Method for intelligently discriminating street-side outdoor fitness space type and potential

Abstract

The invention discloses a method for intelligently discriminating the type and potential of a roadside outdoor fitness space, and relates to the technical field of urban planning, landscape gardens and artificial intelligence, and the method comprises the steps: inputting an urban space image data set of a target region into a pre-established outdoor fitness space activity type discrimination model, obtaining an outdoor fitness space potential point of the outdoor fitness space activity type corresponding to the urban space image; inputting the urban space image data set of the target area into a corresponding outdoor fitness space activity potential evaluation model to obtain activity potential evaluation of an outdoor fitness space potential point of an outdoor fitness space activity type corresponding to the urban space image; according to the method, the urban space site can be analyzed by means of the intelligent target detection and deep learning technology, the configuration relation between the fitness behavior of the space site and the site element characteristics is quantified, and the stable derivation relation between the complete visual scene of the urban space and the outdoor fitness activity potential is established.

Description

Technical Field

[0001] The invention relates to the fields of urban planning, landscape gardening and artificial intelligence technology, and specifically to a method for intelligently identifying the types and potentials of streetside outdoor fitness spaces. Background Art

[0002] At present, the industry mainly uses manual analysis as the evaluation method for the exploration of urban fitness space. In the evaluation process, the literature method is often used, that is, the successful cases of fitness spaces in developed countries are used as a guide to analyze their practices and summarize their experiences; or the logical analysis method is used to analyze the venue using relevant theories. For the analysis of urban fitness space, artificial intelligence, big data and other means have been involved in the urban public space planning industry, but the research objects are mostly existing fitness spaces, that is, the distribution characteristics of existing spaces are analyzed. For example, a certain region uses QGIS spatial research and analysis tools, takes the POI information provided by Amap as the research object, and analyzes the distribution characteristics of outdoor leisure and fitness spaces in a certain region. It can be seen that there have been certain attempts at dataization in the industry, but data analysis still remains on the surface.

[0003] In the manual evaluation system, there are generally problems of low efficiency and inaccurate results. At present, the theoretical system for establishing fitness space is still imperfect. Due to its strong randomness and many influencing factors, it is difficult to summarize the relevant laws. Manual analysis also depends on the ability of planners, and different planners have different ideas, which makes the results of manual analysis have a lot of uncertainty. Existing data methods also have many shortcomings. The analysis of urban space data mostly stays in one aspect, the analysis is relatively superficial and lacks comprehensiveness. In the existing technology, the data analysis of fitness space mostly focuses on the fitness space that has been formed, which lacks foresight. Summary of the invention

[0004] In order to solve the deficiencies mentioned in the above-mentioned background technology, the purpose of the present invention is to provide a method for intelligently identifying the types and potential of street-side outdoor fitness spaces, which can analyze urban space venues with the help of intelligent target detection and deep learning technology, quantify the configuration relationship between spatial venue fitness behavior and venue element characteristics, and establish a stable derivation relationship between the complete visual scene of the urban space and the potential for outdoor fitness activities.

[0005] The object of the present invention can be achieved by the following technical solution: A method for intelligently identifying the types and potentials of streetside outdoor fitness spaces comprises: obtaining an urban space image dataset of a target area; The urban space image dataset of the target area is input into the pre-established outdoor fitness space activity type identification model, and the outdoor fitness space potential points corresponding to the outdoor fitness space activity type of the urban space image are obtained through the identification of characteristic elements in the outdoor fitness space activity type identification system; Inputting the urban space image data set of the target area into a pre-established outdoor fitness space activity potential evaluation model corresponding to the outdoor fitness space activity type, and obtaining the activity potential evaluation of the outdoor fitness space potential points corresponding to the outdoor fitness space activity type of the urban space image; According to the activity potential evaluation of the outdoor fitness space potential points corresponding to the outdoor fitness space activity types, the target area is selected as the outdoor fitness venue corresponding to the outdoor fitness space activity type.

[0006] Furthermore, an urban space image dataset of the target area is obtained, including: Obtain the driving trajectory map of the street view collection vehicle in the target area of ​​the map open platform, vectorize the trajectory map in the geographic information system platform, determine the collection interval according to the needs, convert the vector line data into point data, obtain the urban space street view collection points, obtain the surrounding urban space images corresponding to all urban space street view collection points, and form an urban space image dataset; the surrounding urban space images include complete visual image information and geographic space location information.

[0007] Furthermore, the pre-established outdoor fitness space activity type identification model includes: Obtaining an existing picture containing characteristic elements in a characteristic element set in an outdoor fitness space activity type identification system, and performing feature element selection and type labeling in the outdoor fitness space activity type identification system on the existing picture containing characteristic elements in the characteristic element set in the outdoor fitness space activity type identification system, as an outdoor fitness space activity type identification training set; The ResNet50 network built based on the deep learning framework tool is used as the underlying network, and the Faster RCNN network is used as the target detection to build an outdoor fitness space type identification model. The training samples in the outdoor fitness space activity type identification training set are used as input for training. The characteristic elements in the outdoor fitness space activity type identification system in the training samples are identified until the accuracy of the outdoor fitness space type identification model meets the requirements, and the establishment of the outdoor fitness space activity type identification model is completed.

[0008] Furthermore, a system for identifying the types of outdoor fitness space activities is constructed, including: The outdoor fitness activity space image dataset where fitness activities actually occur is extracted according to the characteristic elements of the fitness activity space corresponding to each outdoor fitness space activity type, and the characteristic element set of the fitness activity space corresponding to each outdoor fitness space activity type is obtained; The characteristic elements of fitness activity space are divided into decisive positive decision-making characteristic elements, general positive decision-making characteristic elements and decisive negative decision-making characteristic elements according to the degree of fitting the corresponding outdoor fitness space activity type; By identifying characteristic elements in the outdoor fitness space activity type identification system, the outdoor fitness space potential points corresponding to the outdoor fitness space activity type of the urban space image are obtained, including: if the pre-established outdoor fitness space activity type identification model identifies the decisive positive decision-making elements corresponding to the outdoor fitness activities or only belongs to the general positive decision-making elements, then it is used as the outdoor fitness space potential points corresponding to the outdoor fitness space activity type of the urban space image; If the decisive positive decision-making factors corresponding to the outdoor fitness activities are not identified, and the decisive negative decision-making factors corresponding to the outdoor fitness activities are identified, or the decisive positive decision-making factors, decisive negative decision-making factors and general positive decision-making factors corresponding to the outdoor fitness activities are not identified, then this is not considered as a potential point of outdoor fitness space corresponding to the outdoor fitness space activity type of the urban space image.

[0009] Furthermore, the positive decision-making characteristic factors, general positive decision-making factors and decisive negative decision-making characteristic factors include one or more factors of site size, site shape, site paving, site green plants, fitness facilities and municipal facilities.

[0010] Furthermore, the outdoor fitness activity space image dataset of actual fitness activities includes: an outdoor fitness actual activity space image dataset for each corresponding outdoor fitness space activity type and a set of outdoor fitness actual space image datasets.

[0011] Furthermore, obtaining an outdoor fitness actual activity space image dataset corresponding to each outdoor fitness space activity type includes: The map open platform is used to pick up and record geographic spatial location information, and the static panoramic image service of the map open platform's application programming interface is called to obtain the actual outdoor fitness activity space image, and establish an outdoor fitness space image dataset corresponding to the outdoor fitness space activity type.

[0012] Further, the acquisition of the outdoor fitness actual existence space image data set corresponding to each outdoor fitness space activity type includes: an actual existence space street view image set and an actual existence space intercepted image set; Acquisition of the actual space street view image set includes: utilizing the actual space of outdoor fitness that is spontaneously shared by users on multi-source social media platforms, where activities are taking place and whose spatial geographic location information can be ascertained, obtaining street view images of the actual space of outdoor fitness by calling the static panoramic image service of the application programming interface of the map open platform, and establishing the actual space street view image set; The acquisition of the actual existing space intercepted image set includes: for outdoor fitness spaces whose geographic spatial location information cannot be ascertained, directly capturing platform images to obtain outdoor fitness actual existing space intercepted images, and establishing the actual existing space intercepted image set.

[0013] Furthermore, the pre-established outdoor fitness space activity potential evaluation model corresponding to the outdoor fitness space activity type includes: Taking the outdoor fitness space activity potential evaluation system as the evaluation standard, outdoor fitness space images corresponding to the outdoor fitness space activity types are selected from the outdoor fitness activity space image dataset where there are actual fitness activities as training samples for activity potential annotation, and the training sets are used as the activity potential evaluation training sets for each outdoor fitness activity; Based on the deep learning framework tool, a fully connected image convolutional neural network is built, and the potential evaluation training set of each outdoor fitness activity and each corresponding outdoor fitness space activity are input. Deep learning training is performed on each activity type until the accuracy requirements are met, and an outdoor fitness space activity potential evaluation model for each outdoor fitness space activity type is obtained.

[0014] Furthermore, the construction of the outdoor fitness space activity potential evaluation system includes: Conduct activity potential annotation on outdoor fitness activity spatial images of each type of outdoor fitness spatial activity in the outdoor fitness activity spatial image dataset where actual fitness activities exist; Activity potential annotation, including: The number of descriptive points in each outdoor fitness space activity potential evaluation system is summarized, and the corresponding outdoor fitness activity space images are compared one by one according to the descriptive points. The scores are calculated by cumulative addition. If one descriptive point is met, one point is added, and if it is not met, no points are added. Based on this principle, all training samples are scored as the potential annotation of the outdoor fitness activity space image of each corresponding outdoor fitness space activity type.

[0015] The urban space image is input into a pre-established outdoor fitness space activity potential evaluation model corresponding to the outdoor fitness space activity type to obtain the outdoor fitness space activity potential score of the corresponding outdoor fitness space activity type; the outdoor fitness space activity potential scores of the urban space images of the target area under the same outdoor fitness space activity type are subjected to K-means clustering and grading to obtain the outdoor fitness space activity potential level, which is used to describe the evaluation results of the outdoor fitness space activity potential of this type.

[0016] Beneficial effects: It can analyze urban space venues with the help of intelligent target detection and deep learning technology, quantify the configuration relationship between fitness behavior in space venues and the characteristics of venue elements, establish a stable inference relationship between the complete visual scene of urban space and the potential for outdoor fitness activities; and improve the efficiency of mining urban fitness space and the accuracy of results. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative work: Figure 1 is a flow chart of the method of the present invention; Figure 2 is a schematic diagram of a method according to an embodiment of the present invention; Figure 3 It is a schematic diagram of a flow chart under an embodiment of the present invention; Figure 4 It is an outdoor fitness activity space image related to basketball in the present invention; Figure 5 It is another outdoor fitness activity space image related to basketball in the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the examples of the present disclosure to clearly and completely describe the technical solutions in the examples of the present disclosure. Obviously, the examples described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0019] like Figure 1 and Figure 2 As shown, the method for intelligently identifying the types and potentials of street-side outdoor fitness spaces includes the following steps: Construct an urban spatial image dataset of the target area; The urban space image dataset of the target area is composed of the street view images of the target city space collected based on the map open platform. The specific steps are as follows: Obtain the driving trajectory map of the street view collection vehicle of the map open platform within the target range, vectorize the trajectory map in the geographic information system platform, determine the collection interval according to research needs, convert the vector line data into point data, and obtain the urban space street view collection points; then, call the static panoramic image service of the application programming interface (API) of the map open platform in Python, adjust the parameters based on the urban space street view collection points and research needs, obtain the urban space image, and construct the urban space image dataset. ,The dataset contains complete visual images and geospatial location information of each urban space street view collection point.

[0020] The reasons for choosing the street view images of the open map platform to construct the urban space image dataset of the target area are as follows: 1) The panoramic image service of the open map platform is the largest urban environment image data source with geographic spatial information that can be obtained in batches and has the widest coverage, which can meet the large-scale urban application needs of this technology; 2) The purpose of this technology is to intelligently identify the types and potential evaluation of outdoor fitness spaces along the street. The panoramic image service of the open map platform can meet the multi-angle image acquisition needs of urban streets by personalizing the acquisition parameters of the street view collection vehicle, which matches the goal of this technology.

[0021] In this example, the main urban area is taken as the target area, and Baidu Map is selected as the map open platform. First, the electronic map interception software is used to intercept the urban navigation track map of the planned area in Baidu Map. The trajectory map was vectorized using the GIS platform, and the urban space street view collection points were obtained after the vector line conversion operation. The sampling distance was set to 50m, which can ensure that the street view images of two adjacent collection points fully describe the street space information without omission. Determine the urban space image acquisition parameters, crawl the street view images of the street side at three angles of 45°, 90° and 135° from the driving direction of the street view car at each urban space street view acquisition point, and record the corresponding geographic space coordinates; Urban spatial image dataset Input into the pre-established outdoor fitness space activity type identification model, and obtain the outdoor fitness space potential points corresponding to the urban space image through feature element identification and combination; The construction process of the complete outdoor fitness space activity type identification model includes: Construct an outdoor fitness activity space image dataset of actual fitness activities, and construct an outdoor fitness space activity type identification system, wherein the outdoor fitness space activity type identification system is obtained by summarizing and extracting the outdoor fitness activity space image data; A training set for identifying activity types of outdoor fitness space was constructed. The ResNet50 network built based on the deep learning framework tool was used as the underlying network. The Faster RCNN network was used as the target detection method to build an outdoor fitness space type identification model. The training samples in the outdoor fitness space activity type identification training set were used for training. Among them, the training samples in the outdoor fitness space activity type identification training set were classified, collected and labeled based on the characteristic elements contained in the outdoor fitness activity type identification system.

[0022] The process of constructing an outdoor fitness activity space image set where fitness activities actually occur is as follows: Assume that there are N types of outdoor fitness activities (N=1,2,3,...,N, N∈N+ ), establish an image set of outdoor fitness activity spaces where actual fitness activities exist , define the xth (1<=x<=N, i∈N + ) type of household outdoor fitness activity space image set is ,exist: Among them, the xth (1<=x<=N, i∈N + ) type of household outdoor fitness activity space image set is Includes outdoor fitness actual activity space image dataset A dataset of outdoor fitness images in real space ; Among them, the outdoor fitness actual activity space image dataset The outdoor fitness actual activity space in the data can be understood as follows: for outdoor fitness activity x, the outdoor fitness actual activity space image of outdoor fitness activity x is obtained by picking up and recording the geographic spatial location information based on the known spatial location of the outdoor fitness activity using the map open platform, and calling the static panoramic image service of the application programming interface (API) of the map open platform through Python to obtain the outdoor fitness actual activity space image, and establish the outdoor fitness actual activity space image dataset of outdoor fitness activity x. ; Outdoor Fitness Actual Space Image Dataset The outdoor fitness actual space in the image can be understood as follows: for outdoor fitness activity x, the outdoor fitness actual space image of outdoor fitness activity x is the outdoor fitness actual space spontaneously shared by users on multi-source social media platforms, where activities are taking place and the spatial geographic location information can be ascertained. The static panoramic image service of the application programming interface (API) of the map open platform is called by Python to obtain the street view image of the outdoor fitness actual space, and the actual space street view image set is established. ; Among them, the actual existence space of outdoor fitness that is spontaneously shared by users on multi-source social media platforms, where activities are taking place and whose spatial geospatial location information can be ascertained, includes sports activity videos and images posted by users on Douyin that carry geospatial positioning or text information that specifically describes the geospatial location, and sports videos and images posted by users on Weibo software that carry geospatial positioning or text information that specifically describes the geospatial location.

[0023] For outdoor fitness spaces where the geographic location information cannot be ascertained, the method of directly capturing platform images is chosen to obtain images of the actual outdoor fitness space, and to establish a collection of images of the actual space. The street view image set of the actual space of outdoor fitness and the intercepted image set of the actual space of outdoor fitness together constitute the image set of the actual space of outdoor fitness, and the relationship is as follows: Then we have: The process of constructing an outdoor fitness space activity type identification system is as follows: Based on the outdoor fitness activity spatial image dataset The characteristic elements of the carrying space of each outdoor fitness activity are extracted and logically summarized to obtain the characteristic element set of each fitness activity space, which includes decisive positive decision-making characteristic elements, general positive decision-making characteristic elements and decisive negative decision-making characteristic elements.

[0024] Among them, the three factors are all established based on the standards of the "National Fitness Venue Facility Project Selection Standard"; the decisive positive decision-making characteristic factor is the factor that has a significant positive impact on the occurrence of the fitness activity, that is, as long as the factor exists, the corresponding activity can occur, such as basketball hoops for basketball activities and fitness facilities for personal fitness activities; the general positive decision-making factor is the factor that has a positive impact on the occurrence of the fitness activity, and its positive impact on specific sports is inferior to the decisive positive factor, which refers to the basic condition factors that can meet the needs of sports and have the potential to stimulate activities, such as a regular and flat ground, where basketball activities and personal fitness activities may occur at the same time; the decisive negative decision-making characteristic factor is the factor that has a significant negative impact on the occurrence of the fitness activity, that is, in the absence of In targeted activity facilities and venues with general basic conditions, there are elements that will seriously hinder the progress of sports, such as sand for basketball activities and sand for skateboarding activities; the three together determine the potential points of outdoor fitness space. Since the decisive positive decision-making feature elements have a strong sports type orientation, the identification of elements in the technical solution is also based on real urban space. Their existence must reflect the actual needs of users, so they can make up for or weaken the negative impact of decisive negative decision-making feature elements on sports; and general positive decision-making feature elements can only meet the basic venue needs of sports and are more easily interfered by decisive negative decision-making feature elements. Therefore, the final priority of the three is decisive positive decision elements > decisive negative decision elements > general positive decision elements.

[0025] Positive decision-making characteristic factors, general positive decision-making factors and decisive negative decision-making characteristic factors include one or more of site size, site shape, site paving, site green plants, fitness facilities and municipal facilities factors.

[0026] Set the characteristic element set of the x-th fitness activity space as , the decisive positive decision feature set is , the general positive decision feature set is , the decisive negative decision feature element set is , then: Among them, px represents the decisive positive decision-making characteristic factors of outdoor fitness activity x, and a is its number; gx represents the general positive decision-making characteristic factors of outdoor fitness activity x, and b is its number; ex represents the decisive negative characteristic factors of outdoor fitness activity x, and c is its number.

[0027] In this example, , which contains the following movements: 1.type_football; 2.type_plaza dancing; 3.type_children's activities; 4.type_individual fitness activities; 5.type_basketball; 6.type_volleyball; 7.type_badminton; 8.type_skate boarding; 9.type_roller skating; 10.type_linear-fitness-activities.

[0028] Taking individual fitness activities 4.type_individual fitness activities, basketball 5.type_basketball, and skateboarding 8.type_skate boarding as examples, the type identification framework is as follows: For 4.type_individual fitness activities, its decisive positive decision factor is fitness facilities; its decisive negative decision factor is road line; its general positive decision factors are brick ground, plastic ground, cement ground, and grass ground; In this example, x=4, a=1, b=1, c=4; ; Among them, p41=fitness facilities; e41=road line; g41=brick ground, g42=plastic ground, g43=cement ground, g44=grass.

[0029] For 5.type_basketball, its decisive positive decision factor is basketball hoop; its decisive negative decision factors are sand, grass, water, steps, zebra crossing, and road line; its general positive decision factors are block cement field, block plastic field, lighting facilities, and seating facilities; In this example, x=5, a=1, b=6, c=4; ; Among them, p51=basketball hoop; e51=sand, e52=grass, e53=water surface, e54=steps, e55=zebra crossing, e56=road line; g51=block cement venue, g52=block plastic venue, g53=lighting facilities, g54=seating facilities.

[0030] For 8.type_skate boarding, there is no decisive positive decision factor; the decisive negative decision factors are sand, grass, road line, and vehicle cluster; the general positive decision factors are block cement ground and plastic ground; In this example, x=8, a=0, b=4, c=2; ; Among them, e81=sand, e82=grass, e83=road line, e84=vehicle cluster; g81=cement ground, g82=plastic ground.

[0031] Constructing an outdoor fitness space activity type identification training set includes the following process: Construct a feature element image dataset and annotate the image samples with feature elements; the feature element image dataset is classified and collected based on the feature elements included in the outdoor fitness space activity type identification system. The data source is open image data on the Internet. The number of samples in each feature element image dataset must be consistent. Define the comprehensive spatial feature element set as E, ; Merge the repeated elements in the comprehensive spatial feature set and recode the elements in E, then we have , where k is the total number of spatial characteristic elements; Through the Internet picture collection, we get the image containing the jth ( ) images of spatial feature elements , construct its feature element image dataset , assuming that the number of images in the set is m, then: The feature element labeling process uses the LabelImg tool to manually select and label the feature elements in the feature element image dataset to obtain labeled samples, and the labeled samples are summarized as the outdoor fitness space activity type identification training set.

[0032] The process of building an outdoor fitness space type identification model using the ResNet50 network built based on the deep learning framework tool as the underlying network and the Faster RCNN network as the target detection method is as follows: The Faster RCNN method is selected as the target detection framework, and an outdoor fitness type identification model is constructed. All input spatial feature element images are preprocessed, and ResNet50 based on the residual network is used as the feature extraction network. The RPN network is established to generate the region selection box, and the Classifier classification network is established to classify the images in the candidate box according to the features. In terms of model optimization, the Adam optimizer is used, and the step of adaptively adjusting the learning rate is added on the basis of the stochastic gradient descent method to obtain better training results.

[0033] Finally, the average precision (mAP), accuracy (Precision), recall (Recall) and other parameters are used to evaluate the results of model training.

[0034] Part of the data is randomly selected from the outdoor fitness space activity type identification training set for training, and the rest is used for model testing to ensure the randomness of model training and the universality and generalizability of the model effect.

[0035] After the establishment of the complete target identification training set, the Python program was used to randomly generate a partitioning method, and 90% of the samples were randomly selected from the complete outdoor fitness space data set to form the training set, and the remaining 10% of the samples were used to form the test set. In the training set, 90% of the samples were randomly selected for parameter adjustment during training, and the remaining 10% of the samples were used for accuracy verification during training to evaluate the generalization ability of the model in real time, and finally construct a k-feature element identification model. , , ,..., , k characteristic factor identification models together constitute the outdoor fitness space activity type identification model.

[0036] The process of identifying and combining characteristic elements to obtain the potential points of outdoor fitness space corresponding to the urban space image is as follows: Urban spatial image dataset Input to each feature factor identification model middle( ), perform feature element detection, if the decisive positive decision element of outdoor fitness activity x is identified , or just identify general positive decision factors , then the venue meets the spatial requirements of outdoor fitness activity x, that is, the geographical space corresponding to the urban space image is determined to be the corresponding outdoor fitness venue x; If the decisive positive decision factors of outdoor fitness activity x are not identified , and identified the decisive negative decision factors , or without any relevant elements ( ) is detected, the venue does not meet the spatial requirements of outdoor fitness activity x, that is, the geographic space corresponding to the urban space image is not determined as outdoor fitness venue x.

[0037] In this example, taking 5.type_basketball as an example, if the input urban space image is applied with the outdoor fitness activity type identification model, the result is as follows, then the space is an outdoor fitness venue for basketball activities: 1) Identification of p51; 2) Only one or more of g51, g52, g53, and g54 are identified; If the results are as follows, the space is not an outdoor fitness venue for basketball activities: 1) p51 was not identified, and one or more of e51, e52, e53, e54, e55, and e56 were identified; 2) None of p51, e51, e52, e53, e54, e55, e56, g51, g52, g53, g54, g55, g56 were identified; like Figure 2 As shown in the figure, the method for intelligently identifying the types and potentials of street-side outdoor fitness spaces, and the method for evaluating the potential of outdoor fitness spaces include the following steps: The complete pre-established outdoor fitness space activity potential evaluation model based on deep learning technology is constructed, including building an outdoor fitness space activity potential evaluation system; establishing an outdoor fitness space activity potential evaluation training set; training an outdoor fitness space activity potential evaluation model based on deep learning: Based on the "National Fitness Venue Facility Project Selection Standards", an outdoor fitness space activity potential evaluation system is constructed, which is a collection of potential evaluation standards for all types of outdoor fitness spaces. Each type of potential evaluation standard is composed of a series of potential description points. The description points describe the site characteristics that an excellent activity venue for this type of outdoor fitness activity should meet, including one or more aspects of site space conditions, site landscape conditions, site facility conditions, and site activity conditions. In this example, taking basketball 5.type_basketball as an example, the descriptive points included in its potential evaluation system are: 1. An open space without any obstruction from the ground; 2. The paving surface is smooth and has no bumps; 3. Cement or plastic venues; 4. Activity areas surrounded by buildings or trees on three sides; 5. Public service facilities such as lighting fixtures and rest seats are available; 6. There are recreational service facilities such as basketball hoops; 7. There is no safe place for parking vehicles; 8. A site that is not easily disturbed and has no road intersections or building entrances and exits.

[0038] The complete process of establishing a training set for evaluating the activity potential of outdoor fitness spaces includes: Collect training image samples of various types of outdoor fitness activity spaces. The image samples come from the outdoor fitness activity space image dataset. ,definition For the training set of outdoor fitness activity x, we have: Then we have: ; Taking the outdoor fitness space activity potential evaluation system as the evaluation standard, each outdoor fitness potential evaluation image set is evaluated. The activity potential of the training samples in the evaluation system is annotated. The annotation process is as follows: the number of descriptive points in each outdoor fitness space activity potential evaluation system is summarized, and the corresponding outdoor fitness activity space images are compared one by one according to the descriptive points. The scores are calculated by cumulative addition. If one descriptive point is met, one point is given, and if it is not met, no points are added. Based on this principle, all training samples are scored as potential labels for the outdoor fitness activity space images of each corresponding outdoor fitness space activity type. In this example, take basketball 5.type_basketball as an example: Figure 4 and Figure 5 As shown, two images of outdoor fitness activity spaces are shown. Figure 4 The outdoor fitness activity space image meets the description points 1, 2, 3, 4, 6, 7, and 8, so the potential label of the outdoor fitness activity space image is: 7; Figure 5 The outdoor fitness activity space image meets description points 1, 2, 3, and 6, so the potential label of the outdoor fitness activity space image is: 4.

[0039] The process of evaluating the activity potential of outdoor fitness space based on deep learning is as follows: Based on the deep learning framework tool, a fully connected image convolutional neural network was built. The feature atlas of each outdoor fitness space activity potential evaluation training set and the potential score label set of the corresponding project were input. Deep learning training was performed by type, and the hyperparameters of the fully connected image convolutional neural network were adjusted. The multi-classification ability of the fully connected image convolutional neural network for feature images was trained to obtain a fully connected image convolutional neural network multi-classification model, and the outdoor fitness space activity potential evaluation model based on deep feature perception for each fitness activity type was obtained.

[0040] The activity potential evaluation model of outdoor fitness space is defined as , , , ..., , where N is the number of outdoor fitness activity types.

[0041] The activity potential evaluation of the outdoor fitness space potential point corresponding to the urban space image is obtained, and the urban space image set Input the activity potential evaluation model of the corresponding outdoor fitness venue. For each image in the urban space image set, if it has been determined as a venue for outdoor fitness activity x in the outdoor fitness space activity type screening, then in this step, it is further input , get the outdoor fitness space activity potential score ; K-means cluster analysis was performed on the outdoor fitness space activity potential scores of all urban space images under the same outdoor fitness space activity type, and they were divided into 3, 2, and 1 levels, representing high potential, medium potential, and low potential, respectively, to obtain the outdoor fitness space activity potential level of the target site. .

[0042] Among them, K-means clustering is a typical distance-based clustering algorithm, which uses distance as an evaluation index of similarity, that is, it is believed that the closer the distance between two objects is, the greater their similarity is; when the number of clusters K is specified, K cases are randomly selected from the data set as the starting cluster center points, and the Euclidean distance between the points represented by other cases and the initial cluster center points is calculated, and the cases are divided into the class closest to the cluster center. After all data cases are classified, K data sets (K clusters) are formed, and the mean of the data cases in each cluster is recalculated, and the mean is used as the new cluster center; therefore, the cluster center is in a state of change, and this process is repeated until convergence and the cluster center point no longer changes; in this technical solution, the number of clusters k=3 is specified, and this number of classifications can effectively distinguish different potential levels, meet application requirements, and reduce data redundancy.

[0043] Based on the geospatial location information of the urban space image, further spatial analysis can be conducted on the activity potential level of outdoor fitness space in the target area.

[0044] In this example, assuming that a certain urban space image has been identified as a venue suitable for basketball (x=5) and badminton (x=7) activities in the outdoor fitness space activity type identification model, then in the process of outdoor fitness space activity potential evaluation, this urban space image needs to be input into and In, we get: After K-means clustering and classification, the activity potential level of each type of outdoor activity space in the urban space can be further obtained: The results of potential evaluation can reflect the specificity of the requirements for spatial characteristics of different sports and fitness venues.

[0045] The results of potential evaluation can be applied in the following scenarios: 1) Analyze the spatial distribution of existing outdoor fitness spaces in the city and their impact range. In service blind spots, this technology can be used to identify the types and potential of potential outdoor fitness venues, which can effectively guide the subsequent construction location selection and construction priority sorting, and effectively save resource costs; 2) Based on the resident composition or fitness needs in different blocks in the city, potential spaces can be selectively identified for specific sports; 3) Based on the distribution of specific activity types, urban fitness routes can be planned, and the outdoor fitness space activity potential evaluation model can be used to efficiently focus on areas with low potential and carry out targeted potential improvement construction.

[0046] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0047] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure may have various changes and improvements, and these changes and improvements fall within the scope of the present disclosure to be protected.

Claims

1. A method for intelligently identifying the types and potential of street-side outdoor fitness spaces, characterized in that: include: Obtain an urban spatial image dataset of the target area; Inputting the urban space image data set of the target area into a pre-established outdoor fitness space activity type identification model, and obtaining outdoor fitness space potential points corresponding to the outdoor fitness space activity type of the urban space image through characteristic element recognition in the outdoor fitness space activity type identification system; Inputting the urban space image data set of the target area into a pre-established outdoor fitness space activity potential evaluation model corresponding to the outdoor fitness space activity type, to obtain the activity potential evaluation of the outdoor fitness space potential points of the urban space image corresponding to the outdoor fitness space activity type; According to the activity potential evaluation of the outdoor fitness space potential points corresponding to the outdoor fitness space activity types, the target area is selected as the outdoor fitness venue corresponding to the outdoor fitness space activity type.

2. The method for intelligently identifying the types and potentials of streetside outdoor fitness spaces according to claim 1, characterized in that: The step of obtaining an urban space image dataset of a target area includes: Obtain the driving trajectory map of the street view collection vehicle in the target area of ​​the map open platform, vectorize the trajectory map in the geographic information system platform, determine the collection interval according to the needs, convert the vector line data into point data, obtain the urban space street view collection points, obtain the surrounding urban space images corresponding to all urban space street view collection points, and form an urban space image dataset; the surrounding urban space images include complete visual image information and geographic space location information.

3. The method for intelligently identifying the types and potentials of streetside outdoor fitness spaces according to claim 1, characterized in that: The pre-established outdoor fitness space activity type identification model includes: Obtaining an existing picture containing characteristic elements in a characteristic element set in an outdoor fitness space activity type identification system, and performing feature element selection and type labeling in the outdoor fitness space activity type identification system on the existing picture containing characteristic elements in the characteristic element set in the outdoor fitness space activity type identification system, as an outdoor fitness space activity type identification training set; The ResNet50 network built based on the deep learning framework tool is used as the underlying network, and the Faster RCNN network is used as the target detection to build an outdoor fitness space type identification model. The training samples in the outdoor fitness space activity type identification training set are used as input for training. The characteristic elements in the outdoor fitness space activity type identification system in the training samples are identified until the accuracy of the outdoor fitness space type identification model meets the requirements, and the establishment of the outdoor fitness space activity type identification model is completed.

4. The method for intelligently identifying the types and potentials of streetside outdoor fitness spaces according to claim 1, characterized in that: The construction of an outdoor fitness space activity type identification system includes: The outdoor fitness activity space image dataset where fitness activities actually occur is extracted according to the characteristic elements of the fitness activity space corresponding to each outdoor fitness space activity type, and the characteristic element set of the fitness activity space corresponding to each outdoor fitness space activity type is obtained; The characteristic elements of fitness activity space are divided into decisive positive decision-making characteristic elements, general positive decision-making characteristic elements and decisive negative decision-making characteristic elements according to the degree of fitting the corresponding outdoor fitness space activity type; By identifying characteristic elements in the outdoor fitness space activity type identification system, the outdoor fitness space potential points corresponding to the outdoor fitness space activity type of the urban space image are obtained, including: if the pre-established outdoor fitness space activity type identification model identifies the decisive positive decision-making elements corresponding to the outdoor fitness activities or only belongs to the general positive decision-making elements, then it is used as the outdoor fitness space potential points corresponding to the outdoor fitness space activity type of the urban space image; If the decisive positive decision-making factors corresponding to the outdoor fitness activities are not identified, and the decisive negative decision-making factors corresponding to the outdoor fitness activities are identified, or the decisive positive decision-making factors, decisive negative decision-making factors and general positive decision-making factors corresponding to the outdoor fitness activities are not identified, then this is not considered as a potential point of outdoor fitness space corresponding to the outdoor fitness space activity type of the urban space image.

5. The method for intelligently identifying the types and potentials of streetside outdoor fitness spaces according to claim 4, characterized in that: The positive decision-making characteristic factors, general positive decision-making factors and decisive negative decision-making characteristic factors include one or more factors of site size, site shape, site paving, site green plants, fitness facilities and municipal facilities.

6. The method for intelligently identifying the types and potentials of streetside outdoor fitness spaces according to claim 5, characterized in that: The outdoor fitness activity space image dataset of actual fitness activities includes: an outdoor fitness actual activity space image dataset of each corresponding outdoor fitness space activity type and a set of outdoor fitness actual space image datasets.

7. The method for intelligently identifying the types and potentials of streetside outdoor fitness spaces according to claim 6, characterized in that: The acquisition of outdoor fitness actual activity space image datasets corresponding to each outdoor fitness space activity type includes: The map open platform is used to pick up and record geographic spatial location information, and the static panoramic image service of the map open platform's application programming interface is called to obtain the actual outdoor fitness activity space image, and establish an outdoor fitness space image dataset corresponding to the outdoor fitness space activity type.

8. The method for intelligently identifying the types and potentials of streetside outdoor fitness spaces according to claim 6, characterized in that: The acquisition of the outdoor fitness actual space image dataset corresponding to each outdoor fitness space activity type includes: an actual space street view image set and an actual space intercepted image set; Acquisition of the actual space street view image set includes: utilizing the actual space of outdoor fitness that is spontaneously shared by users on multi-source social media platforms, where activities are taking place and whose spatial geographic location information can be ascertained, obtaining street view images of the actual space of outdoor fitness by calling the static panoramic image service of the application programming interface of the map open platform, and establishing the actual space street view image set; The acquisition of the actual existing space intercepted image set includes: for outdoor fitness spaces whose geographic spatial location information cannot be ascertained, directly capturing platform images to obtain outdoor fitness actual existing space intercepted images, and establishing the actual existing space intercepted image set.

9. The method for intelligently identifying the types and potentials of streetside outdoor fitness spaces according to claim 1, characterized in that: The pre-established outdoor fitness space activity potential evaluation model corresponding to the outdoor fitness space activity type includes: Taking the outdoor fitness space activity potential evaluation system as the evaluation standard, outdoor fitness space images corresponding to the outdoor fitness space activity types are selected from the outdoor fitness activity space image dataset where there are actual fitness activities as training samples for activity potential annotation, and the training sets are used as the activity potential evaluation training sets for each outdoor fitness activity; Based on the deep learning framework tool, a fully connected image convolutional neural network was built. The training set for evaluating the activity potential of each outdoor fitness activity and the label set for the image potential score of each corresponding outdoor fitness space activity type were input. Deep learning training was performed on the corresponding outdoor fitness space activity types until the accuracy requirements were met, and an outdoor fitness space activity potential evaluation model for each outdoor fitness space activity type was obtained.

10. The method for intelligently identifying the types and potentials of streetside outdoor fitness spaces according to claim 9, characterized in that: The construction of the outdoor fitness space activity potential evaluation system includes: Annotate the activity potential of each outdoor fitness activity space image corresponding to the outdoor fitness space activity type in the outdoor fitness activity space image dataset where actual fitness activities exist; Activity potential annotation, including: Summarize the number of descriptive points in each outdoor fitness space activity potential evaluation system, and compare the corresponding outdoor fitness activity space images one by one according to the descriptive points, and score them by cumulative addition. If one descriptive point is met, one point is added, and if it is not met, no point is added. Based on this principle, all training samples are scored as the potential annotation of the outdoor fitness activity space image of each corresponding outdoor fitness space activity type; The urban space image is input into a pre-established outdoor fitness space activity potential evaluation model corresponding to the outdoor fitness space activity type to obtain the outdoor fitness space activity potential score of the corresponding outdoor fitness space activity type; the outdoor fitness space activity potential scores of the urban space images of the target area under the same outdoor fitness space activity type are subjected to K-means clustering and grading to obtain the outdoor fitness space activity potential level, which is used to describe the evaluation results of the outdoor fitness space activity potential of this type.

Images