Three-dimensional Gaussian model anomaly identification method and device, and computing equipment

Abstract

The embodiment of the invention provides a three-dimensional Gaussian model anomaly recognition method and device and computing equipment, and the method comprises the steps: obtaining a target three-dimensional Gaussian model which comprises a plurality of three-dimensional Gaussian points and a plurality of verification regions, the verification area is divided based on a preset spatial range; performing distribution statistics on the visual parameters of the three-dimensional Gaussian points in the plurality of verification areas to obtain visual feature distribution of the plurality of verification areas; and judging whether the target three-dimensional Gaussian model is abnormal or not based on the visual feature distribution of the plurality of verification regions and a preset visual feature distribution threshold. According to the method, local accurate division of the space of the three-dimensional Gaussian model is realized, the statistical law of the local features of the three-dimensional Gaussian model is effectively captured, and automatic and refined recognition of the anomaly of the three-dimensional Gaussian model is realized, so that the accuracy of model anomaly recognition is improved, and an objective basis is provided for quality evaluation of the three-dimensional Gaussian model.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of computer technology, and in particular to a method for identifying anomalies in a three-dimensional Gaussian model, a device for identifying anomalies in a three-dimensional Gaussian model, and a computing device. Background Technology

[0002] With the rapid development of 3D modeling and digital rendering technologies, 3D models are increasingly widely used in fields such as virtual reality interaction and scene display.

[0003] Currently, in the practical application of 3D Gaussian models, rendering and displaying the model is a crucial step. However, if the model contains anomalies, such as those resulting from secondary editing or missing data, it may contain hidden, potentially unsafe information or exhibit uncontrollable view rendering trajectories. Therefore, it is necessary to identify anomalies in 3D Gaussian models. Traditional methods for identifying model anomalies typically rely on global consistency checks or rule-based matching, extracting the overall features of the 3D model and comparing them with a preset template to determine the presence of anomalies.

[0004] However, the aforementioned anomaly identification methods all analyze the 3D model from a holistic perspective, failing to effectively capture subtle anomalies in certain local areas. Furthermore, they rely on manual judgment, making them susceptible to subjective influences, resulting in poor accuracy and low efficiency in anomaly identification. Therefore, a more accurate and efficient method for anomaly identification in 3D Gaussian models is urgently needed. Summary of the Invention

[0005] In view of this, embodiments of this specification provide a method for anomaly identification in a three-dimensional Gaussian model. One or more embodiments of this specification also relate to a device for anomaly identification in a three-dimensional Gaussian model, a computing device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0006] According to a first aspect of the embodiments of this specification, a method for anomaly identification in a three-dimensional Gaussian model is provided, comprising:

[0007] Obtain the target 3D Gaussian model, which includes multiple 3D Gaussian points and multiple verification regions, which are divided based on a preset spatial range.

[0008] The visual parameters of three-dimensional Gaussian points in multiple verification regions are statistically analyzed to obtain the visual feature distribution of multiple verification regions.

[0009] Based on the visual feature distribution of multiple verification regions and the preset visual feature distribution threshold, it is determined whether the target 3D Gaussian model has any anomalies.

[0010] According to a second aspect of the embodiments of this specification, a three-dimensional Gaussian model anomaly identification device is provided, comprising:

[0011] The acquisition module is configured to acquire a target 3D Gaussian model, wherein the target 3D Gaussian model includes multiple 3D Gaussian points and multiple verification regions, and the verification regions are divided based on the spatial range of the target 3D Gaussian model.

[0012] The statistics module is configured to perform distribution statistics on the visual parameters of three-dimensional Gaussian points in multiple verification regions to obtain the visual feature distribution of multiple verification regions.

[0013] The judgment module is configured to determine whether the target 3D Gaussian model has anomalies based on the visual feature distribution of multiple verification regions and a preset visual feature distribution threshold.

[0014] According to a third aspect of the embodiments of this specification, a computing device is provided, comprising:

[0015] Memory and processor;

[0016] The memory is used to store computer-executable instructions, and the processor is used to execute computer programs / instructions. When the computer programs / instructions are executed by the processor, they implement the steps of the above-described three-dimensional Gaussian model anomaly recognition method.

[0017] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores a computer program / instructions that, when executed by a processor, implement the steps of the above-described three-dimensional Gaussian model anomaly recognition method.

[0018] According to a fifth aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described three-dimensional Gaussian model anomaly recognition method.

[0019] One embodiment of this specification implements a method for anomaly identification of a three-dimensional Gaussian model, comprising: acquiring a target three-dimensional Gaussian model, wherein the target three-dimensional Gaussian model includes multiple three-dimensional Gaussian points and multiple verification regions, the verification regions being divided based on a preset spatial range; performing distribution statistics on the visual parameters of the three-dimensional Gaussian points within the multiple verification regions to obtain the visual feature distribution of the multiple verification regions; and determining whether the target three-dimensional Gaussian model has anomalies based on the visual feature distribution of the multiple verification regions and a preset visual feature distribution threshold.

[0020] By acquiring the target 3D Gaussian model, the target 3D Gaussian model is divided into multiple verification regions based on a preset spatial range, achieving accurate local division of the 3D Gaussian model space. By statistically analyzing the distribution of visual parameters of 3D Gaussian points within each verification region, the distribution of visual features is obtained, which can effectively capture the statistical regularity of local features of the 3D Gaussian model. By comparing the distribution of visual features with a preset threshold, automated and refined identification of anomalies in the 3D Gaussian model is achieved, thereby improving the accuracy of model anomaly identification and avoiding the bias of subjective human judgment, providing an objective basis for the quality assessment of the 3D Gaussian model. Attached Figure Description

[0021] Figure 1 This is a flowchart of a three-dimensional Gaussian model anomaly identification method provided in one embodiment of this specification;

[0022] Figure 2 This is a schematic diagram of the structure of a three-dimensional Gaussian model anomaly recognition device provided in one embodiment of this specification;

[0023] Figure 3 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0024] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0025] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0026] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0027] Furthermore, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are obtained through open-source datasets or public datasets that comply with their license agreements, or are obtained with full authorization from the relevant parties. Moreover, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0028] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0029] 3D Gaussian Splatting (3DGS) is a representation method that uses Gaussian ellipsoids to model 3D scenes. It achieves efficient rendering by rasterizing the Gaussian ellipsoid into an image. Compared with implicit representation methods such as neural radiation fields, its explicit representation facilitates downstream tasks such as dynamic reconstruction, geometric editing, and physical simulation.

[0030] Polygon File Format (.PLY) is a file format used to store 3D mesh model data. It supports binary encoding and is commonly used to store point cloud or triangular mesh data. It is flexible, readable, and concise, and allows users to customize file attributes such as color, transparency, normal vectors, and texture coordinates.

[0031] Splat Format (.SPLT) is also a file format used to store 3D Gaussian model data. It contains visual parameters of 3D Gaussian points, such as coordinates, size, rotation parameters, opacity, and spherical harmonics. It is commonly used for the storage and transmission of 3D Gaussian models. It is similar to the .PLY format but is specifically used for Gaussian point cloud data.

[0032] K-Nearest Neighbors (KNN) is a machine learning algorithm used for classification and regression. It calculates the distance between samples and selects the K nearest neighbors to vote on the category. It supports multiple distance metrics and is suitable for classification and regression tasks.

[0033] The Pearson correlation coefficient is a statistic used to measure the linear correlation between two variables X and Y. Its value ranges from -1 to 1. The larger the absolute value, the stronger the correlation. It is suitable for linear relationship analysis of continuous variables and requires that the data roughly follow a normal distribution.

[0034] Spearman's rank correlation coefficient is a statistical indicator used to measure the strength of the monotonic relationship between two variables. It does not depend on whether the data follows a normal distribution and is suitable for ordinal variables or continuous data that does not follow a normal distribution. It has a higher tolerance for outliers than Pearson's correlation coefficient.

[0035] With the rapid development of 3D modeling and digital rendering technologies, 3D models are increasingly widely used in fields such as virtual reality interaction and scene display.

[0036] Currently, in the practical application of 3D Gaussian models, rendering and displaying the model is a crucial step. However, if the model contains anomalies, such as those resulting from secondary editing or missing data, it may contain hidden, potentially unsafe information or exhibit uncontrollable view rendering trajectories. Therefore, it is necessary to identify anomalies in 3D Gaussian models. Traditional methods for identifying model anomalies typically rely on global consistency checks or rule-based matching, extracting the overall features of the 3D model and comparing them with a preset template to determine the presence of anomalies.

[0037] However, the above-mentioned anomaly detection methods all analyze the 3D model from an overall perspective, which cannot effectively capture subtle anomalies in some local areas of the 3D model. Furthermore, they rely on manual judgment and are easily affected by subjective factors, resulting in poor accuracy and low efficiency in anomaly detection.

[0038] In view of this, this specification provides a method for identifying anomalies in a three-dimensional Gaussian model. This specification also relates to a device for identifying anomalies in a three-dimensional Gaussian model, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0039] See Figure 1 , Figure 1 A flowchart of a three-dimensional Gaussian model anomaly identification method according to an embodiment of this specification is shown, which specifically includes the following steps.

[0040] Step 102: Obtain the target 3D Gaussian model, which includes multiple 3D Gaussian points and multiple verification regions, which are divided based on a preset spatial range.

[0041] The anomaly identification method for 3D Gaussian models provided in one or more embodiments of this specification can be applied to various scenarios, including but not limited to virtual reality interaction, scene display, e-commerce, game development, and film and television production, for anomaly identification of 3D Gaussian models. It can be applied to content production platforms, 3D model sharing platforms, cloud computing platforms, and various mobile terminals.

[0042] A 3D Gaussian model is digital content built upon a 3D scene, allowing users to switch between multiple viewpoints, dynamically adjust parameters, or interact with scene elements. Specifically, a 3D Gaussian model can be based on 3D Gaussian points, and algorithms can be used to generate renderable 3D models that support user observation or manipulation from different perspectives. For example, it can be constructed using 3D Gaussian Splatting (3DGS) technology.

[0043] The target 3D Gaussian model is a specific 3D Gaussian model that needs to be analyzed for anomaly identification. It can include various 3D information and visual attribute parameters of a 3D scene or 3D space.

[0044] A 3D Gaussian point is a set of 3D spatial data composed of a large number of discrete points. It is the basic unit for constructing a 3D Gaussian model. Each 3D Gaussian point contains parameters such as position coordinates, size parameters, rotation parameters, opacity, and spherical harmonic coefficients, used to accurately represent the visual features and spatial distribution of a specific location in a 3D scene. Specifically, 3D Gaussian points can exist in the form of a Gaussian point cloud, serving as the foundation for constructing a 3D Gaussian model. Each 3D Gaussian point in the Gaussian point cloud can be a Gaussian sphere, i.e., a sphere, ellipsoid, or other shape with certain attributes such as volume, color, and opacity.

[0045] A verification region is a local detection unit or window that divides the target 3D Gaussian model according to its spatial location. It can be used to analyze and evaluate the consistency of local features within a specific spatial range of the model. Each verification region covers a portion of the model's 3D space and contains a certain number of 3D Gaussian points, thereby enabling localized analysis of the 3D Gaussian model space. This allows anomaly identification to focus on the feature distribution patterns within a specific spatial range. For example, a relatively complex 3D Gaussian model can be divided into multiple cube-shaped or spherical verification regions, and these regions can overlap or be closely adjacent to each other to fully cover the entire 3D Gaussian model space and avoid missing any anomaly regions.

[0046] The preset spatial range is a predefined set of spatial scale and shape parameters used to divide the verification area. It can include configuration parameters such as the size, shape, distribution density, and overlap ratio of the area. The preset spatial range can be a fixed value, such as a cube with a side length of a certain length (e.g., 10cm), or a relative value adaptively calculated based on the overall size of the 3D Gaussian model, such as a cube with a size of 1 / 10 of the overall model size. It can also be adaptively adjusted according to the scene complexity of the model. The preset spatial range provides a standard and basis for dividing the verification area, ensuring the systematic and comprehensive nature of the analysis of the 3D Gaussian model, and balancing computational efficiency and detection accuracy. A smaller range can capture more subtle anomalies, but the corresponding computational cost is higher; a larger range can cover a wider area, but may ignore small-scale anomalies. The specific preset spatial range can be flexibly determined according to the size and complexity of the actual 3D Gaussian model or the detection accuracy required by the applied scene. This specification does not impose specific limitations on this.

[0047] In practical applications, the target 3D Gaussian model can be obtained through various methods. For example, it can include loading the 3D Gaussian model file from a local storage device, downloading the 3D Gaussian model data to be detected from a server via a network interface, or acquiring it in real time using a 3D scanning device or reconstructing the input multi-view image sequence using algorithms such as 3D Gaussian Splatting.

[0048] Specifically, the target 3D Gaussian model can typically be stored in standard file formats, such as .PLY or .SPLT. These files can include the visual parameters of the 3D Gaussian points in the model. Once the target 3D Gaussian model is obtained, the model file can be read and parsed to obtain various attribute information such as the coordinate parameters, size parameters, color attributes, and transparency of each 3D Gaussian point.

[0049] In some optional implementation schemes, a model preprocessing process can be set up, such as performing integrity verification and format standardization on models downloaded from the network, and performing coordinate system normalization on models generated from training. In other optional schemes, metadata information from the model's source, such as camera parameters and training configuration, can be combined to help determine the model's credibility and provide more comprehensive contextual information for subsequent anomaly identification.

[0050] Multiple verification regions in the target 3D Gaussian model can be divided based on a preset spatial range. Specifically, the division of verification regions can employ regular mesh partitioning or an adaptive spatial segmentation strategy.

[0051] For example, for the regular mesh generation method, the overall spatial bounding box of the target 3D Gaussian model can be calculated first. Then, according to preset spatial range parameters, the model space can be uniformly divided into multiple adjacent verification regions. A preset overlap ratio between each verification region can also be configured to avoid boundary effects. Specifically, the minimum bounding cube of the 3D Gaussian model can be determined first. Then, according to preset step size and overlap ratio, mesh points can be generated in the X, Y, and Z dimensions, with each mesh point defining a verification region centered on it.

[0052] For adaptive spatial segmentation strategies, the size and position of the validation region can be dynamically adjusted based on the spatial distribution density of each 3D Gaussian point in the target 3D Gaussian model. For example, a smaller validation region is used in areas with denser point clouds to capture subtle anomalies, while a larger validation region is used in areas with sparser point clouds to ensure statistical significance. Furthermore, a hybrid segmentation strategy can be employed, applying different spatial range parameters to different parts of the 3D Gaussian model. For instance, a finer segmentation can be used for foreground objects, while a coarser segmentation can be used for background regions, thereby optimizing computational resource allocation while maintaining detection accuracy.

[0053] In this step, by acquiring the target 3D Gaussian model, which includes a verification region based on a preset spatial range, a structured data foundation is laid for subsequent anomaly identification. This allows anomaly identification to be performed at the local scale of the model, effectively capturing subtle anomalies in local areas. Dividing the verification region based on the preset spatial range avoids the subjective bias of manually selecting analysis areas. Through reasonable configuration of spatial range parameters, an optimal balance can be achieved between computational complexity and identification accuracy, adapting to the needs of 3D scenes with varying complexity. This provides an objective basis for anomaly identification of the 3D Gaussian model and improves the efficiency of anomaly identification.

[0054] Step 104: Perform distribution statistics on the visual parameters of the three-dimensional Gaussian points in multiple verification regions to obtain the visual feature distribution of multiple verification regions.

[0055] Visual parameters are a set of attributes related to the visual appearance of a 3D Gaussian point. These can include coordinate parameters, size parameters, color attributes, and transparency parameters, describing the spatial location, shape, size, color, and other characteristics of the 3D Gaussian point. Visual parameters can accurately characterize the visual appearance of a 3D Gaussian point, providing a data foundation for subsequent distribution statistics.

[0056] Distribution statistics involves statistically analyzing and quantifying the visual parameters of three-dimensional Gaussian points within a specific validation region. This analysis is used to extract the distribution patterns and statistical characteristics of these visual parameters within the validation region. Distribution statistics transforms raw visual parameters into statistically significant feature indicators, thereby characterizing the local data distribution patterns within the validation region.

[0057] Visual feature distribution is a set of quantified features obtained by statistically analyzing the distribution of visual parameters of 3D Gaussian points within a verification region. It can be used to characterize the distribution patterns and statistical properties of visual parameters within that region. Specifically, visual feature distribution can include local coordinate uniformity, local density abrupt change, and local attribute correlation. Local coordinate uniformity measures the uniformity of the coordinate distribution of 3D Gaussian points and can be calculated using the mean and variance of the coordinate parameters. Local density abrupt change assesses the magnitude of density variation in 3D Gaussian points and can be obtained through density statistics of neighboring point sets. Local attribute correlation analyzes the correlation between visual parameters of 3D Gaussian points and can be determined by calculating the correlation coefficient. Visual feature distribution can effectively capture local anomalies in 3D Gaussian models caused by human editing or data gaps, and can be used for subsequent comparison with preset visual feature distribution thresholds, providing a scientific basis for anomaly identification.

[0058] In practical applications, the distribution statistics of visual parameters of three-dimensional Gaussian points in multiple verification areas can be carried out, and various technical solutions can be adopted according to different types of visual parameters and statistical objectives.

[0059] One possible approach is to calculate the mean vector and covariance matrix of all three-dimensional Gaussian point coordinates within the current verification area based on the distribution statistics of coordinate parameters, thereby determining the local coordinate uniformity. For example, the ratio of the variance to the mean of the coordinate distribution can be used as a uniformity index. The smaller the ratio, the more concentrated the point distribution and the higher the coordinate uniformity.

[0060] Another alternative approach is to construct a spatial density field based on the joint distribution statistics of coordinate and size parameters. By using local density estimation or voxelization methods, the Gaussian point distribution density within each verification region can be calculated to determine the local density parameters. Subsequently, by comparing the density differences between adjacent verification regions, the local density abruptness can be calculated. For example, the absolute value or relative rate of change of the density difference between adjacent regions can be used as the abruptness index.

[0061] Another possible approach is to calculate the correlation coefficient of visual parameters between any two three-dimensional Gaussian points within the current verification area, such as the Pearson correlation coefficient, based on the distribution statistics of visual attribute parameters such as color and transparency. This correlation coefficient can then be used to form the local attribute correlation through averaging or weighted aggregation.

[0062] In addition, information entropy analysis can be used to calculate the degree of uncertainty in the distribution of visual parameters within the verification area; histogram statistics can be used to analyze the frequency of parameter values ​​in different intervals; principal component analysis can be used to extract the main directions and intensities of change within the area; and clustering algorithms can be used to identify the parameter distribution patterns within the area.

[0063] Different distribution statistics methods can be flexibly determined according to the actual anomaly identification scenario or needs, and the embodiments in this specification do not make specific limitations on this.

[0064] In this step, by statistically analyzing the distribution of visual parameters of 3D Gaussian points within multiple verification regions, the visual feature distribution of multiple verification regions is obtained, enabling quantitative analysis of local features of the 3D Gaussian model and effectively capturing the statistical regularity of the spatial distribution of visual parameters. Through distribution statistics, the high-dimensional original visual parameters are transformed into statistical features with clear physical meaning, realizing the quantitative expression of local characteristics of the 3D Gaussian model and improving the objectivity and repeatability of anomaly identification. By performing distribution statistics on different verification regions separately, a refined spatial analysis of the 3D Gaussian model is achieved, effectively capturing subtle anomaly features within local regions and improving the spatial resolution of anomaly detection. By constructing a multi-dimensional visual feature distribution system based on distribution statistics of multiple visual parameters, the anomaly identification results are more comprehensive, providing a data foundation for anomaly identification of the 3D Gaussian model and improving the accuracy and efficiency of anomaly identification.

[0065] Step 106: Based on the visual feature distribution of multiple verification regions and the preset visual feature distribution threshold, determine whether there is anomaly in the target 3D Gaussian model.

[0066] The preset visual feature distribution threshold is a pre-defined set of critical criteria used to determine whether a 3D Gaussian model is abnormal. It specifies the statistical boundary conditions that a normal 3D Gaussian model should satisfy in terms of visual feature distribution. The preset visual feature distribution threshold provides an objective quantitative standard for anomaly identification, freeing the evaluation of 3D Gaussian models from the limitations of subjective experience. Specifically, the preset visual feature distribution threshold can be set according to the obtained visual feature distribution. For example, preset visual feature distribution thresholds include coordinate anomaly thresholds, density mutation thresholds, and attribute correlation thresholds. The preset visual feature distribution threshold can be determined through statistical analysis of a large number of normal models or adjusted according to the security requirements of actual application scenarios, providing a scientific basis for model anomaly identification.

[0067] In practical applications, once the visual feature distributions of multiple verification regions are obtained, it is possible to determine whether the target 3D Gaussian model has any anomalies based on the visual feature distributions of multiple verification regions and the preset visual feature distribution threshold.

[0068] One method for determining whether a target 3D Gaussian model is abnormal is to determine whether the visual feature distribution exceeds a preset visual feature distribution threshold. If it does, the target 3D Gaussian model is abnormal.

[0069] Specifically, for different types of visual feature distributions, corresponding preset visual feature distribution thresholds can be used for judgment.

[0070] One possible method is to compare the local coordinate uniformity of each verification region with a preset coordinate anomaly threshold. If the local coordinate uniformity (such as the ratio of coordinate variance to mean) of any verification region exceeds the preset threshold, the model is determined to be abnormal. This method is suitable for regions with uneven coordinate distribution caused by human editing.

[0071] Another option is to calculate the local density abruptness between adjacent verification regions and compare it with a preset density abruptness threshold. If the local density abruptness (such as the relative rate of change of point density in adjacent regions) of any region exceeds the threshold, the model is determined to be abnormal. This method is suitable for density discontinuities caused by the addition or deletion of point cloud data.

[0072] Another alternative method is to evaluate the local attribute correlation of each verification region against a preset attribute correlation threshold. If the local attribute correlation of any verification region (such as the Pearson correlation coefficient between the color parameters of two three-dimensional Gaussian points in the region) is lower than the threshold, the model is judged to be abnormal. This method is suitable for detecting parameter correlation breaks caused by local modification of visual attributes.

[0073] In practical applications, a single threshold judgment strategy can be adopted, or a multi-threshold fusion decision mechanism can be constructed. If there is an anomaly in the distribution of any type of visual feature, the three-dimensional Gaussian model is judged to be abnormal. Alternatively, different visual feature distributions can be comprehensively judged through weighted voting, machine learning classifiers, and other methods to form a more reliable anomaly recognition result.

[0074] In this step, by systematically comparing the visual feature distribution of multiple verification areas with preset visual feature distribution thresholds, an objective judgment of the abnormal state of the target 3D Gaussian model is achieved, significantly improving the scientificity and reliability of model quality assessment. By using preset thresholds as an objective benchmark, the consistency and repeatability of detection results are ensured. Through anomaly identification analysis at the local scale, anomalies in subtle areas of the 3D Gaussian model are effectively captured, improving the sensitivity to hidden anomalies. The identification process through preset threshold comparison is computationally efficient and can be applied to large-scale 3D Gaussian model identification and real-time application scenarios.

[0075] In the embodiments of this specification, by acquiring the target 3D Gaussian model, the target 3D Gaussian model is divided into multiple verification regions based on a preset spatial range, thereby achieving accurate local division of the 3D Gaussian model space. By statistically analyzing the distribution of visual parameters of 3D Gaussian points within each verification region, the distribution of visual features is obtained, which can effectively capture the statistical regularity of local features of the 3D Gaussian model. By comparing the distribution of visual features with a preset threshold, automated and refined identification of anomalies in the 3D Gaussian model is achieved, thereby improving the accuracy of model anomaly identification and avoiding the bias of subjective human judgment, providing an objective basis for the quality assessment of the 3D Gaussian model.

[0076] In one optional embodiment of this specification, a preset overlap ratio exists between multiple verification areas;

[0077] The visual parameters of 3D Gaussian points within multiple verification regions are statistically analyzed to obtain the visual feature distribution of multiple verification regions, including:

[0078] For the current verification area, the visual parameters of the three-dimensional Gaussian points in the current verification area are statistically analyzed to obtain the visual feature distribution of the current verification area;

[0079] Based on a preset overlap ratio, update the verification region that overlaps with the current verification region as the current verification region;

[0080] Return to the current verification area and perform statistical analysis on the distribution of visual parameters of the three-dimensional Gaussian points within the current verification area to obtain the visual feature distribution of the current verification area. Continue until the preset stopping condition is met to obtain the visual feature distribution of multiple verification areas.

[0081] The preset overlap ratio is a predefined parameter indicating the degree of spatial overlap between multiple verification regions. It can be expressed as a percentage or a decimal and is used to ensure sufficient overlap between adjacent verification regions to avoid boundary effects. The preset overlap ratio provides a basis for the sliding switching and continuous coverage of verification regions, ensuring the continuity and integrity of local feature statistics for the target 3D Gaussian model. Specifically, the preset overlap ratio can be set to a fixed value, such as 10%-30%; or dynamically adjusted according to the complexity of the 3D Gaussian model, for example, setting a 25% overlap ratio in dense point cloud regions and a 15% overlap ratio in point cloud coefficient regions, to ensure that feature statistics do not abruptly change at region boundaries, thereby effectively capturing subtle anomalies caused by editing.

[0082] The current validation region is the specific validation area currently being processed during the distribution statistics process; it is the current processing object of the distribution statistics operation. Specifically, the current validation region is the basic operational unit for realizing local feature statistics. By progressively updating the current validation region, continuous statistical coverage of the entire 3D Gaussian model space is achieved, ensuring that each local space is systematically analyzed, thereby avoiding the omission of anomalies.

[0083] A preset stopping condition is a predefined criterion used to determine the end of the distribution statistics process. This can include the extent of the area covered by the distribution statistics, the number of processed verification regions, and the convergence of the statistical results. Specifically, the preset stopping condition controls the execution scope and termination timing of the distribution statistics process, ensuring its integrity and efficiency. In practical applications, the preset stopping condition can be set to "traverse all verification regions," thus guaranteeing full coverage of the 3D Gaussian model. Optionally, the preset stopping condition can also be set to "the verified regions that have undergone distribution statistics cover 95% of the model space," thereby avoiding unnecessary redundant calculations and improving processing efficiency.

[0084] In practical applications, distribution statistics can be performed on multiple verification regions. The distribution statistics can be performed separately on the current verification region. The current verification region can be any one of the multiple verification regions, which is usually the verification region starting from the origin of the three-dimensional Gaussian model.

[0085] Once the distribution statistics of the current verification area are completed and the visual distribution characteristics of the current verification area are obtained, the verification areas adjacent to and overlapping with the current verification area can be determined based on a preset overlap ratio. These overlapping verification areas are then updated as the current verification area, and the process of performing distribution statistics returns to normal. This ensures the continuity of the statistical process and avoids statistical gaps between verification areas.

[0086] When the preset stopping conditions are met, such as when all verification regions have been traversed and the entire 3D Gaussian model space has been covered, the verification regions can be updated, and the visual feature distribution of multiple verification regions can be obtained.

[0087] In the embodiments described in this specification, by switching and updating the current verification region based on a preset overlap ratio and performing iterative distribution statistics, the completeness and accuracy of anomaly identification for the 3D Gaussian model are improved. This achieves a systematic traversal of the 3D Gaussian model space, ensuring that each local region can be analyzed, thereby capturing subtle anomalies. By controlling the loop termination through preset stopping conditions, the processing efficiency is optimized, redundant calculations and resource waste are avoided, and the robustness and automation of detection are enhanced. While ensuring detection accuracy, a balance between computational complexity and coverage is achieved.

[0088] In one optional embodiment of this specification, the visual parameters of a three-dimensional Gaussian point include coordinate parameters;

[0089] The visual parameters of 3D Gaussian points within multiple verification regions are statistically analyzed to obtain the visual feature distribution of multiple verification regions, including:

[0090] For the current verification area, based on the coordinate parameters of the three-dimensional Gaussian points within the current verification area, determine the mean and variance of the coordinate distribution of the three-dimensional Gaussian points;

[0091] Based on the mean and variance of the coordinate distribution, the local coordinate uniformity of the current verification area is determined.

[0092] Coordinate parameters are the spatial location attributes of a 3D Gaussian point, specifically including 3D coordinates (x, y, z), which can be used to accurately describe the position of a 3D Gaussian point in 3D space. In particular, coordinate parameters are the core attributes characterizing the spatial location of a 3D Gaussian point, serving as the foundational data for local spatial analysis and distribution statistics, and providing data support for subsequent coordinate distribution statistics.

[0093] The coordinate distribution mean is the statistical average of the coordinate parameters of all three-dimensional Gaussian points within the current validation region, used to characterize the center position of points within that region. Specifically, the coordinate distribution mean is typically a three-dimensional vector, which can include the mean of each of the three-dimensional coordinates (x, y, z), thus measuring the spatial distribution center of each three-dimensional Gaussian point within the current validation region. For example, if the mean x-coordinate of all three-dimensional Gaussian points within a validation region is 2.3, the mean y-coordinate is 1.8, and the mean z-coordinate is 0.5, it means that the three-dimensional Gaussian points within that region are roughly concentrated around the coordinate position (2.3, 1.8, 0.5).

[0094] Coordinate distribution variance is a measure of the dispersion of the coordinate parameters of all 3D Gaussian points within the current validation region relative to the mean of the coordinate distribution. It is used to characterize the degree of concentration of points within the current region. Specifically, coordinate distribution variance can typically be a scalar value or a covariance matrix, representing the range of fluctuations and uniformity of the point coordinates. Coordinate distribution variance can be used to assess the uniformity of the spatial distribution of 3D Gaussian points within the validation region; a larger variance value indicates a more dispersed point distribution, while a smaller variance value indicates a more concentrated point distribution.

[0095] Local coordinate uniformity is a quantitative indicator calculated based on the mean and variance of the coordinate distribution within the current verification region. It measures the uniformity and consistency of the coordinate distribution of each 3D Gaussian point within the current verification region. Local coordinate uniformity provides a metric for evaluating whether the coordinates of local regions of the model are abnormal, identifying coordinate distribution anomalies caused by manual editing or missing data, such as point cloud position shifts or spatial discontinuities. The local coordinate uniformity is compared with a preset visual feature distribution threshold to determine whether the 3D Gaussian model has anomalies. If the local coordinate uniformity exceeds the preset coordinate anomaly threshold, it indicates that the verification region is abnormal, meaning the target 3D Gaussian model is abnormal.

[0096] In practical applications, for the current verification area, the coordinate parameters of all three-dimensional Gaussian points in the area can be determined. The mean values ​​of the three coordinate axes x, y, and z can be calculated respectively. Furthermore, the variance of the coordinate distribution can be determined based on the average of the squared differences between the coordinates of each three-dimensional Gaussian point on each coordinate axis and the mean value of the coordinates.

[0097] The mean and variance of a coordinate distribution can be calculated in various ways. For example, the sample mean and variance can be calculated directly, or a weighted average can be used to calculate the mean and variance. Alternatively, the statistical characteristics of a multidimensional coordinate distribution can be calculated using the covariance matrix.

[0098] Once the mean and variance of the coordinate distribution are calculated, the local coordinate uniformity of the current verification area can be further determined, thereby measuring the uniformity of the coordinate distribution of points within the verification area.

[0099] Specifically, various calculation methods can be used to determine local coordinate uniformity. For example, the ratio of variance to mean, the ratio of standard deviation to mean, or other statistical indicators such as the coefficient of variation can be used. For instance, the ratio of variance to mean for each coordinate axis can be calculated, and then the average value can be taken as the local coordinate uniformity index. Alternatively, the ratio can be calculated separately for each of the three coordinate axes, and then a weighted average can be used to obtain the overall local coordinate uniformity.

[0100] For example, local coordinate uniformity can be calculated and determined in the following way:

[0101] Let the mean value of the coordinate distribution within the current verification region be μ. P The variance of the coordinate distribution is Where P represents the coordinate parameters, including (x, y, z), and the corresponding μ P It can include μ X μ Y μ Z ,Can include

[0102] The local coordinate uniformity 'a' can then be expressed as:

[0103]

[0104] Among them, P i Let x be the coordinates of the i-th 3D Gaussian point, including (x, y, y) i ,y i ,z i ), where ∈ is a hyperparameter.

[0105] In the embodiments of this specification, the mean and variance of the coordinate distribution are determined by coordinate parameters, and the local coordinate uniformity is further calculated and determined. The mean and variance of the coordinate distribution provide basic statistics, while the local coordinate uniformity provides an intuitive anomaly indicator. This enables an accurate assessment of the uniformity of the local coordinate distribution of the 3D Gaussian model. The coordinates of the 3D Gaussian points are used to objectively measure spatial consistency, avoiding the bias of subjective human judgment. At the same time, the sensitivity to subtle anomalies is improved through local scale analysis, thereby improving the accuracy and reliability of anomaly identification for the target 3D Gaussian model.

[0106] In one optional embodiment of this specification, the preset visual feature distribution threshold includes a coordinate anomaly threshold;

[0107] Based on the visual feature distribution of multiple verification regions and a preset visual feature distribution threshold, determine whether the target 3D Gaussian model has anomalies, including:

[0108] The local coordinate uniformity and coordinate anomaly threshold are judged. If the local coordinate uniformity of any verification area exceeds the coordinate anomaly threshold, the target 3D Gaussian model is judged to have an anomaly.

[0109] The coordinate anomaly threshold is a pre-set critical standard used to determine whether the coordinate distribution of a 3D Gaussian model, i.e., the local coordinate uniformity, is abnormal. It specifies the statistical boundary conditions that a normal 3D Gaussian model should satisfy in terms of coordinate distribution. Specifically, the coordinate anomaly threshold can provide an objective quantitative standard for anomaly identification based on local coordinate uniformity, making the anomaly identification process automated and repeatable, and avoiding biases caused by subjective experience.

[0110] In practical applications, the coordinate anomaly threshold can be set as the upper limit of local coordinate uniformity, meaning that under normal circumstances, the local coordinate uniformity of three-dimensional Gaussian points within the verification area should not exceed this value.

[0111] Once the local coordinate uniformity for the current verification area is determined, a judgment can be made based on the local coordinate uniformity and the coordinate anomaly threshold.

[0112] Specifically, by comparing the local coordinate uniformity of multiple verification regions with a coordinate anomaly threshold, the local coordinate uniformity of each verification region is extracted. If this value exceeds the coordinate anomaly threshold, the verification region is determined to have coordinate anomalies. If any verification region is determined to have coordinate anomalies among multiple verification regions, then the entire target 3D Gaussian model can be judged to have anomalies.

[0113] Optionally, the method for determining the coordinate anomaly threshold can be set by statistical analysis of multiple normal 3D Gaussian models as samples, or adjusted according to the security requirements of the actual application scenario.

[0114] In practical applications, coordinate distribution data of multiple normal 3D Gaussian models can be collected first. The ratio of their coordinate distribution variance to mean can be calculated, and a suitable value can be determined as the coordinate anomaly threshold. Furthermore, the coordinate anomaly threshold can be dynamically adjusted according to the application scenario and security requirements of the 3D Gaussian model. For example, in scenarios with high security requirements, the coordinate anomaly threshold can be set lower to improve the sensitivity of anomaly detection.

[0115] Optionally, in the process of judging the uniformity of local coordinates and the threshold of coordinate anomalies, a single threshold judgment strategy can be adopted, or a multi-threshold fusion decision mechanism can be constructed to improve the accuracy and reliability of anomaly identification by comprehensively judging the anomalies in the coordinate distribution of different verification areas.

[0116] In the embodiments of this specification, by comparing the local coordinate uniformity with the coordinate anomaly threshold, it is possible to effectively capture areas of uneven coordinate distribution caused by human editing, improve the sensitivity to subtle anomalies, and achieve objective and quantitative identification and judgment of coordinate distribution anomalies in 3D Gaussian models, thereby improving the accuracy and reliability of anomaly identification. In particular, by determining the coordinate anomaly threshold through statistical analysis based on multiple normal models, the scientific and objective nature of the judgment standard can be guaranteed, avoiding interference from subjective experience. Furthermore, the coordinate anomaly threshold also provides an important basis for the quality assessment of 3D Gaussian models.

[0117] In one optional embodiment of this specification, the visual parameters of a three-dimensional Gaussian point include coordinate parameters and size parameters;

[0118] The visual parameters of 3D Gaussian points within multiple verification regions are statistically analyzed to obtain the visual feature distribution of multiple verification regions, including:

[0119] For the current verification area, the distribution density parameters of the current verification area are determined based on the coordinate and size parameters of the three-dimensional Gaussian points within the current verification area.

[0120] Based on the distribution density parameters corresponding to multiple validation regions, the local density abruptness between multiple validation regions is determined.

[0121] Coordinate parameters are the spatial location attributes of a 3D Gaussian point, specifically including 3D coordinates (x, y, z), which can be used to accurately describe the position of a 3D Gaussian point in 3D space. In particular, coordinate parameters are the core attributes characterizing the spatial location of a 3D Gaussian point, serving as the foundational data for local spatial analysis and distribution statistics, and providing data support for subsequent coordinate distribution statistics.

[0122] Dimensional parameters are attributes of a 3D Gaussian point that characterize its spatial size and shape. They can include geometric features such as the semi-major axis, semi-minor axis, and radius, describing the point's volume and morphology in space. Specifically, dimensional parameters accurately characterize the size properties of a 3D Gaussian point, providing fundamental data for subsequent distribution density calculations, and together with coordinate parameters, constitute the spatial representation of the point. In a 3D Gaussian model, the uniformity of the dimensional parameter distribution reflects the model's construction quality and editing traces.

[0123] The distribution density parameter is a quantitative indicator describing the spatial density of three-dimensional Gaussian points within a validation region. It can be determined by coordinate and size parameters and is used to characterize the distribution density of points within that region. Specifically, the distribution density parameter transforms the spatial distribution of three-dimensional Gaussian points into quantifiable statistical characteristics, providing a basis for calculating local density abrupt changes. For example, the distribution density parameter can be expressed as the average density value of three-dimensional Gaussian points within the validation region, i.e., the number of three-dimensional Gaussian points within the validation region divided by the volume of the region; it can also be expressed as the local density value calculated using the K-nearest neighbor algorithm, i.e., the reciprocal of the average nearest neighbor distance of each three-dimensional Gaussian point within the validation region.

[0124] Local density abruptness is a quantitative indicator that measures the degree of difference in distribution density between adjacent validation regions. It can be used to assess the spatial variation of 3D Gaussian point density. Specifically, local density abruptness can transform the distribution density differences between adjacent regions into comparable statistical features, providing a key basis for identifying model anomalies. For example, local density abruptness can be expressed as the relative rate of change of distribution density between adjacent validation regions, i.e., |(d1-d2) / d1|, where d1 and d2 are the distribution densities of two adjacent regions, respectively; it can also be expressed as the absolute value of the difference in distribution density between adjacent regions, i.e., |d1-d2|.

[0125] In practical applications, the distribution density parameter can be determined by calculating the average density of points within the region or by constructing a spatial density field, based on the coordinate and size parameters of the three-dimensional Gaussian points within the current verification region.

[0126] Specifically, statistical calculations can be performed based on the coordinate and size parameters of the three-dimensional Gaussian points within the current verification area, which can be done through various methods.

[0127] One possible approach is to determine the bounding box or volume of the current verification region by calculating the coordinates of all three-dimensional Gaussian points within the region, and then calculate the ratio of the number of three-dimensional Gaussian points in the region to the volume of the region as a distribution density parameter.

[0128] Another alternative approach is to use the K-nearest neighbor algorithm to calculate the nearest neighbor distance for each 3D Gaussian point, and then calculate the reciprocal of the average nearest neighbor distance of points within that region as the distribution density parameter.

[0129] Another alternative approach is to use the voxel method to divide each validation region into multiple small voxels, count the number of points within each voxel, and then calculate the average density as the distribution density parameter.

[0130] The specific calculation method can be flexibly selected according to the actual characteristics of the three-dimensional Gaussian model and the requirements of the anomaly identification scenario to ensure the accuracy of the distribution density parameter and the calculation efficiency. The embodiments in this specification do not make specific limitations on this.

[0131] For example, the density distribution parameters can be calculated and determined in the following way:

[0132] Let R be the current verification region, and let N be the set of neighboring points of the i-th 3D Gaussian point within R. R (i).

[0133] Then the density distribution parameter d of the current verification region j It can be represented as:

[0134]

[0135] Where j represents the j-th verification region in the target 3D Gaussian model.

[0136] Once the distribution density parameters corresponding to multiple validation regions are determined, the local density abruptness between multiple validation regions can be determined based on the distribution density parameters corresponding to multiple validation regions.

[0137] Specifically, the method for determining the local density abruptness can be based on the distribution density parameters corresponding to multiple validation regions, and can be achieved by comparing the distribution density differences between adjacent validation regions. In practical applications, various methods can be used to determine this.

[0138] One possible method is to calculate the relative rate of change of distribution density between adjacent verification regions, i.e., |(d1-d2) / d1|, where d1 and d2 are the distribution densities of two adjacent regions, respectively; another possible method is to calculate the absolute value of the difference in distribution density between adjacent regions, i.e., |d1-d2|; yet another possible method is to calculate the ratio of distribution density between adjacent regions, i.e., d1 / d2, and then compare it with a preset threshold.

[0139] In practical applications, a single-indicator judgment strategy can be adopted, or a multi-indicator fusion decision mechanism can be constructed to comprehensively consider the density changes in different adjacent areas, and improve the accuracy and reliability of local density abruptness judgment through weighted averaging or machine learning methods.

[0140] In the embodiments of this specification, by determining the distribution density parameters based on coordinate and size parameters, and further determining the local density abruptness based on the distribution density parameters of multiple verification regions, an objective quantitative analysis of local density changes in a 3D Gaussian model is achieved, improving the accuracy and reliability of anomaly identification. Specifically, determining the distribution density parameters based on coordinate and size parameters accurately reflects the spatial distribution characteristics of 3D Gaussian points, providing a data foundation for calculating the local density abruptness. Determining the local density abruptness by comparing the distribution density differences between adjacent verification regions effectively captures model density discontinuities caused by human editing or data loss, enabling accurate identification of model anomalies at the local scale, improving sensitivity to subtle anomalies, and thus enhancing the accuracy and reliability of anomaly identification for the target 3D Gaussian model.

[0141] In one optional embodiment of this specification, the preset visual feature distribution threshold includes a density abrupt change threshold;

[0142] Based on the visual feature distribution of multiple verification regions and a preset visual feature distribution threshold, determine whether the target 3D Gaussian model has anomalies, including:

[0143] The local density mutation degree and density mutation threshold are compared. If the local density mutation degree in any verification region exceeds the density mutation threshold, the target 3D Gaussian model is judged to be abnormal.

[0144] The density mutation threshold is a pre-defined set of critical criteria used to judge the density mutation characteristics of a 3D Gaussian model, i.e., whether the local density mutation degree is abnormal. It specifies the statistical boundary conditions that a normal 3D Gaussian model should satisfy in terms of density mutation characteristics. Specifically, the density mutation threshold can provide an objective quantitative standard for anomaly identification and judgment, so that the assessment of density change anomalies in 3D Gaussian models can avoid the limitations of subjective experience.

[0145] In practical applications, once the local density abruptness of multiple verification regions is determined, the judgment can be made using the local density abruptness and the density abruptness threshold.

[0146] Specifically, the method for determining the local density abruptness and density abruptness threshold involves comparing the local density abruptness between any two adjacent verification regions across multiple verification regions with the density abruptness threshold. If the abruptness exceeds the density abruptness threshold, then the density between those adjacent verification regions is determined to be abnormal. If any verification region is determined to have a density abruptness across multiple verification regions, then the entire target 3D Gaussian model is considered to have an anomaly.

[0147] Optionally, the method for determining the density mutation threshold can be set by statistical analysis of multiple normal three-dimensional Gaussian models as Yang, or adjusted according to the security requirements of the actual application scenario.

[0148] In practical applications, density distribution data of multiple normal 3D Gaussian models can be collected first. The relative rate of change or density difference between their adjacent verification regions can be calculated, and a suitable value can be determined as the density mutation threshold. Furthermore, the density mutation threshold can be dynamically adjusted according to the application scenario and security requirements of the 3D Gaussian model. For example, in scenarios with high security requirements, the density mutation threshold can be set lower to improve the sensitivity of anomaly detection.

[0149] Alternatively, various methods can be used to determine the density mutation threshold. For example, the threshold range can be determined by statistical methods, the threshold can be automatically optimized based on historical data by machine learning methods, or the threshold can be calibrated by combining the experience of domain experts.

[0150] In the embodiments of this specification, by comparing the local density mutation degree with the density mutation threshold, objective quantitative identification of local density changes in a 3D Gaussian model is achieved, improving the accuracy and reliability of anomaly identification. Specifically, the density mutation threshold is determined through statistical analysis based on multiple normal models, ensuring the scientific rigor and objectivity of the judgment criteria. By comparing the local density mutation degree with the density mutation threshold, discontinuities in model density caused by human editing or data loss can be captured, avoiding the limitations of traditional global analysis methods. This enables precise identification of model anomalies at the local scale, improving sensitivity to subtle anomalies and thus enhancing the accuracy and reliability of anomaly identification for the target 3D Gaussian model.

[0151] In one optional embodiment of this specification, the visual parameters of three-dimensional Gaussian points within multiple verification regions are statistically analyzed to obtain the visual feature distribution of multiple verification regions, including:

[0152] For the current verification area, determine the parameter correlation coefficient based on the visual parameters of any two 3D Gaussian points within the current verification area;

[0153] Based on the parameter correlation coefficient, the local attribute correlation of the current verification region is determined.

[0154] Visual parameters are a set of attribute parameters related to the visual representation of a 3D Gaussian point. These can include coordinate parameters, size parameters, color attributes, transparency parameters, etc., used to describe the characteristics of the 3D Gaussian point in terms of spatial location, shape, size, color appearance, etc. Specifically, visual parameters may also include other attribute parameters, such as the normal direction of the 3D Gaussian point.

[0155] Visual parameters provide multidimensional data on the visual characteristics of 3D Gaussian points, laying the data foundation for subsequent distribution statistics. This allows for the analysis of point distribution patterns from multiple perspectives, leading to a more comprehensive identification of anomalies. For example, coordinate parameters accurately describe the position of a 3D Gaussian point in 3D space, color attributes describe its color value, and transparency parameters describe its opacity. These parameters collectively constitute the attribute information of a 3D Gaussian point.

[0156] The parameter correlation coefficient is a quantitative indicator that measures the correlation between visual parameters of two three-dimensional Gaussian points. It can be calculated using statistical methods and is used to characterize the strength of the association between two visual parameters. Specifically, the parameter correlation coefficient can use Pearson correlation coefficient or Spearman rank correlation coefficient to quantify the degree of association between parameters, and can transform the degree of association into a quantifiable statistical feature, providing a basis for calculating the correlation of local attributes.

[0157] Local attribute correlation is an aggregated index of parameter correlation coefficients calculated based on the visual parameters of any two 3D Gaussian points within the current validation region. It is used to characterize the strength of the correlation between visual parameters within the validation region. Specifically, local attribute correlation can transform the parameter correlation within a region into comparable statistical features, providing statistical results of local regional attribute consistency and offering crucial evidence for identifying model anomalies. For example, local attribute correlation can be expressed as the average of all parameter correlation coefficients within the region, or it can be achieved by weighted averaging to consider the importance of different parameters, thus forming a comprehensive local attribute correlation index.

[0158] In practical applications, for the current validation region, the correlation coefficient of parameters can be determined by various statistical methods, such as Pearson correlation coefficient and Spearman rank correlation coefficient.

[0159] Specifically, for any two 3D Gaussian points within the current verification area, the correlation between parameters of the same type in the visual parameters of the two 3D Gaussian points can be calculated, such as calculating the Pearson correlation coefficient between color parameters or the Spearman rank correlation coefficient between transparency parameters.

[0160] For example, the parameter correlation coefficient τ can be calculated as follows:

[0161] τ s,r =corr{{s i},{r i}}

[0162] Where s and r represent any two three-dimensional Gaussian points within the current verification region, {s i} represents the set of visual parameters for a 3D Gaussian point s, {r i} represents the set of visual parameters for a 3D Gaussian point r, and corr{} is the correlation metric.

[0163] Once the parameter correlation coefficient is determined, the local attribute correlation of the current verification region can be determined based on the parameter correlation coefficient.

[0164] Specifically, determining local attribute correlation based on parameter correlation coefficients can be achieved by averaging or weighted averaging all parameter correlation coefficients within the validation region. In practical applications, the correlation coefficients between all parameters between any two 3D Gaussian points within the current validation region can be calculated, and the average value can be used as an indicator of local attribute correlation. Alternatively, a weighted average can be used, assigning different weights to the correlations of different types of visual parameters to reflect the importance of different parameters in the model. Furthermore, principal component analysis and other methods can be used to extract the main correlation patterns within the current validation region to form local attribute correlation.

[0165] In the embodiments of this specification, by determining the parameter correlation coefficient based on the visual parameters of any two 3D Gaussian points within the current verification area, and by determining the local attribute correlation based on the parameter correlation coefficient, an objective quantitative analysis of the local attribute correlation of the 3D Gaussian model is achieved, improving the accuracy and reliability of anomaly identification. By determining the parameter correlation coefficient based on visual parameters, the attribute characteristic relationship of the 3D Gaussian points can be accurately reflected, providing a data foundation for the calculation of local attribute correlation. By aggregating the parameter correlation coefficient to determine the local attribute correlation, the phenomenon of model attribute correlation breakage caused by human editing or data loss can be effectively captured, enabling accurate identification of model anomalies at the local scale, improving the sensitivity to subtle anomalies, and thus improving the accuracy and reliability of anomaly identification for the target 3D Gaussian model.

[0166] In one optional embodiment of this specification, the preset visual feature distribution threshold includes an attribute correlation threshold;

[0167] Based on the visual feature distribution of multiple verification regions and a preset visual feature distribution threshold, determine whether the target 3D Gaussian model has anomalies, including:

[0168] The local attribute correlation is compared with the attribute correlation threshold. If the local attribute correlation in any verification region is lower than the attribute correlation threshold, then the target 3D Gaussian model is judged to be abnormal.

[0169] The attribute correlation threshold is a pre-defined set of critical criteria used to determine whether the attribute correlation of a 3D Gaussian model, i.e., whether the local attribute correlation is abnormal, defines the statistical boundary conditions that a normal 3D Gaussian model should satisfy in terms of attribute correlation. Specifically, the attribute correlation threshold can provide an objective quantitative standard for anomaly identification based on local attribute correlation, making the anomaly identification process automated and repeatable, and avoiding bias caused by subjective experience.

[0170] Specifically, comparing the attribute correlation threshold with the local attribute correlation is a key basis for determining whether there are anomalies in the target 3D Gaussian model. If the local attribute correlation in any verification region is lower than the threshold, it indicates that there may be a break in the correlation of attribute parameters in that threshold region due to human editing.

[0171] In practical applications, once the local attribute correlation of multiple verification regions is determined, a judgment can be made based on the local attribute correlation and the attribute correlation threshold.

[0172] Specifically, the method for determining the correlation between local attributes and the attribute correlation threshold involves comparing the local attribute correlation of multiple validation regions with the attribute correlation threshold. For each validation region, if the local attribute correlation of that region is lower than the attribute correlation threshold, then that region is considered to have an attribute anomaly. If any validation region is determined to have an attribute anomaly across multiple validation regions, then the entire target 3D Gaussian model is considered to have an anomaly.

[0173] In practical applications, a single threshold judgment strategy can be adopted, or a multi-threshold fusion decision mechanism can be constructed. By comprehensively judging the abnormality of attribute correlation in different verification regions, the accuracy and reliability of anomaly identification can be improved. For example, the attribute correlation of multiple verification regions can be comprehensively evaluated by weighted voting or machine learning classifiers.

[0174] Optionally, the method for determining the attribute correlation threshold can be set by statistical analysis of multiple normal 3D Gaussian models as samples, or adjusted according to the security requirements of the actual application scenario.

[0175] In practical applications, attribute correlation data from multiple normal 3D Gaussian models can be collected first, their local attribute correlation indices can be calculated, and a suitable value can be determined as the attribute correlation threshold. Furthermore, statistical methods can be used to determine the threshold range, machine learning methods can be used to automatically optimize the threshold based on historical data, or threshold calibration can be performed in conjunction with domain expert experience to adapt to the security requirements of different scenarios. For example, in scenarios with high security requirements, the attribute correlation threshold can be set lower to improve the sensitivity of anomaly detection.

[0176] In the embodiments of this specification, by comparing the local attribute correlation with the attribute correlation threshold, objective quantitative identification of the local attribute correlation of a 3D Gaussian model is achieved, improving the accuracy and reliability of anomaly identification. Specifically, determining the attribute correlation threshold through statistical analysis based on multiple normal models ensures the scientific and objective nature of the judgment criteria, avoiding interference from subjective experience. Comparing the local attribute correlation with the attribute correlation threshold effectively captures the loss of visual parameters in a 3D Gaussian model due to human editing or data omissions, achieving precise identification of model anomalies at the local scale and improving sensitivity to hidden anomalies. This avoids the limitations of traditional global analysis methods, providing an objective basis for the quality assessment of 3D Gaussian models, making the anomaly identification process more automated and refined, and achieving precise identification of model anomalies at the local scale, thereby improving the accuracy and reliability of anomaly identification for target 3D Gaussian models.

[0177] Corresponding to the above method embodiments, this specification also provides embodiments of a three-dimensional Gaussian model anomaly recognition device, see [link to documentation]. Figure 2 , Figure 2 A schematic diagram of a three-dimensional Gaussian model anomaly recognition device according to one embodiment of this specification is shown. Figure 2 As shown, the three-dimensional Gaussian model anomaly identification device includes:

[0178] The acquisition module 202 is configured to acquire a target 3D Gaussian model, wherein the target 3D Gaussian model includes multiple 3D Gaussian points and multiple verification regions, and the verification regions are divided based on the spatial range of the target 3D Gaussian model.

[0179] The statistics module 204 is configured to perform distribution statistics on the visual parameters of three-dimensional Gaussian points in multiple verification regions to obtain the visual feature distribution of multiple verification regions.

[0180] The judgment module 206 is configured to determine whether there is an anomaly in the target 3D Gaussian model based on the visual feature distribution of multiple verification regions and a preset visual feature distribution threshold.

[0181] The acquisition module is the data input unit in the 3D Gaussian model anomaly recognition device, responsible for acquiring and pre-processing the target 3D Gaussian model. It can acquire, parse, and pre-process the target 3D Gaussian model, providing a structured data foundation for subsequent anomaly recognition. For example, the acquisition module can load 3D Gaussian model files from local storage devices, download the 3D Gaussian model data to be detected from a server via a network interface, or obtain the target 3D Gaussian model through real-time acquisition by a 3D scanning device or by reconstructing input multi-view image sequences using algorithms such as 3DGS.

[0182] The statistics module, within the 3D Gaussian model anomaly detection device, is a data processing unit responsible for statistically analyzing the distribution of visual parameters within the verification area. It can perform statistical analysis on the visual parameters of 3D Gaussian points across multiple verification areas, extracting visual feature distributions and providing a data foundation for anomaly detection. For example, the statistics module can calculate the mean vector and covariance matrix of the coordinates of all 3D Gaussian points within the current verification area, thereby determining the local coordinate uniformity. It can also calculate the nearest neighbor distance for each 3D Gaussian point using the K-nearest neighbor algorithm, and then calculate the reciprocal of the average nearest neighbor distance for points within that area as the distribution density parameter.

[0183] The judgment module is the decision-making unit responsible for anomaly detection in the 3D Gaussian model anomaly identification device. It can systematically compare the visual feature distribution of multiple verification regions with preset visual feature distribution thresholds to determine whether the target 3D Gaussian model has anomalies. For example, the judgment module can compare the local coordinate uniformity of each verification region with a preset coordinate anomaly threshold. If the local coordinate uniformity of any verification region exceeds the preset threshold, the model is determined to be anomaly. It can also calculate the local density abruptness between adjacent verification regions and compare it with a preset density abruptness threshold to determine whether the model has anomalies.

[0184] The three-dimensional Gaussian model anomaly recognition device provided in this specification achieves systematization and automation of three-dimensional Gaussian model anomaly recognition through the collaboration of the acquisition module, statistics module, and judgment module, thereby improving recognition accuracy and efficiency. Specifically, the acquisition module ensures accurate acquisition of the target three-dimensional Gaussian model, providing a data foundation for local scale analysis and avoiding the bias and delays of manual data processing; the statistics module, through distribution statistics, transforms high-dimensional visual parameters into quantified feature distributions, capturing subtle anomaly patterns in local areas and enhancing the precision and comprehensiveness of feature extraction; the judgment module uses preset thresholds for objective comparison, eliminating the interference of subjective judgment and ensuring the consistency and repeatability of anomaly recognition results; and by recognizing anomalies in local verification areas, the sensitivity to hidden anomalies is improved, thereby enhancing the accuracy and reliability of anomaly recognition for the target three-dimensional Gaussian model.

[0185] Optionally, there is a preset overlap ratio among multiple verification regions; the statistics module 204 is further configured to: perform distribution statistics on the visual parameters of the three-dimensional Gaussian points in the current verification region to obtain the visual feature distribution of the current verification region; based on the preset overlap ratio, update the verification region that overlaps with the current verification region as the current verification region; return to execute the step of performing distribution statistics on the visual parameters of the three-dimensional Gaussian points in the current verification region to obtain the visual feature distribution of the current verification region, until the preset stopping condition is reached, and obtain the visual feature distribution of multiple verification regions.

[0186] Optionally, the visual parameters of the three-dimensional Gaussian points include coordinate parameters; the statistics module 204 is further configured to: for the current verification area, determine the mean and variance of the coordinate distribution of the three-dimensional Gaussian points based on the coordinate parameters of the three-dimensional Gaussian points in the current verification area; and determine the local coordinate uniformity of the current verification area based on the mean and variance of the coordinate distribution.

[0187] Optionally, the preset visual feature distribution threshold includes a coordinate anomaly threshold; the judgment module 206 is further configured to: judge the local coordinate uniformity and the coordinate anomaly threshold, and if the local coordinate uniformity of any verification area exceeds the coordinate anomaly threshold, then the target 3D Gaussian model is judged to be abnormal.

[0188] Optionally, the visual parameters of the three-dimensional Gaussian points include coordinate parameters and size parameters; the statistics module 204 is further configured to: determine the distribution density parameters of the current verification region based on the coordinate parameters and size parameters of the three-dimensional Gaussian points in the current verification region; and determine the local density abruptness between multiple verification regions based on the distribution density parameters corresponding to multiple verification regions.

[0189] Optionally, the preset visual feature distribution threshold includes a density mutation threshold; the judgment module 206 is further configured to: judge the local density mutation degree and the density mutation threshold, and if the local density mutation degree corresponding to any verification area exceeds the density mutation threshold, then the target three-dimensional Gaussian model is judged to be abnormal.

[0190] Optionally, the statistics module 204 is further configured to: determine the parameter correlation coefficient based on the visual parameters of any two three-dimensional Gaussian points within the current verification region; and determine the local attribute correlation of the current verification region based on the parameter correlation coefficient.

[0191] Optionally, the preset visual feature distribution threshold includes the attribute correlation threshold; the judgment module 206 is further configured to: judge the local attribute correlation and the attribute correlation threshold, and if the local attribute correlation in any verification area is lower than the attribute correlation threshold, then the target three-dimensional Gaussian model is judged to be abnormal.

[0192] The above is a schematic diagram of a three-dimensional Gaussian model anomaly identification device according to this embodiment. It should be noted that the technical solution of this three-dimensional Gaussian model anomaly identification device and the technical solution of the aforementioned three-dimensional Gaussian model anomaly identification method belong to the same concept. Details not described in detail in the technical solution of the three-dimensional Gaussian model anomaly identification device can be found in the description of the technical solution of the aforementioned three-dimensional Gaussian model anomaly identification method.

[0193] Figure 3 A structural block diagram of a computing device 300 according to one embodiment of this specification is shown. The components of the computing device 300 include, but are not limited to, a memory 310 and a processor 320. The processor 320 is connected to the memory 310 via a bus 330, and a database 350 is used to store data.

[0194] The computing device 300 also includes an access device 340, which enables the computing device 300 to communicate via one or more networks 360. Examples of these networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 340 may include one or more of any type of wired or wireless network interface (e.g., a network interface controller (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0195] In one embodiment of this specification, the aforementioned components of the computing device 300 and Figure 3 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 3 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0196] The computing device 300 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 300 can also be a mobile or stationary server.

[0197] The processor 320 is used to execute the following computer program / instructions, which, when executed by the processor, implement the steps of the above-described three-dimensional Gaussian model anomaly recognition method.

[0198] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the above-described three-dimensional Gaussian model anomaly recognition method belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the above-described three-dimensional Gaussian model anomaly recognition method.

[0199] An embodiment of this specification also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the above-described three-dimensional Gaussian model anomaly recognition method.

[0200] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the technical solution of the above-described three-dimensional Gaussian model anomaly recognition method. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the above-described three-dimensional Gaussian model anomaly recognition method.

[0201] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described three-dimensional Gaussian model anomaly recognition method.

[0202] The above is an illustrative scheme of a computer program according to this embodiment. It should be noted that the technical solution of this computer program belongs to the same concept as the technical solution of the above-described three-dimensional Gaussian model anomaly recognition method. For details not described in detail in the technical solution of the computer program, please refer to the description of the technical solution of the above-described three-dimensional Gaussian model anomaly recognition method.

[0203] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0204] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0205] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0206] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0207] The preferred embodiments disclosed above are merely illustrative of this specification. Optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A three-dimensional Gaussian model anomaly identification method, characterized in that, The method comprises: acquiring a target three-dimensional Gaussian model, wherein the target three-dimensional Gaussian model comprises a plurality of three-dimensional Gaussian points, and the target three-dimensional Gaussian model comprises a plurality of verification regions, and the verification regions are divided based on a preset spatial range; performing distribution statistics on visual parameters of the three-dimensional Gaussian points in the plurality of verification regions to obtain visual feature distributions of the plurality of verification regions; judging whether the target three-dimensional Gaussian model is abnormal based on the visual feature distributions of the plurality of verification regions and a preset visual feature distribution threshold.

2. The method of claim 1, wherein, The plurality of verification regions have a preset overlap ratio. The performing of the distribution statistics on the visual parameters of the three-dimensional Gaussian points in the plurality of verification regions to obtain the visual feature distributions of the plurality of verification regions comprises: for a current verification region, performing distribution statistics on visual parameters of three-dimensional Gaussian points in the current verification region to obtain a visual feature distribution of the current verification region; based on the preset overlap ratio, updating a verification region overlapping with the current verification region as the current verification region; returning to the step of performing the distribution statistics on the visual parameters of the three-dimensional Gaussian points in the current verification region to obtain the visual feature distribution of the current verification region until a preset stop condition is reached, and obtaining the visual feature distributions of the plurality of verification regions.

3. The method according to claim 1 or 2, characterized in that, The visual parameters of the three-dimensional Gaussian points comprise coordinate parameters, and the visual feature distributions comprise local coordinate system degrees. The performing of the distribution statistics on the visual parameters of the three-dimensional Gaussian points in the plurality of verification regions to obtain the visual feature distributions of the plurality of verification regions comprises: for a current verification region, determining a coordinate distribution mean and a coordinate distribution variance of three-dimensional Gaussian points in the current verification region based on coordinate parameters of the three-dimensional Gaussian points in the current verification region; determining a local coordinate system degree of the current verification region based on the coordinate distribution mean and the coordinate distribution variance.

4. The method of claim 3, wherein, The preset visual feature distribution threshold comprises a coordinate abnormality threshold. The judging of whether the target three-dimensional Gaussian model is abnormal based on the visual feature distributions of the plurality of verification regions and the preset visual feature distribution threshold comprises: judging whether the local coordinate system degree and the coordinate abnormality threshold, and if the local coordinate system degree of any verification region exceeds the coordinate abnormality threshold, judging that the target three-dimensional Gaussian model is abnormal.

5. The method according to claim 1 or 2, characterized in that, The visual parameters of the three-dimensional Gaussian points comprise coordinate parameters and size parameters, and the visual feature distributions comprise local density mutation degrees. The performing of the distribution statistics on the visual parameters of the three-dimensional Gaussian points in the plurality of verification regions to obtain the visual feature distributions of the plurality of verification regions comprises: for a current verification region, determining a distribution density parameter of the current verification region based on coordinate parameters and size parameters of three-dimensional Gaussian points in the current verification region; determining local density mutation degrees between the plurality of verification regions based on the distribution density parameters corresponding to the plurality of verification regions.

6. The method of claim 5, wherein, The preset visual feature distribution threshold comprises a density mutation threshold. The method comprises the following steps: The method comprises the following steps:

7. The method according to claim 1 or 2, characterized in that, The visual feature distribution comprises a local attribute correlation degree. The method comprises the following steps: For the current verification area, a parameter correlation degree coefficient is determined based on the visual parameters of any two three-dimensional Gaussian points in the current verification area. The local attribute correlation degree of the current verification area is determined based on the parameter correlation degree coefficient.

8. The method of claim 7, wherein, The preset visual feature distribution threshold comprises an attribute correlation degree threshold. The method comprises the following steps: The method comprises the following steps:

9. A three-dimensional Gaussian model anomaly identification apparatus characterized by comprising: The method comprises the following steps: A memory and a processor are comprised. The memory is used for storing computer programs / instructions, and the processor is used for executing the computer programs / instructions. The memory stores computer programs / instructions, which are executed by the processor to realize the steps of the three-dimensional Gaussian model anomaly identification method according to any one of claims 1 to 8.

10. A computing device, comprising: The computer programs / instructions are executed by the processor to realize the steps of the three-dimensional Gaussian model anomaly identification method according to any one of claims 1 to 8. ​ ​ 11. A computer readable storage medium, characterized in that, ​ 12. A computer program product, characterised in that, ​

Images