Intelligent interaction method and device based on virtual reality

Abstract

The present application relates to the field of virtual reality technology, and discloses a method and device for intelligent interaction based on virtual reality. The method comprises: obtaining first virtual interaction video data and performing image frame segmentation and interaction part feature point detection, performing feature point motion acceleration analysis and nonlinear space mapping, obtaining nonlinear space mapping data, and performing clustering and denoising to obtain target clustering results; performing virtual interaction intention recognition and intelligent interaction control modeling to obtain an initial virtual environment interaction control model; obtaining second virtual interaction video data to perform image key point recognition to obtain a target center key point heat map; performing virtual reality environment interaction control and feature extraction to obtain feedback response evaluation parameter features; generating a control model parameter update strategy and performing model parameter update to obtain a customized virtual environment interaction control model, thereby improving the accuracy of virtual reality interaction terminal control.

Description

Technical Field

[0001] The present application relates to the field of virtual reality technology, and in particular to an intelligent interaction method and device based on virtual reality. Background Art

[0002] As a human-computer interaction method, virtual reality technology has attracted more and more attention. Virtual reality interactive terminals are widely used in entertainment and other fields. The quality of interactive experience in a virtual reality environment depends largely on how accurately and efficiently the system can interpret and respond to the actions and intentions of the target user. Traditional interaction methods are often limited by hardware performance and algorithm limitations and cannot fully utilize the rich interactive information provided by the target user.

[0003] Traditional solutions face more challenges in virtual reality scenarios. Target users are unable to customize the interactive modeling process according to their own characteristics, which leads to low accuracy in controlling virtual reality interactive terminals. Summary of the invention

[0004] The present application provides a virtual reality-based intelligent interaction method and device for improving the accuracy of virtual reality interaction terminal control.

[0005] The first aspect of the present application provides an intelligent interaction method based on virtual reality, which includes: obtaining first virtual interaction video data of a target user, and performing image frame segmentation and interaction part feature point detection on the first virtual interaction video data to obtain a plurality of standard interaction part feature images and interaction part feature point motion data; performing feature point motion acceleration analysis and nonlinear space mapping on the interaction part feature point motion data to obtain nonlinear space mapping data, and clustering and denoising the nonlinear space mapping data to obtain a target clustering result; performing virtual interaction intention recognition on the plurality of standard interaction part feature images to obtain a virtual interaction intention set, and performing intelligent interaction control modeling on the target clustering result and the virtual interaction intention set to obtain an initial virtual environment interaction. The method comprises the steps of: obtaining second virtual interaction video data of the target user, performing image key point recognition on the second virtual interaction video data to obtain a plurality of target center key points, performing drift correction on the plurality of target center key points, and obtaining a target center key point heat map; performing virtual reality environment interaction control on the target center key point heat map through the initial virtual environment interaction control model, obtaining a feedback response evaluation parameter set of the virtual reality interaction terminal, and performing feature extraction on the feedback response evaluation parameter set to obtain feedback response evaluation parameter features; generating a control model parameter update strategy for the feedback response evaluation parameter features, and updating the model parameters of the initial virtual environment interaction control model according to the control model parameter update strategy to obtain a customized virtual environment interaction control model for the target user.

[0006] The second aspect of the present application provides a virtual reality-based intelligent interactive device, which includes: an acquisition module, which is used to acquire first virtual interactive video data of a target user, and perform image frame segmentation and interaction part feature point detection on the first virtual interactive video data to obtain a plurality of standard interaction part feature images and interaction part feature point motion data; a mapping module, which is used to perform feature point motion acceleration analysis and nonlinear space mapping on the interaction part feature point motion data to obtain nonlinear space mapping data, and cluster and denoise the nonlinear space mapping data to obtain a target clustering result; a modeling module, which is used to perform virtual interaction intention recognition on the plurality of standard interaction part feature images to obtain a virtual interaction intention set, and perform intelligent interaction control modeling on the target clustering result and the virtual interaction intention set to obtain an initial virtual environment interaction control. model; an identification module, used to obtain the second virtual interaction video data of the target user, perform image key point recognition on the second virtual interaction video data to obtain multiple target center key points, and perform drift correction on the multiple target center key points to obtain a target center key point heat map; a feature extraction module, used to perform virtual reality environment interactive control on the target center key point heat map through the initial virtual environment interactive control model, obtain a feedback response evaluation parameter set of the virtual reality interactive terminal, and perform feature extraction on the feedback response evaluation parameter set to obtain feedback response evaluation parameter features; a customized update module, used to generate a control model parameter update strategy for the feedback response evaluation parameter features, and update the model parameters of the initial virtual environment interactive control model according to the control model parameter update strategy to obtain a customized virtual environment interactive control model of the target user.

[0007] In the technical solution provided by the present application, by acquiring and analyzing the virtual interactive video data of the target user, the method can accurately identify the feature points and motion data of the interactive parts. This meticulous data processing ensures higher interaction accuracy and responsiveness, thereby improving the target user experience. By continuously learning and adapting to the interaction habits and intentions of the target user, a customized virtual environment interactive control model can be generated. This personalized adaptation enables each target user to obtain an interactive experience that is more in line with their personal needs and preferences. Through feature point motion acceleration analysis and nonlinear spatial mapping, as well as subsequent clustering and denoising processing, the method can effectively reduce errors and noise interference in the data processing process, thereby improving the accuracy of the interaction. Through data processing and model update strategies, it can adapt to various complex interactive environments, maintain the stability of interactive control, and maintain good performance even under changing or unforeseen conditions. Through the processing of the second virtual interactive video data and the real-time update of the model, the system can quickly feedback the target user's interaction results, so that the target user can immediately see the effect of its operation, enhancing the immediacy and coherence of the interaction. It allows it to adapt to a variety of different virtual reality applications and devices, showing good technical adaptability and scalability, and can be applied to a variety of different virtual reality scenarios. Through image processing and key point recognition technology, the target user's natural movements and intentions can be intuitively converted into responses in the virtual environment, making the interaction more natural and intuitive, thereby improving the accuracy of virtual reality interactive terminal control. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0009] Figure 1 A schematic diagram of an embodiment of an intelligent interaction method based on virtual reality in an embodiment of the present application;

[0010] Figure 2 This is a schematic diagram of an embodiment of an intelligent interactive device based on virtual reality in an embodiment of the present application. DETAILED DESCRIPTION

[0011] The embodiment of the present application provides a method and device for intelligent interaction based on virtual reality, which is used to improve the accuracy of control of virtual reality interactive terminals. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a preset order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0012] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiments of the present application, an embodiment of the intelligent interaction method based on virtual reality includes:

[0013] S101, obtaining first virtual interaction video data of a target user, and performing image frame segmentation and interaction part feature point detection on the first virtual interaction video data to obtain a plurality of standard interaction part feature images and interaction part feature point motion data;

[0014] It is understandable that the execution subject of the present application may be a virtual reality-based intelligent interactive device, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0015] Specifically, the camera parameters of the target user are calibrated by presetting the virtual reality interaction terminal, and the interaction part video of the target user is acquired by the image acquisition device according to the preset sampling frequency to obtain the first virtual interaction video data. This helps to capture the dynamic changes of the target. Camera parameter calibration is part of ensuring data accuracy. The first virtual interaction video data is subjected to video frame processing and divided into multiple frames to obtain multiple original interaction part feature images. This step helps to decompose the virtual interaction target into a series of static images for better analysis and detection of the movement of the interaction part feature points. For example, if the target user makes a target in front of the virtual reality interaction terminal, such as waving or clicking, the target will be captured and decomposed into a series of static images. The change rate of these multiple original interaction part feature images is calculated to obtain the target change frame rate. This is to determine the speed and degree of change of the target. According to the target change frame rate, the multiple original interaction part feature images are subjected to image standardization processing to obtain multiple standard interaction part feature images. For example, if the target user's target is a slow finger movement, the target change frame rate will be relatively low. This will cause multiple original interaction part feature images to be standardized into a series of static images for subsequent analysis. The saliency detection of the interaction part feature points is performed on multiple standard interaction part feature images to determine the location of the interaction part feature points. For example, if the target user's goal is to click on the virtual interactive environment, the saliency detection of the interaction part feature points will help determine the location of the finger. The regularization weight corresponding to the interaction part feature point saliency detection area of ​​two adjacent standard interaction part feature images is calculated according to the preset Gaussian kernel function. This is to take into account the correlation between different frames so as to more accurately detect the location of the interaction part feature points. For example, if the target user makes similar goals continuously, the Gaussian kernel function can help capture the correlation between these goals so as to better detect the interaction part feature points. The interaction part feature points are detected and the position is updated for multiple standard interaction part feature images according to the regularization weights to obtain the interaction part feature point motion data.

[0016] S102, performing feature point motion acceleration analysis and nonlinear space mapping on the feature point motion data of the interaction part to obtain nonlinear space mapping data, and clustering and denoising the nonlinear space mapping data to obtain a target clustering result;

[0017] Specifically, the zero-speed interval detection is performed on the motion data of the feature points of the interactive parts by using a preset generalized likelihood ratio detection algorithm to obtain the zero-speed interval of the feature point motion. For example, the speed of the feature points of the interactive parts is zero at the moment of clicking. This moment will be identified as the zero-speed interval. According to the detection result of the zero-speed interval, the junction of the zero-speed interval and the non-zero-speed interval is determined, and acceleration extraction is performed on these junctions. This is to capture the acceleration information of the target, because acceleration can provide important features about the virtual interactive target. For example, when the target user waves his arms quickly, the speed of the feature points of the interactive parts increases rapidly from zero. In this case, acceleration extraction captures this speed change and uses it for subsequent analysis. Nonlinear spatial mapping is performed on the extracted feature point motion acceleration data. Nonlinear spatial mapping is to map the original data into a more informative space for better target recognition. For example, consider that the target user's target is a spiral motion. Nonlinear spatial mapping can convert this spiral motion data into data that is easier to distinguish and classify. After the mapping is completed, data amplification is performed to highlight the key features in the target so that subsequent analysis is easier. This step helps to further improve the accuracy of target recognition. For example, if the target user's goal is a subtle rotation virtual interaction, data amplification can highlight the details of this rotation, making it easier to analyze and identify. Cluster analysis is performed on the nonlinear spatial mapping data to classify the target data into different target types. The purpose of this step is to group similar targets into one category. The preliminary clustering results need to be followed by noise removal to ensure that accurate target clustering results are finally obtained. For example, if the target user's goals include multiple virtual interactions such as clicking, sliding, and rotating, cluster analysis can classify them into different target categories such as "clicking", "sliding", and "rotating".

[0018] S103, performing virtual interaction intention recognition on multiple standard interaction part feature images to obtain a virtual interaction intention set, and performing intelligent interaction control modeling on the target clustering results and the virtual interaction intention set to obtain an initial virtual environment interaction control model;

[0019] It should be noted that the state variables and measurement variables of the virtual interaction targets in multiple standard interaction part feature images are defined, and a state transition model of the virtual interaction target is established. Various aspects of the virtual interaction, such as the shape, speed, direction, etc. of the target, are determined, and a model is established to describe how they change over time. For example, the virtual interaction targets include "click", "slide" and "zoom in". The state variables may include the position and speed of the target, and the state transition model describes how these states evolve over time. An observation model of the virtual interaction target is established based on the state variables and the measurement variables. The observation model is used to associate the actual observed data with the state variables. This is external observation data used to explain the target in a preset state. For example, if the target user "clicks" the target, the observation model will use the feature point position in the interaction part feature image to explain this virtual interaction. Through the state transition model, the virtual interaction intention is estimated for multiple standard interaction part feature images, and the first virtual interaction intention estimation result is obtained. This is the process of associating the state variables with the virtual interaction target. For example, if the target user's target is a sliding target, the state transition model can help estimate the position and speed of the target. According to the observation model, the virtual interaction intention is estimated for the standard processing live video to obtain the second virtual interaction intention estimation result. This step is to estimate the target based on the actual observation data. For example, if the target user slides the target in front of the outdoor billboard, the observation model will use the actual acquired interaction part feature image to estimate the virtual interaction target. The first virtual interaction intention estimation result and the second virtual interaction intention estimation result are compared and updated to obtain the virtual interaction intention set. This process helps to improve the accuracy and robustness of the virtual interaction intention. For example, if the first estimation result shows that the target user "clicks" the target, and the second estimation result also reaches the same conclusion, then it can be determined that "click" is one of the virtual interaction intentions of the target user. Define the model control variables, and establish the first association rule between the target clustering results and the model control variables, and at the same time establish the second association rule between the virtual interaction intention set and the model control variables. These rules are used to associate the target and the virtual interaction intention with the actual control variables. For example, if the target user "clicks" the target, the first association rule can map it to the "play" operation, while the second association rule can associate the "click" target with the "play" control of the billboard. Finally, intelligent interactive control modeling is performed through these rules and model control variables to obtain the initial virtual environment interactive control model. This model can control the virtual reality interactive terminal according to the target user's goals and virtual interaction intentions.

[0020] S104, obtaining second virtual interactive video data of the target user, performing image key point recognition on the second virtual interactive video data to obtain multiple target center key points, and performing drift correction on the multiple target center key points to obtain a target center key point heat map;

[0021] Specifically, the second virtual interactive video data of the target user is obtained through the virtual reality interactive terminal. This is the starting point for obtaining the original video data for target identification. In this process, abnormal exposure area detection is required to determine the abnormal exposure problem in the video. Abnormal exposure area information will be helpful for subsequent processing. For example, external light can cause overexposure of certain areas of the interactive part, resulting in a decrease in image quality. Abnormal exposure area detection can determine these problem areas. According to the abnormal exposure area information, the second virtual interactive video data is subjected to multi-level video decomposition to obtain multiple sub-exposure area video images and a basic area video image. This is to separate the abnormal exposure area from other areas for better processing. For example, if a part of the interactive part is under strong external light, the multi-level video decomposition can separate this abnormal exposure area from other parts to generate multiple video images, each of which represents a different area. Subsequently, exposure area repair and video image fusion are performed on multiple sub-exposure area video images and the basic area video image to generate a target area fused video image. Exposure area repair is to reduce the impact of abnormal exposure areas, while fusion merges images of different areas together to form a comprehensive video image. For example, if a certain area of ​​the interaction part is overexposed, exposure repair can alleviate this problem, and then a more balanced video image can be obtained by fusing images of different areas. Target center recognition is performed on the target area fused video image to obtain multiple target center key points. These key points mark the important positions of the target. For example, if the target user makes a sliding target in front of an outdoor billboard, the target center key point will mark the center position and motion trajectory of the interaction part. Neighbor point recognition is performed on each target center key point in the initial target key point cloud map based on the K-nearest neighbor algorithm. This helps find the nearest neighbor points of each target center key point to measure the position offset between them. For example, if a target center key point represents the palm of the target user, neighbor point recognition can find other key points closest to the palm to calculate the relative position between them. The initial target key point cloud map is drift corrected according to the position offset to obtain a target center key point heat map. This heat map shows the position and motion trajectory of the target center for subsequent target analysis and recognition.

[0022] S105, performing virtual reality environment interactive control on the target center key point heat map through the initial virtual environment interactive control model, obtaining a feedback response evaluation parameter set of the virtual reality interactive terminal, and performing feature extraction on the feedback response evaluation parameter set to obtain feedback response evaluation parameter features;

[0023] Specifically, the target center key point heat map is matched with virtual interactive control through the initial virtual environment interactive control model to obtain the virtual interactive control operation strategy. The initial model learns how to map the target center key point heat map to the specific control operation on the virtual reality interactive terminal through a pre-trained algorithm or machine learning technology. For example, if the target center key point heat map indicates that the target user's palm is on the left side of the virtual reality interactive terminal, the model will generate a corresponding control operation strategy, such as lowering the volume. Subsequently, the virtual reality interactive terminal responds to the control operation strategy and performs the corresponding operation. This includes the recognition results of the target user's target, such as volume control, content switching, etc. The virtual reality interactive terminal will actually operate according to these operation strategies to meet the needs of the target user. By changing the matching method of the target center key point heat map, different control strategies can be implemented, thereby providing more interactive options. In order to evaluate the performance of the virtual reality interactive terminal, it is necessary to create an evaluation index system, which includes parameters such as response time, accuracy, and stability. These indicators are used to measure the response efficiency and accuracy of the virtual reality interactive terminal to the target user's target. For example, the response time of the virtual reality interactive terminal can be measured, that is, the time from the target user's target to the actual control operation execution. Subsequently, a set of feedback response evaluation parameters is obtained, which are key data for evaluating the performance of the virtual reality interactive terminal. These parameters include information on response time, accuracy, and stability. These data are mapped into a target parameter feature space for more in-depth analysis and processing. Feature extraction is performed to extract key features from the target parameter feature space. These features can be used for further analysis, optimization, and visualization to better understand the performance of the virtual reality interactive terminal.

[0024] S106: Generate a control model parameter update strategy that responds to the evaluation parameter characteristics of the feedback, and update the model parameters of the initial virtual environment interactive control model according to the control model parameter update strategy to obtain a customized virtual environment interactive control model for the target user.

[0025] Specifically, the model parameter update range analysis is performed. Multiple model parameters in the initial virtual environment interactive control model are analyzed to determine their acceptable value ranges. These parameters include parameters in terms of speed and accuracy of control operations. Through the model parameter update range, random initial values ​​are generated for multiple parameters of the initial virtual environment interactive control model. A random initial value is determined for each parameter, and these initial values ​​are within the acceptable range of the parameters. This helps to provide initial values ​​for the particle swarm algorithm. Subsequently, the preset particle swarm algorithm is used to construct a parameter update strategy for the random initial values ​​corresponding to each model parameter update range. The particle swarm algorithm is an optimization algorithm used to determine the best combination of parameter values ​​within a given range to meet a preset objective function. In this case, the objective function is an indicator of the control performance of the virtual reality interactive terminal, such as accuracy or response time. The particle fitness calculation is performed on the control model parameter update strategy. The fitness of each particle (parameter combination) is evaluated, that is, their performance on the objective function. The higher the fitness value, the better the parameter combination. Iterative calculation and optimization solution are performed to find the optimal model parameter update strategy. The particle swarm algorithm will continuously adjust the position of the particles until the best parameter combination is found to meet the performance target. The parameters of the initial virtual environment interactive control model are updated according to the optimal model parameter update strategy to obtain a customized virtual environment interactive control model for the target user. This model will better meet the needs of the target user and provide more accurate target recognition and control performance.

[0026] In the embodiment of the present application, by acquiring and analyzing the virtual interactive video data of the target user, the method can accurately identify the feature points and motion data of the interactive parts. This meticulous data processing ensures higher interaction accuracy and responsiveness, thereby improving the target user experience. By continuously learning and adapting to the interaction habits and intentions of the target user, a customized virtual environment interactive control model can be generated. This personalized adaptation enables each target user to obtain an interactive experience that is more in line with their personal needs and preferences. Through feature point motion acceleration analysis and nonlinear spatial mapping, as well as subsequent clustering and denoising processing, the method can effectively reduce errors and noise interference in the data processing process, thereby improving the accuracy of the interaction. Through data processing and model update strategies, it can adapt to various complex interactive environments, maintain the stability of interactive control, and maintain good performance even under changing or unforeseen conditions. Through the processing of the second virtual interactive video data and the real-time update of the model, the system can quickly feedback the target user's interaction results, so that the target user can immediately see the effect of its operation, and enhance the immediacy and coherence of the interaction. It allows it to adapt to a variety of different virtual reality applications and devices, showing good technical adaptability and scalability, and can be applied to a variety of different virtual reality scenarios. Through image processing and key point recognition technology, the target user's natural movements and intentions can be intuitively converted into responses in the virtual environment, making the interaction more natural and intuitive, thereby improving the accuracy of virtual reality interactive terminal control.

[0027] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0028] (1) Calibrate the camera parameters of the target user through the preset virtual reality interaction terminal, and obtain the video of the target user's interaction part through the image acquisition device according to the preset sampling frequency to obtain the first virtual interaction video data;

[0029] (2) performing video frame processing on the first virtual interaction video data to obtain a corresponding plurality of original interaction part feature images, and calculating the change rate of the plurality of original interaction part feature images to obtain a target change frame rate;

[0030] (3) performing image standardization processing on multiple original interaction part feature images according to the target change frame rate to obtain multiple standard interaction part feature images;

[0031] (4) performing saliency detection of interaction part feature points on multiple standard interaction part feature images to obtain a saliency detection region of interaction part feature points of each standard interaction part feature image;

[0032] (5) Calculating the regularization weights corresponding to the saliency detection regions of the interaction part feature points of two adjacent standard interaction part feature images by using a preset Gaussian kernel function;

[0033] (6) Perform interaction part feature point detection and position update on multiple standard interaction part feature images according to the regularization weights to obtain interaction part feature point motion data.

[0034] Specifically, the camera parameters of the target user are calibrated through a preset virtual reality interaction terminal. According to the preset calibration process, information about the target user is collected for subsequent acquisition and processing of the interaction part video. For example, the calibration process may include identifying the position and posture of the target user. Subsequently, the interaction part video of the target user is acquired through the image acquisition terminal according to the preset sampling frequency. This will generate first virtual interaction video data, including the virtual interaction and posture of the interaction part of the target user. These data will be used for subsequent target recognition and control. The first virtual interaction video data is subjected to video frame processing, and is divided into multiple frames to obtain multiple original interaction part feature images. Each image represents a frame in the video and reflects the state of the interaction part. These original images will serve as the basis for recognition and analysis. After obtaining multiple original interaction part feature images, the change rate is calculated to obtain the target change frame rate. This can help identify the speed and frequency of the virtual interaction of the interaction part. The multiple original interaction part feature images are subjected to image standardization processing according to the target change frame rate. This step ensures that all images have similar feature representations, thereby improving the accuracy and robustness of target recognition. The interaction part feature point saliency detection is performed on multiple standard interaction part feature images. This step helps to determine the key feature points of the interaction part in each image, such as finger joints or palm centers. These feature points will be used to track the virtual interaction of the interaction parts. Subsequently, the regularized weights corresponding to the saliency detection areas of the interaction part feature points of two adjacent standard interaction part feature images are calculated by a preset Gaussian kernel function. These weights can be used to measure the correlation between different feature points in order to better understand the virtual interaction of the interaction parts. The interaction part feature points are detected and their positions are updated for multiple standard interaction part feature images according to the regularized weights. The weight information is used to determine the motion trajectory of the interaction part feature points, thereby obtaining the motion data of the interaction part feature points. These data will provide key information for subsequent target recognition and virtual reality interactive terminal control.

[0035] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0036] (1) Using a preset generalized likelihood ratio detection algorithm, the zero-speed interval detection is performed on the motion data of the feature points of the interaction part to obtain the zero-speed interval of the feature point motion;

[0037] (2) According to the zero-speed interval of the feature point motion, determine the boundary between the zero-speed interval and the non-zero-speed interval in the feature point motion data of the interaction part, and extract the acceleration at the boundary to obtain the feature point motion acceleration data;

[0038] (3) performing nonlinear spatial mapping on the motion acceleration data of the feature points to obtain nonlinear spatial mapping data, and amplifying the nonlinear spatial mapping data to obtain target amplified data;

[0039] (4) Perform cluster analysis on the nonlinear spatial mapping data to obtain the initial clustering results, and remove noise from the initial clustering results to obtain the target clustering results.

[0040] Specifically, the zero-speed interval detection is performed on the motion data of the feature points of the interaction parts by using a preset generalized likelihood ratio detection algorithm. The purpose is to determine the zero-speed interval in the motion of the interaction parts, that is, the time period when the interaction parts are stationary or moving at a low speed. The detection of these zero-speed intervals helps to separate the key virtual interactions of different targets. For example, the motion data of the target user's interaction parts include different states such as stationary, fast movement and slow movement. The generalized likelihood ratio detection algorithm can be used to detect the changes between these states. According to the zero-speed interval of the feature point motion, the junction between the zero-speed interval and the non-zero-speed interval in the feature point motion data of the interaction parts is determined. At these junctions, acceleration extraction is performed to obtain feature point motion acceleration data. This step helps to identify the acceleration changes of the virtual interaction of the interaction parts, so as to better distinguish different targets. For example, when the target user's interaction part changes from a stationary state to a fast movement, the junction is a key point, marking the acceleration change. This acceleration information helps to classify the target. The feature point motion acceleration data is nonlinearly mapped in space. This step converts the acceleration data into a nonlinear mapping form to better capture the target features. Nonlinear mapping helps to improve the accuracy and robustness of target recognition. For example, nonlinear mapping can transform acceleration data into a form that is easier to analyze and emphasize important features in the target. Data amplification is performed on nonlinear spatial mapping data to enhance the features of the target. Data amplification helps to highlight important information in the target, thereby better performing subsequent analysis and classification. For example, by amplifying the key features of the target, the target can be more clearly identified and classified. Cluster analysis is performed on the nonlinear spatial mapping data to obtain initial clustering results, and noise is removed from the initial clustering results to finally obtain target clustering results. This step groups similar virtual interactive targets together and removes uncertainty caused by noise. For example, the system clusters similar virtual interactive targets together, such as sliding targets and clicking targets. Noise removal can ensure the accuracy of target classification.

[0041] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0042] (1) Define the state variables and measurement variables of the virtual interaction target in multiple standard interaction part feature images, and establish a state transition model of the virtual interaction target;

[0043] (2) Establishing an observation model of the virtual interaction target based on state variables and measurement variables;

[0044] (3) performing virtual interaction intention estimation on multiple standard interaction part feature images through a state transition model to obtain a first virtual interaction intention estimation result;

[0045] (4) Estimating the virtual interaction intention of the standard processing scene video according to the observation model to obtain a second virtual interaction intention estimation result;

[0046] (5) comparing the first virtual interaction intention estimation result and the second virtual interaction intention estimation result and updating the state to obtain a virtual interaction intention set;

[0047] (6) Define the model control variables and establish the first association rule between the target clustering results and the model control variables. At the same time, establish the second association rule between the virtual interaction intention set and the model control variables.

[0048] (7) Intelligent interactive control modeling is performed through the first association rule, the second association rule, and the model control variables to obtain an initial virtual environment interactive control model.

[0049] Specifically, state variables and measurement variables of virtual interaction targets in multiple standard interaction part feature images are defined, and a state transition model of the virtual interaction target is established. The purpose is to establish a model for tracking the state changes of virtual interaction of interaction parts, such as the position, speed and direction of the interaction parts. The state transition model will describe how the state variables change over time. For example, the state variables may include the position and speed of the interaction part, and the state transition model will tell the system how the interaction part moves at different time points. An observation model of the virtual interaction target is established based on the state variables and measurement variables. The observation model will describe how to map the state variables to the observation values, that is, the interaction part feature image. This helps to link the state of the interaction part with the feature image. For example, the observation model can tell the system how to map the interaction part position to the interaction part shape and position on the feature image. The virtual interaction intention is estimated for multiple standard interaction part feature images through the state transition model to obtain a first virtual interaction intention estimation result, and the system predicts the virtual interaction intention of the target user, that is, the target that the target user intends to execute. For example, if the target user interaction part is at a preset position, the state transition model can help the system predict the target that the target user intends to execute. The virtual interaction intention is estimated for the standard processing scene video according to the observation model to obtain a second virtual interaction intention estimation result. This step will help the system identify the virtual interaction intention of the target user from the actual video data obtained. For example, the system analyzes the feature image of the interaction part and tries to understand the target actually performed by the target user. The first virtual interaction intention estimation result and the second virtual interaction intention estimation result are compared and updated to obtain a virtual interaction intention set. This step will combine the information obtained from the model and the actual data to obtain a more accurate virtual interaction intention. For example, if the model predicts that the target user is to perform a goal that does not match the actual observation, the system updates the virtual interaction intention according to the observed data. Define the model control variables and establish the first association rule between the target clustering result and the model control variable. At the same time, establish the second association rule between the virtual interaction intention set and the model control variable. These rules will help the system map the virtual interaction of the target user to the actual virtual interaction control operation strategy. For example, the system defines the model control variables, such as the interaction part position and speed, and the target clustering results related to these variables. By establishing the association rules, the system maps the virtual interaction intention of the target user to the actual virtual reality interaction terminal control operation. Intelligent interaction control modeling is performed through the first association rule, the second association rule and the model control variable to obtain the initial virtual environment interaction control model. This model will allow the system to implement the target user's goal control and translate the target user's virtual interaction into a VR interactive terminal operation strategy. For example, if the target user wants to zoom in on the content on the VR interactive terminal, the system will identify the target user's virtual interaction goal, map it to a zoom-in operation, and execute the operation.

[0050] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0051] (1) Obtaining second virtual interactive video data of the target user through the virtual reality interactive terminal, and performing abnormal exposure area detection on the second virtual interactive video data to obtain abnormal exposure area information;

[0052] (2) performing multi-level video decomposition on the second virtual interactive video data according to the abnormal exposure area information to obtain a plurality of sub-exposure area video images and a basic area video image, and performing exposure area repair and video image fusion on the plurality of sub-exposure area video images and the basic area video image to generate a target area fused video image;

[0053] (3) Perform target center recognition on the fused video image of the target area to obtain multiple target center key points, and construct an initial target key point cloud map based on the multiple target center key points;

[0054] (4) Based on the K-nearest neighbor algorithm, neighboring points of each target center key point in the initial target key point cloud map are identified to obtain the K nearest neighbor points corresponding to each target center key point;

[0055] (5) Calculate the position offsets between the K nearest neighbor points and the corresponding target center key points respectively, and perform drift correction on the initial target key point cloud map according to the position offsets to obtain the target center key point heat map.

[0056] Specifically, the second virtual interactive video data of the target user is obtained through the virtual reality interactive terminal. The corresponding device or camera is used to capture the interactive part video of the target user. The second virtual interactive video data is subjected to abnormal exposure area detection to obtain abnormal exposure area information. This step helps to determine abnormal lighting or exposure problems in the video, which affect target recognition. For example, the system detects an abnormal exposure area caused by direct external light. According to the abnormal exposure area information, the second virtual interactive video data is subjected to multi-level video decomposition to obtain multiple sub-exposure area video images and basic area video images. This helps to separate the abnormal exposure area from other parts to improve the accuracy of target recognition. For example, the area where the interactive part is located is separated from the background. Exposure area repair and video image fusion are performed on multiple sub-exposure area video images and basic area video images to generate a target area fused video image. The purpose is to repair the abnormal exposure area and merge all images into one at the same time. For example, the abnormal exposure area is repaired by an exposure compensation algorithm, and all images are fused to create a target area image. Subsequently, the target area fused video image is subjected to target center recognition to obtain multiple target center key points. The purpose is to detect the target center in the image to determine the target user target. Subsequently, an initial target key point cloud map is constructed based on multiple target center key points. This cloud map describes all detected target centers. Based on the K-nearest neighbor algorithm, neighboring point identification is performed on each target center key point in the initial target key point cloud map to obtain the K nearest neighbor points corresponding to each target center key point. This helps to determine the surrounding area of ​​each target center key point. The position offsets between the K nearest neighbor points and the corresponding target center key points are calculated respectively, and the initial target key point cloud map is drift corrected according to the position offsets. This step helps to correct the position deviation of the target center caused by abnormal exposure or other problems to obtain an accurate target center key point heat map.

[0057] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0058] (1) Perform virtual interactive control matching on the target center key point heat map through the initial virtual environment interactive control model to obtain the virtual interactive control operation strategy;

[0059] (2) Respond to the virtual interaction control operation strategy through the virtual reality interaction terminal and create an evaluation index system for the virtual reality interaction terminal;

[0060] (3) Obtaining a set of feedback response evaluation parameters of the virtual reality interactive terminal according to the evaluation index system, and performing feature mapping on the set of feedback response evaluation parameters to obtain a target parameter feature space;

[0061] (4) Extract features from the target parameter feature space to obtain feedback response evaluation parameter features.

[0062] Specifically, the target center key point heat map is matched with virtual interactive control through the initial virtual environment interactive control model. The purpose is to associate the position of the target center with the operation response of the virtual reality interactive terminal so as to realize the control operation according to the target. For example, assuming that a target user's target slides to the right, the system will identify the position of the target center on the heat map and then map it to the corresponding control operation. The virtual reality interactive terminal responds to the virtual interactive control operation strategy. Once the system identifies the target user's target, it needs to perform corresponding operations according to the type of the target, such as moving the virtual interactive environment content, zooming or triggering a preset function. For example, if the target user's target indicates "slide up", the virtual reality interactive terminal will scroll the displayed content upward. At the same time, an evaluation index system for the virtual reality interactive terminal is created. A set of indicators is established to evaluate the effectiveness of the virtual reality interactive terminal control and the target user's satisfaction. For example, the evaluation indicators may include response time, control accuracy, feedback quality, etc. According to the evaluation index system, a set of feedback response evaluation parameters of the virtual reality interactive terminal is obtained. These parameters include various indicators for evaluating the control of the virtual reality interactive terminal, such as response time, accuracy, precision, etc. For example, the system records the actual response time and accuracy of the virtual reality interactive terminal after the target user's target. Subsequently, feature mapping is performed on the feedback response evaluation parameter set. The purpose is to map the feedback response parameters to a feature space for further analysis and processing. For example, response time and accuracy are mapped to the feature space to better understand the control performance. Feature extraction is performed on the target parameter feature space. This step helps to extract key information about the control of the virtual reality interactive terminal from the feature space. For example, the system analyzes the data in the feature space to identify opportunities for performance improvement or potential problems.

[0063] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0064] (1) Analyze the model parameter update range of the initial virtual environment interactive control model and obtain multiple model parameter update ranges;

[0065] (2) Generate random initial values ​​for multiple model parameters of the initial virtual environment interactive control model through multiple model parameter update ranges to obtain random initial values ​​corresponding to each model parameter update range, and construct a parameter update strategy for the random initial values ​​corresponding to each model parameter update range through a preset particle swarm algorithm to obtain a control model parameter update strategy;

[0066] (3) Calculate the particle fitness of the control model parameter update strategy to obtain the particle fitness set corresponding to the particle population, and iteratively calculate and optimize the particle fitness set to obtain the optimal model parameter update strategy;

[0067] (4) The parameters of the initial virtual environment interactive control model are updated according to the optimal model parameter updating strategy to obtain a customized virtual environment interactive control model for the target user.

[0068] Specifically, the model parameter update range analysis is performed on the initial virtual environment interactive control model. The reasonable value range of each model parameter is analyzed to ensure that the parameter update is within a reasonable range. For example, if a model parameter represents speed, the range analysis determines that the value range of the parameter is between 0 and 100. Subsequently, random initial values ​​are generated for multiple model parameters of the initial virtual environment interactive control model through the update ranges of multiple model parameters. These random initial values ​​will be used to start the process of parameter update strategy. For example, if a model has three parameters, namely speed, sensitivity and accuracy, different initial values ​​can be generated to cover the range of each parameter. The parameter update strategy is constructed using the preset particle swarm algorithm for the random initial values ​​corresponding to each model parameter update range. The particle swarm algorithm is an optimization algorithm used to find parameters that optimize the objective function within a given range. For example, through the particle swarm algorithm, the best value can be searched in the parameter space to meet the preset performance indicators, such as minimum error or maximum response speed. The particle fitness calculation is performed on the control model parameter update strategy. This step involves evaluating the performance of each parameter combination, which is usually achieved by comparing the actual output of the model with the target output. For example, if the model parameter update strategy results in a significant reduction in the response time of the model, then the strategy is considered to have a high fitness. Perform iterative calculations and optimization solutions to obtain the optimal model parameter update strategy. Continuously improve the parameter update strategy until the best parameter combination that meets the performance criteria is found. Update the parameters of the initial virtual environment interactive control model according to the optimal model parameter update strategy. Apply the optimal parameter values ​​to the initial model to generate a customized virtual environment interactive control model for the target user. For example, if the optimal strategy significantly improves the accuracy of the virtual reality interactive terminal control, the new model will better meet the needs of the target user.

[0069] The above describes the intelligent interaction method based on virtual reality in the embodiment of the present application. The following describes the intelligent interaction device based on virtual reality in the embodiment of the present application. Figure 2 In the embodiments of the present application, an embodiment of a virtual reality-based intelligent interactive device includes:

[0070] The acquisition module 201 is used to acquire the first virtual interaction video data of the target user, and perform image frame segmentation and interaction part feature point detection on the first virtual interaction video data to obtain a plurality of standard interaction part feature images and interaction part feature point motion data;

[0071] A mapping module 202 is used to perform feature point motion acceleration analysis and nonlinear space mapping on the feature point motion data of the interaction part to obtain nonlinear space mapping data, and to cluster and denoise the nonlinear space mapping data to obtain a target clustering result;

[0072] A modeling module 203 is used to perform virtual interaction intention recognition on the plurality of standard interaction part feature images to obtain a virtual interaction intention set, and to perform intelligent interaction control modeling on the target clustering result and the virtual interaction intention set to obtain an initial virtual environment interaction control model;

[0073] The recognition module 204 is used to obtain the second virtual interaction video data of the target user, perform image key point recognition on the second virtual interaction video data to obtain a plurality of target center key points, and perform drift correction on the plurality of target center key points to obtain a target center key point heat map;

[0074] The feature extraction module 205 is used to perform virtual reality environment interactive control on the target center key point heat map through the initial virtual environment interactive control model, obtain a feedback response evaluation parameter set of the virtual reality interactive terminal, and perform feature extraction on the feedback response evaluation parameter set to obtain feedback response evaluation parameter features;

[0075] The customized update module 206 is used to generate a control model parameter update strategy based on the feedback response evaluation parameter characteristics, and update the model parameters of the initial virtual environment interactive control model according to the control model parameter update strategy to obtain a customized virtual environment interactive control model for the target user.

[0076] Through the synergy of the above components, by acquiring and analyzing the virtual interaction video data of the target user, the method can accurately identify the feature points and motion data of the interaction parts. This meticulous data processing ensures higher interaction accuracy and responsiveness, thereby improving the target user experience. By continuously learning and adapting to the interaction habits and intentions of the target user, a customized virtual environment interaction control model can be generated. This personalized adaptation enables each target user to obtain an interaction experience that is more in line with their personal needs and preferences. Through feature point motion acceleration analysis and nonlinear spatial mapping, as well as subsequent clustering and denoising processing, the method can effectively reduce errors and noise interference in the data processing process, thereby improving the accuracy of the interaction. Through data processing and model update strategies, it can adapt to various complex interaction environments, maintain the stability of interactive control, and maintain good performance even under changing or unforeseen conditions. Through the processing of the second virtual interaction video data and the real-time update of the model, the system can quickly feedback the interaction results of the target user, so that the target user can immediately see the effect of his operation, enhancing the immediacy and coherence of the interaction. It allows it to adapt to a variety of different virtual reality applications and devices, showing good technical adaptability and scalability, and can be applied to a variety of different virtual reality scenarios. Through image processing and key point recognition technology, the target user's natural movements and intentions can be intuitively converted into responses in the virtual environment, making the interaction more natural and intuitive, thereby improving the accuracy of virtual reality interactive terminal control.

[0077] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0078] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0079] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent interaction method based on virtual reality, characterized in that: The intelligent interaction method based on virtual reality includes: Acquire first virtual interaction video data of the target user, and perform image frame segmentation and interaction part feature point detection on the first virtual interaction video data to obtain a plurality of standard interaction part feature images and interaction part feature point motion data; Performing feature point motion acceleration analysis and nonlinear space mapping on the feature point motion data of the interaction part to obtain nonlinear space mapping data, and clustering and denoising the nonlinear space mapping data to obtain a target clustering result; Performing virtual interaction intention recognition on the plurality of standard interaction part feature images to obtain a virtual interaction intention set, and performing intelligent interaction control modeling on the target clustering result and the virtual interaction intention set to obtain an initial virtual environment interaction control model; Acquire second virtual interaction video data of the target user, perform image key point recognition on the second virtual interaction video data to obtain a plurality of target center key points, and perform drift correction on the plurality of target center key points to obtain a target center key point heat map; Performing virtual reality environment interactive control on the target center key point heat map through the initial virtual environment interactive control model, obtaining a feedback response evaluation parameter set of the virtual reality interactive terminal, and performing feature extraction on the feedback response evaluation parameter set to obtain feedback response evaluation parameter features; A control model parameter update strategy based on the feedback response evaluation parameter feature is generated, and model parameters of the initial virtual environment interactive control model are updated according to the control model parameter update strategy to obtain a customized virtual environment interactive control model for the target user.

2. The intelligent interaction method based on virtual reality according to claim 1, characterized in that: The step of acquiring the first virtual interaction video data of the target user, performing image frame segmentation and interaction part feature point detection on the first virtual interaction video data, and obtaining a plurality of standard interaction part feature images and interaction part feature point motion data includes: Calibrate the camera parameters of the target user by presetting the virtual reality interaction terminal, and acquire the interactive part video of the target user by the image acquisition device according to the preset sampling frequency to obtain the first virtual interaction video data; Performing video frame processing on the first virtual interaction video data to obtain a corresponding plurality of original interaction part feature images, and calculating the change rate of the plurality of original interaction part feature images to obtain a target change frame rate; Performing image standardization processing on the plurality of original interaction part feature images according to the target change frame rate to obtain a plurality of standard interaction part feature images; Performing interaction part feature point saliency detection on the plurality of standard interaction part feature images to obtain an interaction part feature point saliency detection region of each standard interaction part feature image; The regularization weight corresponding to the significant detection area of ​​the interaction part feature point of two adjacent standard interaction part feature images is calculated by a preset Gaussian kernel function; Interaction part feature point detection and position update are performed on the plurality of standard interaction part feature images according to the regular weights to obtain interaction part feature point motion data.

3. The intelligent interaction method based on virtual reality according to claim 2, characterized in that: The step of performing feature point motion acceleration analysis and nonlinear space mapping on the feature point motion data of the interaction part to obtain nonlinear space mapping data, and clustering and denoising the nonlinear space mapping data to obtain a target clustering result includes: Performing zero-speed interval detection on the motion data of the feature points of the interaction part by using a preset generalized likelihood ratio detection algorithm to obtain a zero-speed interval of the feature point motion; According to the zero-speed interval of the feature point movement, determining the junction of the zero-speed interval and the non-zero-speed interval in the feature point movement data of the interaction part, and extracting the acceleration of the junction to obtain the feature point movement acceleration data; Performing nonlinear spatial mapping on the feature point motion acceleration data to obtain nonlinear spatial mapping data, and amplifying the nonlinear spatial mapping data to obtain target amplified data; Cluster analysis is performed on the nonlinear spatial mapping data to obtain an initial clustering result, and noise is removed from the initial clustering result to obtain a target clustering result.

4. The intelligent interaction method based on virtual reality according to claim 1, characterized in that: The virtual interaction intention recognition is performed on the plurality of standard interaction part feature images to obtain a virtual interaction intention set, and the intelligent interaction control modeling is performed on the target clustering result and the virtual interaction intention set to obtain an initial virtual environment interaction control model, including: Defining state variables and measurement variables of virtual interaction targets in the plurality of standard interaction part feature images, and establishing a state transition model of the virtual interaction targets; Establishing an observation model of the virtual interactive target according to the state variables and the measurement variables; Performing virtual interaction intention estimation on the plurality of standard interaction part feature images through the state transition model to obtain a first virtual interaction intention estimation result; Performing virtual interaction intention estimation on the standard processing scene video according to the observation model to obtain a second virtual interaction intention estimation result; Comparing and updating the state of the first virtual interaction intention estimation result and the second virtual interaction intention estimation result to obtain a virtual interaction intention set; defining a model control variable, and establishing a first association rule between the target clustering result and the model control variable, and at the same time, establishing a second association rule between the virtual interaction intention set and the model control variable; Intelligent interactive control modeling is performed through the first association rule, the second association rule and the model control variables to obtain an initial virtual environment interactive control model.

5. The intelligent interaction method based on virtual reality according to claim 4, characterized in that: The step of acquiring the second virtual interactive video data of the target user, performing image key point recognition on the second virtual interactive video data to obtain a plurality of target center key points, and performing drift correction on the plurality of target center key points to obtain a target center key point heat map includes: Acquire second virtual interactive video data of the target user through a virtual reality interactive terminal, and perform abnormal exposure area detection on the second virtual interactive video data to obtain abnormal exposure area information; Performing multi-level video decomposition on the second virtual interactive video data according to the abnormal exposure area information to obtain a plurality of sub-exposure area video images and a basic area video image, and performing exposure area repair and video image fusion on the plurality of sub-exposure area video images and the basic area video image to generate a target area fused video image; Performing target center recognition on the target area fused video image to obtain a plurality of target center key points, and constructing an initial target key point cloud map according to the plurality of target center key points; Based on the K-nearest neighbor algorithm, neighboring points of each target center key point in the initial target key point cloud map are identified to obtain K nearest neighbor points corresponding to each target center key point; The position offsets between the K nearest neighbor points and the corresponding target center key points are calculated respectively, and the initial target key point cloud map is drift-corrected according to the position offsets to obtain a target center key point heat map.

6. The intelligent interaction method based on virtual reality according to claim 1, characterized in that: The method of performing virtual reality environment interactive control on the target center key point heat map through the initial virtual environment interactive control model, obtaining a feedback response evaluation parameter set of the virtual reality interactive terminal, and performing feature extraction on the feedback response evaluation parameter set to obtain feedback response evaluation parameter features includes: Performing virtual interactive control matching on the target center key point heat map through the initial virtual environment interactive control model to obtain a virtual interactive control operation strategy; Responding to the virtual interaction control operation strategy through a virtual reality interaction terminal, and creating an evaluation index system for the virtual reality interaction terminal; Acquire a feedback response evaluation parameter set of the virtual reality interactive terminal according to the evaluation index system, and perform feature mapping on the feedback response evaluation parameter set to obtain a target parameter feature space; Feature extraction is performed on the target parameter feature space to obtain feedback response evaluation parameter features.

7. The intelligent interaction method based on virtual reality according to claim 6, characterized in that: The generating of the control model parameter updating strategy of the feedback response evaluation parameter feature, and updating the model parameters of the initial virtual environment interactive control model according to the control model parameter updating strategy to obtain the customized virtual environment interactive control model of the target user, comprises: Performing a model parameter update range analysis on the initial virtual environment interactive control model to obtain multiple model parameter update ranges; By using the multiple model parameter update ranges, generating random initial values ​​for the multiple model parameters of the initial virtual environment interactive control model, obtaining the random initial values ​​corresponding to each model parameter update range, and constructing a parameter update strategy for the random initial values ​​corresponding to each model parameter update range by using a preset particle swarm algorithm, to obtain a control model parameter update strategy; Performing particle fitness calculation on the control model parameter update strategy to obtain a particle fitness set corresponding to the particle population, and performing iterative calculation and optimization on the particle fitness set to obtain an optimal model parameter update strategy; The parameters of the initial virtual environment interactive control model are updated according to the optimal model parameter updating strategy to obtain a customized virtual environment interactive control model for the target user.

8. An intelligent interactive device based on virtual reality, characterized in that: The virtual reality-based intelligent interactive device includes: An acquisition module is used to acquire first virtual interaction video data of a target user, and perform image frame segmentation and interaction part feature point detection on the first virtual interaction video data to obtain a plurality of standard interaction part feature images and interaction part feature point motion data; A mapping module, used for performing feature point motion acceleration analysis and nonlinear space mapping on the feature point motion data of the interaction part to obtain nonlinear space mapping data, and clustering and denoising the nonlinear space mapping data to obtain a target clustering result; A modeling module, used to perform virtual interaction intention recognition on the plurality of standard interaction part feature images to obtain a virtual interaction intention set, and perform intelligent interaction control modeling on the target clustering results and the virtual interaction intention set to obtain an initial virtual environment interaction control model; A recognition module, used to obtain second virtual interactive video data of the target user, perform image key point recognition on the second virtual interactive video data to obtain a plurality of target center key points, and perform drift correction on the plurality of target center key points to obtain a target center key point heat map; A feature extraction module is used to perform virtual reality environment interactive control on the target center key point heat map through the initial virtual environment interactive control model, obtain a feedback response evaluation parameter set of the virtual reality interactive terminal, and perform feature extraction on the feedback response evaluation parameter set to obtain feedback response evaluation parameter features; The customized update module is used to generate a control model parameter update strategy based on the feedback response evaluation parameter characteristics, and update the model parameters of the initial virtual environment interactive control model according to the control model parameter update strategy to obtain a customized virtual environment interactive control model for the target user.

Images