Method and system for identifying non-formal outdoor fitness space based on feature intelligent detection

Abstract

The application discloses a non-regular outdoor fitness space recognition method and system based on feature intelligent detection, relates to the technical fields of city planning, landscape gardening and artificial intelligence, and comprises the following steps: receiving a city space street view image; inputting the city space street view image into a trained outdoor fitness space service capability identification model; outputting space site features corresponding to the city space street view image and the matching degree of the space site features and each type of outdoor fitness activity; and matching the corresponding space site of the outdoor fitness activity according to the matching degree, wherein the corresponding space site is site information in the city space street view image corresponding to the space site features. The application can analyze city space sites by means of a deep learning model and quantize the configuration relationship between the space sites and facilities.

Description

Technical Field

[0001] This invention relates to the fields of urban planning, landscape architecture and artificial intelligence technology, specifically to a method and system for identifying informal outdoor fitness spaces based on feature-based intelligent detection. Background Technology

[0002] Currently, the industry's exploration of informal fitness spaces in cities primarily relies on manual analysis for evaluation. This evaluation often employs literature review, using successful case studies of fitness spaces in developed countries as a guide to analyze their practices and summarize their experiences; or it uses logical analysis, applying relevant theories to analyze the sites. While the urban public space planning industry has incorporated artificial intelligence and big data into the analysis of informal fitness spaces, the research objects are mostly existing informal fitness spaces, analyzing their distribution characteristics. For example, a certain region used QGIS spatial research and analysis tools, with POI information provided by Gaode Maps as the research object, to analyze the distribution characteristics of outdoor recreational fitness spaces in that region. This shows that there have been some data-driven attempts within the industry, but data analysis remains superficial.

[0003] In systems relying on manual evaluation, inefficiency and inaccurate results are common problems. Currently, the theoretical framework for establishing informal fitness spaces remains incomplete. Due to their high degree of randomness and numerous influencing factors, summarizing relevant patterns is difficult. Manual analysis also depends on the planner's capabilities, and differing perspectives among planners lead to significant uncertainty in the results. Existing data-driven methods also have many shortcomings. Analysis of urban spatial data often remains focused on a single aspect, resulting in superficial analysis lacking comprehensiveness. Current data-driven analyses of informal fitness spaces primarily focus on existing spaces, lacking foresight. Summary of the Invention

[0004] To address the shortcomings mentioned in the background art, the present invention aims to provide a method and system for identifying informal outdoor fitness spaces based on feature-based intelligent detection. This method and system can analyze urban spaces using deep learning models, quantify the configuration relationship between spaces and facilities, and assess their potential and probability of developing into informal fitness spaces.

[0005] Firstly, the objective of this invention can be achieved through the following technical solution: a method for identifying informal outdoor fitness spaces based on feature-based intelligent detection, the method comprising the following steps:

[0006] Receive urban street view images, input the urban street view images into the trained outdoor fitness space service capability identification model, and output the spatial site features corresponding to the urban street view images, as well as the suitability of the spatial site features to match each type of outdoor fitness activity.

[0007] The outdoor fitness activity is matched with the appropriate space based on suitability, and the corresponding space is the site information in the urban street view image corresponding to the space site features.

[0008] The outdoor fitness space service capability identification model is trained by inputting a complete outdoor fitness space dataset into a pre-established outdoor fitness space service capability identification model.

[0009] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of obtaining the complete outdoor fitness space dataset is as follows:

[0010] Receive an initial outdoor fitness space image set, extract features from the initial outdoor fitness space image set to obtain spatial features, crawl corresponding spatial images based on the extracted spatial features to supplement the data of the initial outdoor fitness space image set, and label and mark the supplemented initial outdoor fitness space image set to obtain a complete outdoor fitness space dataset.

[0011] The initial outdoor fitness space image set includes the actual activity space coordinates of outdoor fitness activities, behavioral images within the outdoor fitness space, and the actual spatial coordinates of the outdoor fitness space.

[0012] The initial outdoor fitness space image set includes various types of outdoor fitness activities. The types of outdoor fitness activities are classified through an outdoor fitness space feature classification system. This system is established by surveying the types of outdoor fitness behaviors that actually occur in urban spaces, extracting the spatial characteristics of the specific activities in each behavior space, and referring to spatial classification standards.

[0013] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of acquiring the initial outdoor fitness space image set is as follows:

[0014] For the actual spatial coordinates of outdoor fitness activities, the coordinate picker tool of Baidu Street View Big Data Platform is used to record the spatial coordinates. Then, using the spatial coordinates and road direction information, images are obtained by crawling outdoor fitness space street view images through Python web crawler to establish dataset P_x1.

[0015] For behavioral images in outdoor fitness spaces, we use outdoor fitness spaces that are currently active and whose spatial coordinates can be identified by users sharing them on multiple social media platforms. We then use the spatial coordinates to crawl street view images of outdoor fitness spaces to obtain images and build a dataset P_x2.

[0016] To determine the actual spatial coordinates of outdoor fitness spaces, we used images of outdoor fitness spaces where activities were taking place and whose spatial coordinates could not be determined, which were shared by users on multiple social media platforms. We then directly captured images from these platforms to create a dataset P_x3.

[0017] Then, an initial outdoor fitness space image set of type x is established, denoted as P_x, then:

[0018] P_x=P_x1∪P_x2∪P_x3

[0019] For the initial set of outdoor fitness space images, feature extraction is performed on the features of the x-th type of outdoor fitness space. Outdoor fitness space images that conform to the activity features of the x-th type of outdoor fitness space are collected by using image cropping software, and a dataset P_x4 is established.

[0020] By supplementing the initial outdoor fitness space image set with dataset P_x4, we obtain the x-th class of outdoor fitness space dataset, which serves as the supplemented initial outdoor fitness space image set, denoted as Pic_x. Then:

[0021] Pic_x = P_x ∪ P_x 4.

[0022] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of labeling and marking the supplemented initial outdoor fitness space image set to obtain a complete outdoor fitness space dataset:

[0023] The initial outdoor fitness space image set was supplemented with overall annotations of the activity sites and labels of key site elements to obtain a complete outdoor fitness space dataset. The calculation process is as follows:

[0024] For the supplemented initial set of outdoor fitness space images, there are a total of N outdoor fitness space images, denoted as px for the i-th image. i Then we have:

[0025] Pic_x = {px1,px2,…,px} N}

[0026] For image px i The activity area was selected and labeled using the LabelImg image annotation tool, and the site marking information was denoted as lg_x. i If the site is labeled as the first label set L_x1, then:

[0027] L_x1={lg_x1,lg_x2,…,lg_x N}

[0028] For outdoor fitness spaces of type x, and their key site elements of type n, the LabelImg image annotation tool is used to mark the key site elements present within them. Then, the image px... i Let lk_x be the labeling status of the j-th key element of the site. ij The key elements of the site are marked as lk_x i If the key elements of the site are labeled using the second label set L_x2, then:

[0029] lk_x i ={lk_x i1 ,lk_x i2 ,…,lk_x in}

[0030] L_x2={lk_x1,lk_x2,…,lk_x N}

[0031] For the x-th type of outdoor fitness space, let the outdoor fitness space site dataset be D1_x and the special element dataset be D2_x, then we have:

[0032] D1_x=(P_x,L_x1)

[0033] D2_x = (P_x, L_x2).

[0034] Thus, the outdoor fitness space dataset D1_x and the special feature dataset D2_x together constitute a complete outdoor fitness space dataset.

[0035] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the pre-established outdoor fitness space service capability identification model, which is trained using a ResNet50 network built based on deep learning framework tools as the underlying network and a Faster RCNN network as the object detection method, wherein:

[0036] We selected Faster R-CNN as the object detection framework to construct a deep learning model for identifying the service capabilities of outdoor fitness spaces. All input images were preprocessed and uniformly converted to 600x600 pixel, RGB 3-channel JPG images. ResNet50 based on residual networks was used as the feature extraction network, and an RPN network was established to generate region selection boxes. A Classifier network was built to classify the images within the candidate boxes according to their features.

[0037] The loss function consists of two parts: classification loss and bounding box regression loss, as shown in the following formula:

[0038]

[0039] The first item For classification loss, N cls This is the mini-batch size. i Represents the predicted value. Represents the true value; when the current Anchor is a positive sample, When the Anchor is a negative sample The log loss for the two classes is as follows:

[0040]

[0041] Second item N represents the bounding box regression loss. reg The number of Anchor Locations For the regression loss at the two bounding box locations, t i This represents the predicted anchor position. This represents the actual anchor position. This represents the offset of the Anchor relative to the true value. R is the Smooth L1 function, and its analytical expression is as follows:

[0042]

[0043] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the pre-established outdoor fitness space service capability identification model is used to establish a stable derivation relationship between the complete visual scene information of the outdoor fitness space and the outdoor fitness space service potential identification system, including: establishing a stable derivation relationship between the complete visual scene information of the outdoor space and the fitness space-related elements, and a stable derivation relationship between the complete visual scene information of the outdoor space and the location and service capability of the outdoor fitness space.

[0044] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of establishing a stable inferential relationship between complete visual scene information of the outdoor space and related elements of the fitness space is achieved by training a target detection model that identifies and labels specific elements related to the xth type of fitness activity in the image, as follows:

[0045] Input the outdoor fitness space image dataset in the deep learning training set for the x-th type of outdoor fitness space and the second label set, perform deep learning training, adjust the model parameters using the model learning ability, train the deep learning model to distinguish feature images, and obtain the first deep learning image feature discrimination model. On this basis, train and adjust the structure and hyperparameters of the first deep learning image feature discrimination model multiple times until a model with high accuracy and good training effect is obtained, which serves as the target detection model for identifying and labeling specific elements related to the x-th type of fitness activity in the image.

[0046] The process of establishing a stable derivation relationship between complete visual scene information of outdoor spaces and the location and service capabilities of outdoor fitness spaces is achieved by obtaining a classification model for the x-th type of outdoor fitness spaces based on deep feature perception, as follows:

[0047] Input the outdoor fitness space image dataset and the first label set of outdoor fitness spaces in the deep learning training set of the x-th type of outdoor fitness space, perform deep learning training, adjust the parameters of the deep learning model using the learning ability of the deep learning model, train the deep learning model to distinguish feature images, and obtain the second deep learning image feature discrimination model. On this basis, train and adjust the structure and hyperparameters of the second deep learning image feature discrimination model multiple times until a model with high accuracy and good training effect is obtained, which serves as the x-th type of outdoor fitness space discrimination model based on deep feature perception.

[0048] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of inputting the urban street view image into the trained outdoor fitness space service capability identification model and outputting the spatial site features corresponding to the urban street view image includes:

[0049] The image is input into the Faster R-CNN deep learning model, and feature extraction is performed using a pre-trained ResNet50 convolutional neural network model to obtain the image's feature map. A selective random search algorithm is used to obtain the regions of interest (ROIs). For each region, the RPN is used to predict whether it contains an object and its bounding box offset. After detecting each region, it is determined whether the region contains outdoor fitness space, its spatial category, and the bounding box is fine-tuned. Finally, the image and object detection results are output.

[0050] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of matching the spatial venue corresponding to the outdoor fitness activity based on suitability.

[0051] The suitability scores for each type of activity are compared with the set suitability threshold 'a':

[0052] To determine whether a venue is suitable for the xth sport, a suitability score is assigned. A suitability threshold a is set. If the score ≤ a, the venue is deemed unsuitable for the sport.

[0053] If score > a, then the venue is deemed suitable for this sport. Based on the image information of the venue, a suitable location for the sport is selected.

[0054] Secondly, in order to achieve the above objectives, this invention discloses an informal outdoor fitness space recognition system based on feature-based intelligent detection, comprising:

[0055] The discrimination and detection module is used to receive urban street scene images, input the urban street scene images into the trained outdoor fitness space service capability discrimination model, and output the spatial site features corresponding to the urban street scene images, as well as the suitability of the spatial site features to match each type of outdoor fitness activity.

[0056] The suitability matching module is used to match the spatial venues corresponding to outdoor fitness activities based on suitability, wherein the corresponding spatial venues are the venue information in the urban street view image corresponding to the spatial venue features.

[0057] The outdoor fitness space service capability identification model is trained by inputting a complete outdoor fitness space dataset into a pre-established outdoor fitness space service capability identification model.

[0058] The beneficial effects of this invention are:

[0059] This invention: (1) Establishes a dataset of urban informal fitness spaces and activity types based on data from social media platforms and street view platforms. By capturing the spaces and types of activities carried out by people based on social media platforms, the collected samples are obtained based on objective circumstances, thus making the samples more objective.

[0060] (2) Based on deep learning technology, a technique was designed to classify the suitability of specific types of activities in informal fitness spaces, and a technique to overlay the suitability of multiple activity types onto multiple objectives. This technique solves the problem of high cost of manual judgment and can efficiently judge informal fitness spaces.

[0061] (3) Based on the prediction results of the deep learning model, a system for evaluating the potential of informal fitness spaces was designed. This system comprehensively analyzes the suitability of different activity types, the suitable population, and the per capita area, and provides a relatively comprehensive assessment of the potential of informal fitness spaces.

[0062] (4) Based on the non-formal fitness space potential assessment system, it is possible to efficiently assess the potential of non-formal fitness spaces in a large area. Attached Figure Description

[0063] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0065] Figure 2 This is a schematic diagram of the model training process of the present invention;

[0066] Figure 3 This is a schematic diagram of the prediction process of the model in this invention;

[0067] Figure 4 This is a schematic diagram of the workflow of the present invention;

[0068] Figure 5 This is a schematic diagram of the system structure of the present invention;

[0069] Figure 6 This is a specific example diagram illustrating how the present invention identifies the potential for individual sports activities to occur on a given site.

[0070] Figure 7 This is a specific example diagram illustrating how the present invention identifies the potential for badminton activities to occur on a venue;

[0071] Figure 8 This is a specific example diagram illustrating how the present invention identifies the potential for volleyball activities to occur on a court. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Example 1: The following is a description of the relevant terms used in the embodiments of this application:

[0074] ResNet50 is a residual network, a classic model in the ResNet series. Proposed in 2015 by Kaiming He and others from Microsoft Research, it has been widely used in computer vision tasks such as image classification, object detection, and image segmentation.

[0075] Outdoor fitness facilities: Outdoor fitness facilities refer to various public sports and fitness venues and facilities located in the outdoor environment, including the venue for physical activities and the auxiliary equipment used for these activities. Outdoor fitness facilities are flexible in scale and form, and have a relatively low barrier to entry. They can be integrated with urban parks, squares, and other public spaces. Especially in densely populated urban areas with limited space resources, developing small and medium-sized outdoor fitness facilities in conjunction with public spaces has become a major way to promote moderate to high-intensity physical activity among residents and improve their health.

[0076] Threshold: Threshold means limit or limit, hence it is also called critical value. It refers to the lowest or highest value at which an effect can be produced. This term is widely used in various fields, including architecture, biology, aviation, chemistry, telecommunications, electrical engineering, and psychology, such as ecological threshold.

[0077] Informal outdoor fitness spaces differ from formal outdoor fitness spaces, which are centrally located, occupy independent land, are specially planned and designed, and offer specialized and singular activities. They rely on urban public spaces and are characterized by their dispersed and small-scale nature. The activities conducted in these spaces are dynamic and diverse.

[0078] like Figure 1 , 4 As shown, a method for identifying informal outdoor fitness spaces based on feature-based intelligent detection includes the following steps:

[0079] Receive urban street view images, input the urban street view images into the trained outdoor fitness space service capability identification model, output the spatial site features corresponding to the urban street view images, and the suitability of the spatial site features with each type of outdoor fitness activity, and match the spatial site corresponding to the outdoor fitness activity according to the suitability. The corresponding spatial site is the site information in the urban street view image corresponding to the spatial site features.

[0080] The outdoor fitness space service capability identification model is trained by inputting a complete outdoor fitness space dataset into a pre-established outdoor fitness space service capability identification model.

[0081] Specifically, the present invention will be further illustrated below through embodiments:

[0082] In this embodiment, outdoor fitness activities are divided into thirteen categories, referring to the spatial classification standards in documents such as "Requirements for the Configuration of Sports Parks" and existing relevant outdoor fitness space classification systems;

[0083] The process of obtaining the complete outdoor fitness space dataset is as follows:

[0084] Receive an initial outdoor fitness space image set, extract features from the initial outdoor fitness space image set to obtain spatial features, crawl corresponding spatial images based on the extracted spatial features to supplement the data of the initial outdoor fitness space image set, and label and mark the supplemented initial outdoor fitness space image set to obtain a complete outdoor fitness space dataset.

[0085] The initial outdoor fitness space image set includes the actual activity space coordinates of outdoor fitness activities, behavioral images within the outdoor fitness space, and the actual spatial coordinates of the outdoor fitness space.

[0086] The initial outdoor fitness space image set includes various types of outdoor fitness activities. The types of outdoor fitness activities are classified through an outdoor fitness space feature classification system. This system is established by surveying the types of outdoor fitness behaviors that actually occur in urban spaces, extracting the spatial characteristics of the specific activities in each behavior space, and referring to spatial classification standards.

[0087] The process of acquiring the initial outdoor fitness space image set is as follows:

[0088] For the actual spatial coordinates of outdoor fitness activities, the coordinate picker tool of Baidu Street View Big Data Platform is used to record the spatial coordinates. Then, using the spatial coordinates and road direction information, images are obtained by crawling outdoor fitness space street view images through Python web crawler to establish dataset P_x1.

[0089] For behavioral images within outdoor fitness spaces, we utilize outdoor fitness spaces where activities are taking place and whose spatial coordinates can be identified, shared by users within multiple software applications. We then crawl street view images of these outdoor fitness spaces using spatial coordinates to obtain images and establish a dataset P_x2. The software used is a social media platform.

[0090] For the actual spatial coordinates of outdoor fitness spaces, we used images of outdoor fitness spaces where activities were taking place but whose spatial coordinates could not be determined, which were shared by users in multiple software applications. We then selected the method of directly capturing images from the platform to obtain images and established a dataset P_x3.

[0091] Then, an initial outdoor fitness space image set of type x is established, denoted as P_x, then:

[0092] P_x=P_x1∪P_x2∪P_x3

[0093] For the initial set of outdoor fitness space images, feature extraction is performed on the features of the x-th type of outdoor fitness space, where x = {1, 2…12, 13}. Outdoor fitness space images that conform to the activity features of the x-th type of outdoor fitness space are collected by using image cropping software, and a dataset P_x4 is established.

[0094] By supplementing the initial outdoor fitness space image set with dataset P_x4, we obtain the x-th class of outdoor fitness space dataset, which serves as the supplemented initial outdoor fitness space image set, denoted as Pic_x. Then:

[0095] Pic_x = P_x ∪ P_x 4.

[0096] The process of annotating and labeling the supplemented initial outdoor fitness space image set to obtain a complete outdoor fitness space dataset:

[0097] The initial outdoor fitness space image set was supplemented with overall annotations of the activity sites and labels of key site elements to obtain a complete outdoor fitness space dataset. The calculation process is as follows:

[0098] For the supplemented initial set of outdoor fitness space images, there are a total of N outdoor fitness space images, denoted as px for the i-th image. i Then we have:

[0099] Pic_x = {px1,px2,…,px} N}

[0100] In this embodiment, for the image px i The activity area was selected using the LabelImg image annotation tool, and the site marking information was recorded as lg_x. i Let the site marking situation be the first label set L_x1, then we have:

[0101] L_x1={lg_x1,lg_x2,…,lg_x N}

[0102] For outdoor fitness spaces of type x, key elements of type y are selected, and the LabelImg image annotation tool is used to mark the key elements present in the area. Then, the image pixels (px) are... i Let lk_x be the labeling status of the j-th key element of the site. ij The key elements of the site are marked as lk_x i Let the labeling of the key elements of the site be the second label set L_x2, then we have:

[0103] lk_x i ={lk_x i1,lk_x i2 ,…,lk_x in}

[0104] L_x2={lk_x1,lk_x2,…,lk_x N}

[0105] For the x-th type of outdoor fitness space, let the outdoor fitness space site dataset be D1_x and the special element dataset be D2_x, then we have:

[0106] D1_x=(P_x,L_x1)

[0107] D2_x = (P_x, L_x2).

[0108] Thus, the outdoor fitness space dataset D1_x and the special feature dataset D2_x together constitute a complete outdoor fitness space dataset.

[0109] The complete outdoor fitness space dataset is input into a pre-established outdoor fitness space service capability identification model, and the trained outdoor fitness space service capability identification model is output. Urban street view images are received and input into the trained outdoor fitness space service capability identification model, and the suitability scores of various activities are output. A suitability threshold is set, and the suitability scores of various activities are compared with the suitability threshold. If the scores are not greater than the suitability threshold, the site is judged to be unsuitable for exercise. If the scores are greater than the suitability threshold, the space for various activities is determined.

[0110] like Figure 2 , 3 As shown, a pre-established outdoor fitness space service capability identification model can be used to establish and train various different models. In this embodiment, the ResNet50 network model is selected.

[0111] The advantages of the ResNet50 network model are: compared to other traditional neural network models, it effectively solves the vanishing and exploding gradient problems, and it can train and converge faster, facilitating the training of deep learning models. Given the complex characteristics of outdoor fitness activity spaces, more complex models are required, and ResNet50's deeper network structure and stronger model performance are beneficial for analyzing more complex spatial images. Finally, because ResNet50 uses residual connections, it has fewer parameters and stronger model generalization ability, thus better adapting to the diverse features of spatial venues.

[0112] In this embodiment, a ResNet50 network built using deep learning framework tools is used as the underlying network, and a Faster R-CNN network is used for training as the object detection method, wherein:

[0113] ResNet50 based on residual network is used as the feature extraction network, and RPN network is established to select region selection boxes. At the same time, Classifier network is established to classify the image within the candidate box according to features.

[0114] The loss function consists of two parts: classification loss and bounding box regression loss, as shown in the following formula:

[0115]

[0116] The first item For classification loss, N cls This is the mini-batch size. i Represents the predicted value. Represents the true value; when the current Anchor is a positive sample, When the Anchor is a negative sample The log loss for the two classes is as follows:

[0117]

[0118] Second item N represents the bounding box regression loss. reg The number of Anchor Locations For the regression loss at the two bounding box locations, t i This represents the predicted anchor position. This represents the actual anchor position. This represents the offset of the Anchor relative to the true value. R is the Smooth L1 function, and its analytical expression is as follows:

[0119]

[0120] In terms of model optimization, the Adam optimizer is used, which adds an adaptive learning rate adjustment step to the stochastic gradient descent method to obtain better training results.

[0121] Finally, the model training results are evaluated using parameters such as average precision (mAP), precision, and recall.

[0122] A portion of the complete outdoor fitness space dataset is randomly selected for training, while the remainder is used for model testing. This ensures the randomness of model training and the universality and generalizability of the model's performance.

[0123] After establishing a complete outdoor fitness space dataset, a Python program was used to randomly generate a partitioning method. 90% of the samples from the dataset were randomly selected to form the training set, and the remaining 10% formed the test set. In the training set, 90% of the randomly selected samples were used for parameter tuning during training, and the remaining 10% were used for accuracy validation during training to evaluate the model's generalization ability in real time.

[0124] Furthermore, the pre-established outdoor fitness space service capacity identification model is used to establish a stable derivation relationship between the complete visual scene information of outdoor fitness spaces and the outdoor fitness space service potential identification system, including: establishing a stable derivation relationship between the complete visual scene information of outdoor spaces and the relevant elements of fitness spaces, and establishing a stable derivation relationship between the complete visual scene information of outdoor spaces and the location and service capacity of outdoor fitness spaces.

[0125] The process of establishing a stable inferential relationship between complete visual scene information of outdoor space and relevant elements of fitness space is achieved by training a target detection model in images to identify and label specific elements related to the xth type of fitness activity, as follows:

[0126] Input the outdoor fitness space image dataset in the deep learning training set for the x-th type of outdoor fitness space and the second label set, perform deep learning training, adjust the model parameters using the model learning ability, train the deep learning model to distinguish feature images, and obtain the first deep learning image feature discrimination model. On this basis, train and adjust the structure and hyperparameters of the first deep learning image feature discrimination model multiple times until a model with high accuracy and good training effect is obtained, which serves as the target detection model for identifying and labeling specific elements related to the x-th type of fitness activity in the image.

[0127] The process of establishing a stable derivation relationship between complete visual scene information of outdoor spaces and the location and service capabilities of outdoor fitness spaces is achieved by obtaining a classification model for the x-th type of outdoor fitness spaces based on deep feature perception, as follows:

[0128] Input the outdoor fitness space image dataset and the first label set of outdoor fitness spaces in the deep learning training set of the x-th type of outdoor fitness space, perform deep learning training, adjust the parameters of the deep learning model using the learning ability of the deep learning model, train the deep learning model to distinguish feature images, and obtain the second deep learning image feature discrimination model. On this basis, train and adjust the structure and hyperparameters of the second deep learning image feature discrimination model multiple times until a model with high accuracy and good training effect is obtained, which serves as the x-th type of outdoor fitness space discrimination model based on deep feature perception.

[0129] The process of inputting urban street view images into the trained outdoor fitness space service capability identification model and outputting the spatial site features corresponding to the urban street view images includes:

[0130] The image is input into the Faster R-CNN deep learning model, and feature extraction is performed using a pre-trained ResNet50 convolutional neural network model to obtain the image's feature map. A selective random search algorithm is used to obtain the region of interest (ROI). For each RPN, the presence of an object and its bounding box offset are predicted. After detecting each region, it is determined whether the region contains outdoor fitness space, its space category, and the bounding box is fine-tuned. Finally, the image and object detection results are output. The process of matching outdoor fitness activity spaces based on suitability is as follows:

[0131] The suitability scores for each type of activity are compared with the set suitability threshold 'a':

[0132] To determine whether a venue is suitable for the xth sport, a suitability score is assigned. A suitability threshold a is set. If the score ≤ a, the venue is deemed unsuitable for the sport.

[0133] If score > a, then the venue is deemed suitable for this sport. Based on the image information of the venue, a suitable location for the sport is selected.

[0134] In this embodiment, Score∈[0,1] and a=0.5 are set.

[0135] Based on the above method, a specific urban block was selected for a real-world case study to obtain the suitability values ​​of urban street view images with area selection and the occurrence of different types of activities:

[0136] Specifically, the present invention will be further illustrated below through embodiments:

[0137] Example 2: Second aspect, such as Figure 5As shown in this embodiment, in order to achieve the above-mentioned objective, the present invention discloses an informal outdoor fitness space recognition system based on feature-based intelligent detection, comprising:

[0138] The discrimination and detection module is used to receive urban street scene images, input the urban street scene images into the trained outdoor fitness space service capability discrimination model, and output the spatial site features corresponding to the urban street scene images, as well as the suitability of the spatial site features to match each type of outdoor fitness activity.

[0139] The suitability matching module is used to match the spatial venues corresponding to outdoor fitness activities based on suitability, wherein the corresponding spatial venues are the venue information in the urban street view image corresponding to the spatial venue features.

[0140] The outdoor fitness space service capability identification model is trained by inputting a complete outdoor fitness space dataset into a pre-established outdoor fitness space service capability identification model.

[0141] Based on the same inventive concept, this invention also provides a computer device, comprising: one or more processors, and a memory for storing one or more computer programs; the programs include program instructions, and the processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., serving as the computing and control core of the terminal, and is used to implement one or more instructions, specifically for loading and executing one or more instructions stored in a computer storage medium to implement the aforementioned method.

[0142] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, performs the above-described method. This storage medium can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0143] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0144] The foregoing has shown and described the basic principles, main features, and advantages of this disclosure. Those skilled in the art should understand that this disclosure is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this disclosure. Various changes and modifications can be made to this disclosure without departing from its spirit and scope, and all such changes and modifications fall within the scope of this disclosure as claimed.

Claims

1. A method for identifying informal outdoor fitness spaces based on feature-based intelligent detection, characterized in that, The method includes the following steps: Receive urban street view images, input the urban street view images into the trained outdoor fitness space service capability identification model, and output the spatial site features corresponding to the urban street view images, as well as the suitability of the spatial site features to match each type of outdoor fitness activity. The outdoor fitness activity is matched with the appropriate space based on suitability, and the corresponding space is the site information in the urban street view image corresponding to the space site features. The outdoor fitness space service capability identification model is trained by inputting the complete outdoor fitness space dataset into the pre-established outdoor fitness space service capability identification model. The process of obtaining the complete outdoor fitness space dataset is as follows: Receive an initial set of outdoor fitness space images, extract features from the initial set of outdoor fitness space images to obtain spatial features, crawl corresponding spatial images based on the extracted spatial features, supplement the initial set of outdoor fitness space images with data, and label and mark the supplemented initial set of outdoor fitness space images to obtain a complete outdoor fitness space dataset. The initial outdoor fitness space image set includes the actual activity space coordinates of outdoor fitness activities, behavioral images within the outdoor fitness space, and the actual spatial coordinates of the outdoor fitness space. The initial outdoor fitness space image set includes various types of outdoor fitness activities. The types of outdoor fitness activities are classified through an outdoor fitness space feature classification system. This system is established by surveying the types of outdoor fitness behaviors that actually occur in urban spaces, extracting the spatial characteristics of the specific activities in each behavior space, and referring to spatial classification standards. The process of acquiring the initial outdoor fitness space image set is as follows: For the actual spatial coordinates of outdoor fitness activities, the coordinate picker tool of Baidu Street View Big Data Platform was used to record the spatial coordinates. Then, using the spatial coordinates and road orientation information, images of outdoor fitness space street view were obtained by crawling with Python web crawlers to build a dataset. ; For behavioral images within outdoor fitness spaces, we utilize user-shared images from multiple social media platforms showing ongoing activities and with identifiable spatial coordinates. We then acquire images by crawling street view images of these outdoor fitness spaces using spatial coordinates to build a dataset. ; To determine the actual spatial coordinates of outdoor fitness spaces, we used images shared by users on multiple social media platforms showing ongoing activities where spatial coordinates were not yet known. We directly captured images from these platforms to create a dataset. ; Then, an initial outdoor fitness space image set of type x is established, denoted as . Then we have: For the initial set of outdoor fitness space images, features of the x-th type of outdoor fitness space are extracted. By cropping social media images, images of outdoor fitness spaces that match the activity characteristics of the x-th type of outdoor fitness space are collected to build a dataset. ; Using datasets The initial outdoor fitness space image set is supplemented to obtain the x-th type of outdoor fitness space dataset, which is used as the supplemented initial outdoor fitness space image set, denoted as . Then we have: The pre-established model for identifying the service capacity of outdoor fitness spaces uses a ResNet50 network built on a deep learning framework as the underlying network and a Faster R-CNN network for object detection during training. We selected Faster RCNN as the object detection framework, constructed a deep learning model for identifying the service capabilities of outdoor fitness spaces, preprocessed all input images, used ResNet50 based on residual networks as the feature extraction network, established an RPN network to generate region selection boxes, and established a Classifier classification network to classify the images within the candidate boxes according to their features. The loss function consists of two parts: classification loss and bounding box regression loss, as shown in the following formula: The first item For classifying losses, For mini-batch size, Represents the predicted value. Represents the true value; when the current Anchor is a positive sample, When the Anchor is a negative sample, , The log loss for the two classes is as follows: Second item For bounding box regression loss, The number of Anchor Locations The regression loss is for the two bounding box locations. This represents the predicted anchor position. This represents the actual anchor position. This represents the offset of the Anchor relative to the true value. R is the SmoothL1 function, and its analytical expression is as follows: 。 2. The method for identifying informal outdoor fitness spaces based on feature-based intelligent detection according to claim 1, characterized in that, The process of annotating the supplemented initial outdoor fitness space image set to obtain a complete outdoor fitness space dataset: The initial outdoor fitness space image set was supplemented with overall annotations of the activity sites and labels of key site elements to obtain a complete outdoor fitness space dataset. The calculation process is as follows: For the supplemented initial set of outdoor fitness space images, there are a total of N outdoor fitness space images, denoted as the i-th image. Then we have: For images The activity area was selected and labeled using the LabelImg image annotation tool, and the site markings were recorded as follows. The site marking status is the first tag set. Then we have: For outdoor fitness spaces of type x, and their key site elements of type n, the LabelImg image annotation tool is used to mark the key site elements present in the image. Let the labeling of the j-th key element of the site be denoted as Key elements of the site are marked as follows: The labeling of key site elements is the second label set. Then we have: For the x-th type of outdoor fitness space, let the outdoor fitness space site dataset be denoted as . Special feature dataset is Then we have: Therefore, outdoor fitness space site dataset With special feature datasets Together, they constitute a complete dataset of outdoor fitness spaces.

3. The method for identifying informal outdoor fitness spaces based on feature-based intelligent detection according to claim 1, characterized in that, The pre-established outdoor fitness space service capacity identification model is used to establish a stable derivation relationship between the complete visual scene information of the outdoor fitness space and the outdoor fitness space service potential identification system, including: establishing a stable derivation relationship between the complete visual scene information of the outdoor space and the relevant elements of the fitness space, and establishing a stable derivation relationship between the complete visual scene information of the outdoor space and the location and service capacity of the outdoor fitness space.

4. The method for identifying informal outdoor fitness spaces based on feature-based intelligent detection according to claim 3, characterized in that... The process of establishing a stable inference relationship between complete visual scene information of outdoor space and related elements of fitness space involves training a target detection model that identifies and labels specific elements related to the xth type of fitness activity in images, as follows: Input the outdoor fitness space image dataset and the second label set in the xth type outdoor fitness space deep learning training set, perform deep learning training, adjust the model parameters using the model's learning ability, train the deep learning model's ability to distinguish feature images, and obtain the first deep learning image feature discrimination model. On this basis, the structure and hyperparameters of the first deep learning image feature discrimination model are trained and adjusted multiple times until a model with high accuracy and good training effect is obtained, which serves as the target detection model for identifying and labeling specific elements related to the xth type of fitness activity in images. The process of establishing a stable derivation relationship between complete visual scene information of outdoor spaces and the location and service capabilities of outdoor fitness spaces is achieved by obtaining a classification model for the x-th type of outdoor fitness spaces based on deep feature perception, as follows: Input the outdoor fitness space image dataset and the first label set of outdoor fitness spaces in the deep learning training set of the x-th type of outdoor fitness space, perform deep learning training, adjust the parameters of the deep learning model using the learning ability of the deep learning model, train the deep learning model to distinguish feature images, and obtain the second deep learning image feature discrimination model. On this basis, train and adjust the structure and hyperparameters of the second deep learning image feature discrimination model multiple times until a model with high accuracy and good training effect is obtained, which serves as the x-th type of outdoor fitness space discrimination model based on deep feature perception.

5. The method for identifying informal outdoor fitness spaces based on feature-based intelligent detection according to claim 1, characterized in that, The process of inputting urban street view images into the trained outdoor fitness space service capability identification model and outputting the spatial site features corresponding to the urban street view images includes: The image is input into the Faster RCNN deep learning model, and features are extracted using a pre-trained ResNet50 convolutional neural network model to obtain the feature map of the image. The region of interest (ROI) of the image is obtained through a selective random search algorithm. For each region, the RPN is used to predict whether it contains an object and its bounding box offset. After the detection of each region is completed, it is determined whether the region contains an outdoor fitness space and its space category, and the bounding box is adjusted. Finally, the image and object detection results are output.

6. The method for identifying informal outdoor fitness spaces based on feature-based intelligent detection according to claim 1, characterized in that, The process of matching outdoor fitness activities with suitable spaces based on suitability: The suitability scores for each type of activity are compared with the set suitability threshold 'a': To determine whether a venue is suitable for the xth sport, a suitability score is assigned. A suitability threshold a is set. If the score ≤ a, the venue is deemed unsuitable for the sport. If score > a, then the venue is deemed suitable for this sport. Based on the image information of the venue, a suitable location for the sport is selected.

7. A non-formal outdoor fitness space identification system based on feature-intelligent detection, employing the non-formal outdoor fitness space identification method based on feature-intelligent detection as described in claim 1, characterized in that... include: The discrimination and detection module is used to receive urban street scene images, input the urban street scene images into the trained outdoor fitness space service capability discrimination model, and output the spatial site features corresponding to the urban street scene images, as well as the suitability of the spatial site features to match each type of outdoor fitness activity. The suitability matching module is used to match the spatial venues corresponding to outdoor fitness activities based on suitability, wherein the corresponding spatial venues are the venue information in the urban street view image corresponding to the spatial venue features. The outdoor fitness space service capability identification model is trained by inputting a complete outdoor fitness space dataset into a pre-established outdoor fitness space service capability identification model.

Images