A Smart Evaluation Method and System for Community Fitness Spaces Based on Artificial Intelligence Algorithms

Abstract

This invention relates to the technical field of artificial intelligence and urban design, and in particular to an intelligent evaluation method and system for community fitness spaces based on artificial intelligence algorithms. The method includes the following steps: acquiring three-dimensional spatial data, constructing a multivariate dataset, and establishing a three-dimensional model of the fitness space; identifying and disassembling digital assets of the fitness space; restructuring the fitness space assets; and training a benchmark evaluation model for the fitness space; and constructing an intelligent evaluation model for the fitness space. This invention proposes an intelligent evaluation method and system for community fitness spaces using artificial intelligence algorithms in spatial quality assessment. It provides an innovative research framework for the scientific planning and optimized layout of community fitness space construction. This invention is applicable to the quality evaluation of fitness spaces during the design phase, allowing for pre-assessment of spatial quality. Based on the assessment results, targeted design measures can be implemented, reducing subsequent maintenance costs.

Description

Technical Field

[0001] This invention relates to the technical field of artificial intelligence and urban design, and in particular to an intelligent evaluation method and system for community fitness spaces based on artificial intelligence algorithms. Background Technology

[0002] In recent years, people have paid increasing attention to the construction of community fitness spaces, resulting in facilities layout, configuration, and environmental quality that meet people's requirements and are suitable for fitness use. However, current community fitness space construction is generally based on human experience in design and construction.

[0003] For example, Chinese patent CN118279731A discloses a method and system for identifying informal outdoor fitness spaces based on feature-based intelligent detection. The method includes: receiving urban street view images; inputting the urban street view images into a trained outdoor fitness space service capability discrimination model; outputting the spatial site features corresponding to the urban street view images, and the suitability of the spatial site features matching each type of outdoor fitness activity; matching the spatial site corresponding to the outdoor fitness activity based on the suitability; wherein the corresponding spatial site is the site information in the urban street view image corresponding to the spatial site features. This invention can analyze urban spatial sites using a deep learning model and quantify the configuration relationship between spatial sites and facilities.

[0004] Although the above methods can identify and explore spaces to assist in site planning, the layout of facilities in community fitness spaces is still designed based on human experience, and cannot take into account the different needs of different types of people for the layout of facilities in community fitness spaces. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method and system for intelligent evaluation of community fitness spaces based on artificial intelligence algorithms.

[0006] This invention provides an intelligent evaluation method for community fitness spaces based on artificial intelligence algorithms, employing the following technical solution:

[0007] A method for intelligent evaluation of community fitness spaces based on artificial intelligence algorithms includes the following steps:

[0008] Acquire three-dimensional spatial data, construct a multivariate dataset, and establish a three-dimensional model of the public fitness space;

[0009] Identification and dismantling of digital assets in public fitness spaces;

[0010] Restructuring of public fitness space assets and training of benchmark evaluation models for public fitness spaces;

[0011] Construct an intelligent evaluation model for public fitness spaces.

[0012] In a specific feasible implementation, the acquisition of 3D spatial data includes methods such as UAV laser scanning, IoT sensors, and user terminals.

[0013] Three-dimensional spatial data includes multi-view images, point clouds, mesh models, current spatial conditions, facility distribution, and landscape configuration.

[0014] In a specific feasible implementation, the identification and dismantling of digital assets in public fitness spaces includes the following steps:

[0015] By using preset prompts, perspective projection changes are used to map them onto a multi-view 2D image plane, generating 2D segmentation prompts that are compatible with the SAM model.

[0016] Based on the multi-view 2D segmentation results of SAM, coarse-grained global segmentation is generated by inverse projection fusion and 3D conditional random field optimization.

[0017] Based on the global segmentation results, local 3D cue words of target components in the public fitness space are received, and attention-guided graph cut algorithm is used to extract fine-grained component masks. The topological integrity of the segmentation results is then verified.

[0018] By optimizing the segmentation results through surface smoothing, volume conservation, and physical rationality constraints, we can complete the multi-format object output and rendering of independent mesh files or semantic voxel datasets.

[0019] In one specific feasible implementation, the target component is one or more semantic blocks in the global segmentation result;

[0020] Fine-grained components refer to the smallest functional or structural units that are further segmented from the target component using a refined segmentation algorithm.

[0021] In a specific feasible implementation, the identification and dismantling of digital assets in public fitness spaces also includes the following steps:

[0022] Based on the three-dimensional model of the public fitness space, a dual-channel neural network is used to perform high-precision semantic segmentation and three-dimensional feature extraction of spatial elements, and to accurately count the number of each spatial element; multi-dimensional feature vectors with geometric parameters, topological relationships and environmental attributes are generated, and based on this, combined with a predefined coding rule library, spatial elements are mapped into standardized digital assets.

[0023] Clustering algorithms are used to perform unsupervised clustering based on functional similarity, dividing the data into different categories. The classification boundaries are then manually verified and optimized to achieve spatial element clustering. Asset nodes are dynamically stored using a graph database.

[0024] The feature of the segmented spatial elements is extracted by point cloud geometric analysis algorithm, and the position of each spatial element is calculated by combining three-dimensional spatial coordinates. Based on the spatial element coordinate data, the distance between facilities is calculated. At the same time, the topological relationship network is constructed and stored by combining graph database, and the facility layout is dynamically analyzed.

[0025] Spatial feature characteristics are integrated into multi-dimensional vectors and associated with functional and security labels in the coding rule base to form a quantifiable and traceable spatial feature database.

[0026] In a specific feasible implementation plan, the restructuring of public fitness space assets includes the following steps:

[0027] Based on the current data of public fitness spaces, a variety of virtual space schemes are generated by perturbing the parameters with random factors.

[0028] In a specific feasible implementation plan, the training of the benchmark evaluation model for public fitness spaces includes the following steps:

[0029] Collect user behavior data and subjective rating data, associate the behavior data and subjective rating data with the feature parameters of the corresponding spatial scheme, and construct a "user-spatial feature-evaluation" triple sample set;

[0030] Multilayer perceptron and convolutional neural network are used to train the samples. A dynamic feedback correction mechanism is introduced during the training process. Backpropagation optimization is performed on samples with large deviations between predicted scores and subjective scores. Through multiple rounds of iterative training, a benchmark evaluation model for national fitness space is generated and spatial optimization suggestions are output.

[0031] In a specific feasible implementation, behavioral data includes dwell time and route selection;

[0032] Subjective rating data includes comfort and aesthetics;

[0033] Characteristic parameters include facility density and greening rate.

[0034] In a specific feasible implementation plan, the construction of an intelligent evaluation model for public fitness spaces includes the following steps:

[0035] By analyzing sample data using a mixed-effects model, we can distinguish between fixed and random effects and quantify the contribution of each spatial element to spatial evaluation. We can also use cross-validation to assess the model's generalization ability and construct an intelligent evaluation model for national fitness spaces that takes into account both commonalities and individual characteristics.

[0036] This invention provides an intelligent evaluation system for community fitness spaces based on artificial intelligence algorithms, employing the following technical solution:

[0037] A smart evaluation system for community fitness spaces based on artificial intelligence algorithms, including

[0038] The sensing module is used to acquire three-dimensional spatial data of the community's fitness space;

[0039] The processing module stores multiple computer instructions. It acquires the three-dimensional spatial data obtained by the perception module and executes the computer instructions to perform the aforementioned intelligent evaluation method for community fitness spaces based on artificial intelligence algorithms.

[0040] The output module is used to output a report of the spatial scoring data.

[0041] In summary, the present invention has the following beneficial effects:

[0042] 1. By integrating 3D spatial modeling, virtual reality interaction, and deep learning algorithms, an intelligent evaluation system for community fitness spaces was constructed, demonstrating significant technological advantages in the quantitative evaluation and optimized design of fitness space quality. The spatial element reorganization technology based on the UE5 engine breaks through the limitations of traditional two-dimensional planar analysis, achieving three-dimensional simulation of fitness facilities through 3D digital twins. Combined with a VR real-time experience feedback mechanism, it enables dynamic and visual evaluation of indicators such as spatial accessibility and functional integration.

[0043] 2. By employing a mixed-effects model to integrate multi-source user data, the interference of individual preference differences on evaluation results is effectively addressed. Deep learning networks are used to analyze the correlation between user behavior data and spatial parameters, enabling the evaluation model to possess adaptive iterative optimization characteristics, thus improving accuracy compared to traditional questionnaire surveys. Furthermore, by placing spatial quality assessment at the design stage, the system can generate a multi-dimensional library of spatial optimization solutions, reducing costs during later renovations and improving the scientific nature of fitness facility layouts, significantly enhancing the human-centered design level of community sports spaces. This technical framework provides a scalable technological paradigm for the intelligent governance of public spaces in the context of smart cities. Attached Figure Description

[0044] Figure 1 This is a flowchart of the overall process for an intelligent evaluation method for community fitness spaces based on artificial intelligence algorithms.

[0045] Figure 2 This is a diagram illustrating the steps involved in AI-based identification and segmentation.

[0046] Figure 3 This is a flowchart of the assetization and classification process for public fitness spaces.

[0047] Figure 4 This is a diagram illustrating the steps involved in generating a database of features for public fitness spaces.

[0048] Figure 5This is a diagram illustrating the training steps of the basic model for evaluating public fitness spaces.

[0049] Figure 6 This is a flowchart illustrating the steps involved in constructing an intelligent evaluation model for public fitness spaces. Detailed Implementation

[0050] The following combination Figures 1-6 The present invention will be described in further detail below.

[0051] The intelligent evaluation method for community fitness spaces based on artificial intelligence algorithms includes the following steps:

[0052] S1: Acquire three-dimensional spatial data, construct a multivariate dataset, and establish a three-dimensional model of the public fitness space.

[0053] Three-dimensional spatial data of public fitness spaces can be collected through drone laser scanning, IoT sensors, and user terminals. This 3D spatial data includes multi-view images, point clouds, mesh models, current spatial conditions, facility distribution, and landscape configuration, thereby constructing a multivariate dataset for storage. A 3D model of the public fitness space is then built based on the data in this multivariate dataset.

[0054] For example, in this embodiment, a DJI Matrice 300 RTK drone equipped with a LiDAR module is used to perform high-precision 3D scanning of the target area. The scanning accuracy is 5cm and the point cloud density is 80pts / m². Point cloud data and a 3D model are generated. Point cloud registration algorithms, such as ICP and NDT, are used to align the drone scanning data, accurately record the spatial layout, terrain undulations and facility distribution, and further optimize the 3D model to realize the reconstruction of the 3D model of the public fitness space.

[0055] S2, Digital Asset Identification and Disassembly of Public Fitness Spaces.

[0056] Digital assets of public fitness spaces are three-dimensional models of public fitness spaces.

[0057] Using preset prompts, perspective projection transformations are applied to map the images onto a multi-view 2D image plane, generating 2D segmentation prompts adapted to the SAM model. The SAM model is then used to segment the 3D model, achieving 3D-to-2D cross-modal prompt alignment. Based on the SAM-based multi-view 2D segmentation results, coarse-grained global segmentation is generated through inverse projection fusion and 3D conditional random field optimization.

[0058] Based on the global segmentation results, local 3D cue words for target components in the fitness space are received. A fine-grained component mask is extracted using an attention-guided graph cut algorithm, and the segmentation results are checked for topological integrity. Layered 3D segmentation and optimization are then completed. Furthermore, the segmentation results are optimized through surface smoothing, volume conservation, and physical rationality constraints, resulting in multi-format object output and rendering of independent mesh files or semantic voxel datasets. The target component is one or more semantic blocks in the global segmentation results, such as a 3D region identified as "fitness equipment." A fine-grained component refers to the smallest functional or structural unit further segmented from the "target component" using a refined segmentation algorithm (attention-guided graph cut), such as "fitness equipment."

[0059] Based on the three-dimensional model of the public fitness space, a dual-channel neural network is used to perform high-precision semantic segmentation and three-dimensional feature extraction on spatial elements such as fitness equipment, trails, and greenery. The quantity of each spatial element is accurately counted, and multi-dimensional feature vectors with geometric parameters, topological relationships, and environmental attributes are generated. Based on this, combined with a predefined coding rule library, the spatial elements are mapped into standardized digital assets.

[0060] Based on the standardized digital assets mapped to the spatial elements of public fitness spaces, unsupervised clustering is performed using clustering algorithms such as K-means, based on functional similarity (e.g., exercise intensity, service type). This results in categories such as "strength training area" and "aerobic activity area." The classification boundaries are then manually verified and optimized to achieve spatial element clustering. Based on this spatial element clustering and the manually verified and optimized classification boundaries, asset nodes are dynamically stored in a graph database.

[0061] The features of segmented spatial elements are extracted using point cloud geometric analysis algorithms, and the positions of each spatial element are calculated using 3D spatial coordinates. Based on the spatial element coordinate data, the spacing between facilities is calculated. Simultaneously, a topological relationship network is constructed and stored using a graph database, such as Neo4j, for dynamic analysis of facility layout. Topological relationships include, for example, the accessibility of equipment and walkways.

[0062] Spatial feature characteristics are integrated into multi-dimensional vectors and associated with functional and security tags in the coding rule base to form a quantifiable and traceable spatial feature database, providing accurate data support for subsequent spatial evaluation and optimization.

[0063] S3, asset restructuring of public fitness spaces, and training of benchmark evaluation models for public fitness spaces.

[0064] The restructuring of public fitness space assets involves randomly allocating and rearranging digital assets within public fitness spaces based on random factors. Specifically:

[0065] Based on the current data of public fitness spaces, a random factor is added, and the asset elements of public fitness spaces are reorganized using the UE5 Unreal Engine. Using a multilayer perceptron (MLP) or CNN neural network model, a user is connected, and combined with the user's real-time VR experience evaluation, the data is corrected, and the user is reconnected for evaluation. The rating model is trained cyclically. Artificial intelligence evaluation samples are constructed to fit the preferences of a single user for fitness spaces, and the benchmark evaluation model for public fitness spaces is completed.

[0066] The specific steps for training the benchmark evaluation model for public fitness spaces are as follows:

[0067] Based on the 3D model of the public fitness space, a variety of virtual space schemes are generated using Unreal Engine. These schemes are created by randomly perturbing parameters such as facility layout and vegetation density.

[0068] Users immerse themselves in virtual space solutions using VR devices. Simultaneously, user behavior data and subjective ratings are collected in real time. These data are then correlated with the characteristic parameters of the corresponding space solutions to construct a "user-space feature-evaluation" triplet sample set. Behavioral data includes dwell time and path selection, while subjective ratings include comfort and aesthetics. Characteristic parameters include facility density and green space ratio.

[0069] Multilayer perceptron and convolutional neural network are used to train the samples, and a dynamic feedback correction mechanism is introduced during the training process. For samples with large deviations between predicted scores and subjective scores, backpropagation optimization is performed. At the same time, generative adversarial network (GAN) is combined to expand the data diversity of low coverage scenarios. Through multiple rounds of iterative training, an accurate benchmark evaluation model for national fitness space is generated, and spatial optimization suggestions are output, realizing closed-loop optimization of "data collection-model training-solution correction".

[0070] Among them, the deviation between the predicted score and the subjective score data can be judged by the statistical threshold method, which calculates the absolute error between the predicted score and the subjective score data and compares it with the error threshold; or by the distribution analysis method, which judges the deviation based on the error distribution of the sample set and the range of the residual.

[0071] S4, complete the construction of an intelligent evaluation model for public fitness spaces.

[0072] User evaluation samples were accumulated through multiple rounds of experiments, covering groups of different ages, genders, and fitness needs. Mixed-effects models were used to analyze the sample data, distinguishing between fixed effects (such as facility type and spatial scale) and random effects (such as individual preference differences), quantifying the contribution of each spatial element to the spatial evaluation. Cross-validation was used to assess the model's generalization ability, ultimately constructing an intelligent evaluation model for national fitness spaces that considers both commonalities and individual preferences. This model can automatically generate scoring reports for new design schemes and output optimization suggestions, such as adding sunshades and adjusting equipment distribution, achieving pre-design spatial quality assessment and reducing later renovation costs. Specifically, it includes the following steps:

[0073] Based on the sample set of users of different age groups, such as teenagers, middle-aged and elderly, collected in step S3, fixed effect variables (such as spatial elements) and random effect variables (such as age groups) are extracted.

[0074] Based on the hierarchical structure of the data and statistical significance tests of the random effects variance (such as likelihood ratio tests), mathematical modeling is used to explicitly separate group commonalities (fixed effects) from individual differences (random effects). In mixed-effects models, fixed effects model the global regularities of spatial inherent properties (such as facility density), while random effects model the individual specificities of user group differences (such as age group preferences). Maximum likelihood estimation or restricted maximum likelihood estimation is used to solve for the fixed effects coefficients and random effects variance. In mixed-effects models, cross-validation is used to evaluate the model's generalization ability; its core objective is to verify the model's predictive reliability for unknown users and unknown spatial scenarios.

[0075] Residual analysis is used to examine the distribution of residuals between the model's predicted values ​​and actual user ratings (e.g., normality, heteroscedasticity) to determine whether the model assumptions hold. If the residuals show systematic bias (e.g., nonlinear trends, outliers) or the coefficients of fixed and random effects variables are not significant (p-value > 0.05), redundant fixed and random effects variables need to be removed, or interaction / nonlinear terms need to be introduced to improve the model's simplicity and prediction accuracy.

[0076] Based on the commonalities and individual characteristics of evaluation rules for people of different age groups, and by generating dynamic evaluation functions, the system supports multi-dimensional scoring prediction of new design schemes and generates spatial scoring data, which can be output in the form of reports. Optimization suggestions are generated from aspects such as facility layout and landscape configuration, thereby completing the construction of an interpretable and adaptable intelligent evaluation model for public fitness spaces, providing data-driven decision-making basis for space optimization.

[0077] A smart evaluation system for community fitness spaces based on artificial intelligence algorithms includes:

[0078] The sensing module is used to acquire three-dimensional spatial data of the community's fitness space; it can be drones, IoT sensors, and user terminals, etc.

[0079] The processing module stores multiple computer instructions. It acquires the three-dimensional spatial data obtained by the perception module and executes the computer instructions to perform the aforementioned intelligent evaluation method for community fitness spaces based on artificial intelligence algorithms.

[0080] The output module is used to output a report of the spatial scoring data.

[0081] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. An artificial intelligence algorithm-based community fitness space intelligent evaluation method, characterized in that: Comprise the following steps: Obtain three-dimensional space data, build a multi-element data set, and establish a three-dimensional model of the national fitness space; Specifically, the three-dimensional space data includes multi-view images, point clouds, mesh models, space status, facility distribution, and landscape configuration, the multi-element data set stores the three-dimensional space data, and a three-dimensional model of the national fitness space is constructed according to the data in the multi-element data set; National fitness space digital asset identification and disassembly; National fitness space asset restructuring, and national fitness space benchmark evaluation model training; Constructing an intelligent evaluation model of the national fitness space; The national fitness space digital asset identification and disassembly comprises the following steps: Through the preset prompt word, the perspective projection change is used to map it to the multi-view two-dimensional image plane to generate a 2D segmentation prompt suitable for the SAM model; Based on the multi-view 2D segmentation result of SAM, a coarse-grained global segmentation is generated by inverse projection fusion and three-dimensional conditional random field optimization; On the basis of the global segmentation result, the local three-dimensional prompt word of the target component in the national fitness space is received, and a fine-grained component mask is further extracted by using an attention-guided graph cut algorithm, and the segmentation result is topologically integrity checked; Through surface smoothing, volume conservation and physical rationality constraints, the segmentation result is optimized, and multi-format object output and rendering of independent mesh files or semantic voxel data sets are completed; The target component is one or more semantic blocks in the global segmentation result; The fine-grained component refers to the smallest functional unit or structural unit further segmented from the target component by a fine segmentation algorithm; The national fitness space digital asset identification and disassembly further comprises the following steps: Based on the three-dimensional model of the national fitness space, a double-channel neural network is used for high-precision semantic segmentation and three-dimensional feature extraction of space elements, and the number of each space element is accurately counted; a multi-dimensional feature vector with geometric parameters, topological relations and environmental attributes is generated, and on this basis, the space elements are mapped to standardized digital assets by combining a pre-defined coding rule library; A clustering algorithm is used for unsupervised clustering according to functional similarity, different categories are divided, the classification boundary is optimized by manual checking, and space element clustering is realized; the asset nodes are dynamically stored through a graph database; The features of the segmented space elements are extracted through a point cloud geometry analysis algorithm, and the positions of the space elements are calculated in combination with three-dimensional space coordinates; based on the space element coordinate data, the distances between facilities are calculated, and a topological relation network is constructed and stored in combination with the graph database, and the facility layout is dynamically analyzed; The space element features are integrated into multi-dimensional vectors, and the functional and safety labels in the coding rule library are associated to form a quantifiable and traceable space element feature database.

2. The community fitness space intelligent evaluation method based on an artificial intelligence algorithm according to claim 1, characterized in that: The acquisition method of three-dimensional space data includes unmanned aerial vehicle laser scanning, IoT sensor and user terminal.

3. The community fitness space intelligent evaluation method based on an artificial intelligence algorithm according to claim 1, characterized in that: The national fitness space asset restructuring comprises the following steps: Based on the current data of the national fitness space, a virtual space scheme is generated, and a diversified virtual space scheme is generated by randomly disturbing parameters.

4. The community fitness space intelligent evaluation method based on an artificial intelligence algorithm according to claim 1, characterized in that: The national fitness space benchmark evaluation model training comprises the following steps: The behavior data and subjective score data of the user are collected, the behavior data and subjective score data are associated with the characteristic parameters of the corresponding space scheme, and a "user-space feature-evaluation" three-tuple sample set is constructed; The sample is trained by using a multilayer perception and a convolutional neural network, a dynamic feedback correction mechanism is introduced in the training process, samples with large deviations between predicted scores and subjective score data are optimized by back propagation, the national fitness space benchmark evaluation model is generated through multiple rounds of iterative training, and space optimization suggestions are output.

5. The community fitness space intelligent evaluation method based on an artificial intelligence algorithm according to claim 4, characterized in that: The behavior data includes the length of stay and path selection; The subjective score data includes comfort and aesthetics; The characteristic parameters include facility density and greening rate. 6.The method of claim 5, wherein the method further comprises: The construction of the intelligent evaluation model of the national fitness space includes the following steps: The sample data is analyzed by using a mixed effect model to distinguish fixed effects and random effects and to quantify the contribution of each space element to space evaluation; the generalization ability of the model is evaluated by using cross-validation to construct an intelligent evaluation model of the national fitness space that takes into account commonality and individuality.

7. An artificial intelligence algorithm-based community fitness space intelligent evaluation system, characterized in that: The perception module is configured to acquire three-dimensional space data of the community national fitness space; The processing module stores a plurality of computer instructions, acquires the three-dimensional space data acquired by the perception module, and executes the computer instructions to execute the intelligent evaluation method of the community national fitness space based on the artificial intelligence algorithm according to any one of claims 1-6; The output module is configured to output a report of the space score data. ​

Images