Three-dimensional scene real-time rendering optimization system and method based on 3D gaussian sputtering

Abstract

The application relates to the technical field of scene rendering processing, and discloses a three-dimensional scene real-time rendering optimization system and method based on 3D Gaussian sputtering, which comprises a real-time rendering three-dimensional scene identification module, a real-time rendering three-dimensional scene point cloud model establishment module and a real-time rendering three-dimensional scene optimal point cloud model generation module. Through independent and efficient real-time rendering three-dimensional scene sparse point cloud model modeling processing based on real-time rendering three-dimensional scene two-dimensional image parameters and real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control parameters, a plurality of sparse point cloud models of the real-time rendering three-dimensional scene can be scientifically established based on different sparse point cloud model modeling software objects. Based on real-time rendering three-dimensional scene required standard sparse point cloud model parameters and real-time rendering three-dimensional scene real-time modeling sparse point cloud model parameters, the optimal sparse point cloud model of the real-time rendering three-dimensional scene can be scientifically matched by combining an artificial intelligence recognition algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of scene rendering processing, in particular to a three-dimensional scene real-time rendering optimization system and method based on 3D Gaussian sputtering. BACKGROUND

[0002] The core of 3D Gaussian sputtering is to use a set of points to represent a 3D scene, each point with specific attributes for parameterizing a set of anisotropic 3D Gaussian distributions; these Gaussian distributions overlap when rendering an image, forming the final image; the rendering process 3D Gaussian sputtering uses alpha-blending techniques, integrating along rays emitted from the camera, through the image plane and the scene; all intersecting Gaussian distributions contribute to the color of the final pixel. Unlike neural radiance fields, 3D Gaussian sputtering does not require deep learning models, but generates images by overlapping Gaussian distributions of different sizes and colors; the 3D Gaussian sputtering three-dimensional scene rendering steps include the following steps: 1. generating a sparse point cloud structure from multi-view 2D images, 2. converting the point cloud structure into 3D Gaussian functions containing parameters such as position, color, and covariance matrix; 3. optimizing Gaussian parameters using stochastic gradient descent, combined with adaptive density control to enhance details; 4. optimizing Gaussian projection through backpropagation, achieving efficient real-time rendering; the existing 3D Gaussian sputtering three-dimensional scene real-time rendering process cannot realize intelligent and accurate modeling of the sparse point cloud model of the three-dimensional scene, reducing the efficiency and quality of the 3D Gaussian sputtering three-dimensional scene real-time rendering.

[0003] Chinese patent application for invention with publication number CN119169171A and publication date of 2024.12.20 discloses a three-dimensional scene rendering processing method and device, by obtaining the transformation matrix corresponding to each three-dimensional model included in the three-dimensional scene, based on the transformation matrix corresponding to the three-dimensional model to be rendered, the initial rendering pose of the three-dimensional scene is transformed to obtain the target rendering pose corresponding to the three-dimensional model to be rendered; based on the three-dimensional model to be rendered, the target picture corresponding to the target rendering pose is rendered; however, the above technical solution cannot realize intelligent and accurate collection of the transformation matrix parameters of the three-dimensional model. SUMMARY

[0004] (I) Technical problems solved

[0005] To solve the problems that the existing 3D Gaussian sputtering three-dimensional scene real-time rendering process cannot realize intelligent and accurate modeling of sparse point cloud model of three-dimensional scene, and reduces the efficiency and quality of 3D Gaussian sputtering three-dimensional scene real-time rendering, the above dynamic identification real-time rendering three-dimensional scene type, accurate matching real-time rendering three-dimensional scene standard sparse point cloud model, accurate analysis real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameter, efficient construction real-time rendering three-dimensional scene sparse point cloud model, intelligent construction real-time rendering three-dimensional scene optimal sparse point cloud model are realized.

[0006] (II) Technical scheme

[0007] The application is implemented by the following technical scheme: a three-dimensional scene real-time rendering optimization method based on 3D Gaussian sputtering, the method comprising the following steps:

[0008] S1, collecting real-time rendering three-dimensional scene feature text data and real-time rendering three-dimensional scene two-dimensional image data;

[0009] S2, three-dimensional scene type identification processing based on the real-time rendering three-dimensional scene feature text data and different types of real-time rendering three-dimensional scene feature text data, to generate real-time rendering three-dimensional scene type identification data;

[0010] S3, standard sparse point cloud model matching processing of the real-time rendering three-dimensional scene type based on the real-time rendering three-dimensional scene type identification data and different types of real-time rendering three-dimensional scene standard sparse point cloud model data, to construct the required standard sparse point cloud model data of real-time rendering three-dimensional scene;

[0011] S4, sparse point cloud model modeling software modeling control parameter analysis processing of the real-time rendering three-dimensional scene type based on the real-time rendering three-dimensional scene type identification data and real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data, to generate the required sparse point cloud model modeling software modeling control data of real-time rendering three-dimensional scene;

[0012] S5, sparse point cloud model modeling processing of real-time rendering three-dimensional scene based on the real-time rendering three-dimensional scene two-dimensional image data and the required sparse point cloud model modeling software modeling control data of real-time rendering three-dimensional scene, to generate real-time modeling sparse point cloud model data of real-time rendering three-dimensional scene

[0013] S6, sparse point cloud model data collection processing of real-time rendering three-dimensional scene based on the required standard sparse point cloud model data of real-time rendering three-dimensional scene and the real-time modeling sparse point cloud model data of real-time rendering three-dimensional scene, to construct sparse point cloud model collection data of real-time rendering three-dimensional scene;

[0014] S7, according to the real-time rendering three-dimensional scene sparse point cloud model data collection data for real-time rendering three-dimensional scene optimal sparse point cloud model generation processing, generate real-time rendering three-dimensional scene optimal sparse point cloud model data.

[0015] Preferably, the operation steps of collecting real-time rendering three-dimensional scene feature text data and real-time rendering three-dimensional scene two-dimensional image data are as follows:

[0016] S11, through the voice input module online collection target real-time rendering three-dimensional scene scene feature text information, and generate real-time rendering three-dimensional scene feature text data J, the real-time rendering three-dimensional scene feature text data includes target real-time rendering three-dimensional scene name feature information, structure feature information, appearance feature information and use feature information;

[0017] Through the camera from different space direction online collection target real-time rendering three-dimensional scene two-dimensional image information, and generate real-time rendering three-dimensional scene two-dimensional image data set A=(a1,…,a m ,…,a ε ), m=1,2,3,…, epsilon; wherein a m The mth space direction of the real-time rendering three-dimensional scene two-dimensional image data collected, epsilon represents the maximum value of the number of space direction.

[0018] Preferably, based on the real-time rendering three-dimensional scene feature text data and different types of real-time rendering three-dimensional scene feature text data for real-time rendering three-dimensional scene type identification processing, the operation steps of generating real-time rendering three-dimensional scene type identification data are as follows:

[0019] S21, establish different types of real-time rendering three-dimensional scene feature text data set K=(k1,…,k n ,…,k φ ), n=1,2,3,…, phi; wherein k n The nth real-time rendering specific three-dimensional scene type corresponding to different types of real-time rendering three-dimensional scene feature text data, phi represents the maximum value of the number of real-time rendering specific three-dimensional scene type, real-time rendering specific three-dimensional scene type includes game development specific three-dimensional scene type, animation specific three-dimensional scene type, industrial design specific three-dimensional scene type, AR augmented reality specific three-dimensional scene type, VR virtual reality specific three-dimensional scene type, meteorological visualization specific three-dimensional scene type and geological visualization specific three-dimensional scene type; the different types of real-time rendering three-dimensional scene feature text data indicates that the three-dimensional scene feature information is set for different real-time rendering specific three-dimensional scene type standard;

[0020] S22, using the XGBoost search algorithm to search the real-time rendering three-dimensional scene feature text data J and the different types of real-time rendering three-dimensional scene feature text data k in the different types of real-time rendering three-dimensional scene feature text data set K n Performing three-dimensional scene feature keyword matching, searching out the different types of real-time rendering three-dimensional scene feature text data k matched with the real-time rendering three-dimensional scene feature text data J n Corresponding real-time rendering specific three-dimensional scene type text information, and generating real-time rendering three-dimensional scene type identification data J through data identification leixing .

[0021] Preferably, according to the real-time rendering three-dimensional scene type identification data and the different types of real-time rendering three-dimensional scene standard sparse point cloud model data, the operation steps of constructing the standard sparse point cloud model data required by the real-time rendering three-dimensional scene are as follows:

[0022] S31, establishing a different type of real-time rendering three-dimensional scene standard sparse point cloud model data set K'=(k'1,…,k' n ,…,k' φ ), wherein k' n represents the different types of real-time rendering three-dimensional scene standard sparse point cloud model data corresponding to the nth real-time rendering specific three-dimensional scene type, and the different types of real-time rendering three-dimensional scene standard sparse point cloud model data represent the standard real-time rendering three-dimensional scene sparse point cloud model parameters set for different real-time rendering specific three-dimensional scene types;

[0023] S32, using a bidirectional search algorithm to search the real-time rendering three-dimensional scene type identification data J leixing and the different types of real-time rendering three-dimensional scene standard sparse point cloud model data k' n in the different types of real-time rendering three-dimensional scene standard sparse point cloud model data set K' leixing Performing real-time rendering specific three-dimensional scene type keyword matching, searching out the real-time rendering three-dimensional scene type identification data J n corresponding to the different types of real-time rendering three-dimensional scene standard sparse point cloud model data k' mubiao , and constructing the standard sparse point cloud model data k required by the real-time rendering three-dimensional scene through data identification mubiao When the standard sparse point cloud model data k required by the real-time rendering three-dimensional scene is not constructed, continue to perform the standard sparse point cloud model data k required by the real-time rendering three-dimensional scene mubiao Construction operation instruction.

[0024] Preferably, the steps for analyzing and processing the sparse point cloud modeling control parameters of the real-time rendered 3D scene type based on the real-time rendered 3D scene type identification data and the real-time rendered 3D scene sparse point cloud modeling software modeling control data to generate the sparse point cloud modeling software modeling control data required for real-time rendered 3D scene are as follows:

[0025] S41, when the standard sparse point cloud model data k required for the real-time rendering of the 3D scene is... mubiao Upon completion, a real-time rendering 3D scene sparse point cloud model is established, and the modeling software controls the data set matrix. Q p This represents the set of modeling control data for real-time rendering of sparse point cloud models in 3D scenes, corresponding to the p-th type of sparse point cloud model modeling software. This represents the maximum number of software types for modeling sparse point cloud models of 3D scenes; these software types include COLMAP, OpenMVG, VisualSFM, and SatX; where Q... p =(q p,1 ,…,q p,n ,…,q p,φ ), where q p,n Q represents the modeling control data set Q of the real-time rendering 3D scene sparse point cloud model modeling software. p The modeling control data of the sparse point cloud model modeling software for the nth type of real-time rendering 3D scene corresponds to the nth type of real-time rendering 3D scene. The modeling control data of the sparse point cloud model modeling software for the real-time rendering of 3D scene represents the optimal sparse point cloud model modeling control parameters set by the sparse point cloud model modeling software for different types of real-time rendering 3D scene. The sparse point cloud model modeling control parameters include feature type control parameters, feature point data limitation control parameters, matching threshold control parameters, minimum viewpoint number control parameters, reprojection error threshold control parameters, and camera model selection control parameters of the sparse point cloud model model modeling software for 3D scene.

[0026] S42. The depth-constrained search algorithm is used to identify the real-time rendered 3D scene type data J. leixing Within the modeling control data set matrix Q of the real-time rendered 3D scene sparse point cloud model modeling software, all the modeling control data sets Q of the real-time rendered 3D scene sparse point cloud model modeling software are searched in an orderly manner according to the type number of the 3D scene sparse point cloud model modeling software. p The real-time rendered 3D scene type recognition data J leixing The real-time rendering software uses sparse point cloud models of 3D scenes that match the specific 3D scene type for modeling control data q.p,n and through data identification generate real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data set where q′ p representing the real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data corresponding to the pth three-dimensional scene sparse point cloud model modeling software type.

[0027] Preferably, based on the real-time rendering three-dimensional scene two-dimensional image data, the real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data for real-time rendering three-dimensional scene sparse point cloud model modeling processing, generate real-time rendering three-dimensional scene real-time modeling sparse point cloud model data operation steps as follows:

[0028] S51, according to the real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data set Q′ in the real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data q′1 to control corresponding three-dimensional scene sparse point cloud model modeling software combined with the real-time rendering three-dimensional scene two-dimensional image data set A in the real-time rendering three-dimensional scene two-dimensional image data a1 to a ε real-time rendering three-dimensional scene sparse point cloud model modeling processing, and generate real-time rendering three-dimensional scene real-time modeling sparse point cloud model data set where h p representing the real-time rendering three-dimensional scene real-time modeling sparse point cloud model data generated by the pth three-dimensional scene sparse point cloud model modeling software type through modeling.

[0029] Preferably, based on the real-time rendering three-dimensional scene required standard sparse point cloud model data, the real-time rendering three-dimensional scene real-time modeling sparse point cloud model data for real-time rendering three-dimensional scene sparse point cloud model data collection processing, to build real-time rendering three-dimensional scene sparse point cloud model collection data operation steps as follows:

[0030] S61, to generate the real-time rendering three-dimensional scene required standard sparse point cloud model data k mubiao , the real-time rendering three-dimensional scene real-time modeling sparse point cloud model data set H data combination to build real-time rendering three-dimensional scene sparse point cloud model collection data L, where L=(k mubiao ,H).

[0031] Preferably, according to the real-time rendering three-dimensional scene sparse point cloud model collection data for real-time rendering three-dimensional scene optimal sparse point cloud model establishment processing, generate real-time rendering three-dimensional scene optimal sparse point cloud model data operation steps as follows:

[0032] S71. Collect the standard sparse point cloud model data k required for real-time rendering of the sparse point cloud model in the data L of the real-time rendering 3D scene sparse point cloud model collection data. mubiao The data is ordered according to the software type number of the 3D scene sparse point cloud model and the real-time rendering 3D scene sparse point cloud model data set H. p Perform sparse point cloud model feature matching to search for standard sparse point cloud model data k required for the real-time rendering of the 3D scene. mubiao The most matching sparse point cloud model data h for real-time rendering of 3D scenes p And through data identification, the optimal sparse point cloud model data h for real-time rendering of the 3D scene is generated. zuiyou The process generates the optimal sparse point cloud model data h for the real-time rendered 3D scene. zuiyou The specific operating steps are as follows:

[0033] S711. Initialization phase: Update the point cloud model to search for whale population location and the maximum iteration algorithm T. The formula for updating the point cloud model to search for whale population location is as follows: Among them U i This represents the position of the whale individual i in the search space of the sparse point cloud model data set H in the real-time rendering 3D scene. and These represent the upper and lower bounds of the search space for individual whales in the real-time rendering 3D scene sparse point cloud model data set H, respectively, and rand represents a random number in the interval [0,1].

[0034] S712, In the prey-surrounding stage, when surrounding prey, the point cloud model search whale will either choose to swim towards the point cloud model search whale in the optimal position or randomly swim towards a point cloud model search whale. The point cloud model search whale searches for the standard sparse point cloud model data k required for the real-time rendering 3D scene within the search space of the real-time rendering sparse point cloud model data set H. mubiao The most matching sparse point cloud model data h for real-time rendering of 3D scenes p Alternatively, a random search can be performed to find the standard sparse point cloud model data k required for the real-time rendering of the 3D scene. mubiao Matching the real-time rendering 3D scene real-time modeling sparse point cloud model data h p ;

[0035] S7121, when selecting to swim toward the optimal position of the point cloud model search whale individual, that is, the point cloud model search whale individual searches in the search space of the real-time modeling sparse point cloud model data set H of the real-time rendering three-dimensional scene to search out the required standard sparse point cloud model data k of the real-time rendering three-dimensional scene mubiao The most matched real-time modeling sparse point cloud model data h of the real-time rendering three-dimensional scene p The position update formula of the point cloud model search whale individual is as follows: Wherein U represents the position of the point cloud model search whale individual j in the search space of the real-time modeling sparse point cloud model data set H of the real-time rendering three-dimensional scene after the t+1th iteration, U j t U represents the position of the point cloud model search whale individual j in the search space of the real-time modeling sparse point cloud model data set H of the real-time rendering three-dimensional scene after the tth iteration, U best t U represents the optimal position of the point cloud model search whale individual in the search space of the real-time modeling sparse point cloud model data set H of the real-time rendering three-dimensional scene after the tth iteration, and represents a random number in the value (-Π, Π), and the initial value of Π is 2, which is linearly decreased to 0 with the number of iterations; represents a random number in the interval (0, 2), and || represents the absolute value symbol;

[0036] S7122, when selecting to swim toward the position of the random point cloud model search whale individual, the point cloud model search whale individual randomly searches in the search space of the real-time modeling sparse point cloud model data set H of the real-time rendering three-dimensional scene to search out the required standard sparse point cloud model data k of the real-time rendering three-dimensional scene mubiao The most matched real-time modeling sparse point cloud model data h of the real-time rendering three-dimensional scene p The position update formula of the point cloud model search whale individual is as follows: Wherein U rand t U represents the position of the randomly selected point cloud model search whale individual in the search space of the real-time modeling sparse point cloud model data set H of the real-time rendering three-dimensional scene after the tth iteration;

[0037] When Φ<1, the point cloud model search whale individual selects to swim toward the optimal point cloud model search whale individual, and the point cloud model search whale individual searches in the search space of the real-time modeling sparse point cloud model data set H of the real-time rendering three-dimensional scene to search out the required standard sparse point cloud model data k of the real-time rendering three-dimensional scene mubiao The most matched real-time modeling sparse point cloud model data h of the real-time rendering three-dimensional scene p ;

[0038] When Φ≥1, the point cloud model search whale individual selects to swim towards a random point cloud model search whale individual, and the point cloud model search whale individual randomly searches in the search space of the real-time modeling sparse point cloud model data set H of the real-time rendering three-dimensional scene to obtain the required standard sparse point cloud model data k of the real-time rendering three-dimensional scene mubiao The matched real-time modeling sparse point cloud model data h of the real-time rendering three-dimensional scene p

[0039] S713, bubble net stage, the point cloud model search whale individual will spray steam pocket to form bubble net when hunting to drive the prey, the point cloud model search whale individual uses the bubble net to drive the prey, and the point cloud model search whale individual constantly updates its own position in the search space of the real-time modeling sparse point cloud model data set H of the real-time rendering three-dimensional scene, when using the bubble net, the position updating formula of the point cloud model search whale individual is as follows:

[0040] U j t+1 = |U best t -U j t | × e ψζ × cos (2πΩ) + U best t Wherein ψ represents a constant with a value of 1, ζ represents a random number in the interval [-1, 1], e ψζ represents the exponential power of the product of ψ and ζ with e as the base number; 2πΩ is the total angle value of the bubble net, Ω represents the number of the bubble net, and cos(2πΩ) represents the cosine value of 2πΩ;

[0041] S714, when the maximum iteration algorithm is satisfied, output the required standard sparse point cloud model data k of the real-time rendering three-dimensional scene mubiao The most matched real-time modeling sparse point cloud model data h of the real-time rendering three-dimensional scene p

[0042] S715, output the real-time modeling sparse point cloud model data h of the real-time rendering three-dimensional scene in step S714 p The optimal sparse point cloud model data h of the real-time rendering three-dimensional scene generated through data identification zuiyou

[0043] The three-dimensional scene real-time rendering optimization system based on 3D Gaussian sputtering is used to realize the three-dimensional scene real-time rendering optimization method based on 3D Gaussian sputtering, and the system comprises a real-time rendering three-dimensional scene identification module, a real-time rendering three-dimensional scene point cloud model establishment module, and a real-time rendering three-dimensional scene optimal point cloud model generation module; ​​​

[0044] The real-time rendering three-dimensional scene recognition module comprises a real-time rendering three-dimensional scene feature information collection unit, a real-time rendering three-dimensional scene two-dimensional image collection unit, a different type real-time rendering three-dimensional scene feature information storage unit, and a real-time rendering three-dimensional scene type recognition unit.

[0045] The real-time rendering three-dimensional scene feature information collection unit collects real-time rendering three-dimensional scene feature text data through a voice input module. The real-time rendering three-dimensional scene two-dimensional image collection unit collects real-time rendering three-dimensional scene two-dimensional image data through a shooting camera. The different type real-time rendering three-dimensional scene feature information storage unit is used for storing different type real-time rendering three-dimensional scene feature text data. The real-time rendering three-dimensional scene type recognition unit performs real-time rendering three-dimensional scene type recognition processing based on the real-time rendering three-dimensional scene feature text data and different type real-time rendering three-dimensional scene feature text data, and generates real-time rendering three-dimensional scene type recognition data.

[0046] The real-time rendering three-dimensional scene point cloud model establishment module comprises a different type real-time rendering three-dimensional scene standard sparse point cloud model storage unit, a real-time rendering three-dimensional scene standard sparse point cloud model matching unit, a real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameter storage unit, a real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameter analysis unit, and a real-time rendering three-dimensional scene sparse point cloud model construction unit.

[0047] The different type real-time rendering three-dimensional scene standard sparse point cloud model storage unit is used for storing different type real-time rendering three-dimensional scene standard sparse point cloud model data;The real-time rendering three-dimensional scene standard sparse point cloud model matching unit carries out real-time rendering three-dimensional scene type standard sparse point cloud model matching processing according to the real-time rendering three-dimensional scene type identification data and different type real-time rendering three-dimensional scene standard sparse point cloud model data, and constructs the required standard sparse point cloud model data of the real-time rendering three-dimensional scene;The real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameter storage unit is used for storing real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data;The real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameter analysis unit carries out real-time rendering three-dimensional scene type sparse point cloud model modeling software modeling control parameter analysis processing according to the real-time rendering three-dimensional scene type identification data and real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data, and generates the required sparse point cloud model modeling software modeling control data of the real-time rendering three-dimensional scene;The real-time rendering three-dimensional scene sparse point cloud model construction unit carries out sparse point cloud model modeling processing of the real-time rendering three-dimensional scene based on the real-time rendering three-dimensional scene two-dimensional image data and the required sparse point cloud model modeling software modeling control data of the real-time rendering three-dimensional scene, and generates real-time modeling sparse point cloud model data of the real-time rendering three-dimensional scene;

[0048] The real-time rendering three-dimensional scene optimal point cloud model generation module comprises a real-time rendering three-dimensional scene sparse point cloud model collection unit and a real-time rendering three-dimensional scene optimal sparse point cloud model establishment unit.

[0049] The real-time rendering three-dimensional scene sparse point cloud model collection unit carries out sparse point cloud model data collection processing of the real-time rendering three-dimensional scene based on the required standard sparse point cloud model data of the real-time rendering three-dimensional scene and the real-time modeling sparse point cloud model data of the real-time rendering three-dimensional scene, and constructs sparse point cloud model collection data of the real-time rendering three-dimensional scene;The real-time rendering three-dimensional scene optimal sparse point cloud model establishment unit carries out optimal sparse point cloud model establishment processing of the real-time rendering three-dimensional scene according to the sparse point cloud model collection data of the real-time rendering three-dimensional scene, and generates optimal sparse point cloud model data of the real-time rendering three-dimensional scene.

[0050] (Three) beneficial effects

[0051] The application provides a three-dimensional scene real-time rendering optimization system and method based on 3D Gaussian sputtering.

[0052] I. Accurate and efficient acquisition of feature information and two-dimensional images of real-time rendering three-dimensional scene through voice input module and shooting camera, to provide reliable data support for accurate establishment of sparse point cloud model of real-time rendering three-dimensional scene; based on feature information of real-time rendering three-dimensional scene, intelligent search algorithm and scientific preset different types of feature information of real-time rendering three-dimensional scene, intelligent identification of specific three-dimensional scene type of real-time rendering is realized, the quality of real-time rendering of 3D Gaussian sputtering three-dimensional scene is improved.

[0053] II. According to the type information of real-time rendering three-dimensional scene, the intelligent search algorithm and the different types of real-time rendering three-dimensional scene standard sparse point cloud model parameters based on big data storage are combined to accurately match the real-time rendering three-dimensional scene standard sparse point cloud model, and the scientific establishment of real-time rendering three-dimensional scene sparse point cloud model standard reference is realized, and the standardized data support for accurate construction of real-time rendering three-dimensional scene sparse point cloud model is provided; according to the type identification parameters of real-time rendering three-dimensional scene, the intelligent search algorithm and the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameters stored in the standard are combined to dynamically screen the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameters, and the modeling control parameters of different sparse point cloud model software are accurately adjusted based on the type of real-time rendering three-dimensional scene; based on the two-dimensional image parameters of real-time rendering three-dimensional scene and the sparse point cloud model modeling software modeling control parameters required for real-time rendering three-dimensional scene, the sparse point cloud model modeling processing of real-time rendering three-dimensional scene is carried out independently and efficiently, and the multiple sparse point cloud models of real-time rendering three-dimensional scene are scientifically established based on different sparse point cloud model software objects, and the precision of real-time rendering of 3D Gaussian sputtering three-dimensional scene is improved.

[0054] III. Based on the standard sparse point cloud model parameters required for real-time rendering three-dimensional scene and the real-time modeling sparse point cloud model parameters of real-time rendering three-dimensional scene, the optimal sparse point cloud model of real-time rendering three-dimensional scene is scientifically matched by combining artificial intelligence recognition algorithm, the optimal sparse point cloud model of real-time rendering three-dimensional scene based on 3D Gaussian sputtering is efficiently and accurately screened based on artificial intelligence, and the reliability and efficiency of real-time rendering of 3D Gaussian sputtering three-dimensional scene are improved. BRIEF DESCRIPTION OF DRAWINGS

[0055] Fig. 1 The module schematic diagram of the three-dimensional scene real-time rendering optimization system based on 3D Gaussian sputtering provided by the application;

[0056] Fig. 2 The flow chart of the three-dimensional scene real-time rendering optimization method based on 3D Gaussian sputtering provided by the application. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0058] The implementation of the three-dimensional scene real-time rendering optimization system and method based on 3D Gaussian sputtering is as follows:

[0059] Embodiment 1:

[0060] Please refer to Figs. 1-2 The three-dimensional scene real-time rendering optimization method based on 3D Gaussian sputtering includes the following steps:

[0061] S1, collecting real-time rendering three-dimensional scene feature text data and real-time rendering three-dimensional scene two-dimensional image data;

[0062] S2, performing three-dimensional scene type identification processing based on the real-time rendering three-dimensional scene feature text data and different types of real-time rendering three-dimensional scene feature text data, to generate real-time rendering three-dimensional scene type identification data;

[0063] S3, performing standard sparse point cloud model matching processing of the real-time rendering three-dimensional scene type based on the real-time rendering three-dimensional scene type identification data and different types of real-time rendering three-dimensional scene standard sparse point cloud model data, to construct the required standard sparse point cloud model data of the real-time rendering three-dimensional scene;

[0064] S4, performing sparse point cloud model modeling software modeling control parameter analysis processing of the real-time rendering three-dimensional scene type based on the real-time rendering three-dimensional scene type identification data and real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data, to generate the required sparse point cloud model modeling software modeling control data of the real-time rendering three-dimensional scene;

[0065] S5, performing sparse point cloud model modeling processing of the real-time rendering three-dimensional scene based on the real-time rendering three-dimensional scene two-dimensional image data and the required sparse point cloud model modeling software modeling control data of the real-time rendering three-dimensional scene, to generate real-time modeling sparse point cloud model data of the real-time rendering three-dimensional scene

[0066] S6, performing sparse point cloud model data collection processing of the real-time rendering three-dimensional scene based on the required standard sparse point cloud model data of the real-time rendering three-dimensional scene and the real-time modeling sparse point cloud model data of the real-time rendering three-dimensional scene, to construct sparse point cloud model collection data of the real-time rendering three-dimensional scene;

[0067] S7, according to the real-time rendering three-dimensional scene sparse point cloud model data collected for real-time rendering three-dimensional scene optimal sparse point cloud model generation process, generate real-time rendering three-dimensional scene optimal sparse point cloud model data.

[0068] Further, please refer to Figs. 1-2 , the operation steps of collecting real-time rendering three-dimensional scene feature text data and real-time rendering three-dimensional scene two-dimensional image data are as follows:

[0069] S11, through the voice input module online collection target real-time rendering three-dimensional scene scene feature text information, and generate real-time rendering three-dimensional scene feature text data J, real-time rendering three-dimensional scene feature text data includes target real-time rendering three-dimensional scene name feature information, structure feature information, appearance feature information and use feature information;

[0070] Through the camera from different space direction online collection target real-time rendering three-dimensional scene two-dimensional image information, and generate real-time rendering three-dimensional scene two-dimensional image data set A=(a1,…,a m ,…,a ε ), m=1,2,3,…,ε; Wherein a m Indicates the mth space direction of real-time rendering three-dimensional scene two-dimensional image data collected, and ε represents the maximum value of the number of space directions.

[0071] Based on real-time rendering three-dimensional scene feature text data and different types of real-time rendering three-dimensional scene feature text data for real-time rendering three-dimensional scene type identification processing, the operation steps of generating real-time rendering three-dimensional scene type identification data are as follows:

[0072] S21, establish different types of real-time rendering three-dimensional scene feature text data set K=(k1,…,k n ,…,k φ ), n=1,2,3,…,φ; Wherein k n Indicates the nth real-time rendering specific three-dimensional scene type corresponding to different types of real-time rendering three-dimensional scene feature text data, φ represents the maximum value of the number of real-time rendering specific three-dimensional scene type, real-time rendering specific three-dimensional scene type includes game development specific three-dimensional scene type, animation specific three-dimensional scene type, industrial design specific three-dimensional scene type, AR augmented reality specific three-dimensional scene type, VR virtual reality specific three-dimensional scene type, meteorological visualization specific three-dimensional scene type and geological visualization specific three-dimensional scene type; Different types of real-time rendering three-dimensional scene feature text data represent three-dimensional scene feature information set for different real-time rendering specific three-dimensional scene type standard;

[0073] S22, using the XGBoost search algorithm to search the real-time rendering three-dimensional scene feature text data J and different types of real-time rendering three-dimensional scene feature text data k in the different types of real-time rendering three-dimensional scene feature text data set K n Performing three-dimensional scene feature keyword matching, searching for different types of real-time rendering three-dimensional scene feature text data k matching the real-time rendering three-dimensional scene feature text data J n Corresponding real-time rendering specific three-dimensional scene type text information, and generating real-time rendering three-dimensional scene type identification data J through data identification leixing .

[0074] Through the cooperation of the real-time rendering three-dimensional scene feature information acquisition unit and the real-time rendering three-dimensional scene two-dimensional image acquisition unit, the voice input module and the camera are used to accurately and efficiently obtain the feature information and two-dimensional image of the real-time rendering three-dimensional scene, providing reliable data support for accurately establishing the real-time rendering three-dimensional scene sparse point cloud model; the different types of real-time rendering three-dimensional scene feature information storage unit and the real-time rendering three-dimensional scene type identification unit cooperate with each other, based on the real-time rendering three-dimensional scene feature information, combining the intelligent search algorithm with the scientifically preset different types of real-time rendering three-dimensional scene feature information to perform real-time rendering specific three-dimensional scene type intelligent identification, realizing the scientific construction of the sparse point cloud model based on the real-time rendering three-dimensional scene type, and improving the quality of the real-time rendering three-dimensional scene based on the 3D Gaussian sputtering.

[0075] Further, please refer to Figs. 1-2 , according to the real-time rendering three-dimensional scene type identification data and the different types of real-time rendering three-dimensional scene standard sparse point cloud model data, the operation steps of the standard sparse point cloud model matching processing of the real-time rendering three-dimensional scene type are as follows:

[0076] S31, establishing a different type of real-time rendering three-dimensional scene standard sparse point cloud model data set K' = (k'1,..., k' n ,..., k' φ ), where k' n represents the different types of real-time rendering three-dimensional scene standard sparse point cloud model data corresponding to the nth real-time rendering specific three-dimensional scene type, and the different types of real-time rendering three-dimensional scene standard sparse point cloud model data represent the standard real-time rendering three-dimensional scene sparse point cloud model parameters set for different real-time rendering specific three-dimensional scene types;

[0077] S32, using a bidirectional search algorithm to search the real-time rendering three-dimensional scene type identification data J leixing and the different types of real-time rendering three-dimensional scene standard sparse point cloud model data k' nThe real-time rendering specific three-dimensional scene type keyword matching is performed to search for real-time rendering three-dimensional scene type identification data J leixing The corresponding different type real-time rendering three-dimensional scene standard sparse point cloud model data k' n And the data identification is constructed to obtain the real-time rendering three-dimensional scene required standard sparse point cloud model data k mubiao When the real-time rendering three-dimensional scene required standard sparse point cloud model data k mubiao is not constructed, the real-time rendering three-dimensional scene required standard sparse point cloud model data k mubiao is continuously constructed, and the construction operation instruction is executed.

[0078] According to the real-time rendering three-dimensional scene type identification data and the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data, the sparse point cloud model modeling software modeling control parameter analysis processing of the real-time rendering three-dimensional scene type is performed, and the operation steps of generating the real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data are as follows:

[0079] S41, when the real-time rendering three-dimensional scene required standard sparse point cloud model data k mubiao is constructed, the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data set matrix is established Wherein Q p represents the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data set corresponding to the pth three-dimensional scene sparse point cloud model modeling software type, represents the maximum value of the number of three-dimensional scene sparse point cloud model modeling software types; wherein the three-dimensional scene sparse point cloud model modeling software types include COLMAP, OpenMVG, VisualSFM and SatX; wherein Q p =(q p,1 ,…,q p,n ,…,q p,φ ), wherein q p,n represents the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data corresponding to the nth real-time rendering specific three-dimensional scene type in the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data set Q p ; the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data represents the optimal sparse point cloud model modeling control parameters set by the three-dimensional scene sparse point cloud model modeling software for different real-time rendering specific three-dimensional scene types in the three-dimensional scene real-time rendering; wherein the sparse point cloud model modeling control parameters include feature type control parameters, feature point data limit control parameters, matching threshold control parameters, minimum view angle number control parameters, re-projection error threshold control parameters and camera model selection control parameters.

[0080] S42, using a depth-limited search algorithm to identify real-time rendering three-dimensional scene type data J leixing In the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data set matrix Q, according to the three-dimensional scene sparse point cloud model modeling software type number in order to search for all real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data set Q p with real-time rendering three-dimensional scene type identification data J leixing Real-time rendering specific three-dimensional scene type matching real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data q p,n , and generate real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data set Where q' p Indicates the real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data corresponding to the pth three-dimensional scene sparse point cloud model modeling software type.

[0081] Based on real-time rendering three-dimensional scene two-dimensional image data, real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data, real-time rendering three-dimensional scene sparse point cloud model modeling processing, generate real-time rendering three-dimensional scene real-time modeling sparse point cloud model data operation steps as follows:

[0082] S51, according to real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data set Q' in real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data q'1 to Control corresponding three-dimensional scene sparse point cloud model modeling software combined with real-time rendering three-dimensional scene two-dimensional image data set A in real-time rendering three-dimensional scene two-dimensional image data a1 to a ε Real-time rendering three-dimensional scene sparse point cloud model modeling processing, and generate real-time rendering three-dimensional scene real-time modeling sparse point cloud model data set Where h p Indicates the real-time rendering three-dimensional scene real-time modeling sparse point cloud model data generated by the pth three-dimensional scene sparse point cloud model modeling software type through modeling.

[0083] The different types of real-time rendering three-dimensional scene standard sparse point cloud model storage unit and real-time rendering three-dimensional scene standard sparse point cloud model matching unit cooperate with each other, and according to the real-time rendering three-dimensional scene type information, the intelligent search algorithm and the different types of real-time rendering three-dimensional scene standard sparse point cloud model parameters based on big data storage are combined to accurately match the real-time rendering three-dimensional scene standard sparse point cloud model, so that the scientific establishment of the real-time rendering three-dimensional scene sparse point cloud model standard reference is realized, and standardized data support is provided for accurately constructing the real-time rendering three-dimensional scene sparse point cloud model; the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameter analysis unit dynamically screens the sparse point cloud model modeling software modeling control parameters of the real-time rendering three-dimensional scene based on the real-time rendering three-dimensional scene type identification parameters, the intelligent search algorithm and the standard stored real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameters, so that the modeling control parameters of different sparse point cloud model modeling software are accurately adjusted based on the real-time rendering three-dimensional scene type; the real-time rendering three-dimensional scene sparse point cloud model construction unit independently and efficiently performs real-time rendering three-dimensional scene sparse point cloud model modeling processing based on the real-time rendering three-dimensional scene two-dimensional image parameters and the real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control parameters, so that various sparse point cloud models of the real-time rendering three-dimensional scene are scientifically established based on different sparse point cloud model modeling software objects, and the accuracy of the 3D Gaussian sputtering three-dimensional scene real-time rendering is improved.

[0084] Further, please refer to Figs. 1-2 , based on the real-time rendering three-dimensional scene required standard sparse point cloud model data and the real-time rendering three-dimensional scene real-time modeling sparse point cloud model data, the sparse point cloud model data collection processing of the real-time rendering three-dimensional scene is performed, and the operation steps of constructing the real-time rendering three-dimensional scene sparse point cloud model collection data are as follows:

[0085] S61, the generated real-time rendering three-dimensional scene required standard sparse point cloud model data k mubiao , the real-time rendering three-dimensional scene real-time modeling sparse point cloud model data set H are combined to construct the real-time rendering three-dimensional scene sparse point cloud model collection data L, wherein L=(k mubiao ,H).

[0086] According to the real-time rendering three-dimensional scene sparse point cloud model collection data, the optimal sparse point cloud model of the real-time rendering three-dimensional scene is established and processed, and the operation steps of generating the real-time rendering three-dimensional scene optimal sparse point cloud model data are as follows:

[0087] S71, the real-time rendering three-dimensional scene required standard sparse point cloud model data k mubiaoThe data is ordered according to the software type number of the sparse point cloud modeling software for real-time rendering of 3D scenes, and includes the real-time rendering sparse point cloud modeling data set H. p Perform sparse point cloud model feature matching to search for standard sparse point cloud model data k required for real-time rendering of 3D scenes. mubiao The most suitable real-time rendering 3D scene real-time modeling sparse point cloud model data h p And through data identification, the optimal sparse point cloud model data h for real-time rendering of the 3D scene is generated. zuiyou Executes the generation of optimal sparse point cloud model data h for real-time rendering of 3D scenes. zuiyou The specific operating steps are as follows:

[0088] S711. Initialization phase: Update the point cloud model to search for whale population location and the maximum iteration algorithm T. The formula for updating the point cloud model to search for whale population location is as follows: Among them U i This represents the position of the individual whale i in the search space of the sparse point cloud model data set H in the real-time rendering 3D scene. and These represent the upper and lower bounds of the search space for individual whales in the real-time rendering 3D scene sparse point cloud model data set H, respectively, and rand represents a random number in the interval [0,1].

[0089] S712, Surrounding the Prey Phase: When surrounding prey, the point cloud model search whale will either swim towards the point cloud model search whale in the optimal position or randomly swim towards another point cloud model search whale. The point cloud model search whale searches within the search space of the real-time rendering 3D scene sparse point cloud model data set H to find the standard sparse point cloud model data k required for real-time rendering of the 3D scene. mubiao The most suitable real-time rendering 3D scene real-time modeling sparse point cloud model data h p Alternatively, a random search can be performed to find the standard sparse point cloud model data k required for real-time rendering of a 3D scene. mubiao Matching real-time rendering 3D scene real-time modeling sparse point cloud model data h p ;

[0090] S7121. When choosing to swim towards the optimal position of the point cloud model to search for individual whales, that is, when searching for individual whales in the point cloud model, the search space of the sparse point cloud model data set H for real-time rendering of the 3D scene searches for the standard sparse point cloud model data k required for real-time rendering of the 3D scene. mubiao The most suitable real-time rendering 3D scene real-time modeling sparse point cloud model data h p The formula for updating the position of an individual whale using a point cloud model is as follows:

[0091] wherein U j t+1 denotes the position of the whale individual j in the search space of the sparse point cloud model data set H for real-time modeling of the real-time rendered three-dimensional scene after the t+1th iteration, U j t denotes the position of the whale individual j in the search space of the sparse point cloud model data set H for real-time modeling of the real-time rendered three-dimensional scene after the tth iteration, U best t denotes the optimal position of the whale individual in the search space of the sparse point cloud model data set H for real-time modeling of the real-time rendered three-dimensional scene after the tth iteration, and Φ denotes a random number in the interval (-Π, Π), the initial value of Π being 2, which linearly decreases to 0 with the number of iterations; denotes a random number in the interval (0, 2), and || denotes the absolute value symbol;

[0092] S7122, when the position of the whale individual is selected to swim towards the random point cloud model search individual, the whale individual randomly searches the sparse point cloud model data k mubiao matching the real-time rendered three-dimensional scene in the search space of the sparse point cloud model data set H for real-time modeling of the real-time rendered three-dimensional scene; p The position updating formula of the whale individual is as follows:

[0093] wherein U rand t denotes the position of the randomly selected whale individual in the search space of the sparse point cloud model data set H for real-time modeling of the real-time rendered three-dimensional scene after the tth iteration;

[0094] When Φ < 1, the whale individual selects to swim towards the optimal whale individual, and the whale individual searches the sparse point cloud model data k mubiao matching the real-time rendered three-dimensional scene in the search space of the sparse point cloud model data set H for real-time modeling of the real-time rendered three-dimensional scene; p

[0095] When Φ ≥ 1, the whale individual selects to swim towards the random whale individual, and the whale individual randomly searches the sparse point cloud model data k mubiao matching the real-time rendered three-dimensional scene in the search space of the sparse point cloud model data set H for real-time modeling of the real-time rendered three-dimensional scene.The matching real-time rendering three-dimensional scene real-time modeling sparse point cloud model data h p ;

[0096] S713, bubble net stage, the point cloud model searches for a whale individual that will spray steam to form a bubble net to drive prey when hunting, the point cloud model searches for a whale individual that uses a bubble net to drive prey, the point cloud model searches for a whale individual that constantly updates its own position in the search space of the real-time rendering three-dimensional scene real-time modeling sparse point cloud model data set H, when using the bubble net, the position update formula of the point cloud model searching for a whale individual is as follows:

[0097] U j t+1 =|U best t -U j t |×e ψζ ×cos(2πΩ)+U best t , wherein ψ represents a constant with a value of 1, ζ represents a random number in the interval [-1, 1], e ψζ represents the exponential power of the product of ψ and ζ with e as the base number; 2πΩ is the total angle value of the bubble net, Ω represents the number of bubble nets, and cos(2πΩ) represents the cosine value of 2πΩ.

[0098] S714, when the maximum iteration algorithm is met, output the standard sparse point cloud model data k mubiao The most matching real-time rendering three-dimensional scene real-time modeling sparse point cloud model data h p ;

[0099] S715, output the real-time rendering three-dimensional scene real-time modeling sparse point cloud model data h p generated by the data identification real-time rendering three-dimensional scene optimal sparse point cloud model data h zuiyou .

[0100] Through the cooperation of the real-time rendering three-dimensional scene sparse point cloud model collection unit and the real-time rendering three-dimensional scene optimal sparse point cloud model establishment unit, based on the real-time rendering three-dimensional scene required standard sparse point cloud model parameters, the real-time rendering three-dimensional scene real-time modeling sparse point cloud model parameters, and the artificial intelligence recognition algorithm, the optimal sparse point cloud model of the real-time rendering three-dimensional scene is scientifically matched, realizing the efficient and accurate screening of the optimal sparse point cloud model of the real-time rendering three-dimensional scene based on 3D Gaussian sputtering based on artificial intelligence, and improving the reliability and efficiency of the real-time rendering of the three-dimensional scene based on 3D Gaussian sputtering.

[0101] Example 2:

[0102] Please refer to Figs. 1-2, a three-dimensional scene real-time rendering optimization system based on 3D Gaussian sputtering, for realizing a three-dimensional scene real-time rendering optimization method based on 3D Gaussian sputtering, the system comprising a real-time rendering three-dimensional scene identification module, a real-time rendering three-dimensional scene point cloud model establishment module, a real-time rendering three-dimensional scene optimal point cloud model generation module;

[0103] The real-time rendering three-dimensional scene identification module comprises a real-time rendering three-dimensional scene feature information acquisition unit, a real-time rendering three-dimensional scene two-dimensional image acquisition unit, a different type real-time rendering three-dimensional scene feature information storage unit, and a real-time rendering three-dimensional scene type identification unit.

[0104] The real-time rendering three-dimensional scene feature information acquisition unit acquires real-time rendering three-dimensional scene feature text data through a voice input module; the real-time rendering three-dimensional scene two-dimensional image acquisition unit acquires real-time rendering three-dimensional scene two-dimensional image data through a shooting camera; the different type real-time rendering three-dimensional scene feature information storage unit is used for storing different type real-time rendering three-dimensional scene feature text data; and the real-time rendering three-dimensional scene type identification unit performs real-time rendering three-dimensional scene type identification processing based on real-time rendering three-dimensional scene feature text data and different type real-time rendering three-dimensional scene feature text data, and generates real-time rendering three-dimensional scene type identification data.

[0105] The real-time rendering three-dimensional scene point cloud model establishment module comprises a different type real-time rendering three-dimensional scene standard sparse point cloud model storage unit, a real-time rendering three-dimensional scene standard sparse point cloud model matching unit, a real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameter storage unit, a real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameter analysis unit, and a real-time rendering three-dimensional scene sparse point cloud model construction unit.

[0106] The different types of real-time rendering three-dimensional scene standard sparse point cloud model storage unit is used for storing different types of real-time rendering three-dimensional scene standard sparse point cloud model data; the real-time rendering three-dimensional scene standard sparse point cloud model matching unit is used for performing real-time rendering three-dimensional scene type standard sparse point cloud model matching processing according to the real-time rendering three-dimensional scene type identification data and the different types of real-time rendering three-dimensional scene standard sparse point cloud model data, and constructing real-time rendering three-dimensional scene required standard sparse point cloud model data; the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameter storage unit is used for storing real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data; the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control parameter analysis unit is used for performing real-time rendering three-dimensional scene type sparse point cloud model modeling software modeling control parameter analysis processing according to the real-time rendering three-dimensional scene type identification data and the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data, and generating real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data; the real-time rendering three-dimensional scene sparse point cloud model construction unit is used for performing real-time rendering three-dimensional scene sparse point cloud model modeling processing based on the real-time rendering three-dimensional scene two-dimensional image data and the real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data, and generating real-time rendering three-dimensional scene real-time modeling sparse point cloud model data.

[0107] The real-time rendering three-dimensional scene optimal point cloud model generation module comprises a real-time rendering three-dimensional scene sparse point cloud model collection unit and a real-time rendering three-dimensional scene optimal sparse point cloud model establishment unit.

[0108] The real-time rendering three-dimensional scene sparse point cloud model collection unit is used for performing real-time rendering three-dimensional scene sparse point cloud model data collection processing based on the real-time rendering three-dimensional scene required standard sparse point cloud model data and the real-time rendering three-dimensional scene real-time modeling sparse point cloud model data, and constructing real-time rendering three-dimensional scene sparse point cloud model collection data; the real-time rendering three-dimensional scene optimal sparse point cloud model establishment unit is used for performing real-time rendering three-dimensional scene optimal sparse point cloud model establishment processing according to the real-time rendering three-dimensional scene sparse point cloud model collection data, and generating real-time rendering three-dimensional scene optimal sparse point cloud model data.

[0109] Although the embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for real-time rendering optimization of a three-dimensional scene based on 3D Gaussian sputtering, characterized in that, The method comprises the following steps: S1, collecting real-time rendering three-dimensional scene feature text data and real-time rendering three-dimensional scene two-dimensional image data; S2, performing real-time rendering three-dimensional scene type identification processing to generate real-time rendering three-dimensional scene type identification data; S3, performing real-time rendering three-dimensional scene type standard sparse point cloud model matching processing to construct real-time rendering three-dimensional scene required standard sparse point cloud model data; S4, performing real-time rendering three-dimensional scene type sparse point cloud model modeling software modeling control parameter analysis processing to generate real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data; S5, performing real-time rendering three-dimensional scene sparse point cloud model modeling processing to generate real-time rendering three-dimensional scene real-time modeling sparse point cloud model data S6, performing real-time rendering three-dimensional scene sparse point cloud model data collection processing to construct real-time rendering three-dimensional scene sparse point cloud model collection data; S7, performing real-time rendering three-dimensional scene optimal sparse point cloud model establishment processing to generate real-time rendering three-dimensional scene optimal sparse point cloud model data; The S7 comprises the following steps: S71、will In According to the three-dimensional scene sparse point cloud model modeling software type number order and In Carry out sparse point cloud model feature matching, search out the most matching The most matching And through data identification, generate real-time rendering three-dimensional scene optimal sparse point cloud model data , the specific operation steps of generating the real-time rendering three-dimensional scene optimal sparse point cloud model data are as follows: S711, initialization stage, updating the point cloud model search whale population position and the maximum iteration algorithm T; S712, the surrounding prey stage, the point cloud model searches for the whale individual to select the point cloud model search whale individual swimming towards the optimal position or randomly swimming towards a point cloud model search whale individual, and the point cloud model search whale individual searches in the search space of the The search space of the The search space of the The search space of the The search space of the The search space of the S7121、when selecting to search for a whale individual swimming towards the optimal position point cloud model, the point cloud model searches for a whale individual in the search space of the ; and ; and ; and S7122、when selecting to search the location of the whale individual toward the random point cloud model, the point cloud model searches the whale individual in the search space at random to search out the ; and ; and ; and when When <1, Indicates the value Random numbers within, The initial value is 2. The point cloud model searches for individual whales and selects to swim towards the optimal point. Search the search space for the match with the The most matching ; When ≥1, the point cloud model search whale individual selects to swim towards a random point cloud model search whale individual, and the point cloud model search whale individual randomly searches in the search space of the to match the . ; S713, Bubble Web Stage: During hunting, individual whales in the point cloud model expel air bubbles to form a bubble web to drive away prey. These whales continuously update their own... The position in the search space, when using bubble nets, the point cloud model searches for and updates the position of individual whales; S714, output the maximum matching algorithm when the maximum iteration algorithm is satisfied the maximum matching algorithm ; S715, output the data in step S714 in step S714 Optimal sparse point cloud model data of real-time rendering three-dimensional scene through data identification ; wherein representing real-time rendering three-dimensional scene sparse point cloud model collection data, said comprising ; wherein representing real-time rendering three-dimensional scene standard sparse point cloud model data required, representing real-time rendering three-dimensional scene real-time modeling sparse point cloud model data collection, said comprising ; wherein representing real-time rendering three-dimensional scene real-time modeling sparse point cloud model data generated by the first type of three-dimensional scene sparse point cloud model modeling software.

2. The method of claim 1, wherein the method further comprises: The S1 comprises the following steps: S11, collecting scene feature text information of the target real-time rendering three-dimensional scene through the voice input module online, and generating real-time rendering three-dimensional scene feature text data ; The two-dimensional image information of the real-time rendering three-dimensional scene is collected online from different spatial orientations by a shooting camera, and a real-time rendering three-dimensional scene two-dimensional image data set is generated , ; wherein represents the real-time rendering three-dimensional scene two-dimensional image data of the i-th spatial orientation collected, represents the maximum value of the number of spatial orientations.​ 3. The method of claim 2, wherein the method further comprises: The S2 comprises the following steps: S21, establishing different type real-time rendering three-dimensional scene feature text data set , ; wherein represents the first type real-time rendering specific three-dimensional scene type corresponding to different type real-time rendering three-dimensional scene feature text data represents the maximum value of the number of real-time rendering specific three-dimensional scene types; S22, using XGBoost search algorithm to search the corresponding real-time rendering specific three-dimensional scene type text information, and generating real-time rendering three-dimensional scene type identification data through data identification corresponding real-time rendering specific three-dimensional scene type text information, and generating real-time rendering three-dimensional scene type identification data through data identification .​​​ 4. The method of claim 3, wherein the method further comprises: The S3 comprises the following steps: S31, establishing a different type of real-time rendering three-dimensional scene standard sparse point cloud model data set wherein represents the first different type of real-time rendering three-dimensional scene standard sparse point cloud model data corresponding to a specific three-dimensional scene type S32, using a bidirectional search algorithm to search the corresponding to the corresponding to the real-time rendering of the specific three-dimensional scene type keyword matching, searching for the corresponding to the , and constructing a standard sparse point cloud model data required for real-time rendering of the three-dimensional scene , when the not completed, continue to execute the construction operation instruction.

5. The method of claim 4, wherein: The S4 comprises the following steps: S41、when the When the construction is completed, the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data set matrix is established , ; wherein indicates the first The real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data set corresponding to the three-dimensional scene sparse point cloud model modeling software type, Indicates the maximum value of the number of three-dimensional scene sparse point cloud model modeling software types; wherein , wherein Indicates the real-time rendering specific three-dimensional scene type corresponding to the real-time rendering three-dimensional scene sparse point cloud model modeling software modeling control data in the ​​ S42, using a depth limited search algorithm to the said In the According to the three-dimensional scene sparse point cloud model building software type number in order to search for all the said In the said Real-time rendering of specific three-dimensional scene type matching the said , and through the data identification generation real-time rendering three-dimensional scene required sparse point cloud model modeling software modeling control data set , wherein The first Three-dimensional scene sparse point cloud model building software type corresponding to the real-time rendering of the three-dimensional scene required sparse point cloud model modeling software modeling control data.

6. The method of claim 5, wherein the method further comprises: The S5 comprises the following steps: S51、in accordance with the S51、in accordance with the S51、in accordance with the S51、in accordance with the S51、in accordance with the S51、in accordance with the S51、in accordance with the S51、in accordance with the S51、in accordance with the S51、in accordance with the 7. The method of claim 6, wherein the method further comprises: The S6 comprises the following steps: S61、to generate the , the data combination is performed to construct a real-time rendering three-dimensional scene sparse point cloud model .

8. A 3D Gaussian Splatting based real-time rendering optimization system for implementing the 3D Gaussian Splatting based real-time rendering optimization method of any one of claims 1-7, characterized in that: The system comprises a real-time rendering three-dimensional scene identification module, a real-time rendering three-dimensional scene point cloud model establishment module, and a real-time rendering three-dimensional scene optimal point cloud model generation module.

Images