An editable image shooting training scene building system

Abstract

The application relates to the field of computer simulation and human-computer interaction technology, and discloses an editable image shooting training scene building system, which comprises a virtual scene component management unit configured to build a parameterized template containing three-dimensional distance field data and morphological connectivity control parameters; an interactive channel analysis unit configured to map shooting instructions into a conical pickup area, and to calculate an interactive operation confidence index by comprehensively considering the depth occlusion layering amount, component distribution density and visual tolerance angle domain in the area; and a geometric morphology evolution unit configured to generate an updated damping coefficient according to the index, and to perform weighted adjustment on the control parameters. The application converts discrete construction changes into a continuous field evolution process under the confidence gating, overcomes the judgment jump defects of the traditional ray determination mechanism in a complex training environment, and realizes stable scene construction regulation.

Description

Technical Field

[0001] This invention relates to the field of computer simulation and human-computer interaction technology, and more specifically, to an editable image shooting training scene construction system. Background Technology

[0002] In the construction of modern tactical drills and shooting training systems, procedural generation technology is often used to quickly build virtual scenarios containing a large number of walls, bunkers, and building complexes in order to simulate the ever-changing battlefield environment. To improve the relevance of training, instructors or trainees often need to interact with the screen through simulated shooting equipment to modify the scene structure in real time, such as opening observation holes in the walls, opening connecting passages, or removing obstacles.

[0003] Most existing mainstream interactive technologies employ raycasting detection mechanisms, which involve the system emitting an idealized, widthless ray based on the landing point of the shooting command on the screen, detecting the geometric intersection of the ray with the virtual scene model, and then performing discrete Boolean shearing operations based on the intersection point to modify the scene model.

[0004] However, when dealing with complex training scenarios involving high density and multiple layers of occlusion, the aforementioned general technical solutions face severe challenges, specifically in the following three aspects: First, there's the contradiction between the limitations of input dimensions and actual operational errors. Existing ray-detection mechanisms treat shooting commands as absolutely precise coordinate points. However, in actual human-computer interaction, due to the natural jitter of human muscle control and the accuracy drift of device sensors, user input is essentially a range of areas with random dispersion properties, rather than a single point. Existing technology ignores this input dispersion property, forcibly performing precise single-point matching, leading to frequent misinterpretations of the user's operational intent by the system.

[0005] Second, target identification is difficult in high-density occlusion environments. In procedurally generated complex training scenes, there are often a large number of densely packed homogeneous components (such as overlapping fences or continuous wall layers). When multiple closely arranged objects exist in the line of sight, the human visual system experiences a crowding effect, making it difficult to distinguish individual objects within a local area. In this situation, single-ray-based detection mechanisms are prone to misjudging in the depth direction, incorrectly selecting foreground occluders or background objects instead of the target layer the user expects, thus reducing the efficiency of scene construction.

[0006] Third, the instability of construction evolution caused by discrete logic. This is the main bottleneck faced by existing technologies in real-time editing. Current scene editing largely relies on binary discrete decision logic of yes or no. When the shot's impact point deviates slightly near the edge of a component, or when the system's analysis of occlusion depth fluctuates between consecutive frames, it can trigger drastically different topology changes (e.g., a sudden jump from non-penetration to complete penetration). This extreme sensitivity to minute input changes causes the scene structure to flicker and jump during editing, making it impossible for users to obtain continuous, smooth, and predictable control feedback, severely affecting the accuracy and controllability of training scene construction. Summary of the Invention

[0007] This invention provides an editable image shooting training scene construction system, which solves the technical problems mentioned in the background art.

[0008] This invention provides an editable image shooting training scene construction system, comprising: A parameterized scene generation module is used to construct and store a virtual scene based on a volumetric generation data model, wherein the volumetric generation data model includes basic shape units defined by continuous functions, and structural connectivity control parameters for adjusting the scene's geometric construction and connectivity state. The interactive area definition module is used to receive click instructions and, in combination with the input positioning deviation and the visual recognition limitation range, map the click instructions into a visual cone interactive area with a regional angle. The interaction stability calculation module is used to calculate the structural control stability index that characterizes the effectiveness of a single click interaction within the view frustum interaction area, based on the depth occlusion level, the generation unit projection density, and the subtended angle of the area. The scene structure control module is used to generate an update step size adjustment coefficient based on the structure control stability index and construct a controlled parameter update scope. The update step size adjustment coefficient is used to dynamically adjust the scope and update amplitude of the controlled parameter update scope, so as to continuously adjust the parameters in the parameterized scene generation module to change the geometric structure of the virtual scene in real time.

[0009] The beneficial effects of this invention include: by constructing a deep correlation mechanism between interactive channel analysis and geometric morphology evolution, it effectively solves the problem of unstable scene construction control caused by shooting jitter and visual congestion in high-density occlusion training environments; this invention no longer forcibly performs discrete judgment on ambiguous shooting commands, but automatically evaluates the operation confidence based on the depth of occlusion stacking and component distribution density, and uses continuous damping suppression logic to smooth unreliable shooting commands, thereby preserving the trainee's intention while shielding input noise, ensuring that the construction evolution of the virtual training scene in the complex editing process remains continuous, deterministic and predictable, and significantly improving the robustness and accuracy of simulation interaction. Attached Figure Description

[0010] Figure 1 This is a block diagram of an editable image shooting training scene construction system according to the present invention; Figure 2 This is a schematic diagram illustrating a specific implementation of the present invention. Detailed Implementation

[0011] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0012] like Figure 1 As shown, an editable image shooting training scene construction system includes: A parameterized scene generation module is used to construct and store a virtual scene based on a volumetric generation data model, wherein the volumetric generation data model includes basic shape units defined by continuous functions, and structural connectivity control parameters for adjusting the scene's geometric construction and connectivity state. The interactive area definition module is used to receive click instructions and, in combination with the input positioning deviation and the visual recognition limitation range, map the click instructions into a visual cone interactive area with a regional angle. The interaction stability calculation module is used to calculate the structural control stability index that characterizes the effectiveness of a single click interaction within the view frustum interaction area, based on the depth occlusion level, the generation unit projection density, and the subtended angle of the area. The scene structure control module is used to generate an update step size adjustment coefficient based on the structure control stability index and construct a controlled parameter update scope. The update step size adjustment coefficient is used to dynamically adjust the scope and update amplitude of the controlled parameter update scope, so as to continuously adjust the parameters in the parameterized scene generation module to change the geometric structure of the virtual scene in real time.

[0013] Preferably, a virtual scene based on a volumetric generated data model is constructed and stored. The volumetric generated data model includes basic shape units defined by continuous functions, and structural connectivity control parameters for adjusting the scene's geometry and connectivity state, including: Constructing the global field function of the volumetric generated data model as follows: in, For spatial coordinates, This is a model parameter vector containing the structural connectivity control parameters. To extract the threshold for the isosurface, For the first The sign distance field function of each basic shape unit; The soft Boolean operation logic Includes soft union operations Soft intersection operation Or soft difference operation : in, The distance field value used in the calculation. The smoothness coefficient; For topology connectivity control, the continuous gated activation function is introduced. Acting on the basic shape unit used for subtraction : in, These are the structural connectivity control parameters. For gate steepness, The center offset, For controlling the depth of penetration. It serves as the basic shape unit for the main body.

[0014] The model parameter vector is a vector containing structural connectivity control parameters, used to control properties such as the position, size, number, and density of basic shape units.

[0015] The isosurface extraction threshold is the threshold for the overall field function of the volumetric data model, used to define the geometric boundaries (isosurfaces).

[0016] The symbolic distance field function is a continuous function of the m-th basic shape unit, used to describe the spatial distribution of the basic shape.

[0017] The smoothness factor is the smoothing coefficient for soft Boolean operations, used to control how closely the result approximates the result of hard Boolean operations. A value of 1 to 10 is preferred, as this range balances smoothness and geometric accuracy. Smaller values ​​result in more pronounced smoothing, while larger values ​​bring the result closer to that of hard Boolean operations.

[0018] Structural connectivity control parameters are parameters that control the topological connectivity state and are used to drive the opening and closing of holes or the disconnection of component connections. The preferred values ​​are 0 to 5, as this range allows for a continuous transition from topological invariance to complete topological switching. 0 corresponds to no topological change, 1 corresponds to the beginning of topological change, and 5 corresponds to complete topological switching.

[0019] Gating steepness is the steepness coefficient of a continuous gating activation function, used to control the rate at which the gating opens or closes. A value of 3 to 7 is preferred, as this range avoids slow gating switching leading to response delays or excessively fast gating switching leading to topology abrupt changes; 3 corresponds to smooth switching, and 7 corresponds to fast switching.

[0020] The center offset is the center threshold of the continuous gating activation function, used to define the critical structural connectivity control parameter value for topology state switching. It is preferably 2 to 3, as this range matches the preferred values ​​of the structural connectivity control parameters, ensuring that topology switching occurs within a reasonable parameter range.

[0021] The penetration depth control is a parameter that adjusts the penetration force of the subtractive basic shape unit to ensure the complete evolution of the hole from a pit to a full penetration. A value of 0.5 to 2 is preferred, as this range accommodates the sizes of common basic shape units; 0.5 corresponds to a shallow pit, and 2 corresponds to a full penetration.

[0022] Spatial coordinates are arbitrary points in three-dimensional space, used for field function calculations to describe the spatial distribution of basic shape units.

[0023] The signed distance field is used as the continuous function description form of the basic shape unit. The signed distance field is a continuous function that can accurately describe the geometry. The function value at each spatial point is equal to the shortest distance from that point to the boundary of the basic shape, and is negative inside the shape and positive outside. The overall field function is constructed by combining soft Boolean operations, including soft union, soft intersection, and soft difference operations. A smoothness coefficient controls the smoothness of the operation. For example, when constructing a combined model of walls and holes, using soft difference operations can avoid the geometric jaggedness caused by hard Boolean operations, making the transition between walls and holes more natural.

[0024] By introducing structural connectivity control parameters into a continuous gating activation function, which is a smooth nonlinear function, the continuous changes of structural connectivity control parameters can be transformed into continuous output values ​​between 0 and 1. For example, when the structural connectivity control parameters increase from 0 to 5, the function output value smoothly rises from 0 to 1, realizing the gradual opening of the gating.

[0025] The output value of the continuous gating activation function is used to dynamically adjust the weight of the target basic shape unit in soft Boolean operation. For example, in the hole generation scenario, the weight of the subtractive basic shape unit increases with the increase of the gating output value. When the weight is small, only a pit is formed on the wall surface. When the weight reaches 1, a complete through hole is formed. This continuous adjustment achieves smooth switching of topological state and avoids the topological abruptness caused by traditional discrete Boolean operation.

[0026] In soft Boolean operations, the smoothness coefficient ranges from 1 to 10. The adjustment rules for different scenarios are as follows: when constructing a scene that requires obvious transition effects, such as an archway, the value is 1 to 3; when constructing a scene that requires clear boundaries, such as a square wall, the value is 7 to 10; and for ordinary scenes, the value is 4 to 6.

[0027] The parameters of the continuous gating activation function are selected based on the following criteria: the gating steepness ranges from 3 to 7. For interactive scenarios requiring rapid response, such as hole generation in real-time shooting training, a value of 5 to 7 is used; for scenarios requiring smooth transitions, such as slowly constructed cover, a value of 3 to 4 is used. The determination of the center offset is related to the scene type. For scenes with small hole diameters, such as observation holes, a value of 2 is used; for scenes with large hole diameters, such as passage doors, a value of 3 is used.

[0028] The specific setting standard for the penetration depth control value is as follows: it is determined according to the thickness of the basic shape unit. When the thickness of the basic shape unit is 1 to 3, the value is 0.5 to 1; when the thickness is 4 to 6, the value is 1 to 1.5; and when the thickness is 7 to 10, the value is 1.5 to 2, to ensure that the penetration depth matches the thickness of the basic shape unit.

[0029] The specific primitive type selection criteria for the symbolic distance field are as follows: a spherical symbolic distance field is used when constructing spherical objects such as ammunition boxes in a bunker; a box-shaped symbolic distance field is used when constructing cuboid objects such as walls; and a cylindrical symbolic distance field is used when constructing cylindrical objects such as pipes.

[0030] Preferably, the click command is received, and combined with the input positioning deviation and visual recognition limitation range, the click command is mapped to a visual cone interaction region with a regional angle, including: Parse the screen coordinates corresponding to the click command Combined with camera internal parameters With external references Calculate the direction of the central ray : Calculate the input positioning deviation half angle : in, The eigenvalues ​​are the eigenvalues ​​of the two-dimensional endpoint distribution covariance matrix of the input device. The confidence interval coefficient is... The focal length of the camera; Calculate the visual recognition restriction half-angle : in, The offset of the clicked point relative to the center of the field of view. It is the critical spacing factor for crowding; Calculate the angle of the effective region : Construct the view frustum interaction region : in, Let be the direction vector of any point in space relative to the camera origin.

[0031] The screen coordinates of a click are the pixel coordinates of the screen corresponding to the firing command. These coordinates can be obtained by collecting the pixel position of the click point using sensors on a simulated firing device.

[0032] Camera intrinsic parameters are a set of internal parameters of a camera, including information such as focal length and pixel size, used for coordinate transformation. They can be calculated by taking an image of a standard calibration board using a camera calibration tool.

[0033] Camera extrinsic parameters are external parameters describing the camera's attitude in world space, specifically rotation matrices. They can be obtained by tracking the camera's position and attitude in real time using a visual SLAM system.

[0034] The two-dimensional endpoint distribution covariance matrix is ​​a statistical matrix of input device endpoint deviations, representing input uncertainty. It can be obtained by statistically calculating after collecting a large amount of user click data on the device.

[0035] The confidence interval coefficient is a factor used in calculating the half-angle of the input positioning deviation to determine the range of the confidence ellipse. A value of 2 is preferred, as this corresponds to approximately 95% of the confidence intervals, covering the vast majority of input deviation cases.

[0036] The camera focal length is the camera's equivalent focal length, measured in pixels, and is used to convert screen deviations into angular deviations. It can be obtained directly from the camera's intrinsic parameter calibration results.

[0037] Field of view eccentricity is the angular deviation of the clicked point relative to the center of the field of view, measured in degrees. It is a calculated variable derived from the camera's intrinsic and extrinsic parameters based on the clicked screen coordinates.

[0038] The crowding critical spacing factor is a core factor in visual recognition limiting half-angle calculation, and it is related to the visual crowding pattern. It is preferably between 0.5 and 1.5, as this range can adapt to the physiological characteristics of visual crowding in the human eye. 0.5 corresponds to dense scenes and 1.5 corresponds to sparse scenes.

[0039] The effective region angle is the result of the fusion of the squares of the input positioning deviation half-angle and the visual recognition constraint half-angle, expressed in radians. It is a calculated variable obtained by orthogonally superimposing the two half-angles.

[0040] The central ray direction is a ray direction vector derived from the intrinsic and extrinsic parameters of the camera at the click coordinates, and it is the axis of the view frustum interaction region. It is a calculated variable derived from the intrinsic and extrinsic parameters of the camera at the click screen coordinates.

[0041] By combining input positioning bias and visual recognition limitations, discrete click commands are mapped to a view cone interaction area, rather than a traditional zero-width ray. Input positioning bias stems from human muscle tremors and device sensor drift, while visual recognition limitations arise from the visual crowding effect of the human eye—both unavoidable factors in actual interaction. For example, in shooting training, users naturally shake when holding a simulated gun, resulting in a click landing point that is actually an area. Simultaneously, dense wall components make it difficult for users to accurately distinguish individual targets. The view cone interaction area can completely cover these uncertainties.

[0042] A square sum-rooted fusion strategy is employed to orthogonally superimpose two half-angles. This strategy enables the unbranched, continuous fusion of two error sources from different dimensions: device input error and human visual limitations. For example, if the input positioning deviation half-angle is 0.1 radians and the visual recognition limitation half-angle is 0.08 radians, the effective area angle obtained after fusion using this strategy can simultaneously reflect the combined impact of both errors.

[0043] Based on the statistical deviations of camera intrinsic and extrinsic input devices and visual physiological laws, a view frustum interaction region is constructed, forming a multi-dimensional error correlation model. Camera intrinsic and extrinsic parameters ensure the accuracy of spatial mapping, input device statistical deviations reflect actual hardware performance, and visual physiological laws align with human perception characteristics. The combination of these three factors makes the view frustum interaction region more closely resemble real-world interaction scenarios. For example, in long-range shooting scenarios, the camera focal length affects the field of view, amplifying the impact of input deviations and visual limitations. This model can dynamically adapt to these changes.

[0044] The value of the crowding critical spacing factor is determined based on the density of components in the scene: 0.5 to 0.8 when the component density is 10 to 20 per square meter; 0.8 to 1.2 when the density is 5 to 10 per square meter; and 1.2 to 1.5 when the density is 1 to 5 per square meter. This value range can match the human eye's resolution capability in different density scenes.

[0045] The method for calibrating the two-dimensional endpoint distribution covariance matrix involves having 10 users click continuously 100 times at the same location on the screen using the target input device, collecting the coordinates of all click points, and calculating the covariance matrix of these coordinates. The sample size meets the statistical significance requirements and can reflect the input bias characteristics of the device.

[0046] The selection criteria for the confidence interval coefficient are as follows: the default value of 2 corresponds to a 95% confidence interval; if the scenario requires high input accuracy, such as precision shooting training, a value of 1.5 corresponds to an 86% confidence interval; if the scenario allows for larger input errors, such as rapid scenario building, a value of 2.5 corresponds to a 98% confidence interval.

[0047] The upper limit threshold for the effective region angle is set to 0.5 radians. When the angle of the fused effective region exceeds this value, it is automatically constrained to 0.5 radians. This avoids excessively large view frustum areas due to extreme input deviations or dense scenes, which could affect the accuracy of interaction.

[0048] Preferably, the depth occlusion level includes: Determine the set of screen projection pixels corresponding to the view frustum interaction region. ; Perform a fixed number of times The continuous depth stripping operation, for any pixel in the set , obtain the Fragment coverage count during secondary peeling ; Using continuous smooth activation function Calculate the first The layer has consecutive hierarchical values: in, This is the kurtosis coefficient of the function. This is the count bias. Calculate pixels Cumulative occlusion depth value : Calculate the depth occlusion level. : in, The total number of pixels in the projected pixel set.

[0049] The fixed number of peel operations is the maximum number of times the continuous depth peel operation can be performed to ensure coverage of all relevant occlusion layers within the view frustum interaction area. A value of 5 to 15 is preferred, as this range balances occlusion layer coverage integrity with system computational efficiency. 5 corresponds to sparse occlusion scenarios, 15 to high-density multi-layer occlusion scenarios, and 8 to 12 for normal scenarios.

[0050] The kurtosis coefficient is the kurtosis parameter of a continuous smooth activation function, used to control the mapping rate from fragment cover counts to consecutive level existence values. A value of 2 to 5 is preferred, as this range avoids low numerical discriminability due to overly slow mapping or discretization due to overly fast mapping; 2 corresponds to a smooth mapping, and 5 corresponds to a fast mapping.

[0051] The count bias is a bias parameter of the continuous smooth activation function, used to avoid numerical anomalies when the fragment coverage count is 0. It is preferably between 0.1 and 0.3, as this range can effectively prevent calculation errors caused by the activation function output value approaching 0 without affecting normal counting mapping.

[0052] Fragment coverage count is the number of fragments covered by a single pixel at the k-th depth stripping, used to characterize the occlusion status of that pixel in the corresponding depth layer. It can be obtained by counting the fragment coverage rate or the actual coverage number of pixels after each depth stripping using the GPU's fragment shader.

[0053] The projected pixel set consists of all pixels corresponding to the projection of the view frustum interaction region onto the screen space, serving as the boundary for depth stripping and occlusion layer calculations. It is a computational variable derived from the effective region angle of the view frustum interaction region and camera parameters.

[0054] The cumulative occlusion depth value is the sum of the existence values ​​of all consecutive layers for a single pixel within a preset number of stripping operations, used to characterize the total occlusion depth corresponding to that pixel. It is a calculated variable obtained by summing the existence values ​​of consecutive layers for each stripping operation of a single pixel.

[0055] The depth occlusion level is the average of the cumulative occlusion depth values ​​of all pixels within the projected pixel set, used to quantify the overall occlusion complexity within the view frustum interaction region. It is a calculated variable derived by averaging the cumulative occlusion depth values ​​of the projected pixel set.

[0056] A continuous depth stripping operation is performed on the projected pixel set. Unlike traditional depth stripping, which only determines whether there is an occlusion layer, this operation performs a fixed number of stripping operations and counts the fragment coverage each time. For example, if the preset fixed number of stripping operations is 10, regardless of the actual number of occlusion layers, 10 stripping operations will be performed to ensure that occlusion information at different depths can be captured, avoiding discrete jumps caused by the determination of the number of occlusion layers.

[0057] By using a continuous smooth activation function, fragment cover counts are mapped to continuous hierarchical existence values. Fragment cover counts are discrete integers, but the activation function can transform them into continuous values ​​between 0 and 1. For example, when the fragment cover count is 1, it is mapped to a continuous hierarchical existence value of 0.8; when the count is 0, it is mapped to a continuous value of 0.1. This avoids discrete determinations that are either 0 or 1, making the representation of occlusion existence smoother.

[0058] The cumulative occlusion depth of a single pixel is obtained by summing the consecutive existence values ​​within a preset number of stripping operations, and then averaging them to obtain the depth occlusion level. This method transforms the occlusion depth of a single pixel from presence or absence to degree. For example, if a pixel is covered by fragments in 6 out of 10 stripping operations, the cumulative occlusion depth value is 4.2, reflecting that the spatial location of the pixel is significantly occluded. The average depth occlusion level can reflect the overall occlusion complexity of the frustum region.

[0059] The value of the fixed number of peels is determined based on the scene's occlusion density: for high-density occlusion scenes such as multiple overlapping walls, the value is 10 to 15; for normal occlusion scenes such as a few overlapping shelters, the value is 5 to 10; and for sparse occlusion scenes such as isolated objects in open areas, the value is 3 to 5. This value range was determined through extensive scene testing, ensuring coverage of the actual occlusion layer without causing computational delay due to excessive peels.

[0060] The parameter adjustment rules for the continuous smooth activation function are as follows: the steepness coefficient β is set to 2 to 5; β is set to 2 to 3 when the fragment coverage count range is 0 to 5; β is set to 3 to 4 when the count range is 6 to 10; and β is set to 4 to 5 when the count range is greater than 10. The count bias ε is set to 0.1 to 0.3; ε is set to 0.1 when the fragment coverage count is generally low and ε is set to 0.3 when the count is generally high, to ensure that the continuous values ​​mapped under different count ranges have reasonable distinguishability.

[0061] The specific method for counting fragment coverage is to use the GPU's fragment shader to count the actual number or coverage rate of each pixel in the projected pixel set that is covered by scene model fragments during each depth stripping rendering. The coverage rate needs to be converted into an equivalent count, for example, a coverage rate of 50% corresponds to a count of 0.5, to ensure that the count can reflect the actual degree of fragment coverage.

[0062] The normalization of the cumulative occlusion depth value is performed as follows: when the preset number of peeling operations is different, the cumulative occlusion depth value needs to be divided by the fixed number of peeling operations to obtain a normalized value between 0 and 1. Then, the average value of all pixels is calculated as the depth occlusion level. For example, if the fixed number of peeling operations is 10 and the cumulative occlusion depth value of a certain pixel is 5, then the normalized value is 0.5, ensuring that the depth occlusion level values ​​under different peeling operations are comparable.

[0063] Preferably, the unit projection density includes: Filter out those located in the view frustum interaction region within For each homogeneous generation module instance, obtain the set of its projection center coordinates on the screen. ; According to the effective area angle The interactive neighborhood radius of the screen space is calculated. ; Calculate the click center using the kernel density estimation algorithm The unit projection density at the location : in, For the first The Euclidean distance between the projection center and the click center of each instance. This is the Gaussian smoothing kernel function.

[0064] The instance projection center coordinates are the coordinates of the projection center position of homogeneous generated module instances within the view frustum interaction area on the screen space. This can be obtained by traversing all generated module instances in the virtual scene, determining whether they are located within the view frustum interaction area, extracting the spatial coordinates of instances that meet the criteria, and projecting them onto the screen space.

[0065] The interactive neighborhood radius is the screen space radius calculated from the effective region's angle, and is used as a bandwidth parameter for kernel density estimation.

[0066] Unit projection density is the number of homogeneous generated module instances within a unit interactive area, used to characterize the local instance density at the click location.

[0067] By using the view frustum interaction area as the filtering scope, this method accurately locates homogeneous generated module instances participating in density calculation. Unlike traditional global density statistics or single-pixel density calculation, this method focuses only on the interaction space corresponding to the click, excluding interference from irrelevant instances. For example, in a shooting training scenario, when a user clicks on a wall, only wall component instances within the view frustum area where the wall is located are counted, excluding irrelevant objects in other areas of the scene, making the density calculation more closely aligned with the interaction intent.

[0068] The bandwidth of kernel density estimation, i.e., the radius of the interaction neighborhood, is determined based on the effective region angle, thus achieving a strong correlation between density calculation and the view frustum interaction region. The effective region angle reflects the combined effects of input bias and visual limitations; determining the bandwidth in this way allows the range of density estimation to match the effective region of actual interaction. For example, when the effective region angle is larger, the radius of the interaction neighborhood increases accordingly, covering more instances that may be involved due to input jitter or visual congestion, ensuring the completeness of density calculation.

[0069] A kernel density estimation algorithm employing a Gaussian smoothing kernel function transforms discrete instance projection center coordinates into a continuous local density response. The Gaussian smoothing kernel function smooths the instance distribution, preventing abrupt changes in density values ​​due to the locational deviation of individual instances. For example, in a region with three densely distributed instances, the density values ​​calculated by this algorithm will exhibit a continuous peak distribution, rather than three discrete, isolated high points, thus better reflecting the true density of local instances.

[0070] The criteria for determining homogeneous generated module instances are based on a comprehensive assessment of template type, structural features, and dimensional parameters. Same template type means originating from the same parametric template; same structural features mean consistent basic shape unit combination methods; and dimensional parameter differences are within 10%. Meeting these three conditions qualifies an instance as homogeneous. For example, wall components generated from the same wall template at different locations, all with box-shaped symbol distance field combinations and dimensional differences within 5%, are considered homogeneous generated module instances.

[0071] In kernel density estimation, the Gaussian smoothing kernel function takes the form of a standard Gaussian kernel function. The function value decreases exponentially as the Euclidean distance between the instance projection center and the click center increases. The coverage of the kernel function is controlled by the radius of the interactive neighborhood.

[0072] The formula for converting the interaction neighborhood radius is: the interaction neighborhood radius equals the tangent of the effective area's angle multiplied by the camera's focal length, then divided by the screen pixel size. For example, if the effective area's angle is 0.1 radians, the camera's focal length is 1000 pixels, and the screen pixel size is 0.01 millimeters, the converted interaction neighborhood radius is approximately 10 pixels, ensuring that the conversion result accurately corresponds to the screen space range.

[0073] The normalization method for local density response is to divide the local density response by the area of ​​the interaction neighborhood (π) multiplied by the square of the radius of the interaction neighborhood, thus obtaining the number of instances per unit interaction area. For example, if the local density response is 15, the radius of the interaction neighborhood is 5 pixels, the area of ​​the interaction neighborhood is approximately 78.5 square pixels, and the normalized unit projection density is approximately 0.19.

[0074] Preferably, based on the depth occlusion level, the generation unit projection density, and the region subtended angle, a structural regulation stability index characterizing the effectiveness of a single click interaction is calculated, including: According to the angle of the region Calculate the area of ​​the effective interaction region : Calculate the effective aliasing size : in, The depth occlusion level is the value. The projection density of the generating unit; Calculate the structural regulation stability index : in, This is a preset, small anti-shake bias used to prevent the denominator from being zero and to maintain numerical stability.

[0075] The micro anti-jitter bias is a preset, extremely small constant used to prevent logarithmic operations from malfunctioning when the effective aliasing scale is 0, thus maintaining numerical stability. It is preferably set between 1e-6 and 1e-4, as values ​​within this range are extremely small and will not affect the accuracy of the effective aliasing scale calculation, while completely avoiding calculation errors caused by a denominator of 0. 1e-5 is the most commonly used value.

[0076] The effective interaction area is the area defined by the effective area on a unit sphere, and is used to quantify the spatial coverage of the view frustum interaction area.

[0077] Effective aliasing scale is the product of depth occlusion level, generation unit projection density, and effective interaction area, used to characterize the degree of aliasing during click interaction.

[0078] The structural regulation stability index is a core indicator for quantifying the effectiveness of a single click interaction, used to characterize the amount of topology control information that can be transmitted in the current interaction state.

[0079] A 3D correlation model is constructed, comprising depth occlusion level, generator unit projection density, and effective interaction area, to define the effective aliasing scale. These three parameters reflect the core characteristics of the interaction scene from three dimensions: occlusion complexity, instance density, and interaction range. The correlation among these three parameters comprehensively characterizes the essence of interaction aliasing. For example, in a shooting scene, the depth occlusion level is 3, the generator unit projection density is 2 pixels per square meter, and the effective interaction area is 5 square pixels. The product of these three parameters yields an effective aliasing scale of 30, which intuitively reflects the aliasing tendency of clicks in this scene.

[0080] A small anti-shake bias is introduced, and the reciprocal of the effective aliasing scale is converted into a structural control stability index through logarithmic calculation. Logarithmic calculation can compress the large-scale variation of the effective aliasing scale into a reasonable numerical range, facilitating subsequent parameter adjustment. For example, when the effective aliasing scale changes from 1 to 100, after logarithmic calculation, the structural control stability index changes from 0 to approximately -6, with a smoother numerical change, adapting to continuous control requirements.

[0081] Based on information theory, this paper treats click interactions as topology control information transmission channels and uses a structural regulation stability index to quantify the upper limit of channel capacity. Unlike traditional binary judgments, this index continuously reflects the effectiveness of the interaction. For example, an index value of 3 indicates that the current interaction can transmit 3 bits of topology control information, while an index value of 1 indicates that only 1 bit of information can be transmitted, providing a clear basis for the intensity of subsequent regulation.

[0082] The specific value of the micro-stabilization bias is between 1e-6 and 1e-4, with 1e-5 being preferred. The reason for choosing this order of magnitude is that the typical range of the effective aliasing scale is 0.1 to 100. A bias of 1e-5 will not have a substantial impact on the value of the effective aliasing scale, while ensuring that the denominator is always greater than 0, thus avoiding errors in logarithmic calculations.

[0083] The reason for choosing a base of 2 for the logarithm is that, in information theory, the bit is the basic unit of information, and the result of a logarithm operation with base 2 directly corresponds to the number of bits, which facilitates the quantification of the amount of information for topology control. For example, when the effective aliasing scale is 0.5, the result of the logarithm operation is 1, which corresponds to 1 bit of information capacity.

[0084] The effective range of the structural regulation stability index is -10 to 10. When the index is greater than 3, it indicates that the interactive information capacity is sufficient and suitable for large-scale topology regulation; when the index is between 0 and 3, it indicates that the information capacity is average and the regulation amplitude needs to be appropriately reduced; when the index is less than 0, it indicates that the interactive aliasing is serious and topology changes need to be significantly suppressed to avoid misoperation.

[0085] A smaller effective aliasing scale value indicates less interactive aliasing, and the more accurately a click can target the desired instance; a larger value indicates more severe aliasing, and the more likely a click is to cover irrelevant instances. For example, when the effective aliasing scale is 1, a click will most likely only cover one target instance; when the scale is 20, a click may cover about 20 homogeneous instances simultaneously, resulting in significant aliasing.

[0086] Preferably, the update step size adjustment coefficient is generated based on the structural control stability index, and a controlled parameter update scope is constructed, including: Based on the preset target information budget value Calculate the information gap. : in, is the kurtosis parameter of the one-sided smooth activation function; Generate the update step size adjustment coefficient : in, The suppression intensity coefficient; Construct the perturbation kernel function corresponding to the control parameter update scope. : in, For the point of action of the click, This refers to the kernel width parameter; The kernel width parameter is dynamically adjusted using the information gap. : in, Based on core width, This is the kernel width adjustment ratio.

[0087] The target information budget value is the system's preset expected information budget for topology operations, providing a benchmark for calculating the information gap. It is preferably 2 to 4 bits, as this range can match the information requirements of common topology adjustments (such as opening and closing holes, and disconnecting components). 2 bits correspond to simple topology operations, and 4 bits correspond to complex topology operations.

[0088] The kurtosis of a one-sided smooth activation function is a parameter used to control the smoothness of the information gap. A value of 1 to 3 is preferred, as this range avoids response lag due to overly slow calculation of the information gap or numerical fluctuations due to overly fast calculation. 1 corresponds to slow calculation, and 3 corresponds to a fast response.

[0089] The suppression intensity coefficient is the intensity coefficient of the negative exponential decay function, used to control the decay rate of the update step size adjustment coefficient. A value of 1.5 to 3.5 is preferred, as this range balances the sensitivity and stability of topology control; 1.5 corresponds to weak suppression, and 3.5 corresponds to strong suppression.

[0090] The click point of action is the actual location where the click occurs in the virtual scene, obtained by intersecting the ray with the current isosurface.

[0091] The base kernel width is the fundamental width of the Gaussian smoothing perturbation kernel, providing a benchmark for dynamic adjustment of the kernel width. It is preferably 5 to 15 pixels, as this range can adapt to common screen resolutions and interaction precision requirements; 5 corresponds to precise interaction, and 15 corresponds to a wide range of interaction.

[0092] The kernel width adjustment ratio is a ratio of the information gap to the kernel width parameter, used to control the rate at which the kernel width changes with the information gap. A value of 2 to 5 is preferred, as this range allows the kernel width adjustment to match the degree of information capacity insufficiency; 2 corresponds to slow adjustment, and 5 corresponds to rapid adjustment.

[0093] The kernel width parameter is the real-time width of the Gaussian smoothed perturbation kernel, which changes dynamically with the amount of information gap.

[0094] The information gap is the difference between the target information budget value and the structural regulation stability index, representing the degree of current insufficient interaction capacity.

[0095] The update step size adjustment coefficient is a continuous coefficient with a value between zero and one, used to dynamically scale the basic evolution step size.

[0096] A one-sided smoothing activation function is used to calculate the information gap. This function transforms the difference between the target information budget value and the structural control stability index into a continuous quantitative indicator. Unlike traditional linear difference calculations, the one-sided smoothing activation function outputs near zero when the difference is negative (i.e., sufficient information capacity) and smoothly increases when the difference is positive (i.e., insufficient information capacity). For example, when the target information budget value is 3 bits and the structural control stability index is 1 bit, the difference is 2. This function calculates a smoothed information gap, avoiding abrupt numerical changes.

[0097] By using a negative exponential decay function to map the information gap to an update step size adjustment coefficient, a nonlinear correlation is achieved where the larger the information gap, the smaller the adjustment coefficient. For example, when the information gap is 0, the adjustment coefficient is 1, allowing for maximum topology control; when the information gap is 3, the adjustment coefficient is approximately 0.05, significantly suppressing topology changes. This nonlinear mapping can accurately match the control requirements under different interaction qualities, avoiding over- or under-control caused by linear mapping.

[0098] A Gaussian smooth perturbation kernel centered at the click point is constructed as the controlled parameter update scope, and the kernel width is dynamically adjusted according to the information gap. The Gaussian smooth perturbation kernel allows the impact of parameter updates to decay smoothly from the click center to the periphery; for example, the update impact is strongest at the click point and weaker the further away from the center. When the information gap increases, the kernel width expands, making the update scope wider, reducing the sensitivity to local topological changes, and avoiding local mutations caused by poor interaction quality.

[0099] The target information budget value is set based on the complexity of the topology operation: 2 bits for simple topology operations such as small hole generation; 3 bits for medium-complexity operations such as medium-sized channel creation; and 4 bits for complex operations such as multi-component disconnection and reassembly. This range of values ​​was obtained through numerous topology control experiments and can match the information requirements of different operations.

[0100] The kurtosis of the one-sided smooth activation function is determined by the following rules: for scenarios with high requirements for interactive response speed, such as rapid training scenarios, a value of 2 to 3 is used; for scenarios with high requirements for stability, such as precise building scenarios, a value of 1 to 2 is used. This ensures that the calculation of information gaps in different scenarios can respond quickly while avoiding fluctuations.

[0101] The selection criteria for the suppression intensity coefficient are as follows: the default value of 2.5 is suitable for most scenarios; if more robust topology control is desired, such as in high-density occlusion scenarios, a value of 3 to 3.5 is used; if more sensitive control is desired, such as in sparse scenarios, a value of 1.5 to 2 is used. This coefficient can be adjusted to suit the stability requirements of different scenarios.

[0102] The matching relationship between the base kernel width and the kernel width adjustment ratio coefficient is as follows: when the base kernel width is 5 to 8 pixels, the kernel width adjustment ratio coefficient is 2 to 3; when the base kernel width is 9 to 12 pixels, the coefficient is 3 to 4; and when the base kernel width is 13 to 15 pixels, the coefficient is 4 to 5. This ensures that the dynamic adjustment range of the kernel width matches the base kernel width, avoiding over-adjustment or under-adjustment.

[0103] Preferably, the update step size adjustment coefficient is used to dynamically adjust the range and magnitude of the controlled parameter update domain, continuously adjusting the parameters in the parameterized scene generation module to change the geometric structure of the virtual scene in real time, including: Calculate the evolution step size of the actual parameters : in, The preset basic evolution step size, The update step size adjustment coefficient; Construct the regional response weight field : in, As for the steepness of the weighted field, This represents the implicit field function value for the current scene; Execute the structural connectivity control parameters Continuous weighted updates: in, Update the perturbation kernel function corresponding to the scope of the controlled parameter. This represents the integration operation across the entire space; Perform the fixed pipeline isosurface extraction and output a geometric mesh. : in, Extracting operators for isosurfaces The threshold value for the target isosurface.

[0104] The baseline evolution step size is a preset benchmark step size for updating structural connectivity control parameters, providing a computational basis for the actual parameter evolution step size. It is preferably between 0.1 and 0.5, as this range can balance the speed and stability of topology control. 0.1 corresponds to fine-tuning, 0.5 corresponds to fast-tuning, and a value of 0.2 to 0.3 is used in ordinary scenarios.

[0105] The weight field steepness is the steepness coefficient of the region response weight field, used to control the clarity of the delineation of topologically sensitive regions. A value of 2 to 5 is preferred, as this range avoids overly vague or sharp delineation of sensitive regions; 2 corresponds to a gentle delineation, and 5 corresponds to a clear delineation.

[0106] The actual parameter evolution step size is the parameter update step size scaled by the update step size adjustment factor, used to dynamically control the parameter update magnitude. It is a calculated variable obtained by multiplying the base evolution step size by the update step size adjustment factor.

[0107] The region response weight field is a continuous weight field constructed based on the implicit field numerical distribution of the current scene. It is used to smoothly delineate sensitive regions where topological changes are permitted. It is a computational variable calculated using the implicit field function values ​​and the kurtosis of the weight field.

[0108] The total spatial convolution integral is the result of the spatial convolution of the perturbation kernel function and the regional response weight field within the controlled parameter update domain. It is used to quantify the overall driving force of parameter updates. It is a computational variable obtained through the convolution operation between the perturbation kernel function and the regional response weight field.

[0109] The isosurface extraction operator is a tool operator used to extract geometric meshes from volumetric generated data models, realizing the transformation from continuous fields to discrete geometry.

[0110] The target isosurface threshold is a decision threshold extracted from isosurfaces and used to define the geometric boundaries of the virtual scene. It is fixed at 0, with the overall field function of the volumetric data model using 0 as the boundary reference.

[0111] A regional response weight field is constructed to smoothly define topologically sensitive areas. Unlike traditional hard-boundary selection, this weight field achieves a gradual transition of sensitive areas through continuous numerical distribution. For example, in a wall scenario during shooting training, the weight field value at the edge of the wall is close to 0, while the value in the middle is close to 1. This means that drastic topological changes are prohibited at the edge, while normal adjustment is allowed in the middle, avoiding jagged edges or breaks due to topological changes at the edge.

[0112] The total spatial convolution integral of the perturbation kernel function and the regional response weight field is calculated. This integral combines the perturbation effect of local interactions with the constraint of the global sensitive region, achieving a smooth parameter update driving force. For example, when clicking on the middle of the wall, the perturbation kernel has the strongest effect in the central region, and the weight field has the highest value in the middle. The total integral after convolution of the two is concentrated in the middle, ensuring that the parameter update mainly affects the target region, while also taking into account global stability.

[0113] The structural connectivity control parameters are adjusted through a continuously weighted update mechanism of base step size × adjustment coefficient × convolution integral. For example, if the base evolution step size is 0.3, the update step size adjustment coefficient is 0.8, the total convolution integral is 2.5, and the actual update amount is 0.3 × 0.8 × 2.5 = 0.6. The parameters are updated by accumulating continuous values, avoiding discrete jumps.

[0114] A fixed-pipeline isosurface extraction is performed on the updated volumetric data model. The fixed pipeline ensures consistent parameters during the extraction process, avoiding geometric mesh flickering caused by fluctuations in the extraction algorithm. For example, the same resolution and sampling step size are used for each extraction to keep the scene output of consecutive frames consistent.

[0115] The base evolution step size is determined based on the complexity of the topology operation: 0.1 to 0.2 for simple topology operations such as small hole generation; 0.2 to 0.3 for medium-complexity operations such as medium-sized channel creation; and 0.3 to 0.5 for complex operations such as multi-component disconnection and reassembly. This range was determined through extensive tuning experiments and can match the update requirements of different operations.

[0116] The adjustment rule for the steepness of the weighted field is as follows: a value of 4 to 5 is used for sensitive areas that need clear definition, such as precision shooting training scenarios; a value of 2 to 3 is used for scenarios that require a smooth transition, such as rapid setup scenarios. This ensures that the definition of sensitive areas meets the requirements without causing topological abrupt changes in different scenarios.

[0117] The discretization method for spatial convolution integrals employs a uniform sampling strategy with a sampling stride of 1 pixel. Grid sampling is performed within the corresponding area of ​​the screen space. At each sampling point, the product of the perturbation kernel function value and the weight field value is calculated, and these products are then summed to obtain the total integral. This method balances computational efficiency and accuracy, making it suitable for real-time interactive requirements.

[0118] The parameters for fixed pipeline isosurface extraction are set as follows: the moving cube algorithm is used, the resolution is set to 512×512×512, and the sampling step size is 0.01 times the unit length of the scene to ensure that the extracted geometric mesh is smooth and free of redundant vertices, while also taking into account the computation speed.

[0119] like Figure 2 As shown, Figure 2 The interactive scene of the editable video shooting training scene construction system was demonstrated: a virtual training scene containing dense targets and occlusions was presented on the display screen. After the trainee performed a click operation, the system did not generate a traditional zero-width ray, but instead constructed a view cone interaction area with the click direction as the axis and covering the interaction neighborhood. This area accurately covered the dense targets and occlusions related to the click, intuitively demonstrating the design of mapping discrete click commands into a continuous view cone interaction space, and reflecting the interaction logic to deal with problems such as high-density occlusion and input deviation.

[0120] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. An editable image shooting training scene construction system, characterized in that, include: A parameterized scene generation module is used to construct and store a virtual scene based on a volumetric generation data model, wherein the volumetric generation data model includes basic shape units defined by continuous functions, and structural connectivity control parameters for adjusting the scene's geometric construction and connectivity state. The interactive area definition module is used to receive click instructions and, in combination with the input positioning deviation and the visual recognition limitation range, map the click instructions into a visual cone interactive area with a regional angle. The interaction stability calculation module is used to calculate the structural control stability index that characterizes the effectiveness of a single click interaction within the view frustum interaction area, based on the depth occlusion level, the generation unit projection density, and the subtended angle of the area. The scene structure control module is used to generate an update step size adjustment coefficient based on the structure control stability index and to construct a controlled parameter update scope. The update step size adjustment coefficient is used to dynamically adjust the scope and magnitude of the control parameter update domain, thereby continuously adjusting the parameters in the parameterized scene generation module to change the geometric structure of the virtual scene in real time.

2. The editable image shooting training scene construction system according to claim 1, characterized in that, A virtual scene is constructed and stored based on a volumetric generated data model, wherein the volumetric generated data model includes basic shape units defined by continuous functions, and structural connectivity control parameters for adjusting the scene's geometry and connectivity state, including: The symbolic distance field is used as a continuous function to describe the basic shape unit; The overall field function of the volumetric data model is constructed by combining multiple basic shape units using soft Boolean operation logic. The structural connectivity control parameters are introduced as input variables into the continuous gating activation function; The output value of the continuous gating activation function is used to dynamically adjust the role weight of the target basic shape unit in the soft Boolean operation logic, thereby realizing the topological state switching of hole opening and closing or component connection disconnection in the virtual scene through continuous numerical evolution.

3. The editable image shooting training scene construction system according to claim 1, characterized in that, Upon receiving a click command, and combining the input positioning deviation with the visual recognition limitation range, the click command is mapped to a visual cone interaction region with a regional angular area, including: The direction of the central ray is determined based on the click command; Based on the two-dimensional endpoint distribution covariance matrix of the input device, the half-angle of the input positioning deviation, which characterizes the input uncertainty, is calculated. Based on the offset of the click point in the field of view and the preset crowding critical spacing factor, the visual recognition limitation half angle representing the limitation of human eye resolution is calculated. By employing a square sum-rooted fusion strategy, the input positioning deviation half-angle and the visual recognition restriction half-angle are orthogonally superimposed to obtain a comprehensive effective region angle. Construct a conical space with the central ray direction as the axis and the effective area angle as the aperture boundary, as the view cone interaction region.

4. The editable image shooting training scene construction system according to claim 3, characterized in that, Depth occlusion levels include: Determine the set of projected pixels of the view frustum interaction region in screen space; Perform a continuous depth stripping operation on the pixels in the projected pixel set and obtain the fragment coverage count in each stripping process; The fragment coverage count is mapped to a continuous level existence value using a continuous smooth activation function; The cumulative occlusion depth value of a single pixel is obtained by summing the values ​​of all consecutive layers within a preset number of peeling operations. The average cumulative occlusion depth of all pixels in the projected pixel set is calculated to obtain the depth occlusion level.

5. The editable image shooting training scene construction system according to claim 3, characterized in that, Unit projection density, including: Identify all homogeneous generated module instances located within the spatial range of the view frustum interaction region; Obtain the projection center coordinates of the homogeneous generation module instance in screen space; The radius of the interactive neighborhood in the screen space is determined based on the angle of the effective region. Using a kernel density estimation algorithm, with the coordinates of the projection center as the sample point and the radius of the interactive neighborhood as the bandwidth, the local density response at the click location is calculated using a Gaussian smoothing kernel function; The local density response is normalized to the number of instances within a unit interaction area, and the number of instances within a unit interaction area is used as the unit projection density.

6. The editable image shooting training scene construction system according to claim 1, characterized in that, Based on the depth occlusion hierarchy, the generated unit projection density, and the region subtended angle, a structural regulation stability index characterizing the effectiveness of a single click interaction is calculated, including: Calculate the area of ​​the effective interactive region on a unit sphere based on the angle of the region. The effective aliasing scale, which characterizes the degree of aliasing, is obtained by multiplying the depth occlusion level, the generation unit projection density, and the effective interaction area. A preset small anti-shake bias is introduced, and the logarithm of the reciprocal of the effective aliasing scale is performed. The result of the logarithmic operation is determined as the structural regulation stability index, which is used to quantify the amount of topology control information that can be transmitted under the current interactive state.

7. The editable image shooting training scene construction system according to claim 6, characterized in that, Based on the structural control stability index, an update step size adjustment coefficient is generated, and a controlled parameter update scope is constructed, including: Calculate the difference between the preset target information budget value and the structural control stability index; The difference is input into a one-sided smoothing activation function to calculate the information gap that characterizes the degree of current interaction capacity insufficiency. The information gap is mapped using a negative exponential decay function to generate an update step size adjustment coefficient with a value between zero and one. Construct a Gaussian smooth perturbation kernel centered at the click point as the scope for updating the controlled parameters; The kernel width parameter of the Gaussian smoothing perturbation kernel is dynamically adjusted according to the amount of information gap, so that when the amount of information gap increases, the spatial range of the controlled parameter update domain expands, thereby reducing the sensitivity to local topological changes.

8. The editable image shooting training scene construction system according to claim 7, characterized in that, The update step size adjustment coefficient is used to dynamically adjust the range and magnitude of the controlled parameter update domain, continuously adjusting the parameters in the parameterized scene generation module to change the geometric structure of the virtual scene in real time, including: The preset basic evolution step size is scaled using the update step size adjustment coefficient to obtain the current actual parameter evolution step size; Based on the implicit field numerical distribution of the current virtual scene, a regional response weight field is constructed to smoothly delineate sensitive areas where topological changes are allowed. Calculate the total spatial convolution integral of the perturbation kernel function corresponding to the control parameter update scope and the region response weight field; The product of the total spatial convolution integral and the actual parameter evolution step size is accumulated into the structural connectivity control parameter to complete the continuous weighted update; The updated volumetric data model is subjected to fixed pipeline isosurface extraction to generate an updated virtual scene geometric mesh.

Images