High-precision automatic recognition and classification method for street landscape based on three-dimensional topological network

Abstract

The application discloses a kind of along street style high-precision automatic identification and classification method based on three-dimensional topological network, including present situation street view image data acquisition, three-dimensional morphological point cloud data preprocessing, three-dimensional topological network construction, street type identification, type identification result verification and man-machine interaction display feedback.The method uses street three-dimensional space morphological point cloud data, uses graph neural network technology to automatically identify and classify street three-dimensional space form.The present application can cope with street space type identification in the field of urban planning, establish a street three-dimensional space form type database based on graph neural network technology, realize the automatic identification and classification of street three-dimensional space form type, improve the accuracy of urban street three-dimensional space identification and classification in a precise and quantitative manner, and provide strong support and reliable basis for the fine management and control of urban street space three-dimensional form.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of urban planning, in particular to a high-precision automatic identification and classification method for street style based on a three-dimensional topological network. BACKGROUND

[0002] Under the background of China's current rapid development, the urbanization process is advancing at an unprecedented speed, bringing economic growth and urban prosperity, but also a series of urban problems such as "thousand cities one face" and dilapidated style. As an important part of urban public space, urban street space is not only used as the main place for residents' daily public activities, but also an important window for people to understand the historical changes of the city and the life of citizens, which is related to the quality of life of residents. This makes the identification and classification of urban street three-dimensional spatial form particularly critical. Urban street three-dimensional spatial form identification and classification technology can provide detailed urban spatial information, help planners make more scientific decisions when designing cities, and to some extent, avoid unreasonable planning and construction. In the face of the risk of historical buildings and traditional blocks being demolished and renovated due to rapid urbanization, urban street three-dimensional spatial form identification and classification technology can record the form and style of existing streets and buildings in detail, helping to develop protection programs and effectively protect the historical style and cultural heritage of the city. Urban street three-dimensional spatial form identification and classification technology can provide strong support whether in new city planning and design or old city protection and renewal, and this widespread applicability makes urban street three-dimensional spatial form identification and classification technology an important tool in urban planning and management.

[0003] The current common urban street space three-dimensional form identification and classification method is one of the traditional field research, through on-site measurement of road length, building height and other spatial form indicators and taking photos of the two sides of the street, to obtain the corresponding street three-dimensional spatial form data, and then use three-dimensional modeling to summarize the street three-dimensional spatial form type. This method has the problems of scattered data, large workload, and large manual measurement error, and cannot accurately restore the three-dimensional form of the street space. Another method is to identify the three-dimensional spatial form type of the street by image sampling and semantic segmentation of street view pictures. This method has image recognition and semantic segmentation errors, and also cannot fully express the characteristics of the three-dimensional spatial form type of the street, with large randomness and lack of explainability. SUMMARY

[0004] To solve the problems mentioned in the background art, the purpose of the present application is to provide a high-precision automatic identification and classification method for street style based on a three-dimensional topological network. The present application can efficiently analyze and process the data of street style and automatically determine its style type.

[0005] The purpose of the present application is achieved by the following technical solutions:

[0006] Step one: data collection

[0007] Use a three-dimensional scanner with GPS positioning, effective pixels 6400 DPI or more, accuracy 100 mm or more to collect the three-dimensional form point cloud data of each street side of the target urban area. Use a portable panoramic image collector with GPS positioning, color depth 48 bits or more, resolution 8K or more to collect street image data on both sides of the target urban area.

[0008] Step two: data preprocessing

[0009] Use the collected GPS positioning data of each street side to perform initial alignment of three-dimensional form point cloud data and street image data. Extract feature points from three-dimensional form point cloud data and street image data, stretch and deform street image data, and perform feature point alignment of three-dimensional form point cloud data and street image data. Use IC P algorithm to map street image data to each three-dimensional point coordinate of three-dimensional form point cloud data, and perform point array alignment of three-dimensional form point cloud data and street image data.

[0010] Step three: three-dimensional topology network construction

[0011] Use a graphics workstation with 24 GB or more of video memory and 384 bits or more of video memory bit width to establish nodes of the three-dimensional topology network according to the three-dimensional point coordinates of each street side based on three-dimensional form point cloud data. Each node extracts image color and image texture information of street image data at that coordinate after point array alignment, as well as surface curvature and normal direction information of three-dimensional form point cloud data at that coordinate based on point coordinate information. Each node is connected to another to establish a connection line of the three-dimensional topology network. Each connection line contains the difference values of its two end nodes in three-dimensional point coordinates, image color, image texture, surface curvature, and normal direction. Use global optimization method to filter node and connection line information.

[0012] Step four: street type identification

[0013] Establish node topology feature matrix and connection line topology feature matrix of each street side based on nodes and connection lines in three-dimensional topology network through graph neural network. Construct graph convolution layer to aggregate topology similarity features of nodes and their adjacent nodes, and stack graph convolution layer according to adjacency relationship to further obtain more adjacent information. Add output layer to output the number of nodes and their proportion of adjacent relationship of each type of topology feature in each street side. Establish street three-dimensional form classification model through fully connected neural network, input the number of nodes and their proportion of adjacent relationship of each type of topology feature in each street side according to training of street three-dimensional form type training database, and then determine the specific type of each street side.

[0014] Step five: type recognition result verification

[0015] According to the proportion of not less than 20%, the data of each street side of each street is extracted as the street three-dimensional form type verification data set. The difference degree of each item of the whole map topological feature of each street side of each street in the verification data set and the whole map topological feature of each sample data in the training database is calculated, and the average difference degree is calculated. When the average difference degree is the smallest, if the street three-dimensional form type recognition result of the verification data set to be verified is the same as the corresponding sample of the training database, it is verified; otherwise, it is not verified, and step four is returned and the interval value of each item of the whole map topological feature parameter of the sample data is adjusted.

[0016] Step six: human-computer interaction display feedback

[0017] The data of the target urban street space three-dimensional form type spectrum database is imported into the holographic sand table display device equipped with a wearable three-dimensional motion capture system and an 8K resolution display screen. The type composition of the overall street space three-dimensional form of the target city and the specific type of each street side are displayed, and the human-computer virtual interaction is carried out and the feedback information is recorded.

[0018] Further, in the step two, the ICP algorithm is used to map the street view image data to the three-dimensional point cloud data of the three-dimensional form. The target points are extracted from the street view image data with a spatial interval of not more than 100 mm, and the three-dimensional point coordinates in the three-dimensional point cloud data closest to the spatial coordinates of the target points are obtained. When multiple target points are mapped to the same three-dimensional point coordinate, only the target point closest to the three-dimensional point coordinate is retained.

[0019] Further, in the step three, the global optimization method is used for data screening of node and connection information. The constraint condition is constructed, that is, the values of the three-dimensional point coordinates, image color, image texture, surface curvature, and normal direction of the node cannot be empty, otherwise the node and related connection are deleted; the absolute length of the connection is not more than 200 mm, otherwise the connection is deleted; the connection cannot have intersection points with other connections in the same plane, otherwise the connection with larger absolute length is deleted.

[0020] Further, in the step four, the node topological feature matrix is established according to the nodes and connections in the three-dimensional topological network. The image color, image texture, surface curvature, and normal direction indicators of the adjacent nodes with connections are collected. When the difference degree of each indicator of the adjacent nodes is less than 15%, it is determined that the two nodes have the same specific node topological feature type, and the topological similarity between each node topological feature type is calculated according to the indicator interval. The node topological feature type is recorded in the node attribute, and the node topological feature matrix is established.

[0021] Advantages of the present application:

[0022] 1、The present application combines three-dimensional morphological data with neural network technology to construct a three-dimensional morphological training database, determines the three-dimensional morphological type of the street space through morphological parameters, and breaks through the problem of large three-dimensional morphological data errors in the traditional space modeling recognition and classification; reduces the amount of data that needs to be collected in traditional space research, and simplifies the process of street three-dimensional morphological recognition and classification. In step four, the graph neural network is combined with the three-dimensional topological network to automatically identify and classify the three-dimensional space form of the street. The application of graph neural network technology in street three-dimensional space form recognition and classification can effectively break through the subjectivity of human judgment in traditional space typology, making the whole process more intelligent and automated.

[0023] 2、The present application establishes a three-dimensional topological network through three-dimensional morphological point cloud data and street image data, and uses ICP algorithm to map street image data on three-dimensional morphological point cloud data in step two, with an error controlled within 100mm, and combines distance constraint conditions for data screening and adjustment, realizing fine simulation of urban street space three-dimensional form in computer three-dimensional network, and maximizing the accuracy and efficiency of three-dimensional morphological recognition and classification.

[0024] 3、The present application uses graph convolutional neural network technology to aggregate features of nodes and lines in the three-dimensional topological network. In step three, when the difference of each index of adjacent nodes with lines is less than 15%, data integration is performed in the form of mean value to form a feature point cluster. When the difference of each index of nodes in each feature point cluster is less than 15%, they are classified into one feature point cluster. Through the integration of graph convolutional neural network, the data structure is optimized, the data operation amount is reduced, and the efficiency of street three-dimensional morphological recognition and classification is improved under the condition of ensuring data integrity.

[0025] 4、In step five of the present application, not less than 20% of each street side data and street three-dimensional morphological training database are combined, through interactive feedback and round-by-round optimization, when the difference of each full graph topological feature of the sample data in the verification data set and the training database is the smallest, the street three-dimensional morphological type of the street to be verified is the same as the street three-dimensional morphological type of the corresponding sample in the training database.

[0026] 5、In step six of the present application, a wearable three-dimensional motion capture system and an 8K resolution or above display holographic sand table display device are used to display the spatial distribution of the three-dimensional morphological type of the city street, improve the display effect, and record feedback information through human-computer virtual interaction, enrich the way for the public to participate in urban planning. BRIEF DESCRIPTION OF DRAWINGS

[0027] Fig. 1 is a flow chart of the method of the present application;

[0028] Fig. 2 is a two-side street view image data collection and preprocessing schematic diagram. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0030] A high-precision automatic identification and classification method for street views based on a three-dimensional atlas network, as shown in FIG. 1, comprises the following steps: Figs. 1-2

[0031] I. Data collection. A three-dimensional scanner with GPS positioning, effective pixels of 6400DPI or more, and accuracy of 100mm or more is used to collect three-dimensional morphological point cloud data of each street side in the target urban area. A portable panoramic image collector with GPS positioning, color depth of 48 bits or more, and resolution of 8K or more is used to collect two-side street view image data in the target urban area.

[0032] In the process of data collection, the device should be calibrated to ensure the accuracy of the data and remove any debris that may affect the scanning quality. Data collection should be performed in cloudy days as much as possible to avoid the influence of shadows and light changes on data collection. The data collection scanning path should be designed to ensure that all important areas are covered, avoiding repeated scanning or missing areas.

[0033] II. Data preprocessing. The collected GPS positioning data of each street side is used to perform initial alignment of the three-dimensional morphological point cloud data and the street view image data. Feature points are extracted from the three-dimensional morphological point cloud data and the street view image data, the street view image data is stretched and deformed, and the feature points of the three-dimensional morphological point cloud data and the street view image data are aligned. ICP algorithm is used to map the street view image data to the three-dimensional point coordinates of the three-dimensional morphological point cloud data, and the point array alignment of the three-dimensional morphological point cloud data and the street view image data is performed.

[0034] The ICP algorithm is used to map the street view image data to the three-dimensional point coordinates of the three-dimensional morphological point cloud data, which extracts target points from the street view image data according to a spatial interval of not more than 100mm, and respectively obtains the three-dimensional point coordinates in the three-dimensional morphological point cloud data closest to the spatial coordinates of the target points. When multiple target points are mapped to the same three-dimensional point coordinate, only the target point closest to the three-dimensional point coordinate is retained. ​

[0035] Three, three-dimensional topological network construction. Using a graphics workstation with a graphics memory of 24 GB or more and a graphics memory bit width of 384 bits or more, a node of a three-dimensional topological network is established according to the three-dimensional point coordinates of the three-dimensional morphological point cloud data of each street side. According to the point coordinate information, the image color and image texture information of the street view image data after lattice alignment at the coordinate, and the surface curvature and normal direction information of the three-dimensional morphological point cloud data at the coordinate are extracted. Each node is connected to each other to establish a connection line of the three-dimensional topological network. In each connection line, the difference values of the three-dimensional point coordinates, image color, image texture, surface curvature, and normal direction of the two end nodes are included. A global optimization method is used to filter the node and connection line information.

[0036] The data filtering of the node and connection line information using the global optimization method is to construct the constraint conditions, that is, the values of the three-dimensional point coordinates, image color, image texture, surface curvature, and normal direction of the node cannot be empty, otherwise the node and related connection lines are deleted; the absolute length of the connection line is not greater than 200 mm, otherwise the connection line is deleted; the connection line cannot have intersection points with other connection lines in the same plane, otherwise the connection line with the larger absolute length is deleted.

[0037] Four, street type identification. Through a graph neural network, a node topological feature matrix and a connection line topological feature matrix are established according to the nodes and connection lines in the three-dimensional topological network of each street side. A graph convolution layer is constructed to aggregate the topological similarity features of the nodes and their adjacent nodes, and the graph convolution layer is stacked according to the adjacent relationship to further obtain more adjacent information. An output layer is added to output the number of nodes and their proportion of the adjacent relationship of each type of topological feature in each street side. A street three-dimensional morphological classification model is established through a fully connected neural network, and the number of nodes and their proportion of the adjacent relationship of each type of topological feature in each street side are input according to the training of the street three-dimensional morphological type training database, so as to determine the specific type of each street side.

[0038] The node topological feature matrix established according to the nodes and connection lines in the three-dimensional topological network is to collect the image color, image texture, surface curvature, and normal direction indicators of the adjacent nodes with connection lines. When the difference degree of each indicator of the adjacent nodes is less than 15%, it is determined that the two nodes have the same specific node topological feature type, and the topological similarity between each node topological feature type is calculated according to the indicator interval. The node topological feature type is recorded in the node attribute to establish the node topological feature matrix.

[0039] In the process of establishing the street three-dimensional form classification model through the fully connected neural network, first, the data is preprocessed, the node number and its proportion of the adjacency relationship of each type of topological feature in each street and each street side are extracted and standardized, then the fully connected neural network is constructed, the input layer is designed to accept these feature vectors, the appropriate activation function and regularization technique are used in the middle layer to prevent overfitting, and the output layer is configured to output the probability distribution of each three-dimensional form type, then the model is trained using the training database, cross-validation and early stopping mechanism are used to optimize the model performance, the classification effect of the model can be analyzed through confusion matrix, precision, recall and F1 score according to the performance of the model on the validation set, and the network structure and hyperparameters can be adjusted as needed to improve the classification accuracy and generalization ability of the model.

[0040] V. Type identification result verification. According to the proportion of not less than 20%, the data of each street and each street side is extracted as the street three-dimensional form type verification data set. The difference degree of each item of the full graph topological feature of each street and each street side in the verification data set and the difference degree of each item of the full graph topological feature of each sample data in the training database is calculated, and the average difference degree is calculated. When the average difference degree is the smallest, if the street three-dimensional form type identification result of the verification data set to be verified is the same as the street three-dimensional form type of the corresponding sample in the training database, it passes the verification; otherwise, it does not pass the verification, returns to step four and adjusts the interval value of each item of the full graph topological feature parameter of the sample data.

[0041] The difference degree of each item of the full graph topological feature of each street and each street side in the verification data set and the difference degree of each item of the full graph topological feature of each sample data in the training database is calculated, and the average difference degree is calculated. In the process of calculating the average difference degree, the importance of each item of the full graph topological feature is calculated according to the weight, which can be calculated by inviting more than 30 experts with more than ten years of experience in urban and rural planning industry to score or rank the importance of each feature, normalizing the score to a weight range, and calculating the weight and further calculating the average difference degree based on the expert evaluation results.

[0042] Six, human-computer interaction display feedback. The data of the target urban street space three-dimensional form type spectrum database is imported into the holographic sand table display equipment equipped with a wearable three-dimensional motion capture system and an 8K resolution display screen, the type composition of the target city's street space three-dimensional form is displayed, as well as the specific type of each street and each street side, human-computer virtual interaction is carried out, and feedback information is recorded.

[0043] The human-computer virtual interaction and feedback information recording, in the process, will utilize the barrier-free user interaction interface supporting sign language and voice interaction, have a data processing module to process the street space three-dimensional form type spectrum data imported from the database, have a data display module to display the overall street space three-dimensional form of the target city and the specific style type of each street side of each street through the holographic sand table device, and have a feedback recording module to record the interactive behavior and feedback information of the user.

[0044] Embodiment

[0045] The following will take the high-precision automatic identification and classification of the street style of a certain area in Nanjing as an example to explain the technical solutions of the present application.

[0046] The following will take the high-precision automatic identification and classification of the street style of a certain area in Nanjing as an example to explain the technical solutions of the present application.

[0047] (1) Taking the streets in a certain area of Nanjing as target streets, collect three-dimensional form point cloud data of each street side, and use a portable panoramic image collector with GPS positioning, color depth of 48 bits or more, and resolution of 8K or more to collect street image data on both sides of the target area. The collection of three-dimensional form point cloud data of each street side should include:

[0048] (1.1) In the data collection process, the device should be calibrated to ensure the accuracy of the data obtained, and the debris on the street side that may affect the scanning process and data accuracy should be removed. In the selection of scanning environment, it is necessary to choose cloudy sunny day to collect data as much as possible, and to avoid the influence of shadow, light change, climate change and other factors on the data collection process.

[0049] (1.2) In the data collection process, a clear and clear data collection scanning path should be developed to ensure that all important areas of the street side are covered, and to avoid scanning duplication or omission of areas, resulting in inaccurate and incomplete street side data.

[0050] (2) Use the collected GPS positioning data of each street side to align the three-dimensional form point cloud data and street image data of each street side. Specifically includes:

[0051] (2.1) When aligning the obtained three-dimensional form point cloud data and street image data of the street side, the corresponding feature points should be extracted in the two kinds of data, and the street image data should be stretched and deformed to ensure that the feature points of the three-dimensional form point cloud data and the street image data are aligned.

[0052] (2.2) After the feature points are aligned, the IPC algorithm is used to map the data carried by the street view image to the coordinates of each three-dimensional point of the three-dimensional morphological point cloud data, and the point array alignment of the three-dimensional morphological point cloud data and the street view image data is performed.

[0053] (2.3) The IPC algorithm used to map the street view image data to the three-dimensional point coordinates of the three-dimensional morphological point cloud data extracts target points from the street view image data with a spatial distance of not more than 100 mm, and simultaneously obtains the three-dimensional point coordinates of the three-dimensional morphological point cloud data closest to the spatial coordinates of the target points. When multiple target points are mapped to the same three-dimensional point coordinates of the three-dimensional morphological point cloud data, only the target point closest in spatial distance to the three-dimensional point coordinates is retained.

[0054] (3) The three-dimensional topological network nodes are established using the three-dimensional point coordinates of the aligned three-dimensional morphological point cloud data along the street side of each street. Specifically, it includes:

[0055] (3.1) A graphics workstation with a graphics processing unit with a display memory of 24 GB or more and a display memory bit width of 384 bits or more is used to establish three-dimensional topological network nodes using the spatial coordinates of each three-dimensional point of each street along the street side three-dimensional morphological point cloud data.

[0056] (3.2) In the established three-dimensional network topology, according to the spatial coordinate information of each node, the street view image data information after point array alignment is extracted, including the image color and image texture information corresponding to the spatial point in the street view image. At the same time, the surface curvature and normal direction information of the three-dimensional morphological point cloud data at the spatial coordinate point should also be extracted.

[0057] (3.3) The three-dimensional topological network nodes after extracting the street view image data and the three-dimensional morphological point cloud data are connected two by two to establish three-dimensional network topology lines. In the process of connecting two by two, global optimization techniques are used to filter the node and line information data.

[0058] (3.4) The global optimization method used is to filter the nodes and lines according to the constraints constructed. The three-dimensional topological network node coordinates, image color, image surface texture information, surface curvature, and normal direction data cannot be empty, otherwise the node and its related lines are deleted. When the absolute length of the spatial line is greater than 200 mm, the spatial line is deleted. When the line intersects with other lines in the same plane, the line with the larger absolute length is deleted.

[0059] (4) The graph neural network technology is used to establish a street three-dimensional morphological classification model database, and through database training, the specific types of each street along the street side are determined.

[0060] (4.1) By using graph neural network technology, the node topology feature matrix and the connection topology feature matrix of each street side according to the nodes and connections in the three-dimensional topological network are established, that is, the image color, image texture, surface curvature, and normal direction indicators of adjacent nodes with connections are collected. When the difference between each indicator is less than 15%, it can be determined that the two nodes have the same specific node topology feature type. The topological similarity between each node topology feature type is calculated according to the indicator interval. The node topology feature type is recorded in the node attribute, and the node topology feature matrix is established.

[0061] (4.2) Construct a graph convolution layer to aggregate the topological similarity features of nodes and their adjacent nodes, and stack the graph convolution layer according to the adjacent relationship to obtain more adjacent information, so as to output the number and proportion of nodes of each type of topological feature in each street side.

[0062] (4.3) Establish a street three-dimensional form classification model through a full neural network. First, the data needs to be preprocessed, and the number and proportion of nodes of each type of topological feature in each street side are extracted and standardized. Then, a fully connected neural network is constructed. After that, the model is trained using a training database, and cross-validation and early stopping mechanism are used to optimize the model performance. The network structure and hyperparameters are adjusted as needed to improve the classification accuracy and generalization ability of the model. Finally, the street three-dimensional form classification model database is used to determine the specific type of each street side.

[0063] (5) Establish a street three-dimensional form type verification data set to test the recognition results. Specifically including:

[0064] (5.1) Extract data of each street side according to a proportion of not less than 20% to establish a street three-dimensional form type verification data set for testing.

[0065] (5.2) Calculate the difference between each item of full graph topology feature in each street side in the verification data set and each item of full graph topology feature in each sample data in the training database, and calculate the average difference. The calculation will involve calculating the average difference according to the weight of each item of full graph topology feature. The importance of each feature can be scored or ranked by inviting more than 30 experts with more than ten years of experience in urban and rural planning industry, and the scores are normalized to a weight range. The weights are assigned based on the expert evaluation results, and the average difference is further calculated.

[0066] (5.3) When the average of the difference is the smallest, if the three-dimensional spatial form type identification result of the street to be verified in the verification data set is consistent with the three-dimensional spatial form type of the corresponding sample in the three-dimensional form type training database, the verification is passed, otherwise the verification is failed, and step four is returned and the interval values of each full map topological feature parameter of the sample data of this type are adjusted.

[0067] (6) The street space three-dimensional form type genealogy database of the target urban area is displayed in a human-computer interaction manner, specifically including:

[0068] (6.1) The data of the street space three-dimensional form type genealogy database of the target urban area is imported into a holographic sand table display device equipped with a wearable three-dimensional motion capture system and an 8K resolution display large screen to display the type composition of the overall street space three-dimensional form of the target city.

[0069] (6.2) In the process of human-computer virtual interaction, an accessible user interface supporting sign language and voice interaction is used, a data processing module is used to process the street space three-dimensional form type genealogy data imported from the database, a data display module is used to display the overall street space three-dimensional form of the target city and the specific style type of each street side of each street through the holographic sand table device, and a feedback recording module is used to record the interactive behavior and feedback information of the user.

[0070] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application.

Claims

1. A method for high-precision automatic recognition and classification of street styles based on a three-dimensional topological network, characterized in that, Comprising the following steps: Step one: data collection Use a three-dimensional scanner with GPS positioning, effective pixels 6400 DPI or more, accuracy 100 mm or more to collect three-dimensional form point cloud data of each street side of the target urban area; use a portable panoramic image collector with GPS positioning, color depth of 48 bits or more, resolution of 8K or more to collect street image data on both sides of the target urban area; Step two: data preprocessing Use the collected GPS positioning data of each street side to perform initial alignment of three-dimensional form point cloud data and street image data; Extract feature points from three-dimensional form point cloud data and street image data, stretch and deform street image data, and align feature points of three-dimensional form point cloud data and street image data; Use ICP algorithm to map street image data to three-dimensional point coordinates of three-dimensional form point cloud data, and perform point array alignment of three-dimensional form point cloud data and street image data; Step three: three-dimensional topology network construction Use a graphics workstation with a graphics processing unit with 24GB or more of video memory and 384 bits or more of video memory bit width to establish nodes of the three-dimensional topology network according to three-dimensional point coordinates of three-dimensional form point cloud data of each street side; Each node extracts image color and texture information of street image data at the coordinate after point array alignment, and surface curvature and normal direction information of three-dimensional form point cloud data at the coordinate according to point coordinate information; Each node is connected to each other to establish the connection line of the three-dimensional topology network; each connection line contains the difference value of the three-dimensional point coordinates, image color, image texture, surface curvature, and normal direction of its two end nodes; use global optimization method to filter node and connection line information data; Step four: street type identification Through the graph neural network, establish node topology feature matrix and connection line topology feature matrix according to the nodes and connection lines in the three-dimensional topology network of each street side; construct a graph convolution layer to aggregate the topology similarity features of the nodes and their adjacent nodes, and stack the graph convolution layer according to the adjacent relationship to further obtain more adjacent information; add an output layer to output the number of nodes and their proportion of adjacent relationship of each type of topology feature in each street side; through the fully connected neural network, establish a street three-dimensional form classification model, train the database according to the street three-dimensional form type, input the number of nodes and their proportion of adjacent relationship of each type of topology feature in each street side, and then determine the specific type of each street side; Step five: type identification result verification According to the proportion of not less than 20%, the data of each street side of each street is extracted as the street three-dimensional form type verification data set; the difference degree of each item of the whole map topological feature of each street side of each street in the verification data set and the difference degree of each item of the whole map topological feature of each sample data in the training database is calculated, and the average difference degree is calculated; when the average difference degree is the smallest, if the street three-dimensional form type identification result of the verification data set to be verified is the same as the street three-dimensional form type of the corresponding sample in the training database, it is verified; otherwise, it is not verified, and step four is returned and the interval value of each item of the whole map topological feature parameter of the sample data is adjusted; Step six: man-machine interactive display feedback The data of the target urban street space three-dimensional form type spectrum database is imported into a holographic sand table display device equipped with a wearable three-dimensional motion capture system and an 8K resolution display screen, the type composition of the street space three-dimensional form of the target city is displayed, and the specific type of each street side is displayed, man-machine virtual interaction is performed, and feedback information is recorded.

2. The three-dimensional topological network-based street three-dimensional morphology type identification method according to claim 1, characterized in that, In step two, the ICP algorithm is used to map the street view image data to the three-dimensional form point cloud data, and the three-dimensional point coordinates are extracted from the street view image data according to a spatial interval of not more than 100 mm, and the three-dimensional point coordinates in the three-dimensional form point cloud data closest to the spatial coordinates of the target point are obtained; when multiple target points are mapped to the same three-dimensional point coordinates, only the target point closest to the three-dimensional point coordinates is retained.

3. The three-dimensional topological network-based street three-dimensional morphology type identification method according to claim 2, characterized in that, In step three, the global optimization method is used for data screening of node and connection information, and the constraint condition is constructed, that is, the three-dimensional point coordinates, image color, image texture, surface curvature, and normal direction of the node data cannot be empty, otherwise the node and related connection are deleted; the absolute length of the connection is not more than 200 mm, otherwise the connection is deleted; the connection cannot have intersection points with other connections in the same plane, otherwise the connection with larger absolute length is deleted.

4. The three-dimensional topological network-based street three-dimensional morphology type identification method according to claim 3, characterized in that, In step four, the node topological feature matrix is established according to the nodes and connections in the three-dimensional topological network, the image color, image texture, surface curvature, and normal direction of the adjacent nodes with connections are collected, when the difference degree of each index of the adjacent nodes is less than 15%, it is determined that the two nodes have the same specific node topological feature type, the topological similarity between each node topological feature type is calculated according to the index interval, the node topological feature type is recorded in the node attribute, and the node topological feature matrix is established. In step six, the data of the target urban street space three-dimensional form type spectrum database is imported into a holographic sand table display device equipped with a wearable three-dimensional motion capture system and an 8K resolution display screen, the type composition of the street space three-dimensional form of the target city is displayed, and the specific type of each street side is displayed, man-machine virtual interaction is performed, and feedback information is recorded.

Images