An artificial intelligence-based scene VR interaction method

Abstract

The application discloses a scene VR interaction method based on artificial intelligence and relates to the technical field of interaction processing, and comprises the following steps: S1, a current visual angle of a user wearing a VR device is acquired; S2, when the user issues a scene switching instruction, a target visual angle of the user is acquired, and a transition visual angle picture is generated according to the current visual angle and the target visual angle of the user; S3, according to the RGB values of a plurality of prop points of the transition visual angle picture, multivariate color values are generated; and S4, the transition visual angle picture is subjected to color updating by using the multivariate color values. The application solves the problems that the visual angle transition is unnatural and the color processing is single in the traditional VR interaction, and provides technical support for high-immersion VR applications (such as games and training simulation).

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of interactive processing, and in particular to a scene VR interaction method based on artificial intelligence. BACKGROUND

[0002] With the rapid development of virtual reality (VR) technology, users' demand for immersive interactive experience is increasing. In VR scenes, view switching and picture transition are key links that affect user experience. Traditional scene switching methods often present unnatural pictures. On the one hand, the picture continuity is insufficient during view switching, which can easily cause visual fatigue or even dizziness in users; on the other hand, the color processing method is single, and only global brightness or contrast adjustment is used to optimize the picture, which cannot adapt to the color detail requirements of local props in complex scenes.

[0003] In recent years, the application of artificial intelligence technology in the field of computer vision provides a new idea for VR interaction optimization. Existing research attempts to use deep learning models to predict user view changes, such as analyzing user head movement trajectories through convolutional neural networks (CNN) to render possible target view pictures in advance. However, this method still has limitations: the prediction model relies on a large amount of training data, and it is not very adaptable to sudden view jumping behavior of users. In addition, in terms of color processing, existing technologies mostly use static adjustment algorithms based on RGB color space, and fail to dynamically optimize color performance in combination with scene content, resulting in insufficient color consistency between transition pictures and target scenes.

[0004] Under the background of continuous improvement of VR device hardware performance, users have higher requirements for the smoothness and realism of scene switching. Traditional transition methods lack intelligent perception of dynamic characteristics of view changes, and are difficult to meet users' needs. SUMMARY

[0005] To solve the above problems, the present application provides a scene VR interaction method based on artificial intelligence.

[0006] The technical solution of the present application is: a scene VR interaction method based on artificial intelligence includes the following steps:

[0007] S1, obtaining the current view of a user wearing a VR device;

[0008] S2, when the user issues a scene switching instruction, obtaining the target view of the user, and generating a transition view picture according to the current view and the target view of the user;

[0009] S3, generating a multi-element color value according to the RGB values of a plurality of prop points of the transition view picture;

[0010] S4, updating the color of the transition view picture using the multi-element color value.

[0011] Further, S3 comprises the following sub-steps:

[0012] S31, extracting a plurality of prop points of the transition view picture;

[0013] S32, generating a multivariate color value by using a plurality of vectors contained in the prop points.

[0014] The beneficial effect of the above further scheme is that in the present application, in the transition picture of the VR scene switching, the pixel area of the key prop (such as object, character or scene element) is recognized and positioned. Through image segmentation or target detection algorithm (such as CNN-based region proposal network), the prop points which have greater influence on visual continuity are screened out. The traditional global color adjustment is upgraded to local optimization based on prop points, and more fine color processing is realized.

[0015] Further, S32 comprises the following sub-steps:

[0016] S321, extracting the centroid of the prop point;

[0017] S322, converting the RGB value of the pixel point where the centroid is located into a center vector, and converting the RGB value of the remaining pixel points into a low-dimensional vector;

[0018] S323, performing dot product operation on the center vector and the low-dimensional vector of the remaining pixel points as center data;

[0019] S324, performing F2 norm operation on the center vector and the low-dimensional vector of the remaining pixel points as sample data;

[0020] S325, generating a multivariate color value for the prop point according to the sample data and the center data of the remaining pixel points.

[0021] The beneficial effect of the above further scheme is that in the present application, the geometric centroid (such as the barycentric coordinate calculated after binarization) of the pixel area of each prop point is calculated to determine its spatial center position, which represents the core area of the prop and provides a spatial reference for subsequent color analysis. The center vector focuses on the core color information, and the low-dimensional vector simplifies the redundant data; and the F2 norm is sensitive to outliers, which is suitable for local detail processing of the prop point. When the user quickly turns his head, the ground texture and other prop points in the transition picture are adjusted by the multivariate color value to avoid color mutation.

[0022] Further, in S324, the multivariate color value of the prop point is calculated according to the following formula:

[0023] ;

[0024] In the formula, represents the first a low-dimensional vector, a center vector representing a prop point, a low-dimensional vector corresponding to the RGB value corresponding to the low-dimensional vector, RGB value corresponding to the center vector of the prop point, an index, a Gaussian kernel parameter, center data corresponding to the low-dimensional vector, center data corresponding to the low-dimensional vector, sample data corresponding to the low-dimensional vector, sample data corresponding to the low-dimensional vector, a number of low-dimensional vectors of the prop point, a minimum operation, a maximum operation.

[0025] The above further scheme has the beneficial effect that in the present application, the multi-element color value is integrated into a color, local color transition is realized, the Gaussian kernel simulates the sensitivity of the human eye to close-range details, nonlinear enhancement is performed, and the sense of reality is improved.

[0026] Further, S4 comprises the following sub-steps:

[0027] S41, clustering a plurality of multi-element color values to obtain a plurality of clusters;

[0028] S42, extracting a centroid of each cluster;

[0029] S43, updating the transition visual picture according to the centroid of each cluster.

[0030] The above further scheme has the beneficial effect that in the present application, the color distribution of the transition picture is optimized through cluster analysis, the local color is precisely adjusted using the dynamically updated coefficient, the visual coherence and the sense of reality are improved. The color update is upgraded from global uniform adjustment to local adaptive optimization based on clustering, and the update strength is dynamically adjusted in combination with statistical characteristics.

[0031] Further, S43 comprises the following sub-steps:

[0032] S431, calculating an update coefficient of each cluster according to the centroid of each cluster;

[0033] S432, multiplying the update coefficient of the cluster corresponding to the prop point by the RGB value of each pixel point in the prop point to complete the transition visual picture update.

[0034] The beneficial effect of the further scheme is that in the application, the unsupervised learning algorithm is used to group the multivariate color values generated in step S3, which can reduce the redundant calculation of color adjustment, focus on the key color area, retain the color difference under different materials or lighting conditions in the scene, and avoid over-smoothing. The coefficients are generated according to the statistical characteristics of the colors in the cluster, and the influence intensity of the centroid on the pixel color is controlled. The coefficients are multiplied by the pixel RGB value to realize local color enhancement or suppression.

[0035] Further, in S431, the update coefficient of the cluster The calculation formula is:

[0036] ;

[0037] In the formula, The centroid of the cluster is represented by The minimum value in the cluster is represented by The maximum value in the cluster is represented by The median in the cluster is represented by.

[0038] The beneficial effect of the further scheme is that in the application, the centroid represents the typical characteristics of the colors in the cluster, and the median centroid can resist the interference of outliers.

[0039] The beneficial effect of the application is that the application extracts the transition perspective picture between the current perspective and the target perspective of the user, extracts the RGB value of the key prop point in the transition picture, generates multivariate color values and dynamically adjusts the picture color, improves the scene realism, reduces the visual discontinuity when the perspective is switched, reduces the complexity of visual interaction through color conversion and local area optimization, guarantees the real-time interaction performance, solves the problem of unnatural perspective transition and single color processing in traditional VR interaction, and provides technical support for high-immersion VR applications such as games and training simulation. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 The flowchart of the scene VR interaction method based on artificial intelligence. DETAILED DESCRIPTION

[0041] The embodiments of the application will be further described below with reference to the accompanying drawings.

[0042] As shown in Figure 1 The application provides a scene VR interaction method based on artificial intelligence, which comprises the following steps:

[0043] S1, obtaining the current perspective of the user wearing the VR device;

[0044] S2, when the user issues a scene switching instruction, obtaining the target perspective of the user, and generating a transition perspective picture according to the current perspective and the target perspective of the user;

[0045] S3, generating a multi-element color value according to RGB values of the several prop points of the transition view picture;

[0046] S4, performing color updating on the transition view picture by using the multi-element color value.

[0047] In the embodiment of the present application, S3 comprises the following sub-steps:

[0048] S31, extracting several prop points of the transition view picture;

[0049] S32, generating a multi-element color value by using several vectors contained in the prop points.

[0050] In the present application, in the transition picture of the VR scene switching, the pixel area of the key prop (such as an object, a character or a scene element) is recognized and positioned. By using an image segmentation or target detection algorithm (such as a region proposal network based on CNN), the prop points which have a greater influence on visual continuity are screened out. The traditional global color adjustment is upgraded to local optimization based on the prop points, so that more fine color processing is realized.

[0051] In the embodiment of the present application, S32 comprises the following sub-steps:

[0052] S321, extracting the centroid of the prop point;

[0053] S322, converting the RGB value of the pixel point where the centroid is located into a center vector, and converting the RGB value of the remaining pixel points into low-dimensional vectors;

[0054] S323, performing dot product operation on the center vector and the low-dimensional vectors of the remaining pixel points as center data;

[0055] S324, performing F2 norm operation on the center vector and the low-dimensional vectors of the remaining pixel points as sample data;

[0056] S325, generating a multi-element color value for the prop point according to the sample data and the center data of the remaining pixel points.

[0057] In the present application, the geometric centroid (such as the barycentric coordinate calculated after binarization) of the pixel area of each prop point is calculated, the spatial center position is determined, the core area of the prop is represented, and the spatial reference is provided for subsequent color analysis. The center vector focuses on the core color information, the low-dimensional vector simplifies the redundant data; and the F2 norm is sensitive to abnormal values, and is suitable for local detail processing of the prop point. When the user quickly turns his head, the prop points such as ground texture in the transition picture are adjusted by using the multi-element color value, so that color mutation is avoided.

[0058] In the embodiment of the present application, in S324, the multi-element color value of the prop point is The calculation formula is:

[0059] ;

[0060] In the formula, Indicates the first item in the item list A low-dimensional vector, Represents the center vector of the item point. Indicates the first item in the item list The RGB values ​​corresponding to each low-dimensional vector. This represents the RGB value corresponding to the center vector of the item point. Indicates an index. Represents the Gaussian kernel parameters. Indicates the first item in the item list The central data corresponding to each low-dimensional vector Indicates the first item in the item list Sample data corresponding to each low-dimensional vector The number of low-dimensional vectors representing item points. This represents the minimum value operation. This indicates the maximum value operation.

[0061] In this invention, multiple color values ​​are combined to achieve smooth local color transitions, and a Gaussian kernel is used to simulate the human eye's sensitivity to close-up details, performing non-linear enhancement to improve realism.

[0062] In this embodiment of the invention, S4 includes the following sub-steps:

[0063] S41. Cluster several multi-dimensional color values ​​to obtain several clusters;

[0064] S42. Extract the centroids of each cluster;

[0065] S43. Update the transition visuals based on the centroid of each cluster.

[0066] In this invention, cluster analysis is used to optimize the color distribution of transitional scenes, and dynamically updated coefficients are used to achieve precise local color adjustments, thereby improving visual coherence and realism. Color updates are upgraded from a globally uniform adjustment to a locally adaptive optimization based on clusters, with the update intensity dynamically adjusted in conjunction with statistical features.

[0067] In this embodiment of the invention, S43 includes the following sub-steps:

[0068] S431. Calculate the update coefficient of each cluster based on the centroid of each cluster;

[0069] S432. Multiply the update coefficient of the cluster corresponding to the prop point by the RGB value of each pixel in the prop point to complete the transition visual update.

[0070] In this invention, an unsupervised learning algorithm is used to group the multi-dimensional color values ​​generated in step S3. This reduces redundant calculations in color adjustment, focuses on key color regions, preserves color differences under different materials or lighting conditions in the scene, and avoids over-smoothing. Coefficients are generated based on the statistical characteristics of color within clusters to control the intensity of the centroid's influence on pixel color. Multiplying the coefficients by the pixel's RGB value achieves local color enhancement or suppression.

[0071] In this embodiment of the invention, in S431, the cluster update coefficient The calculation formula is:

[0072] ;

[0073] In the formula, Represents the centroid of the cluster. Represents the minimum value within the cluster. Represents the maximum value within the cluster. This represents the median within the cluster.

[0074] In this invention, the centroid represents the typical characteristics of the color within a cluster, and the median centroid can resist outlier interference.

[0075] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1.A method for scene VR interaction based on artificial intelligence, characterized in that, The method comprises the following steps: S1, obtaining a current visual angle of a user wearing a VR device; S2, obtaining a target visual angle of the user when the user issues a scene switching instruction, and generating a transition visual angle picture according to the current visual angle and the target visual angle of the user; S3, generating a multi-element color value according to RGB values of a plurality of prop points of the transition visual angle picture; S4, updating the color of the transition visual angle picture by using the multi-element color value; The S3 comprises the following sub-steps: S31, extracting a plurality of prop points of the transition visual angle picture; S32, generating a multi-element color value by using a plurality of vectors contained in the prop points; The S32 comprises the following sub-steps: S321, extracting a centroid of the prop points; S322, converting the RGB value of a pixel point where the centroid is located into a center vector, and converting the RGB value of the remaining pixel points into low-dimensional vectors; S323, performing a dot product operation on the center vector and the low-dimensional vectors of the remaining pixel points as center data; S324, performing an F2 norm operation on the center vector and the low-dimensional vectors of the remaining pixel points as sample data; S325, generating a multi-element color value for the prop points according to the sample data and the center data of the remaining pixel points; In the S324, the multi-element color value of the prop point The calculation formula is: ； In the formula, Indicates the first item in the item list A low-dimensional vector, Represents the center vector of the item point. Indicates the first item in the item list The RGB values ​​corresponding to each low-dimensional vector. This represents the RGB value corresponding to the center vector of the item point. Indicates an index. Represents the Gaussian kernel parameters. Indicates the first item in the item list The central data corresponding to each low-dimensional vector Indicates the first item in the item list Sample data corresponding to each low-dimensional vector The number of low-dimensional vectors representing item points. This represents the minimum value operation. This represents the maximum value operation; The S4 comprises the following sub-steps: S41, clustering a plurality of multi-element color values to obtain a plurality of clusters; S42, extracting a centroid of each cluster; S43, updating the transition visual picture according to the centroid of each cluster; The S43 comprises the following sub-steps: S431, calculating an update coefficient of each cluster according to the centroid of each cluster; S432, multiplying the update coefficient of the cluster corresponding to the prop points and the RGB value of each pixel point in the prop points to complete the update of the transition visual picture. 2.The AI-based scene VR interaction method of claim 1, wherein, In the S431, the update coefficient of the cluster is calculated The calculation formula is as follows. ； wherein denotes the centroid of the cluster, denotes the minimum value within the cluster, denotes the maximum value within the cluster, denotes the median value within the cluster.

Images