Real scene three-dimensional model quality evaluation method

Abstract

The invention relates to the technical field of live-action three-dimensional modeling, in particular to a live-action three-dimensional model quality evaluation method, which comprises the following steps of: firstly, constructing a multi-dimensional evaluation index set covering dimensions such as geometric accuracy and texture quality; an initial weight is obtained through coupling of an analytic hierarchy process and an entropy weight method, and a dynamic weight is formed by combining scene demand coefficient adjustment; then, a subjective-objective linkage relation is established through grey correlation analysis, and index scores and weights are corrected through feedback coefficients; then calculating the total quality score of a single time node; the evaluation effectiveness in the dynamic scene is verified through a time sequence consistency index; and finally outputting the comprehensive quality grade. According to the method, a multi-dimensional index set is constructed to cover full dimensions, dynamic weights are calculated in combination with scene demand coefficients, and the problems of single dimension and poor scene adaptation are solved; building subjective-objective linkage by means of grey relational degree, correcting scores and weights, and solving disjunction; and calculating a time sequence consistency index to verify dynamic coherence, solving static limitation, and adapting to full-life-cycle management and control of the model.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of real scene three-dimensional modeling technology, in particular to a real scene three-dimensional model quality evaluation method. BACKGROUND

[0002] Real scene three-dimensional models have become the core infrastructure in the digital economy era by converting information such as spatial form, texture characteristics, and topological relationships of the physical world into digital models through technologies such as laser radar and oblique photography. The quality evaluation of real scene three-dimensional models, as a key link from production to application, directly determines the application value of the models.

[0003] However, the existing three-dimensional model quality evaluation method has the following defects in actual use:

[0004] Single evaluation dimension, poor scene adaptability: multi-focal geometric accuracy or texture quality, ignoring key dimensions such as topological consistency and semantic integrity, cannot adapt to complex scenes;

[0005] Subjective and objective disconnection, insufficient evaluation accuracy: objective evaluation only relies on data collected by equipment, without considering actual user needs; subjective evaluation relies on expert experience and is easily influenced by personal preferences, without a linkage feedback mechanism;

[0006] Static evaluation limitations, dynamic scene failure: real scene three-dimensional models often need to be updated, and existing methods only evaluate models at a single time node, cannot verify the quality consistency of models at different time sequences, resulting in evaluation results being invalid in dynamic scenes. SUMMARY

[0007] The present application aims to provide a real scene three-dimensional model quality evaluation method to solve the problems raised in the background.

[0008] To achieve the above purpose, the present application provides the following technical solution: a real scene three-dimensional model quality evaluation method, comprising:

[0009] S1, constructing a multi-dimensional evaluation index set: determining a primary index covering geometric accuracy, texture quality, topological consistency, semantic integrity, and model lightweight degree, dividing at least two secondary indexes under each primary index, forming an index system;

[0010] S2, dynamic weight calculation: using the analytic hierarchy process to determine the subjective weight, using the entropy weight method to determine the objective weight, coupling to obtain the initial weight; adjusting the initial weight to obtain the dynamic weight by combining the scene demand coefficient, correcting the initial weight to obtain the dynamic weight, the dynamic weight satisfies: the higher the priority of the index, the larger the weight proportion;

[0011] S3, subjective-objective linkage feedback: collect objective data of each secondary index, organize experts to conduct subjective scoring, establish the correlation between subjective evaluation and objective index through gray correlation analysis, calculate feedback coefficient, and correct objective index score and dynamic weight;

[0012] S4, single time node quality score: based on the corrected objective index score and dynamic weight, the weighted sum method is used to calculate the model quality total score of single time node;

[0013] S5, time sequence consistency evaluation: select the quality total score and each primary index score of at least two time nodes, calculate the time sequence similarity and overall time sequence consistency index of each index; if the consistency index is higher than the preset threshold, it is determined that the evaluation under dynamic scene is effective;

[0014] S6, output comprehensive evaluation result: combine single time node quality total score and time sequence consistency index to output the quality grade of real scene three-dimensional model.

[0015] Preferably, in step S1:

[0016] The secondary index of geometric precision includes plane position mean error and height mean error;

[0017] The secondary index of texture quality includes texture resolution and texture fitting degree;

[0018] The secondary index of topological consistency includes adjacent face connectivity and hole area ratio;

[0019] The secondary index of semantic integrity includes semantic annotation accuracy and key feature missing rate;

[0020] The secondary index of model lightweight degree includes triangular face compression rate and model file volume compression ratio.

[0021] Preferably, in step S2, the calculation formula of dynamic weight is:

[0022] ;

[0023] Wherein, is the dynamic weight of the jth secondary index under the ith primary index;

[0024] is the initial weight of the secondary index, which is obtained by coupling the subjective weight of analytic hierarchy process (AHP) and the objective weight of entropy weight method, and the coupling formula is , is the weight distribution coefficient, the value range is 0.4-0.6;

[0025] The scene demand coefficient is 0.1-0.4, the higher the scene complexity and the stronger the index priority correlation are, The higher the value is.

[0026] The priority coefficient of the secondary index in the target scene is 0-2, the higher the scene priority is, The higher the value is.

[0027] Preferably, the subjective weight Is determined by the analytic hierarchy process.

[0028] The objective weight Is determined by the entropy weight method.

[0029] Preferably, the calculation formula of the grey correlation degree in step S3 is:

[0030] ;

[0031] Wherein, The standardized value of the kth subjective evaluation index;

[0032] The standardized value of the ith secondary index objective data of K;

[0033] The resolution coefficient is 0.5;

[0034] The correlation degree of the ith secondary index and the subjective evaluation;

[0035] The feedback coefficient , the objective index correction score , the modified dynamic weight , wherein The objective index standardized score before correction, and if the modified score exceeds 1, it is 1.

[0036] Preferably, the subjective evaluation index includes visual satisfaction, semantic ease of use, and operation fluency score range 1-10 points, and is converted into a standardized value in the range of 0-1 after standardization.

[0037] The objective data is collected by laser radar and oblique photography equipment, and is converted into a standardized value in the range of 0-1 by min-max standardization.

[0038] Preferably, the objective index score in step S4 based on the correction And the dynamic weight The formula for calculating the total score of the quality of a single time node by using the weighted summation method is: ​

[0039] ;

[0040] wherein, is the number of primary indicators, is the number of secondary indicators corresponding to the primary indicator, is the total score of quality at time node t, and the quality level is divided according to the total score:

[0041] Optimal: ;

[0042] Good: ;

[0043] Medium: ;

[0044] Poor: .

[0045] Preferably, the time sequence similarity in step S5 is calculated by using the cosine of the included angle method, and the formula is:

[0046]

[0047] wherein, , are the scores of the i th primary indicator at time nodes t1 and t2, respectively;

[0048] The calculation formula of the time sequence consistency index is:

[0049]

[0050] wherein, is the corrected dynamic weight of the i th primary indicator.

[0051] Compared with the prior art, the present application has the following beneficial effects:

[0052] 1. The present application constructs a multi-dimensional evaluation index set covering geometric precision, texture quality and other dimensions, calculates the dynamic weight by coupling AHP and entropy weight method and combining with the scene demand coefficient, corrects the index score and weight by means of the gray correlation degree analysis to establish a subjective-objective linkage feedback mechanism, calculates the total score of quality at a single time node and the time sequence consistency index at multiple time nodes, and finally outputs the comprehensive quality level, so as to solve the problems of single dimension, disconnection between subjective and objective, and static evaluation limitation of the existing evaluation method, and is suitable for model full life cycle quality control in digital city and smart city. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 is the overall structure flow block diagram of the real scene three-dimensional model quality evaluation method of the present application. DETAILED DESCRIPTION

[0054] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the scope of protection of the present application.

[0055] With reference to Figure 1 The present application provides a technical solution: a real scene three-dimensional model quality evaluation method, comprising:

[0056] S1, constructing a multi-dimensional evaluation index set: determining a first index covering geometric accuracy, texture quality, topological consistency, semantic integrity and model lightweight degree, dividing 2 second indexes under each first index, forming a full-dimensional index system covering "spatial accuracy-structural integrity-semantic availability-application adaptability". The definition, data source and corresponding problems of each index are as follows:

[0057] Geometric accuracy: the second index is the plane position mean error (the plane deviation mean value of the model point cloud and the GNSS true value point, unit: mm), the height mean error (the height deviation mean value, unit: mm), and the data is collected by laser radar (such as RIEGL VZ-6000) and GNSS receiver. This index ensures the spatial positioning accuracy of the model, solves the problem that the existing method only focuses on geometric accuracy but does not refine the error type;

[0058] Texture quality: the second index is texture resolution (texture image pixel density, unit: dpi), texture fitting degree (texture and model surface deviation value, unit: mm), and the data is obtained by detecting the parameters of the oblique photography camera and the MeshLab software. This index supplements the "texture practicability" (such as poor texture fitting degree will cause visual distortion) ignored by the existing method, and is suitable for ancient building protection, digital twin exhibition hall and other scenes with high requirements for texture;

[0059] Topological consistency: the second index is the adjacent face connectivity rate (the number of adjacent shared edge face pieces accounts for the total number of face pieces, unit: %), the hole area ratio (the ratio of model surface hole area to total area, unit: %), and the data is counted by FME topological detection tool. This index solves the problem that the existing method ignores the model structure integrity, for example, low connectivity rate will cause "chain break" in model path planning, and holes will affect the collision detection function of the model;

[0060] Semantic completeness: The secondary indicators are semantic annotation accuracy (the proportion of correctly annotated features out of the total number of annotated features, in %) and key feature miss rate (the proportion of missed key features such as buildings and roads out of the actual total number, in %). Data is obtained through a combination of LabelMe semantic annotation tool and manual verification. This indicator fills the gap in existing methods that "emphasize data over semantics," ensuring that the model can meet the semantic query needs of scenarios such as planning and emergency response.

[0061] Model lightweighting degree: The secondary indicators are triangle patch compression ratio (the ratio of the number of triangle patches after lightweighting to the original number, in %) and model file volume compression ratio (the ratio of the file volume after lightweighting to the original file volume, in %). The data are obtained through MeshLab lightweighting tool and file attribute verification. This indicator addresses the problem of existing methods ignoring "application adaptability".

[0062] S2. Dynamic Weight Calculation: Dynamic weights are calculated in two steps: "initial weight coupling + scenario requirement adjustment," ensuring that "the higher the scenario priority of the indicator, the greater its weight percentage." The specific process is as follows:

[0063] Initial weight coupling: Subjective weights are determined using the Analytic Hierarchy Process (AHP). A judgment matrix is ​​constructed by inviting 5-8 cross-disciplinary experts (including those from surveying engineering, urban planning, and computer graphics). After passing a consistency test (CR < 0.1), the subjective dynamic weights of each secondary indicator are obtained. The entropy weight method is used to determine objective weights. 30-50 sets of historical model data from similar scenarios (e.g., 30 sets of model indicators for a CBD scenario) are collected. The information entropy and difference coefficient of each indicator are calculated and transformed into objective dynamic weights. The formula is obtained by coupling subjective weights with objective weights of the entropy weight method. To obtain the initial weights ,in The weighting coefficient is set to a value between 0.4 and 0.6. This step avoids the bias of existing methods that rely on a single weight (either subjective or objective) and improves the comprehensiveness of the weights.

[0064] Scenario requirement adjustment: Introduce scenario requirement coefficient and priority coefficient in The value ranges from 0.1 to 0.4. The higher the scenario complexity and the stronger the correlation between the priority of the indicators, the better. The larger the value; The value ranges from 0 to 2, with higher priority metrics for specific scenarios. The larger the value, the better, such as the "texture detail" value in ancient building preservation scenarios. In typical urban scenarios, this indicator ;

[0065] The dynamic weight is calculated by the formula The step solves the problem of "fixed weight and poor scene adaptability" of the existing method, so that the evaluation system can flexibly adapt to different application scenarios.

[0066] The weight distribution coefficient , the scene demand coefficient , the priority coefficient , and the parameter values in different scenes are shown in the following table: The parameter values of , in different scenes

[0067] S3, subjective-objective feedback: through the three steps of "data collection-gray correlation analysis-feedback correction", the linkage between subjective evaluation and objective indicators is established to ensure that the evaluation results take into account "data objectivity" and "application practicality". Specifically,

[0068] Data collection and standardization:

[0069] According to the data source requirements of step S1, collect the original data of each secondary indicator. Objective data is collected by laser radar and oblique photography equipment. The min-max standardization formula is used to convert the score to the 0-1 interval ( is the minimum value of the indicator, is the maximum value of the indicator);

[0070] Subjective evaluation index collection: according to the ratio of "experts + actual users = 1:1", organize scoring personnel (experts need more than 5 years of experience in three-dimensional model field, and users are model target scene users such as planners and emergency command personnel). Anonymously score three indicators "visual satisfaction" (evaluate texture clarity and color authenticity), "semantic ease of use" (evaluate semantic query convenience and labeling accuracy), and "operation fluency" (evaluate model loading speed and interaction delay) (1-10 points). After removing the highest score and the lowest score, take the average value, and then standardize it to the 0-1 interval ;

[0071] Gray correlation analysis: take the average value of subjective evaluation as the reference sequence, and the standardized score of each objective indicator as the comparison sequence. The correlation degree is calculated by the gray correlation degree formula

[0072]

[0073] where is the resolution coefficient,​ The larger, the stronger the correlation between the objective indicator and the subjective demand (e.g., the model with a high semantic ease-of-use score, the semantic annotation accuracy rate is generally larger);

[0074] Feedback correction: calculate the feedback coefficient (ensure in the interval of 0.25-0.75, avoid over-correction), correct the objective indicator score and the dynamic weight;

[0075] Corrected objective score: , if it exceeds 1, take 1;

[0076] Corrected dynamic weight: ;

[0077] Feedback coefficient , objective indicator correction score , corrected dynamic weight , wherein is the normalized score of the objective indicator before correction, and the corrected score is taken as 1 if it exceeds 1;

[0078] Through correlation correction, the subjective demand for "clear texture, easy-to-use semantics" is converted into the improvement of objective score and weight, solving the problem of disconnection between subjective and objective in existing methods.

[0079] S4, single time node quality score: based on the corrected objective indicator score and dynamic weight, the model quality total score of single time node is calculated by weighted summation method, and the formula for calculating the quality total score of single time node is :

[0080] ;

[0081] wherein, is the number of primary indicators is the number of secondary indicators under the corresponding primary indicator, N=5 in this embodiment, , is the quality total score of time node t (0~1 interval), and the quality grade is divided according to the total score:

[0082] Excellent: (the model meets the standard in each dimension and can be directly used for core scenarios);

[0083] Good: (local dimension needs to be fine-tuned, such as slightly low lightweight degree, which can be used for non-core scenarios);

[0084] Medium: (multiple dimensions need to be optimized, such as high semantic omission rate, which needs to be reworked before use);

[0085] Poor: (Core dimension is not up to standard, need to be re-modeled);

[0086] This step makes the model quality comparable and traceable by quantifying the score, solving the problem of fuzzy evaluation results of existing methods.

[0087] S5, Time sequence consistency evaluation: For the model dynamic update scenario, select 2 or more key time nodes (such as t1 before update, t2 after update), and verify the quality continuity through "time sequence similarity calculation-consistency index determination", specifically:

[0088] Time sequence similarity calculation: The cosine method is used to calculate the similarity of each primary index at different time nodes , the formula is:

[0089]

[0090] Among them , are the corrected scores of the ith primary index at t1 and t2 nodes. The closer to 1, the better the quality stability of the index at the two nodes (such as , indicating that the index score is almost unchanged);

[0091] Consistency index calculation: Combined with the dynamic weight of each primary index , the time sequence consistency index C is calculated by the formula ;

[0092] Effectiveness determination: The preset consistency threshold is 0.85 - if , it means that the model quality before and after the update is continuous (such as the topology structure and semantic annotation do not deviate greatly), and the evaluation is effective in the dynamic scenario; if , the model update process needs to be traced back (such as checking whether the lightweight algorithm causes topology errors or whether the semantic annotation is missing), and the reason for the score fluctuation is investigated;

[0093] This step fills the gap of dynamic evaluation of existing methods, ensuring that the model quality is controllable throughout its life cycle.

[0094] S6, Output comprehensive evaluation results: Combine "single time node quality level" and "time sequence consistency index" to output an evaluation report containing the following content:

[0095] Index details: original data, standardized score, and corrected score of each secondary index;

[0096] Weight details: calculation process and numerical value of initial weight, dynamic weight, and corrected weight;

[0097] Subjective-objective linkage result: correlation degree of each index , feedback coefficient ;

[0098] Quality level: total score and level (excellent / good / medium / poor) of each time node;

[0099] Timing consistency: consistency index C and effectiveness determination result;

[0100] Optimization suggestion: improvement scheme for short board dimension.

[0101] It should be noted that, in this article, relationship terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the term "includes", "contains" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or device.

[0102] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the quality of a real-world 3D model, characterized in that: include: S1. Construct a multi-dimensional evaluation index set: Determine the primary indexes covering geometric accuracy, texture quality, topological consistency, semantic integrity, and model lightweightness. Each primary index is further divided into at least two secondary indexes to form an index system. S2. Dynamic weight calculation: Subjective weights are determined by the analytic hierarchy process and objective weights are determined by the entropy weight method, and the initial weights are obtained by coupling them together. The dynamic weight is obtained by adjusting the scenario demand coefficient, and the initial weight is corrected to obtain the dynamic weight. The dynamic weight satisfies the following: the higher the scenario priority of the indicator, the greater the weight ratio. S3. Subjective-Objective Linkage Feedback: Collect objective data for each secondary indicator, organize experts to conduct subjective scoring, establish the correlation between subjective evaluation and objective indicators through grey relational analysis, calculate the feedback coefficient, and correct the objective indicator scores and dynamic weights. S4. Single Time Node Quality Score: Based on the corrected objective indicator scores and dynamic weights, the weighted summation method is used to calculate the total model quality score for a single time node. S5. Temporal Consistency Evaluation: Select the total quality score and the score of each primary indicator at at least two time points, and calculate the temporal similarity of each indicator and the overall temporal consistency index. If the consistency index is higher than the preset threshold, the evaluation in the dynamic scenario is deemed valid. S6. Output comprehensive evaluation results: Combine the total quality score of a single time node with the temporal consistency index to output the quality level of the real scene 3D model.

2. The method for evaluating the quality of a real-scene 3D model according to claim 1, characterized in that: In step S1: The secondary indicators of geometric accuracy include the mean square error of planar position, which includes the mean square error of elevation. The secondary indicators of texture quality include texture resolution and texture fit. The secondary indicators of topological consistency include the connectivity rate of adjacent facets, including the proportion of hole area. The secondary indicators of semantic integrity include semantic annotation accuracy, including the missed detection rate of key features; The secondary indicators of the model's lightweightness include the triangular facet compression ratio, which includes the model file size compression ratio.

3. The method for evaluating the quality of a real-scene 3D model according to claim 1, characterized in that: The formula for calculating the dynamic weight in step S2 is as follows: ； in, Let be the dynamic weight of the j-th secondary indicator under the i-th primary indicator; The initial weights for this secondary indicator are obtained by coupling the subjective weights of the Analytic Hierarchy Process (AHP) with the objective weights of the entropy weight method. The coupling formula is as follows: , The weighting coefficients range from 0.4 to 0.

6. This is the scenario requirement coefficient, ranging from 0.1 to 0.

4. The higher the scenario complexity, the stronger the correlation between the priority of the indicator. The larger the value; This is the priority coefficient of the secondary indicator in the target scenario, with a value ranging from 0 to 2. The higher the scenario priority, the higher the priority of the indicator. The larger the value, the better.

4. The method for evaluating the quality of a real-scene 3D model according to claim 3, characterized in that: Subjective weight Determined using the analytic hierarchy process; The objective weight Determined using the entropy weight method.

5. The method for evaluating the quality of a real-scene 3D model according to claim 1, characterized in that: The formula for calculating the grey relational degree in step S3 is as follows: ； in, Let be the standardized value of the k-th subjective evaluation index; Let be the standardized value of the objective data of the i-th secondary indicator of K; The resolution coefficient is set to 0.

5. Let represent the correlation between the i-th secondary indicator and the subjective evaluation; Feedback coefficient Objective indicator correction score Corrected dynamic weights ,in The standard score is the objective indicator before correction. If the score after correction is greater than 1, it is taken as 1.

6. The method for evaluating the quality of a real-scene 3D model according to claim 1, characterized in that: The subjective evaluation indicators include visual satisfaction, semantic usability, and operational fluency, with a score range of 1 to 10, which are then standardized and converted into standardized values ​​in the range of 0 to 1. The objective data was collected by lidar and oblique photography equipment, and converted into standardized values ​​in the range of 0 to 1 through min-max standardization.

7. The method for evaluating the quality of a real-scene 3D model according to claim 1, characterized in that: In step S4, the scoring is based on the corrected objective indicators. With dynamic weights The weighted summation method is used to calculate the total quality score at a single time point. The formula is: ； in, The number of primary indicators. This refers to the number of secondary indicators under the corresponding primary indicator. The total quality score at time point t is used to classify quality levels. excellent: ; good: ; middle: ; Difference: .

8. The method for evaluating the quality of a real-scene 3D model according to claim 1, characterized in that: In step S5, the temporal similarity is calculated using the cosine similarity method, with the following formula: ； in, , These are the scores of the i-th primary indicator at time points t1 and t2, respectively. The formula for calculating the time series consistency index is: ； in, The dynamic weight is the adjusted weight of the i-th primary indicator.

Images