOD model based on point cloud information entropy to achieve automatic switching of near and far views and resource optimization, thereby outputting a structurally complete, behavior-driven, and efficiently rendered 3D visual model M render ;

[0110] S2302, define all online role sets R on :{(R i )|i=0,1,..,n}, construct the synchronization state matrix S sync (t):

[0111]

[0112] Wherein, i represents the sequence number; represents the identity ID of the i-th character at time t; Represents the position coordinates of the i-th character at time t in three-dimensional space; represents the orientation of the i-th character at time t; represents the action performed by the i-th character at time t;

[0113] S24. All statuses are broadcast to all online clients in each frame period (every 33 ms in this embodiment).

[0114] S3. In the Metaverse large-space cultural tourism scene, users initialize their role information through identity recognition (S1), complete spatial modeling and behavioral graph association (S2), and further conduct intelligent collaboration and task-oriented interaction among multiple users / virtual characters to enhance user interactivity and immersion. Specifically:

[0115] S31. By modeling each character's collaborative capabilities and current spatial state, a dynamic task perception mechanism is implemented, and the types of tasks that the character can participate in are inferred. This means dynamically identifying the set of tasks that each character can participate in within the current semantic space.

[0116] Among them, R i Indicates the current i-th role entity; f task() represents the task perception function, which generates the task perception vector based on the graph neural network (GAT); Indicates the current role's perceptible task list; G act Represents a behavior-driven diagram; S env Represents the semantic point cloud structure, that is, the virtual semantic space modeling structure;

[0117] S32. Decompose complex goals into multiple subtasks according to the current space mission requirements k =f decomp (Task env ) and automatically distribute them to multiple role nodes to form a collaborative execution graph, that is, to build a task collaborative graph based on the collaborative capabilities of multiple roles:

[0118]

[0119] Among them, Task env Represents the original macro task set triggered in the current virtual scene; f decomp () represents the task decomposition function, including but not limited to decomposing macro tasks based on task dependency graph, behavior tree or conditional task graph (HTN); Task k Represents a subtask, the kth decomposed execution unit task; G coop Represents a collaborative graph, where each task and its corresponding set of execution roles form a graph structure to guide the execution of collaborative behaviors; Indicates the jth available task Task k Virtual characters from the current online character set R on , and have the ability to match the task; Representing a role The set of tasks that can be perceived or completed;

[0120] S33. Accordingly: the roles perform behaviors according to the assigned task collaboration diagram, while monitoring the behavior status of each role in real time and performing global feedback updates.

[0121] S4. Obtain multimodal interactions (including but not limited to voice, movement, expression, and touch) between users and characters, events, and scenes in the virtual cultural tourism space, and generate customized interactive story experiences in real time based on user behavior trajectories through an AI-driven story dynamic evolution engine. Specifically:

[0122] S41. Identify the user's multimodal input and perform time series alignment and fusion modeling:

[0123]

[0124] in, Represents the comprehensive interaction state vector of any user u at time t, which is used to judge the subsequent plot evolution; Represents the semantic vector of the speech input signal of user u at time t obtained through speech recognition / natural language processing; Represents the action data of user u, including behavior labels; The spatial motion vector representing user i in the virtual space, including but not limited to real-time position, direction, and speed; Represents user facial expression recognition data and extracts emotional features, including but not limited to happiness, surprise, and confusion; f fuse () represents the multimodal fusion function, which realizes the temporal alignment and fusion encoding of different modalities based on the temporal attention network;

[0125] S42. Based on the multimodal input state vector, task progress status, and behavior trajectory, the AI ​​plot generation engine is called to perform personalized plot dynamic reasoning and content generation. Specifically:

[0126] S4201, call the embedded cultural tourism plot map G plot , containing event node V plot With conditional edge E cond :G plot =(V plot ,E cond );

[0127] S4202, combined with historical trajectory and task status Extract the user's current plot location node C curr ;

[0128] S4203, call the plot reasoning function to generate the next node state

[0129] Among them, f story () represents the graph attention network (GAT) + GPT fine-tuning module, which infers the plot branch according to the current state vector and task conditions;

[0130] Based on this, the status of the plot node is output, including but not limited to [scene ID], [character lines], [plot description], and [whether to branch];

[0131] S43, immersive rendering output and multi-role collaborative feedback, specifically:

[0132] Rendering task combination calculation, generate visual rendering queue R according to the status of plot nodes render :

[0133]

[0134] in, Represents the scene rendering environment parameter set. The environment parameters used are from the semantic point cloud structure S env The feature vector obtained by parsing or converting includes but is not limited to parameters of lighting, weather, time, and background sound environment; R coop Indicates the behavior model (preloaded) of the current role to be participated, including but not limited to action sequence, voice intonation parameters, and expression changes; f render () represents the rendering task generation and resource scheduling function, which is responsible for converting the plot state, environment state and character behavior input into a data instruction set for the rendering engine (including but not limited to action trajectory, scene map, sound path);

[0135] Accordingly: output rendering instruction set, including but not limited to scene loading instructions, character control sequence, interaction node definition, resource scheduling queue.

[0136] S5. Based on the user's behavioral trajectory data, interaction records, multimodal features, task completion status, and preference history in the virtual space of the Cultural Tourism Metaverse, a personalized profile is formed through feature mining and association modeling. Combined with AI recommendation algorithms, task paths, content presentation, and virtual character interaction methods are dynamically optimized to promote the adaptive evolution of the system and enhance the experience. Specifically:

[0137] S51, behavior trajectory collection and multi-dimensional feature modeling, define each user's behavior trajectory in the virtual space as a temporal event sequence T (u) :T (u) ={(t i ,p i ,a i ,r i ,s i ,e i )|i=1,2,...,N};

[0138] Among them, t i Indicates the timestamp of the behavior; p i Indicates the user's position in three-dimensional space; a i Indicates the type of behavior (including but not limited to movement, interaction, and task completion); i Indicates the result of the behavior (including but not limited to task success / failure); i Indicates the duration of the behavior (i.e., the degree of interest); e i represents the emotional state (including but not limited to happy, bored, and nervous); N represents the total number of behaviors;

[0139] Through the deep behavioral feature encoder f enc , convert the behavior sequence into an embedding vector B (u) :B (u)=[b1,b2,...,b i ,...,b N ],b i =f enc (t i ,p i ,a i ,r i ,s i ,e i );

[0140] Among them, b i Represents the embedding representation of the i-th behavior;

[0141] S52. User interest preference profile construction, specifically:

[0142] S5201, hotspot area identification (spatial preference modeling): using DBSCAN clustering algorithm to cluster trajectory coordinates {p i} Perform spatial density analysis:

[0143] Among them, ε represents the maximum neighborhood distance; MinPts represents the minimum number of cluster points, which is used to eliminate noise points; It represents the set of areas frequently visited by users, representing the hotspots of spatial preferences;

[0144] S5202: Constructing content type weight vector

[0145]

[0146] Among them, w k Represents the user's interest weight in the kth category of cultural content (including but not limited to ancient poetry, architecture, and intangible cultural heritage skills); represents the number of interactions with the k-th cultural content; represents the total duration of stay on the k-th type of cultural content; represents the proportion of positive emotions during the interaction with the k-th cultural content; α', β', and λ' represent weight coefficients; K represents the total number of different types of text content;

[0147] S5203. Extract the completion efficiency and interest level of different types of tasks to form task preference vector parameters

[0148] Where M represents the total number of task types (including but not limited to plot exploration, interactive question-answering, and prop collection); j Indicates the user's preference rating for the j-th task (including but not limited to completion rate and score);

[0149] S53. Recommend the next task, interactive character, or virtual route to the user, so that it matches their spatial, content, and task preferences:

[0150]

[0151] Among them, r represents the recommended item; sim() represents the cosine similarity function; r content Represents the cultural content vector corresponding to the recommended item; r task represents the task type vector corresponding to the recommended item; λ1, λ2, and λ3 represent the weighted coefficients of the recommendation factor, reflecting the priority of the current user preference; geoscore represents the geographical proximity score between the recommended item and the user's hot spot area; R all Represents the set of all recommended items;

[0152] S54. The user's actual feedback on the recommended content will be used to update the interest weight:

[0153] in, Represents the content preference vector for the next recommendation period t+1; represents the content preference vector at the current time t; Δω represents the recommendation feedback correction vector; η represents the learning rate, which controls the update amplitude.

[0154] Example 2, as Figure 2 As shown, the present invention proposes a cultural and tourism metaverse virtual character collaboration system based on large-space interaction, which is used to execute a cultural and tourism metaverse virtual character collaboration method based on large-space interaction proposed in Example 1, including: a large-space perception module, a virtual character intelligent collaboration module, a cultural content generation and fusion module, a rendering and data synchronization module, a user behavior analysis and recommendation module, and a cross-module collaborative control and security management module.

[0155] Large-space perception module, using multi-source positioning technologies {including but not limited to UWB (Ultra Wide Band, i.e. wireless carrier communication technology), IMU (Inertial Measurement Unit, inertial measurement unit), and visual SLAM (Simultaneous Localization and Mapping)} to achieve high-precision position tracking of multiple people in large spaces;

[0156] The avatar intelligent collaboration module uses an AI-driven avatar control system to enable multi-level interactions between tourist characters, tour guide characters, and local cultural NPCs, including but not limited to situational dialogue, collaborative tasks, and cultural knowledge Q&A.

[0157] The cultural content generation and integration module, based on digital twin and semantic parsing technology, dynamically models elements in real cultural tourism scenes (including but not limited to architecture and customs) into interactive virtual cultural scenes;

[0158] The rendering and data synchronization module uses cloud computing and edge rendering to ensure consistent scene rendering and real-time interactive response when multiple people enter the large-scale metaverse at the same time;

[0159] User behavior analysis and recommendation module uses deep learning models to analyze user preference behavior paths and generate personalized recommendation tasks;

[0160] The cross-module collaborative control and security management module serves as the control center of the system, coordinating data exchange and process control between various functional modules to ensure operational stability and collaboration in a multi-role, multi-user, and multi-terminal environment. It also reviews user device permissions based on the user device authentication procedure to ensure the security of system data and the privacy of user interactions.

[0161] Example 3, a cultural tourism metaverse virtual character collaboration system based on large-space interaction, which also includes a user device authentication program, and its specific implementation steps are as follows:

[0162] S1. The user device obtains its own device ID, i.e. D i , calculate the application parameter AP=H(D i )∈G2;

[0163] Calculate device connection parameter LS = (AP) s ∈G2, {LS, D i}Transmit to the cross-module collaborative control and security management module and apply for device access application;

[0164] Where H is a predefined hash function, H: {0,1} * →G2; G2 is a predefined prime number p-order cyclic group, the generator is g2; s is a predefined application permission, s∈Z p ; Z p is an integer group of order p.

[0165] S2, when the cross-module collaborative control and security management module receives the user equipment access application, it receives {LS, D i}, calculate the link verification parameter LV=H(D i );

[0166] Calculate the first-level link parameter LPⅠ=e(AA,LV);

[0167] Calculate the secondary link parameter LPⅡ=e(g1,LS);

[0168] Where g1 is the generator of the predefined prime p-order cyclic group G1; AA represents the audit authority parameter, AA = (g1) s ∈G1; e is a predefined bilinear mapping, e:G1×G2→G T ; G T is a prime number p-order cyclic group;

[0169] If LPⅠ=LPⅡ, the user device is allowed to connect to the system; otherwise, it is rejected.

[0170] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for collaborative virtual characters in a cultural and tourism metaverse based on large-space interaction, characterized in that: The specific implementation steps include the following: S1. Identify user identity and verify device legitimacy through multimodal data, build an initial state vector based on spatial positioning and posture information, automatically match role labels and update the collaborative space map, and perform user role initialization and dynamic synchronization with the virtual environment; S2. Generate a semantic point cloud structure by fusing real-time perception data with a semantic graph. Construct a behavior-driven graph by binding semantic nodes to character states. Generate a renderable model using voxel reconstruction and meshing algorithms. Perform real-time rendering of the spatial model, synchronize its state, and perform periodic broadcasting. S3: Dynamically perceive the role task capabilities through graph neural networks and generate a set of participating tasks. Combined with the task decomposition algorithm, the macro goal is broken down into subtasks and then assigned to multiple role nodes to build a task collaboration graph, monitoring the execution status and global feedback updates in real time. S4: Construct user interaction states through multimodal input recognition and time series fusion, call the cultural tourism plot graph, and infer personalized plot branches based on the graph attention network and GPT module to generate node states. Finally, convert the environmental parameters and character behavior model into a rendering instruction set through the rendering task scheduling function. S5. By collecting user behavior trajectories and multimodal data, using DBSCAN clustering and deep feature coding to construct spatial preference and content interest vectors, extracting task completion efficiency and emotional characteristics to form personalized portraits, combining the AI ​​recommendation algorithm of cosine similarity and geographic proximity to dynamically optimize task paths and interaction methods, and updating user preference weights through a feedback mechanism to complete the improvement of the adaptive experience of the cultural and tourism metaverse.

2. A method for coordinating virtual characters in a cultural and tourism metaverse based on large-space interaction according to claim 1, characterized in that: The identification process of user identity through multimodal data is as follows: Collect user multimodal features, including face image I f , voice sample V s 、Device ID(D i ) and the user enters account credentials A k , calculate the user identity matching degree P u : in, Represents the user's real-time face image I f With registered image Similarity score of Represents the user voice sample V s Register sample with the system The matching degree; sim() represents the cosine similarity function; Indicates device ID and registration The degree of match between Indicates that the user enters account credentials A k and registration account credentials The matching degree between them; Bool() represents the judgment function, which outputs 1 if the condition is met, otherwise it outputs 0; α, β, γ, η represent weight coefficients, α+β+γ+η=1.

3. The method for coordinating virtual characters in a cultural and tourism metaverse based on large-space interaction according to claim 1 is characterized in that: The synchronization process of real-time rendering and state synchronization of the spatial model is as follows: S31, collect real-time perception data {scene point cloud P cloud and structural features T struct }, fusion of predefined semantic graph K scene With real-time perception data, generate semantic point cloud structure S env : S env =f map (P cloud ,T struct ,K scene ); Among them, f map () represents the fusion function, which generates the structure through the deep semantic network DGCNN; S32, according to the character state R0 and the spatial structure S env , bind the virtual character to the specific semantic node in the virtual environment to form a behavior-driven graph G act :G act =f bind (R0,S env ); Among them, f bind () represents the semantic matching function; G act Represents a behavior-driven diagram; The behavior graph can be defined as a set of triples as follows: G act ={(R i ,s j ,a k )|R i ∈R on ,s j ∈S env ,a k ∈ActionSet}; Among them, R i Represents a role entity; s j Represents a spatial semantic node; a k Indicates interactive action; R on Represents the set of all currently online roles; ActionSet represents the preset interaction set; S33, perform real-time rendering and state synchronization of the spatial model: convert the semantic point cloud structure S env and behavior-driven graph G act Mapped to renderable model M render , and define the set of all online roles R on :{(R i )|i=0,1,..,n}, construct the synchronization state matrix S sync (t): Wherein, i represents the sequence number; represents the identity ID of the i-th character at time t; Represents the position coordinates of the i-th character at time t in three-dimensional space; represents the orientation of the i-th character at time t; represents the action performed by the i-th character at time t.

4. A method for coordinating virtual characters in a cultural and tourism metaverse based on large-space interaction according to claim 3, characterized in that: Renderable Model M render The generation process is: S41. Perform semantic parsing and spatial stratification of point cloud data through label-driven voxel reconstruction method to construct semantic voxel structure. S42, using Marching Cubes mesh reconstruction algorithm to generate topological geometry M geo , and perform differentiated reconstruction strategies for different structures according to semantic labels; S43, combining semantic labels to map materials and textures, forming a variety of appearances that meet the requirements of realistic rendering; then according to the character behavior graph G act Dynamically mount interactive event points to bind the virtual environment to user behavior logic to achieve controllable interactivity; S44. Build a multi-level LOD model based on point cloud information entropy to achieve automatic switching between near and far views and resource optimization, thereby outputting a structurally complete, behavior-driven, and highly efficient 3D visual model M. render .

5. The method for coordinating virtual characters in a cultural and tourism metaverse based on large-space interaction according to claim 1 is characterized in that: The process of constructing a task collaboration graph is as follows: S51. By modeling each character's collaboration capabilities and current spatial state, a dynamic task perception mechanism is implemented, and the types of tasks that the character can participate in are inferred. This means dynamically identifying the set of tasks that each character can participate in within the current semantic space. Among them, R i Indicates the current i-th role entity; f task () represents the task perception function, which generates the task perception vector based on the graph neural network GAT; Indicates the current role's perceptible task list; G act Represents a behavior-driven diagram; S env Represents a semantic point cloud structure; S52. Decompose complex goals into multiple subtasks according to the current space mission requirements k =f decomp (Task env ) and automatically distribute them to multiple role nodes to form a collaborative execution graph; That is, building a task collaboration map based on the collaborative capabilities of multiple roles: Among them, Task env Represents the original macro task set triggered in the current virtual scene; f decomp () represents the task decomposition function; Task k Represents a subtask, the kth decomposed execution unit task; G coop represents a collaborative graph; Indicates the jth available task Task k virtual characters; Representing a role The set of tasks that can be perceived or completed.

6. The method for coordinating virtual characters in a cultural and tourism metaverse based on large-space interaction according to claim 1 is characterized in that: The node status generation process is as follows: S61, call the embedded cultural tourism plot map G plot , containing event node V plot With conditional edge E cond :G plot =(V plot ,E cond ); S62. Combined with historical trajectory and task status Extract the user's current plot location node C curr ; S63, call the plot reasoning function to generate the next node state Among them, f story () represents the GAT+GPT fine-tuning module based on the graph attention network, which infers the plot branch according to the current state vector and task conditions; S64. Output plot node status 7. The method for coordinating virtual characters in a cultural and tourism metaverse based on large-space interaction according to claim 1 is characterized in that: The user preference weight update process is as follows: S71, behavior trajectory collection and multi-dimensional feature modeling, define each user's behavior trajectory in the virtual space as a temporal event sequence T (u) :T (u) ={(t i ,p i ,a i ,r i ,s i ,e i )|i=1,2,...,N}; Among them, t i Indicates the timestamp of the behavior; p i Indicates the user's position in three-dimensional space; a i Indicates the behavior type; r i Indicates the result of behavior; i Indicates the duration of the behavior; e i represents the emotional state; N represents the total number of behaviors; Through the deep behavioral feature encoder f enc , convert the behavior sequence into an embedding vector B (u) :B (u) =[b1,b2,...,b i ,...,b N ],b i =f enc (t i ,p i ,a i ,r i ,s i ,e i ); Among them, b i Represents the embedding representation of the i-th behavior; S72. Build a user interest preference profile and output a set of user-frequently visited areas Content type weight vector and the task preference vector parameter S73, recommend the next task, interactive role, virtual route to the user, and output the user preference R (u) : Among them, r represents the recommended item; sim() represents the cosine similarity function; r content Represents the cultural content vector corresponding to the recommended item; r task represents the task type vector corresponding to the recommended item; λ1, λ2, and λ3 represent the weighted coefficients of the recommendation factor; geoscore() represents the geographical proximity score between the recommended item and the user's hotspot area; R all Represents the set of all recommended items; S74. The user's actual feedback on the recommended content will be used to update the interest weight: in, Represents the content preference vector for the next recommendation period t+1; represents the content preference vector at the current time t; Δω represents the recommendation feedback correction vector; η represents the learning rate.

8. The method for coordinating virtual characters in a cultural and tourism metaverse based on large-space interaction according to claim 7 is characterized in that: The process of constructing a user interest preference portrait is as follows: S81, hotspot area identification: using DBSCAN clustering algorithm to cluster trajectory coordinates {p i } Perform spatial density analysis: Among them, ε represents the maximum neighborhood distance; MinPts represents the minimum number of cluster points; Indicates the set of areas frequently visited by users; S82. Constructing content type weight vector Among them, w k Represents the user's interest weight in the k-th cultural content; represents the number of interactions with the k-th cultural content; represents the total duration of stay on the k-th type of cultural content; represents the proportion of positive emotions during the interaction with the k-th cultural content; α', β', and λ' represent weight coefficients; K represents the total number of different types of text content; S83. Extract the completion efficiency and interest level of different types of tasks to form task preference vector parameters Where M represents the total number of task types; t j represents the user's preference score for the j-th task.

9. A cultural tourism metaverse virtual character collaboration system based on large space interaction, which is used to execute a cultural tourism metaverse virtual character collaboration method based on large space interaction according to any one of claims 1 to 8, characterized in that: include: Large space perception module, used to achieve high-precision position tracking of multiple people in large spaces using multi-source positioning technology; A virtual character intelligent collaboration module, which is used to conduct multi-level interactions between tourist characters, tour guide characters, and local cultural NPCs through an AI-driven virtual character control system; The cultural content generation and integration module is used to dynamically model elements in real cultural tourism scenes into interactive virtual cultural scenes based on digital twin and semantic parsing technology; The rendering and data synchronization module is used to utilize cloud computing and edge rendering to achieve rendering consistency and real-time interactive response when multiple people enter the large-scale metaverse at the same time; User behavior analysis and recommendation module, used to analyze user preference behavior paths and generate personalized recommendation tasks; The cross-module collaborative control and security management module, the control center of the system, coordinates data exchange and process control between various functional modules, and reviews user device permissions based on the user device authentication procedure to ensure the security of system data and the privacy of user interactions.

10. The large-space interactive cultural and tourism metaverse virtual character collaboration system according to claim 9 is characterized in that: The review process for user device permissions based on the user device authentication procedure is as follows: S101, the user device obtains its own device ID, i.e. D i , calculate the application parameter AP=H(D i )∈G2; Calculate device connection parameter LS = (AP) s ∈G2, {LS, D i }Transmit to the cross-module collaborative control and security management module and apply for device access application; Where H is a predefined hash function, H: {0,1} * →G2; G2 is a predefined prime number p-order cyclic group, the generator is g2; s is a predefined application permission, s∈Z p ; Z p is an integer group of order p; S102, when the cross-module collaborative control and security management module receives the user equipment access application, it receives {LS, D i }, calculate the link verification parameter LV=H(D i ); Calculate the first-level link parameter LPⅠ=e(AA,LV); Calculate the secondary link parameter LPⅡ=e(g1,LS); Where g1 is the generator of the predefined prime p-order cyclic group G1; AA represents the audit authority parameter, AA = (g1) s ∈G1; e is a predefined bilinear mapping, e:G1×G2→G T ; G T is a prime number p-order cyclic group; If LPⅠ=LPⅡ, the user device is allowed to connect to the system, which ensures the legitimacy of the device; otherwise, it is rejected.

Images