Virtual reality content generation and staring interaction system based on artificial intelligence

Abstract

The invention relates to the technical field of virtual reality and artificial intelligence fusion, and discloses an artificial intelligence-based virtual reality content generation and staring interaction system, which comprises a multi-modal sensing module, a virtual reality content generation module and a staring interaction module, a gazing intention analysis module; an AI dynamic content generation module; a virtual reality rendering module; a gazing interaction control module; a user preference modeling module; a rule constraint engine; a data storage module; an edge calculation scheduling module; a security verification module; according to the method, multi-dimensional analysis of the gaze intention is innovatively realized, the limitation that an existing system can only identify simple gaze behaviors is broken through, gaze data and physiological features are fused, three types of gaze intentions are accurately distinguished, confidence verification and historical data correction are combined, the analysis accuracy is remarkably improved, and the error interaction probability is reduced; the method has the dynamic personalized content generation capability, content personalization is ensured, logic consistency is ensured, and the problem of homogenization is solved.

Description

Technical Field

[0001] This invention belongs to the field of virtual reality and artificial intelligence integration technology, specifically a virtual reality content generation and gaze interaction system based on artificial intelligence. Background Technology

[0002] With the rapid development of VR and AI technologies, the efficiency of virtual reality content generation and interactive experience have become core pursuits in the industry. In existing virtual reality systems, content generation largely relies on manual modeling or fixed scripts, resulting in severe content homogenization, long generation cycles, and difficulty in adapting to personalized user needs. Regarding interaction methods, gaze interaction, as a natural and intuitive means of interaction, has been gradually applied to VR devices. However, the gaze interaction accuracy of existing systems is low, only achieving simple gaze triggering functions and failing to accurately interpret the user's gaze intent. Furthermore, the lack of deep linkage between content generation and gaze interaction leads to insufficient user immersion.

[0003] Some existing technologies attempt to combine AI with gaze-based interaction. For example, Meta's smart glasses patent adjusts content presentation using eye-tracking data, but it can only perform simple content optimization based on interest levels and lacks multi-dimensional analysis of gaze intent. Apple's XR interaction patent triggers a digital assistant through gaze, but content generation relies on preset scripts and cannot achieve dynamic personalized generation. Meanwhile, existing systems still have shortcomings in rendering optimization, computational latency control, and content logic consistency, making it difficult to meet the needs of high-end VR application scenarios.

[0004] Therefore, there is an urgent need for a system that can accurately analyze gaze intent, dynamically generate personalized content, and achieve deep interaction between content and gaze, in order to overcome the shortcomings of existing technologies.

[0005] Based on this, an artificial intelligence-based virtual reality content generation and gaze interaction system is designed. Summary of the Invention

[0006] In view of the above situation and to overcome the shortcomings of the prior art, the present invention provides an artificial intelligence-based virtual reality content generation and gaze interaction system, which effectively solves the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a virtual reality content generation and gaze interaction system based on artificial intelligence, comprising a multimodal perception module for real-time collection of user gaze data, physiological characteristic data, and virtual reality environment status data. The gaze data includes gaze point coordinates, gaze duration, and eye movement trajectory. The physiological characteristic data includes pupil dilation and facial micro-expression features. The environment status data includes the current virtual scene element type and scene rendering parameters. The gaze intent parsing module is communicatively connected to the multimodal perception module. It has a built-in pre-trained intent recognition model for fusing and analyzing the collected gaze data and physiological feature data, and outputting user gaze intent results. The intent results include three categories: interest enhancement, content switching, and interaction triggering. The AI ​​dynamic content generation module is connected to the gaze intent parsing module and integrates a diffusion model and a large language model. It is used to generate 3D virtual scenes, character actions and interactive content that meet the requirements based on gaze intent results, user preference data and virtual world rules, and supports real-time iterative updates of content. The virtual reality rendering module is communicatively connected to the AI ​​dynamic content generation module and the gaze interaction control module, respectively. It adopts dynamic gaze point rendering technology to render the generated virtual content in layers, prioritizes improving the rendering accuracy of the user's gaze area, and synchronously outputs immersive images adapted to VR devices. The gaze interaction control module is communicatively connected to the gaze intent parsing module and the virtual reality rendering module to construct a hierarchical gaze interaction mechanism. It executes corresponding interaction operations based on the gaze intent results, including gaze enhancement operations, gaze switching operations, and gaze triggering operations. The user preference modeling module is communicatively connected to the multimodal perception module and the AI ​​dynamic content generation module, respectively, and is used to build a personalized preference model based on the user's historical gaze data and interaction records, and update the user preference tags and weights in real time. The rule constraint engine communicates with the AI ​​dynamic content generation module and has built-in virtual world physical rules, content logic rules and interaction security rules to perform compliance verification on the AI-generated content and eliminate logical contradictions and security risks. The data storage module is used to store multimodal perception data, user preference model data, AI-generated content data, and interaction log data, and supports real-time reading and writing and incremental updates of data. The edge computing scheduling module communicates with each functional module and is used to dynamically schedule computing tasks such as AI content generation, gaze intent parsing, and rendering processing, and allocate computing resources to edge nodes to reduce processing latency. The security verification module is used to perform security verification on user identity information, interaction commands, and generated content to prevent unauthorized access and the generation of malicious content.

[0008] Preferably, the multimodal sensing module includes an infrared eye-tracking unit, a physiological sensing unit, and an environmental acquisition unit, wherein: The infrared eye-tracking unit uses infrared LEDs and a 120 frames per second high-speed camera arranged on the edge of the VR headset lens to capture the user's eye movement data and output the gaze point coordinates and eye trajectory with an accuracy of ≤0.3°. The physiological sensing unit integrates miniature physiological sensors to collect pupil dilation data (sampling frequency 50Hz) and facial micro-expression feature point data in real time, and obtains emotion-related features through feature extraction algorithms; The environment acquisition unit synchronously acquires the element list, rendering resolution, and frame rate parameters of the current virtual scene, providing data support for content generation and rendering optimization.

[0009] Preferably, the working steps of the gaze intent parsing module include: Step S11: Receive gaze data and physiological feature data output by the multimodal perception module, perform data preprocessing, remove abnormal data points, and standardize; Step S12: Input the preprocessed data into the pre-trained intent recognition model. The model adopts a CNN and Transformer fusion architecture to extract the correlation features between gaze features and physiological features. Step S13: Output the intent classification result based on the associated features, and calculate the intent confidence. When the confidence is ≥85%, output the result directly. When the confidence is <85%, trigger secondary verification and correct the result by combining the user's historical interaction data.

[0010] Preferably, the AI ​​dynamic content generation module includes a scene generation unit, a character action generation unit, and an interactive content generation unit, wherein: The scene generation unit adopts an improved Stable Diffusion model. It takes gaze intent, preference labels and rule constraint parameters as input to generate 3D virtual scenes with a resolution of ≥4K, and supports incremental enhancement of scene details and overall switching. The character motion generation unit generates motion description instructions based on a large language model, and transforms them into coherent movements of the virtual character through a motion capture transfer algorithm, with a motion latency of ≤100ms. The interactive content generation unit generates interactive options and dialogue content that are adapted to the current scene based on the gaze-triggered intent, ensuring that the interactive logic is consistent with the rules of the virtual world.

[0011] Preferably, the layered rendering strategy of the virtual reality rendering module includes: The image is divided into a gaze core area, a transition area, and an edge area. The gaze core area is rendered at 8K resolution and the highest quality, the transition area is rendered at 4K resolution and medium quality, and the edge area is rendered at 2K resolution and basic quality. Based on the user's eye movement trajectory, the gaze position is predicted, and the predicted area is pre-processed for rendering 50ms in advance to ensure that the image is smooth and clear when the gaze moves.

[0012] Preferably, the hierarchical interaction mechanism of the gaze interaction control module is as follows: Enhanced interaction through gaze: When the gaze intent is to enhance interest, the display duration of the corresponding virtual element is extended, the element detail rendering and dynamic effects are increased, and related element recommendations are generated at the same time; Gaze-based switching interaction: When the gaze intent is to switch content, alternative content is generated based on the user preference model, and the content is smoothly switched through a fade-in and fade-out effect with a switching time of ≤300ms; Gaze-triggered interaction: When the gaze intent is to trigger an interaction, holding the gaze on the target element for 300-500ms will trigger the preset interaction, generating interactive feedback animation and content.

[0013] Preferably, the user preference modeling module constructs a preference model by including: Step S21: Extract interest elements, gaze duration, and interaction frequency from the user's historical gaze data to generate initial preference labels; Step S22: Assign weights to each preference tag based on the analytic hierarchy process. The weights of interest elements are calculated based on gaze duration and interaction frequency, with weight values ​​ranging from 0 to 1. Step S23: Receive new interaction data in real time, and update the preference labels and weights after each valid interaction to ensure that the model is consistent with the user's current preferences.

[0014] Preferably, the rule constraint engine includes a physical rule base, a logical rule base, and a security rule base, wherein: The physics rule base stores physical laws such as gravity, collision, and lighting in the virtual world to ensure that the generated content conforms to physical logic. The logic rule base stores the relationships between scene elements and the logic of role behavior to prevent the generation of contradictory content; The security rule base stores a list of prohibited content and interaction permission rules, filtering out illegal content and unauthorized interactions.

[0015] Preferably, the scheduling strategy of the edge computing scheduling module includes: Prioritize assigning gaze intent parsing and rendering preprocessing tasks to local edge nodes to reduce transmission latency; When the AI ​​content generation task is highly complex, cloud edge node resources are dynamically invoked, and distributed computing is used to accelerate content generation. Real-time monitoring of the computing load on each node; when the load on a single node is ≥70%, some tasks will be migrated to nodes with lower load.

[0016] Preferably, the security verification module includes an identity verification unit, an instruction verification unit, and a content verification unit, wherein: The identity verification unit verifies user identity through biometric features (iris, face) linked to the VR device; The instruction verification unit verifies the legality of interactive instructions and rejects instructions that exceed the user's permissions. The content verification unit uses a multimodal content review model to review AI-generated text and image content in real time and remove illegal content.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention innovatively achieves multi-dimensional analysis of gaze intent, breaking through the limitation of existing systems that can only recognize simple gaze behaviors. It integrates gaze data and physiological characteristics to accurately distinguish three types of gaze intent. Combined with confidence verification and historical data correction, the analysis accuracy is significantly improved, and the probability of erroneous interaction is reduced. 2. This invention has the ability to dynamically generate personalized content. Based on an improved diffusion model and a large language model, it realizes real-time generation and iteration of content. Combined with a user preference model and a rule constraint engine, it ensures both content personalization and logical consistency, thus solving the problem of homogenization. 3. This invention features deep integration of content and gaze interaction, constructs a hierarchical gaze interaction mechanism, and achieves closed-loop linkage of intent parsing, content generation, and interaction execution. User gaze behavior can directly drive content optimization and interaction triggering, enhancing immersion and naturalness. 4. This invention balances low latency and high image quality by employing dynamic foveated point layered rendering and edge computing scheduling strategies. While improving the image quality of the gaze area, it reduces computing latency and hardware resource consumption, thus meeting the needs of high-performance VR applications. 5. This invention provides comprehensive security and rule protection. Through a rule constraint engine and a security verification module, it constructs a protection system from multiple dimensions, including content logic, interaction permissions, and user identity, thereby improving system stability and security. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0019] In the attached diagram: Figure 1 This is a system block diagram of the present invention; Figure 2 This is a flowchart of the workflow of the present invention; Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] Depend on Figures 1-2 The present invention relates to an artificial intelligence-based virtual reality content generation and gaze interaction system, comprising: 1) Multimodal perception module: As the core of the system's data input, it integrates an infrared eye-tracking unit, a physiological sensing unit, and an environmental acquisition unit. The infrared eye-tracking unit uses infrared LEDs and a high-speed camera arranged at the edge of the lens to avoid the limitations of the pancake optical structure on sensor placement, achieving high-precision eye movement data capture and outputting data such as gaze point coordinates, gaze duration, and eye movement trajectory, with an accuracy of up to 0.3° and a sampling frequency of 120 frames / second. The physiological sensing unit collects pupil dilation and facial micro-expression features, and associates them with the user's emotional state through feature extraction algorithms. The environmental acquisition unit synchronously acquires the current virtual scene parameters to provide an environmental benchmark for subsequent processing.

[0022] 2) Gaze Intent Parsing Module: Based on multimodal data fusion analysis, this module incorporates a pre-trained model with a CNN and Transformer fusion architecture. It extracts the correlation between gaze data and physiological characteristics, accurately distinguishing between three types of intents: interest enhancement, content switching, and interaction triggering. A confidence verification mechanism ensures the accuracy of intent parsing, avoiding false triggering of interactions. Furthermore, it incorporates historical user data for secondary correction, enhancing the reliability of the parsing.

[0023] 3) AI Dynamic Content Generation Module: Integrates an improved diffusion model and a large language model to achieve multi-dimensional content generation; the scene generation unit generates high-resolution 3D scenes through the diffusion model, supporting incremental enhancement of details and overall switching. For example, when a user stares at a virtual building for a long time (interest enhancement intent), the building texture details and surrounding environmental elements can be automatically added; the character action generation unit converts the action commands output by the language model into coherent character actions with a delay controlled within 100ms; the interactive content generation unit generates interactive options and dialogue content adapted to the scene based on the gaze trigger intent, ensuring the coherence of the interaction logic.

[0024] 4) Virtual Reality Rendering Module: Employs dynamic foveated layered rendering technology, dividing the screen into core, transition, and edge zones based on the user's gaze position and allocating rendering resources accordingly. This ensures overall smoothness while enhancing image quality in the gaze area, addressing the hardware resource requirements of high-resolution rendering. The "dynamic foveated layered rendering technology" is an optimized description of existing foveated rendering technology and is a proprietary technology strategy of this system. Specifically, it refers to dynamically rendering the image in zones and levels based on the user's real-time gaze trajectory and predicted position. Unlike traditional fixed-zone rendering technology, this approach achieves a balance between image quality and smoothness. At the same time, by using gaze prediction preprocessing, blurry images and stuttering can be avoided when the gaze moves.

[0025] 5) Gaze Interaction Control Module: Constructs a hierarchical gaze interaction mechanism to execute corresponding operations for different gaze intentions. Specifically, the "hierarchical gaze interaction mechanism" refers to dividing interaction levels based on gaze intention type (interest enhancement, content switching, interaction triggering), configuring exclusive interaction logic, response parameters, and feedback forms for each level, achieving a balance between precision and diversity in interaction, unlike existing single-mode gaze interaction. Interest enhancement interaction strengthens user interest by adding element details and extending display duration; content switching interaction generates alternative content based on a preference model to achieve a smooth transition; gaze triggering interaction sets a gaze duration threshold of 300-500ms to trigger preset interactions, balancing interaction convenience and prevention of accidental triggering.

[0026] 6) User Preference Modeling Module: Based on users' historical gaze data and interaction records, a dynamic preference model is built. The weight of preference tags is allocated through the analytic hierarchy process, and user interest tendencies are updated in real time. For example, if a user gazes at a science fiction style scene multiple times, the system will automatically increase the weight of the "science fiction" tag and give priority to generating content of this style in the future.

[0027] 7) Rule Constraint Engine: It has three built-in rule bases: the physical rule base ensures that the generated content conforms to the laws of gravity and collision in the virtual world; the logic rule base ensures that the logic of scene elements and character behavior is consistent; and the security rule base filters out illegal content and unauthorized interactions to avoid AI-generated logical contradictions or harmful content.

[0028] 8) Data storage module: Adopting a distributed storage architecture, it stores multimodal perception data, user preference models, AI-generated content and interaction logs, supports real-time read and write and incremental updates, and provides data support for intent parsing and preference modeling.

[0029] 9) Edge computing scheduling module: Optimizes computing latency through dynamic resource scheduling. Local edge nodes handle tasks with high real-time requirements such as gaze parsing and rendering preprocessing, while cloud edge nodes collaboratively handle complex content generation tasks. The load balancing mechanism ensures stable system operation.

[0030] 10) Security Verification Module: Verifies user identity through biometrics, verifies the legality of interaction commands, reviews AI-generated content in real time, comprehensively protects system security, and prevents unauthorized access and malicious content generation.

[0031] The workflow of this system is as follows: Step S1: After the system starts, the security verification module verifies the user's identity. After successful verification, the user preference model and the initial scene of the virtual world are loaded. Step S2: The multimodal perception module collects user gaze data, physiological feature data and environmental status data in real time, and transmits them to the gaze intent parsing module after preprocessing. Step S3: The gaze intent parsing module performs fusion analysis on the data, outputs gaze intent results and confidence levels, and corrects them by combining historical data when the confidence level is insufficient. Step S4: The AI ​​dynamic content generation module receives the gaze intent result, combines the user preference model and rule constraint engine to generate personalized virtual content, and outputs it after passing the rule verification. Step S5: The edge computing scheduling module allocates computing resources, and the virtual reality rendering module performs layered rendering of the generated content, prioritizing the optimization of the image quality in the gaze area, and outputs it to the VR device; Step S6: The gaze interaction control module executes the corresponding interactive operation based on the gaze intent result and generates interactive feedback; Step S7: The user preference modeling module updates the user preference model based on the data from this interaction, and the data storage module saves the relevant data; Step S8: Repeat steps S2-S7 to achieve real-time closed-loop optimization of content generation and gaze interaction until the user exits the system.

[0032] Example 1 In this embodiment, the system is applied to a VR education and training scenario, and the specific configuration and operation process are as follows: 1. Hardware configuration: The VR headset adopts a Pancake structure with an integrated Tobii high-precision eye-tracking module. The headset has built-in infrared LEDs, a 120 frames per second high-speed camera, and miniature physiological sensors. The edge computing nodes are deployed in a K8s containerized manner to support distributed computing. The storage module adopts a Redis+PostgreSQL hybrid architecture to ensure real-time read and write and data persistence.

[0033] 2. Software Configuration: The gaze intent parsing model adopts a CNN-Transformer fusion architecture and is pre-trained based on 100,000 levels of user gaze-intent annotation data; the AI ​​content generation module integrates an improved Stable Diffusion model (adapted to educational content generation) and the Llama 3 language model; the rendering module is developed based on Unreal Engine 5 and realizes layered rendering function.

[0034] 3. Operation process: 1) After the user puts on the VR headset, the security verification module verifies the user's identity through iris recognition and loads the user's historical preference model (such as the user's preference for mechanical structure teaching content) and the initial teaching scenario (mechanical parts assembly scenario). 2) The multimodal perception module collects user gaze data in real time (such as gazing at a gear part for 2 seconds), pupil dilation data, and scene status data, and transmits them to the gaze intent parsing module after preprocessing. 3) After analysis by the gaze intent parsing module, the "interest reinforcement" intent is output with a confidence level of 92%, and the result is output directly; 4) The AI ​​dynamic content generation module receives the intent results, combines them with the preference model to generate detailed disassembly diagrams of gear parts and assembly step animations, and outputs them after being verified by the rule constraint engine (conforming to mechanical and physical rules); 5) The edge computing scheduling module allocates rendering preprocessing tasks to local edge nodes. The rendering module renders the gear detail area (stare core area) at 8K resolution and other areas at 4K resolution to output an immersive image. 6) The gaze interaction control module executes interest-enhancing interactions, extends the display time of gear details, adds rotation animation, and generates related part recommendations; 7) The user preference modeling module updates the weight of the "Mechanical Gear" tag, and the data storage module saves the gaze and interaction data for this session. "Mechanical Gear" is a unique preference tag specific to this VR mechanical parts assembly teaching scenario—generated by the user preference modeling module based on the specific virtual elements (gear parts) viewed by the user, with added scene attributes and interest identifiers. This tag is used to accurately quantify the user's immediate interest in the current teaching scenario, distinguishing it from generalized descriptions like "mechanical structure." The tag weight dynamically adjusts based on user interaction behavior, providing a precise basis for the subsequent AI dynamic content generation module to push gear and related parts teaching content, aligning with the core technology solution of personalized system adaptation. 8) When the user subsequently views another bearing part, the system repeats the above process to generate suitable personalized teaching content, thereby achieving a dynamically optimized teaching experience.

[0035] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0036] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A virtual reality content generation and gaze interaction system based on artificial intelligence, characterized in that, include: The multimodal perception module is used to collect user gaze data, physiological feature data and virtual reality environment status data in real time. The gaze data includes gaze point coordinates, gaze duration and eye movement trajectory. The physiological feature data includes pupil dilation and facial micro-expression features. The environment status data includes current virtual scene element type and scene rendering parameters. The gaze intent parsing module is communicatively connected to the multimodal perception module. It has a built-in pre-trained intent recognition model for fusing and analyzing the collected gaze data and physiological feature data, and outputting user gaze intent results. The intent results include three categories: interest enhancement, content switching, and interaction triggering. The AI ​​dynamic content generation module is connected to the gaze intent parsing module and integrates a diffusion model and a large language model. It is used to generate 3D virtual scenes, character actions and interactive content that meet the requirements based on gaze intent results, user preference data and virtual world rules, and supports real-time iterative updates of content. The virtual reality rendering module is communicatively connected to the AI ​​dynamic content generation module and the gaze interaction control module, respectively. It adopts dynamic gaze point rendering technology to render the generated virtual content in layers, prioritizes improving the rendering accuracy of the user's gaze area, and synchronously outputs immersive images adapted to VR devices. The gaze interaction control module is communicatively connected to the gaze intent parsing module and the virtual reality rendering module to construct a hierarchical gaze interaction mechanism. It executes corresponding interaction operations based on the gaze intent results, including gaze enhancement operations, gaze switching operations, and gaze triggering operations. The user preference modeling module is communicatively connected to the multimodal perception module and the AI ​​dynamic content generation module, respectively, and is used to build a personalized preference model based on the user's historical gaze data and interaction records, and update the user preference tags and weights in real time. The rule constraint engine communicates with the AI ​​dynamic content generation module and has built-in virtual world physical rules, content logic rules and interaction security rules to perform compliance verification on the AI-generated content and eliminate logical contradictions and security risks. The data storage module is used to store multimodal perception data, user preference model data, AI-generated content data, and interaction log data, and supports real-time reading and writing and incremental updates of data. The edge computing scheduling module communicates with each functional module and is used to dynamically schedule computing tasks such as AI content generation, gaze intent parsing, and rendering processing, and allocate computing resources to edge nodes to reduce processing latency. The security verification module is used to perform security verification on user identity information, interaction commands, and generated content to prevent unauthorized access and the generation of malicious content.

2. The virtual reality content generation and gaze interaction system based on artificial intelligence according to claim 1, characterized in that: The multimodal perception module includes an infrared eye-tracking unit, a physiological sensing unit, and an environmental acquisition unit, wherein: The infrared eye-tracking unit uses infrared LEDs and a 120 frames per second high-speed camera arranged on the edge of the VR headset lens to capture the user's eye movement data and output the gaze point coordinates and eye trajectory with an accuracy of ≤0.3°. The physiological sensing unit integrates miniature physiological sensors to collect pupil dilation data (sampling frequency 50Hz) and facial micro-expression feature point data in real time, and obtains emotion-related features through feature extraction algorithms; The environment acquisition unit synchronously acquires the element list, rendering resolution, and frame rate parameters of the current virtual scene, providing data support for content generation and rendering optimization.

3. The virtual reality content generation and gaze interaction system based on artificial intelligence according to claim 1, characterized in that: The working steps of the gaze intent parsing module include: Step S11: Receive gaze data and physiological feature data output by the multimodal perception module, perform data preprocessing, remove abnormal data points, and standardize; Step S12: Input the preprocessed data into the pre-trained intent recognition model. The model adopts a CNN and Transformer fusion architecture to extract the correlation features between gaze features and physiological features. Step S13: Output the intent classification result based on the associated features, and calculate the intent confidence. When the confidence is ≥85%, output the result directly. When the confidence is <85%, trigger secondary verification and correct the result by combining the user's historical interaction data.

4. The virtual reality content generation and gaze interaction system based on artificial intelligence according to claim 1, characterized in that: The AI ​​dynamic content generation module includes a scene generation unit, a character action generation unit, and an interactive content generation unit, wherein: The scene generation unit adopts an improved Stable Diffusion model. It takes gaze intent, preference labels and rule constraint parameters as input to generate 3D virtual scenes with a resolution of ≥4K, and supports incremental enhancement of scene details and overall switching. The character motion generation unit generates motion description instructions based on a large language model, and transforms them into coherent movements of the virtual character through a motion capture transfer algorithm, with a motion latency of ≤100ms. The interactive content generation unit generates interactive options and dialogue content that are adapted to the current scene based on the gaze-triggered intent, ensuring that the interactive logic is consistent with the rules of the virtual world.

5. The virtual reality content generation and gaze interaction system based on artificial intelligence according to claim 1, characterized in that: The layered rendering strategy of the virtual reality rendering module includes: The image is divided into a gaze core area, a transition area, and an edge area. The gaze core area is rendered at 8K resolution and the highest quality, the transition area is rendered at 4K resolution and medium quality, and the edge area is rendered at 2K resolution and basic quality. Based on the user's eye movement trajectory, the gaze position is predicted, and the predicted area is pre-processed for rendering 50ms in advance to ensure that the image is smooth and clear when the gaze moves.

6. The virtual reality content generation and gaze interaction system based on artificial intelligence according to claim 1, characterized in that: The hierarchical interaction mechanism of the gaze interaction control module is as follows: Enhanced interaction through gaze: When the gaze intent is to enhance interest, the display duration of the corresponding virtual element is extended, the element detail rendering and dynamic effects are increased, and related element recommendations are generated at the same time; Gaze-based switching interaction: When the gaze intent is to switch content, alternative content is generated based on the user preference model, and the content is smoothly switched through a fade-in and fade-out effect with a switching time of ≤300ms; Gaze-triggered interaction: When the gaze intent is to trigger an interaction, holding the gaze on the target element for 300-500ms will trigger the preset interaction, generating interactive feedback animation and content.

7. The virtual reality content generation and gaze interaction system based on artificial intelligence according to claim 1, characterized in that: The steps for constructing the preference model by the user preference modeling module include: Step S21: Extract interest elements, gaze duration, and interaction frequency from the user's historical gaze data to generate initial preference labels; Step S22: Assign weights to each preference tag based on the analytic hierarchy process. The weights of interest elements are calculated based on gaze duration and interaction frequency, with weight values ​​ranging from 0 to 1. Step S23: Receive new interaction data in real time, and update the preference labels and weights after each valid interaction to ensure that the model is consistent with the user's current preferences.

8. The virtual reality content generation and gaze interaction system based on artificial intelligence according to claim 1, characterized in that: The rule constraint engine includes a physical rule base, a logical rule base, and a security rule base, wherein: The physics rule base stores physical laws such as gravity, collision, and lighting in the virtual world to ensure that the generated content conforms to physical logic. The logic rule base stores the relationships between scene elements and the logic of role behavior to prevent the generation of contradictory content; The security rule base stores a list of prohibited content and interaction permission rules, filtering out illegal content and unauthorized interactions.

9. The virtual reality content generation and gaze interaction system based on artificial intelligence according to claim 1, characterized in that: The scheduling strategy of the edge computing scheduling module includes: Prioritize assigning gaze intent parsing and rendering preprocessing tasks to local edge nodes to reduce transmission latency; When the AI ​​content generation task is highly complex, cloud edge node resources are dynamically invoked, and distributed computing is used to accelerate content generation. Real-time monitoring of the computing load on each node; when the load on a single node is ≥70%, some tasks will be migrated to nodes with lower load.

10. The virtual reality content generation and gaze interaction system based on artificial intelligence according to claim 1, characterized in that: The security verification module includes an identity verification unit, an instruction verification unit, and a content verification unit, wherein: The identity verification unit verifies user identity using biometric features linked to the VR device; The instruction verification unit verifies the legality of interactive instructions and rejects instructions that exceed the user's permissions. The content verification unit uses a multimodal content review model to review AI-generated text and image content in real time and remove illegal content.

Images